CRS Report for Congress Environmental Tobacco Smoke and Lung Cancer Risk
OK, This is the CRS report, scanned in and checked as accurately as I know how. Though they break the flow of the words, I left page numbers in so that material can be compared to the printed report and found quickly. Also, I left the footnotes on the pages on which they occur (except for the tag end of one that ran confusingly onto the next page.) I moved charts to the ends of paragraphs, keeping them on the same page. One graphic chart has been omitted from this version (Page 24). --Gene Borio
CRS Report for Congress Environmental Tobacco Smoke and Lung Cancer Risk
- OVERVIEW
- GENERAL ISSUES
- SOURCES OF UNCERTAINTY
- OCCUPATIONAL RISK
- INTRODUCTION
- ENVIRONMENTAL TOBACCO SMOKE
- MAINSTREAM AND SIDESTREAM SMOKE12
- ETS COMPOSITION AND MEASUREMENT15
- ETS INDOOR AIR CONCENTRATIONS AND EXPOSURE
- ETS CANCER RISK
- ETS AND LUNG CANCER - EPIDEMIOLOGY
- INTRODUCTION
- BACKGROUND
- OVERALL EFFECTS AND PREVIOUS STUDIES
- RESULTS
- ANALYSIS
- Smoker Misclassification
- Exposure Misclassification
- Recall Bias
- Discussion
- Smoker Misclassification - Discussion.
- Exposure Misclassification - Discussion.
- Recall Bias -- Discussion.
- Final Comments
- ETS AND LUNG CANCER DEATH RISK
- OCCUPATIONAL ETS LUNG CANCER RISK
- ESTIMATES OF OCCUPATIONAL ETS LUNG CANCER RISK
- OCCUPATIONAL ETS EXPOSURE
- APPENDIX A -- PASSIVE SMOKING HEART DISEASE RISK AND RESPIRATORY DISEASE RISK IN CHILDREN
- APPENDIX B -- RESIDENTIAL EPIDEMIOLOGICAL STUDIES OF PASSIVE SMOKING AND LUNG CANCER
95-1115 SPR
CRS Report for Congress Environmental Tobacco Smoke and Lung Cancer Risk
C. Stephen Redhead Analyst in Life Sciences and Richard E. Rowberg Senior Specialist in Science and Technology Science Policy Research Division November 14, 1995 Congressional Research Service/The Library of Congress
The Congressional Research Service works exclusively for the Congress, conducting research, analyzing legislation, and providing information at the request of committees, Members, and their staffs. The Service makes such research available, without partisan bias, in many forms including studies, reports, compilations, digests, and background briefings. Upon request, CRS assists committees in analyzing legislative proposals and issues, and in assessing the possible effects of these proposals and their alternatives. The Service's senior specialists and subject analysts are also available for personal consultations in their respective fields of expertise.
TABLE OF CONTENTS OVERVIEW ................................................ 1 GENERAL ISSUES ....................................... 1 SOURCES OF UNCERTAINTY ............................. 2 OCCUPATIONAL RISK ................................... 3 INTRODUCTION ............................................ 5 ENVIRONMENTAL TOBACCO SMOKE .......................... 9 MAINSTREAM AND SIDESTREAM SMOKE ................... 9 ETS COMPOSITION AND MEASUREMENT ................. 11 ETS INDOOR AIR CONCENTRATIONS AND EXPOSURE ...... 12 Stationary Air Samplers .............................13 Personal Monitors ...................................14 Biomarkers ......................................... 16 ETS CANCER RISK ..................................... 16 ETS AND LUNG CANCER - EPIDEMIOLOGY .................... 19 INTRODUCTION ....................................... 19 BACKGROUND ......................................... 19 OVERALL EFFECTS AND PREVIOUS STUDIES .............. 22 RESULTS ............................................. 27 ANALYSIS ............................................. 30 Risk and Exposure Measurement ...................... 30 Confounding ........................................ 31 Misclassification Bias ..............................36 Smoker Misclassification ........................... 36 Exposure Misclassification ......................... 38 Recall Bias ..................................... 38 Discussion ...................................... 40 Smoker Misclassification - Discussion ............. 40 Exposure Misclassification o Discussion ............ 42 Recall Bias -- Discussion ....................... 43 Final Comments ................................. 45 ETS AND LUNG CANCER DEATH RISK ........................ 47 INTRODUCTION ....................................... 47 METHODS ............................................ 47 Population Attributable Risk ........................47 Background ETS .................................... 48 RESULTS ............................................. 49 Exposure Patterns ...................................49 Background Exposure ................................ 50 Lung Cancer Deaths ..................................50 DISCUSSION .......................................... 53 RISK COMPARISON ..................................... 55 OCCUPATIONAL ETS LUNG CANCER RISK .................... 59 ESTIMATES OF OCCUPATIONAL ETS LUNG CANCER RISK . . . 60 OCCUPATIONAL ETS EXPOSURE ......................... 62 APPENDIX A -- PASSIVE SMOKING HEART DISEASE RISK AND RESPIRATORY DISEASE RISK IN CHILDREN ............... 65 HEART DISEASE AND ETS .............................. 65 ETS AND RESPIRATORY DISEASE RISK IN CHILDREN ....... 69 APPENDIX B -- RESIDENTIAL EPIDEMIOLOGICAL STUDIES OF PASSIVE SMOKING AND LUNG CANCER ................... 73
OVERVIEW
GENERAL ISSUES
In response to requests from Congress, this report presents an analysis of the potential health effects of environmental tobacco smoke (ETS). The report concentrates on possible lung cancer risk because of the availability of published literature and resource constraints within CRS. A brief overview of ETS and the risk of heart disease and childhood respiratory illness is also presented.
A substantial body of evidence built up over the last 40 years indicates that smoking is a major cause of illness and premature death. In recent years, several reports have also concluded that exposure to environmental tobacco smoke (ETS) can cause lung cancer in people who have never smoked. In 1992, the Environmental Protection Agency (EPA) classified ETS as a known human carcinogen and estimated that ETS exposure is responsible for about 3000 lung cancer deaths each year among adult nonsmokers. EPA's findings have received much support from the scientific community, but have been criticized by other scientists, statisticians and the tobacco industry.
Environmental tobacco smoke is a highly diluted combination of mainstream smoke exhaled by smokers and sidestream smoke released directly from the burning tips of cigarettes. Researchers have concluded that ETS contains most, if not all, of the carcinogenic and toxic compounds that are present in mainstream smoke. Studies that measured cotinine -- a nicotine derivative -- levels in blood and urine indicate that there is widespread exposure to ETS, and measurable uptake of ETS by nonsmokers. According to the EPA, the chemical similarities between mainstream smoke and ETS, and the evidence of exposure to, and uptake of, ETS among nonsmokers is sufficient to conclude that ETS is a lung-cancer hazard.
The EPA based its estimate of the magnitude of the ETS lung cancer risk among nonsmokers on an analysis of over 30 epidemiologic studies of lung cancer among adult non-smoking women. These studies relied on spousal smoking as a surrogate for ETS exposure and classified the women as exposed or unexposed on the basis of whether their husbands smoked. The lung cancer risk among the exposed women was compared to that of the unexposed women.
Since the EPA report was issued, the largest and most recent case-control epidemiologic study included in the EPA findings has been completed, and three other large, case-control studies have been published. Two of these studies1 show no increased average risk, one2 shows a statistically significant increased
---1 Kabat, G.I., et.al., American Journal of Epidemiology, Vo1.142, No.2, 1995, p.141-148; Brownson,R.C., et.al., American Journal of Public Health, Vol.82, No.11, 1992, p.1525-1530.
---2 Fontham, E.T.H., et.al.,Journal of the American Medical Association, Vol.271,No.22, 1994, p. 1752-1759.
CRS - 2
average risk while the fourth3 shows an increased average risk which is not statistically significant at the 95 percent level.
An extensive review of the literature on ETS and lung cancer risk indicates that any lung cancer risk appears to increase as integrated (time and quantity) exposure to ETS increases. Three of the four recent studies (Fontham, et.al., Brownson, et.al., and Stockwell, et.al.) report statistically significant excess risk values at the highest exposure levels (measured in pack-years [packs per day times years exposed] in two cases and in smoker years in another), and about one-third of the studies reviewed by EPA for dose response behavior show a statistically significant (at the 95 percent level) upward trend. While there is evidence of an upward dose response trend, the results are not definitive. And even at the greatest integrated exposure levels, the measured risks are still subject to uncertainty.
Calculations based on data from the Fontham, et.al., study and assuming an average exposure for the entire population at risk (a no-threshold model) result in a range of 470 to 5500 annual lung cancer deaths in the U .S. from ETS with a mean value of 2780. This compares to a mean value of 3300 calculated by EPA under the same assumption. Data from the Brownson, et.al, study, on the other hand, produce no annual lung cancer deaths from ETS also under the no-threshold assumption. If a threshold model is used to simulate the upper limit of a possible upward dose response behavior, the mean number of lung cancer deaths is 440 calculated from the Fontham, et.al, data and 530 for the Brownson, et.al., data. Over 70 percent of these deaths calculated in the no-threshold example and all those calculated in the threshold model occur to individuals who are exposed to both spousal and background ETS. The remaining deaths in the no-threshold model would result from exposure only to background ETS.
The threshold model results are consequences of the model chosen. It is possible that there may be some exposed to sufficient background ETS to be over the threshold without spousal ETS. An effect like this, however, may be very difficult to detect without very large samples.
Using the results obtained from the Fontham, et.al., data in the no-threshold example, a person exposed to spousal and background ETS has about a 2/10 of one percent chance of dying of lung cancer from the ETS over her lifetime. For a person exposed only to background ETS, the number drops to about 7/100 of one percent.
SOURCES OF UNCERTAINTY
The major sources of uncertainty for interpreting the epi results are confounders -- factors other than ETS which could explain the measured risk values, and misclassification. The latter includes identifying current smokers or
---3 Stockwell, H.G. ,et.al.,Journal of the National Cancer Institute, Vol.84,No. 18, 1992, p. 14 171422.
CRS - 3
recently quit smokers as never smokers (smoker misclassification), identifying a person as exposed to ETS because her spouse smoked when in reality she was not subject to any exposure (exposure misclassification), and under or over estimating the amount of ETS exposure (recall bias).
Evidence from a number of studies examining possible confounders appears inconclusive about whether they may be responsible for the risk values measured in the ETS studies. The statistical uncertainties exhibited in the epi studies of most of these possible confounders suggests that none can be considered a clear cause or inhibitor of lung cancer. Furthermore, there is mixed evidence about the correlation of these confounders with increasing integrated exposure to ETS. The number of studies on confounders is not large, however, and it is possible that other confounders exist which have not been identified. Additional research appears to be important.
There are several types of misclassification errors that could occur in these epi studies. Some of them, such as exposure misclassification, would result in measured relative risk values below the actual values, while others, including smoker misclassification and recall bias would result in the measured risk values being overstated. For the Fontham, et.al., and Brownson, et.al., data, smoker misclassification rates of less than 10 percent would account for all of the measured risk at the highest exposure levels in those studies. An even smaller rate -- less than 3 percent -- would cause those risk values to be no longer statistically significant at the 95 percent level. While accounting for exposure misclassification will raise the measured risk values, simulated calculations using the Fontham, et.al., data indicate that misclassification rates greater than 20 percent would be necessary to increase risk values by as much as 5 percent. Recall bias simulations on the same data indicate that overestimating exposure by 10 to 20 percent would result in a reduction of measured risk by about 20 percent at the higher exposure levels.
Information on misclassification rates is skimpy at best. For the exposure and recall categories, it is virtually non-existent. Nevertheless, these simulated calculations indicate that misclassification can be a potent uncertainty in these ETS epi studies, and could account for the measured risk values. Further research on this issue appears called for.
OCCUPATIONAL RISK
The Occupational Safety and Health Administration (OSHA) assessed the lung cancer risk from workplace exposure to ETS as part of its proposed indoor air quality rule. The agency may choose to make substantial revisions to the ETS risk assessment before releasing a final regulation. Independent scientists and tobacco industry researchers and consultants have submitted new data and analyses to the agency for possible inclusion in a revised risk assessment.
Although there are no specific occupational epi studies, several residential studies also collected data on workplace ETS exposure and reported estimates of occupational lung cancer risk. OSHA based its risk assessment on a
CRS - 4
workplace risk estimate by Fontham et al., which indicated an increased risk, and chose not to use the remaining estimates which found no overall association between workplace exposure and lung cancer. Moreover, it assumed that workplace exposure is comparable to residential exposure, though studies that measured cotinine levels in nonsmokers suggest that residential and other non-workplace exposure may be more important that workplace exposure. If, on average, workplace ETS exposure is lower than residential exposure, then it is likely that relatively few workers would be exposed to sufficient ETS to be at increased risk for lung cancer. More extensive workplace exposure data are required before this issue can be resolved.
CRS - 5
INTRODUCTION
The health effects of cigarette smoking have been the subject of intensive scientific investigation since the 1950s. Smoking is linked to leading causes of chronic illness and premature death, including lung cancer and other malignancies, heart disease and stroke, and chronic obstructive pulmonary disease (e.g., bronchitis and emphysema). The Public Health Service estimates that smoking accounts for 87 percent of all lung cancer deaths, 82 percent of all deaths from chronic obstructive pulmonary disease, and 21 percent of all coronary heart disease deaths.
More recently, there has been concern that nonsmokers may be at risk when exposed to environmental tobacco smoke (ETS) that occurs in indoor environments occupied by smokers. Researchers often refer to the involuntary inhalation of ETS by nonsmokers as passive smoking. In 1986, the National Research Council (NRC) and the Surgeon General of the U.S. Public Health Service both released reports on the health effects of passive smoking/Both reports concluded that ETS can cause lung cancer in adult nonsmokers. That same year, a report by the International Agency for Research on Cancer (IARC) concluded that passive smoking gives rise to some risk of cancer, based on considerations related to biological plausibility.5
A recent review of the health effects of passive smoking in the workplace conducted by the National Institute for Occupational Safety and Health determined that "the collective weight of evidence" indicates that ETS poses an increased risk of lung cancer and possibly heart disease in occupationally exposed workers? An extensive analysis of the health effects of ETS was released by the Environmental Protection Agency (EPA) in January 1993.7 In its report, EPA classified ETS as a Group A (known) human carcinogen under
---4 National Research Council. Environmental Tobacco Smoke.' Measuring Exposures and Assessing Health Effects. National Academy Press, Washington, DC, 1986; U.S. Dept. of Health and Human Services. The Health Consequences of Involuntary Smoking. A Report of the Surgeon General. U.S. DHHS, Public Health Service, Office of the Assistant Secretary of Health, Washington, DC, 1986. DHHS Pub. No. (PHS) 87-8398.
---5 International Agency for Research on Cancer. IARC Monograph on the Evaluation of the Carcinogenic Risk of Chemicals to Man, Volume 38: Tobacco Smoke. 1986. World Health Organization, Lyon, France. The IARC report found the available epidemiological evidence to be equivocal, but stated that "knowledge of the nature of mainstream and sidestream smoke, or the materials absorbed during passive smoking, and of the quantitative relationships between dose and effect that are commonly observed from exposure to carcinogens ... leads to the conclusion that passive smoking gives rise to some risk of lung cancer."
---6 National Institute for Occupational Safety and Health. Environmental Tobacco Smoke in the Workplace: Lung Cancer and Other Health Effects. Current Intelligence Bulletin 54. U.S. Dept. of Health and Human Services, NIOSH, 1991.
---7 National Institutes of Health, Respiratory Health Effects of Passive Smoking: Lung Cancer and Other Disorders; The Report of the Environmental Protection Agency, Monograph 4, NIH Publication No. 93-3605, August 1993, Washington, DC. (Here after referred to as the EPA Report. )
CRS - 6
its carcinogen assessment guidelines and concluded that widespread exposure to environmental tobacco smoke presents a substantial public health risk. The EPA report's conclusions are summarized in the text box. EPA estimated that passive smoking is responsible for about 3000 lung cancer deaths per year in the adult, non-smoking (never smokers and long-ago former smokers) population, and poses a serious threat to the respiratory health of young children.
Environmental Protection Agency -- 1993
Respiratory Health Effects of Passive Smoking
In adults:
ETS exposure is responsible for approximately 3000 lung cancer deaths each year;
ETS exposure has subtle, but significant respiratory health effects among nonsmokers, including chest discomfort and reduced lung function.
In children:
ETS exposure results in 150,000 to 300,000 cases of bronchitis and pneumonia annually among young children up to 18 months of age;
ETS exposure in children irritates the upper respiratory tract and reduces lung function;
ETS exposure increases the prevalence of fluid in the middle ear and contributes to middle ear infection;
ETS exposure increases the frequency of episodes and
severity of symptoms in asthmatic children. Between 200,000 and 1,000,000 asthmatic children are affected by
ETS.
The EPA report received widespread support from the public health community and from the larger scientific community. But it has been criticized by tobacco industry researchers and scientific consultants. A few independent statisticians and epidemiologists have also raised objections to EPA's statistical analysis of the ETS epidemiologic studies? The Congressional Research Service
---8 The reader is referred to two congressional hearing at which researchers who support and criticize the EPA study testified: (i) U.S. Congress, House Committee on Energy and Commerce, Subcommittee on Health and the Environment, Environmental Tobacco Smoke, 103d Congress,
CRS - 7
discussed some of these criticisms in an economic analysis of proposed increases in tobacco taxes.9 In testimony before a Senate subcommittee, CRS concluded that "the statistical evidence does not appear to support a conclusion that there are substantial health effects of passive smoking."10
The controversy over the ETS studies stimulated subsequent requests of the Congressional Research Service to review the issue in more depth. This report is in response to those requests.
The report concentrates on the possible relationship between ETS and lung cancer in non-smokers. The study was carried out by a review and analysis of the major published literature, the preponderance of which is on ETS and lung cancer risk. The analysis was supplemented with a one-day meeting held in June 1995 of independent experts and representatives of the different agency and institutional views on possible health effects of ETS. One finding of the meeting was that detailed analysis of other potential health effects -- heart disease and childhood respiratory illness -- would require substantial additional efforts by CRS. Such efforts are beyond the resources of CRS. As a result, this report only briefly reviews current knowledge about those other topics.
This report is divided into four chapters. The first chapter summarizes the physical and chemical composition of ETS, and the evidence for ETS exposure and uptake among non-smokers. The second chapter examines the results of the various epidemiologic studies, with some emphasis on the implications of the
---1st Session, July 21, 1993; (ii) U.S. Congress, House Committee on Agriculture, Subcommittee on Specialty Crops and Natural Resources, Review of the U.S. Environmental Protection Agency's Tobacco and Smoke Study, 103d Congress, 1st Session, July 21, 1993. Three recent reviews in support of EPA's analysis are (i) Trichopoulos, D., Principles and Practice of Oncology: PPO Updates Volume 8, August 1994, pp. 1-8; (ii) Consumer Reports, January 1995; and (iii) Jinot, J. and S. Bayard, Risk Analysis, Vol. 15, No. 1, 1995, pp. 91-96. For a summary of the tobacco industry's criticism of the EPA report, see The Tobacco Institute, EPA Report Scientifically Deficient. Additional articles critical of EPA's analysis include: (i) The Alexis de Tocqueville Institution, Science, Economics, and Environmental Policy: A Critical Examination, August 1994, pp. 1-13; and (ii) Smith, C.J. et al., Toxicologic Pathology, Vol. 20, No. 2, pp. 289-303. For a critical review of the ETS-lung cancer risk that is written for the layman, see Huber, G.L. et al., Consumers' Research, July 1991, pp. 10-15, 33-34. Finally, see Choices in Risk Assessment: The Role of Science Policy in the Environmental Risk Management Process, Chapter 10, Workplace Indoor Air Quality, Regulatory Impact Analysis Project Inc., Washington, D.C. 1994, for a criticism of OSHA's proposed indoor air quality regulation.
---9 In their report, Cigarette Taxes to Fund Health Care Reform: An Economic Analysis (CRS Report 94-214 E, March 8, 1994), J.G. Gravelle and D. Zimmerman reviewed estimates of the economic costs that smokers impose on nonsmokers. The report reviewed the evidence of a passive smoking health risk because this is a potential component of the cost calculation. It concluded that (i) the evidence that passive smoking causes disease is far less certain than for active smoking, and (ii) the health costs of these potential passive smoking effects, if any, are likely to be quite small.
---10 Testimony of Drs. J.G. Gravelle and D. Zimmerman on May 11, 1994, before the Senate Committee on Environment and Public Works, Subcommittee on Clean Air and Nuclear Regulation.
CRS - 8
dose-response trends for estimating the lung cancer risk among non-smokers. A discussion of confounding, smoker misclassification, and recall bias -- the principal sources of uncertainty in the epi studies -- is presented, including implications for the dose-response observations.
The third chapter discusses the potential lung cancer death risk of ETS including the consequences of an upward dose-response trend. This chapter also puts the potential risk of ETS in the context of other risks faced by the general population. The fourth chapter reviews the Occupational Safety and Health Administration's (OSHA) assessment of occupational ETS lung cancer risk, part of its proposed indoor air quality rule. 11
The report also includes two appendices. Appendix A presents a brief overview of the evidence linking passive smoking with heart disease and childhood respiratory illnesses. Appendix B lists the principal ETS studies reviewed for this report.
---11 U.S. Dept. of Labor, Occupational Safety and Health Administration. Indoor Air Quality. Notice of proposed rulemaking; notice of informal public hearing. Federal Register, v. 59, no. 65, April 5, 1994. p. 15968.
CRS - 9
ENVIRONMENTAL TOBACCO SMOKE
This section of the report briefly describes the chemical and physical characteristics of mainstream and sidestream smoke (the two major components of ETS) and discusses studies which have measured indoor ETS levels, and estimated ETS exposure and uptake among nonsmokers. Researchers have concluded that ETS contains most, if not all, of the carcinogenic and toxic compounds that are present in mainstream smoke. The studies also indicate that there is widespread exposure to ETS, and some measurable uptake of ETS by nonsmokers.
MAINSTREAM AND SIDESTREAM SMOKE12
Environmental tobacco smoke is a combination of mainstream smoke (MS) exhaled by smokers and sidestream smoke (SS) released directly from the burning tip of cigarettes. It is typically highly diluted. Mainstream smoke is comprised of small particles averaging 0.35-0.4 üm in diameter18 (particle phase) and a mixture of gases (vapor phase). The particle phase includes several metals (e.g., cadmium and zinc) and a variety of non-volatile organic compounds of high molecular weight. The vapor phase includes numerous highly volatile compounds such as carbon monoxide and hydrogen cyanide.
Nicotine and many other semi-volatile constituents of tobacco smoke occur both in the particle phase and the vapor phase depending on their volatility and the prevailing conditions. These compounds tend to be present in the particle phase of highly concentrated inhaled MS, but evaporate into the vapor phase as exhaled MS rapidly dilutes during the formation of ETS.
Sidestream smoke is the primary contributor to ETS, providing most of the vapor phase and over half of the particles. It is produced by the same fundamental processes as MS and consists of the same chemical compounds including many known or suspected human carcinogens. However, SS is generated at lower temperatures and at a higher pH than MS, and as a result it has a different relative chemical composition.
Table 1 lists the concentrations of various compounds in both phases of MS delivered by unfiltered cigarettes, as measured by a standard smoking machine. The table also compares the amount of each compound delivered in MS and in SS by computing a SS/MS ratio.14 These ratios indicate that, with the
---12 For a more comprehensive discussion of the physical and chemical characteristics of mainstream and sidestream smoke, see M.R. Guerin et al. The Chemistry of Environmental Tobacco Smoke: Composition and Measurement, 1992, Lewis Publishers, Inc., Chelsea, Michigan.
---13 One micron (m) = 1/1000 millimeter (mm).
---14 There is no standard method for collecting and analyzing SS, unlike MS. Researchers have used a variety of small chambers in which to confine the burning cigarette and collect the SS. These devices produce a somewhat artificial smoking environment compared to that associated with human smoking, and, of course, do not take into account the dilution that occurs during theformation of ETS.
CRS - 10
exception of hydrogen cyanide and organic acids, the majority of compounds are released in greater quantities in SS than in MS. In its analysis of MS and SS emissions data, EPA found that all of the five known human carcinogens, nine probable human carcinogens, and three animal carcinogens are emitted at higher levels in SS than in MS, often by a factor of ten or more.
TABLE 1. Comparison of Mainstream and Sidestream Smoke Deliveries for Selected Compounds Mainstream per Constituent Cigarettes SS/MS Ratio Mainstream vapor phase Carbon monoxide 10-23 mg 2.5-4.7 Carbon dioxide 20-40 mg 8-11 Benzene b 12-48 g 5-10 Acetone 100-250 g 2-5 Hydrogen cyanide 400-500 g 0.1-0.25 Ammonia 50-130 g 40-170 Pyridine 16-40 g 6.5-20 Nitrogen oxides 100-600 g 4-10 N-Nitrosodimethylamine c 10-40 ng 20-100 Mainstream particle phase Nicotine 1-2.5 mg 2.6-3.3 Phenol 60-140 g 1.6-3.0 2-Naphthylamine b 1.7 ng 30 4-Aminobiphenyl b 4.6 ng 31 Cadmium c 100 ng 7.2 Nickel b 20-80 ng 13-30 Lactic acid 63-174 g 0.5-0.7 Succinic acid 110-140 g 0.43-0.62 a The units are in milligrams (1 mg= 1/1000 g), micrograms (1 g = 1/1000 mg), and nanograms (1 ng = 1/1000 g). b Known human carcinogen, according to EPA or IARC. c Probable human carcinogen, according to EPA or IARC. Source: National Research Council, 1986. Table 2-2.
CRS - 11ETS COMPOSITION AND MEASUREMENT15
There is limited information on the chemical composition of ETS. Exhaled MS, which can contribute between 15 percent and 43 percent of the particulate matter in ETS, has yet to be characterized. There is also little data on the impact of dilution on SS emissions. During ETS formation, both SS and exhaled MS are diluted by many orders of magnitude and subsequently undergo physical transformation and alterations in chemical composition.
Numerous studies of the impact of smoking occupancy on indoor air quality have measured several ETS-related compounds of human health concern, including known and suspected carcinogens, in a variety of settings (e.g., residential, office, transportation, etc.). Researchers have concluded (1) that many of the potentially harmful compounds in SS are also present in ETS, and (2) that these ETS contaminants are found above background levels in a wide range of indoor environments in which smoking occurs. These studies indicate that the composition of ETS can be highly variable depending on the smoking rates, the amount and type of ventilation, contact with indoor surfaces, and a host of other environmental conditions.
Given that ETS is a complex mixture of thousands of compounds, many of which change chemically and physically over time, it is necessary to identify a chemical marker to represent the frequency, duration, and magnitude of ETS exposure. An ideal marker would be a compound that is specific to tobacco smoke, easy to measure, and that behaves similarly to ETS as a whole. Several markers have been identified, though none meets all these criteria. However, vapor phase nicotine and respirable suspended particles (RSP)16 are both suitable indicators of exposure to ETS.
A variety of methods have been used to measure indoor nicotine and RSP levels in order to assess ETS exposure. Air sampling devices may be placed a specific indoor locations for varying periods of time (stationary sampling) or worn by individuals (personal monitoring). Researchers have also measured chemicals (biomarkers) in the blood and urine of ETS-exposed nonsmokers.
Tobacco combustion produces significant emissions of respirable suspended particles (RSP). There are a number of accepted methods that permit accurate measurement of RSP concentrations in indoor environments for sampling times ranging from seconds to several days. Studies have shown that RSP levels in smoking environments are usually higher than in non-smoking environments. Leaderer and Hammond conducted a large chamber study using smokers and
---15 For more information on the chemistry of ETS and on chemical markers for ETS, see EPA Report, chapter 3; and Guerin et al., 1992.
---16 Respirable suspended particles (RSP) refers to particles that are small enough to reach the deepest recesses of the lungs during inhalation. There is some disagreement among researchers as to the upper size limit for RSP. Some investigators use a conservative value of 3 m, others use values of 10 or 15 m. However, if one is using RSP as a marker for ETS, choosing among these values is largely irrelevant, because most ETS particles are less than 1 m.
CRS - 12
reported an average RSP emission rate per cigarette of 17.1 mg.17 RSP emission rates among different brands of cigarettes were similar.
Respirable suspended particles are also generated by other types of combustion. At low smoking and high ventilation rates, it might be difficult to distinguish ETS-associated RSP from a background of RSP from other indoor sources (e.g., kerosene heaters) or even outdoor sources. However, studies by Repace indicate that the fraction of indoor RSP attributable to smoking is typically 80 to 90 percent of the total RSP. 18
Vapor phase nicotine is the most common ETS marker. Nicotine is unique to tobacco and can be reliably measured using a variety of methods. Average indoor air concentrations typically range from i to 10 micrograms per cubic meter (üg/m3). Several studies have shown that weekly nicotine concentrations are highly correlated with the number of cigarettes smoked. One of these studies also reported a strong correlation between weekly nicotine concentrations and RSP levels in smoking households. 19 The RSP-to-nicotine ratio in this study was approximately 10:1, which is similar to the ratio seen in chamber studies and other field studies, including a recent California State report.20
Nicotine is not an ideal ETS marker because it is readily adsorbed onto surfaces, thus reducing its concentration relative to other ETS components as ETS ages. Some studies have demonstrated that vapor phase nicotine is depleted from a smoking environment more rapidly than the particulate portion of ETS. This could lead to an underestimation of ETS exposures. Nicotine also evaporates from surfaces onto which it has been adsorbed, which results in measurable concentrations even in the absence of active smoking. The affinity of nicotine for surfaces may limit its use as an ETS marker in environments where ETS concentrations are very low. However, under normally encountered smoking rates, the uncertainties associated with nicotine's high adsorption rate are likely to be small.
ETS INDOOR AIR CONCENTRATIONS AND EXPOSURE
Numerous studies have measured concentrations of nicotine and RSP in a variety of indoor environments. These studies employed a range of sampling devices, sampled over varying timeframes (from minutes to days), and included highly variable information on various factors affecting the measured
---17 Leaderer, B.P. and S.K. Hammond. Environ. Sci. Technol., Vol. 25, 1991, p. 770-777.
---18 See, for example: Repace, J.L. Tobacco Smoke Pollution. In Nicotine Addiction, Principles and Management. Orleans, T. and A.H. Lowrey, eds. Oxford University Press, New York, 1993.
---19 Leaderer, B.P. and S.K. Hammond, 1991.
---20 The California Air Resources Board report, Toxic Volatile Organic Compounds in ETS: Emissions Factors for Modeling Exposures of Californian Populations, was prepared by the Lawrence Berkeley Laboratory and concluded that nicotine and ETS-RSP behave similarly.
CRS - 13
concentrations, such as number of cigarettes smoked and ventilation rates. EPA summarized much of this information in its report, to which the reader is referred for more detailed information.21
Stationary Air Samplers
Most of the studies used stationary air samplers. Although the results were highly variable, nicotine and RSP concentrations in smoking environments were consistently higher than in non-smoking environments. Table 2 shows the range of average values obtained in these studies. The minimum and maximum values are also presented in parentheses. Only studies reporting sampling times over four hours were included in the data on residential and office settings so as to more closely approximate occupancy time. Since occupancy time in restaurants is likely to be shorter than four hours, data from studies using shorter sampling times were included in the table.
TABLE 2. Indoor Nicotine and RSP Concentrations with Smoking Occupancy: Range of Average Values Reported (Min - Max Values) Location Nicotine ( g/m3) RSP (g/m3)a Residential 2-11 18-95 ( < 1-14) (5-560) Office 1-13 <5-62 ( < 1-35) (<5-90) Restaurant 6-18 35-986 ( < 1-70) (10-1370) ----------------------------------------------------------------- a RSP levels associated with smoking occupany were calculated by subtracting background RSP levels associated with non-smoking occupancy. Source: Figures 3-7 and 3-8, EPA, 1992.
The summary nicotine data in the table indicate that average concentrations in residences with smoking occupancy range from 2 üg/m3 to 11 üg/m3, with high values up to 14 üg/m3 and low values down below I üg/m3. Offices with smoking occupancy have average nicotine concentrations that are similar to those in residences, but with significantly higher maximum values. The data from restaurants show even higher maximum values. With regard to RSP concentrations, there is also broad overlap in the average values obtained from residential and office environments. However, the data from restaurants show a much wider range of values.
In a recently published study, Hammond and coworkers measured average weekly nicotine concentrations at 25 diverse worksites including fire stations, newspaper publishers, textile dyeing plants, and a variety of manufacturing
---21 EPA Report, chapter 3.
CRS - 14
companies.22 Between 15 and 25 samplers were placed in each worksite. Worksite smoking policy had a significant effect on the nicotine concentration. The median23 nicotine level in open-plan offices that allowed smoking was 8.6 üg/m3, but only 1.3 üg/m3 in worksites that restricted smoking to designated areas. In worksites that banned smoking, the median nicotine level was 0.3 üg/m3.
Guerin and Jenkins measured the concentrations of ETS constituents, including nicotine and RSP, in "typically encountered" residential and occupational indoor settings and found that low-level concentrations were much more common than higher-level concentrations.24 These results reflect the fact that the researchers included a significant number of non-smoking and smoking-restricted sites. Very high concentrations were generally found in enclosed areas designated for smoking, and in poorly ventilated areas where smoking intensity was high.
Personal Monitors
Measurement of indoor air concentrations of ETS components indicates the potential for exposure, but actual exposure also depends on the amount of time spent in a particular environment. The amount of exposure will depend on the individual's circumstances. A woman who lives with a nonsmoker but works in an office with smokers will receive most of her ETS exposure at work, whereas someone who lives and works with smokers may receive the majority of her exposure in the home where more time is spent.
Personal monitoring allows researchers to estimate individual exposure. Study participants wear a monitor that continuously samples and records the concentration of air contaminants to which individuals are exposed in the course of their daily activities. If subjects use different monitors in different indoor environments (e.g. home vs. workplace) and record the amount of time spent in each setting, then researchers can calculate the contribution of each environment to total exposure.
To date, few studies have measured ETS exposure to nicotine and RSP using personal monitors. Limited published data on nicotine show a wide range of ETS exposures in indoor environments with smoking occupancy, with average concentrations ranging from less than 5 üg/m3 up to 40 üg/m3. Other personal
---22 Hammond, S.K. et al. J. American Medical Association, v. 274, no. 12, 1995. p. 956-960.
---23 The median value is the mid-point of a range of measurements. Half of the values are less than the median, half are greater than the median.
---24 For more information, see Guerin et al., 1992; Guerin, M.R. and R.A. Jenkins. Recent Advances in Tobacco Science, Vol. 18, 1992, p.95-114; and Guerin, M.R. Environmental Tobacco Smoke Exposure Assessment. Paper presented at Japan Indoor Air Research Society, April 1993. Sponsored by U.S. Dept. of Energy. NTIS/DE93015521.
CRS - 15
monitor studies found that ETS exposure increased RSP levels between 18 üg/m3 and 64 üg/m3.25
It is difficult to assess the ETS contribution to nicotine and RSP levels for each indoor environment using these data. In many cases, study participants wore the same monitor for 24 hours, and the reported nicotine and RSP levels represent 24-hour average values. These values may underestimate the contribution of some non-residential indoor environments as they include home sleeping hours when presumably there was little if any ETS exposure.
Unpublished data from a recent multi-city study using personal monitors suggest that typical exposures are low relative to estimates obtained using stationary air samplers. This large study, conducted jointly by Oak Ridge National Laboratory and R.J. Reynolds Tobacco Company, recruited approximately 100 nonsmokers in each of 16 cities nationwide. Study participants were provided with two monitors -- one to wear at work and the other for the remainder of the 24-hour period -- and required to keep a detailed written record of their activities. In addition to nicotine and RSP, the monitors measured five other ETS constituents.
The average nicotine concentration in 415 smoker-occupied homes was 2.16 üg/m3, with a median level of 0.68 üg/m3, indicating that most participants received relatively little ETS exposure. The average and median nicotine levels in workplaces without smoking restrictions were 2.77 üg/m3 and 0.58 üg/m3, respectively. Researchers calculated total daily exposure to nicotine in each indoor environment by multiplying the average nicotine concentration by duration of exposure and breathing rate. Total daily nicotine exposure in smoker-occupied homes was 6.8 üg per day (üg/day), compared to a value of 5.8 üg/day for workplaces without smoking restrictions.
The study's authors suggested two explanations for the fact that average nicotine concentrations recorded in this study lie at the bottom end of the ranges reported in earlier studies. First, fewer smokers are lighting up in the presence of nonsmokers, a response to changing societal attitudes toward smoking. Second, nonsmokers are spending less time in obviously smoky environments. Nonsmokers who come in contact with smokers may receive relatively little exposure depending on their proximity to the smoker and the length of time spent in that indoor environment.
Noting the tobacco industry's involvement in the study, critics claim that it under-represented the amount of ETS exposure among nonsmokers. The study sampled a disproportionately low number of smoker-occupied workplaces. Out of 1,356 workplaces sampled, only 168 (12.4 percent) allowed smoking without restriction. National estimates of workplace smoking prevalence suggest that a significantly higher percentage of workplaces allow smoking (see later section on occupational ETS exposure). However, it is not possible to determine whether the recruitment procedures used in the study led to the
---25 EPA Report, tables 3-5 and 3-6.
CRS - 16
selection of participants whose ETS exposure in smoker-occupied indoor environments was significantly below average exposure levels for nonsmokers nationwide.
Biomarkers
The presence of a biomarker in the blood or urine provides direct evidence of ETS exposure and uptake. The relationship between the biomarker and exposure is complex due to many environmental and physiological factors. The most commonly used and widely accepted ETS biomarker is cotinine, the major metabolite of nicotine inside the body. Nicotine has a half-life of about 2 hours in the blood and is metabolized to cotinine and excreted in the urine. Cotinine has a half-life of approximately 20 hours in smokers, somewhat longer in ETS-exposed nonsmokers, which makes it a good indicator of ETS exposure and uptake over the previous two days.
Studies show that blood and urine cotinine levels in ETS-exposed nonsmokers are generally higher that those in nonsmokers reporting no ETS exposure, but far lower than the levels of cotinine in smokers. Comparisons of cotinine levels in smokers and nonsmokers indicate that ETS-exposed nonsmokers receive approximately 0.7 percent of the nicotine dose of an average smoker.26 Cotinine levels in nonsmokers have also been found to increase with self-reported ETS exposure. There is considerable variation in cotinine levels among smokers and ETS-exposed nonsmokers because of individual differences in the uptake, metabolism, and elimination of nicotine.
ETS CANCER RISK
The EPA classified ETS as a carcinogen based on the chemical similarities between inhaled MS and ETS, and evidence of ETS exposure and uptake by nonsmokers. Studies indicate that tobacco smoke is a lung carcinogen even at the smallest exposures to active smoking, and the risk increases with exposure, as measured either by number of cigarettes smoked per day, or years of cigarette smoking. According to the EPA, exposure to ETS, which is qualitatively similar to MS, therefore, should also increase the risk of lung cancer, and the evidence of widespread exposure to, and uptake of, ETS components in the general population is sufficient to conclude that ETS is a lung-cancer hazard.27
A few researchers have challenged the classification of ETS as a known human carcinogen based on its relationship to MS. They point to the fact that MS contains chemicals at concentrations of up to one million times those found in ETS, and that more of the chemicals are in the particle (tar) phase of MS. Differences between passive smoking (normal inhalation) and active smoking
---26 Jarvis, M.J. Mutation Research, Vol. 222, 1989. p. 101-110.
---27 See, for example, testimony presented by Dr. Douglas Dockery, Harvard School of Public Health, on July 21, 1993, before the House Committee on Agriculture, Subcommittee on Specialty Crops and Natural Resources.
CRS - 17
(deep inhalation) also affect the degree of exposure to vapor phase constituents and the deposition of particles inside respiratory passageways. Based on these considerations, an ETS chemist concluded that the evidence for ETS carcinogenicity remains questionable.28
Asserting that ETS is a lung carcinogen leaves unanswered the question: How great a cancer risk does passive smoking pose? Researchers have used nicotine measurements to calculate ETS exposure in terms of cigarette equivalents, by estimating the number of cigarettes one would have to smoke to receive the same amount of nicotine as breathing ETS in a particular environment for a given period of time.29 For example, the amount of nicotine inhaled by a nonsmoker working in a relatively smoky restaurant for eight hours is equivalent to smoking one-eighth of a cigarette?
Cigarette equivalents calculated for some of the known carcinogens in ETS yield much higher values because these compounds are emitted at higher levels in SS than in MS (see Table 1). About three times as much nicotine is emitted in SS as in MS, whereas approximately 30 times as much 4-aminobiphenyl (4ABP). Thus, a description of exposure in nicotine cigarette equivalents underestimates exposure to a known carcinogens in tobacco smoke by a considerable margin.31
The cigarette equivalent approach can also be applied to cotinine data. If, as stated above, cotinine levels in ETS-exposed nonsmokers average 0.7 percent of the levels found in smokers, and if one assumes that the average smoker smokes 19 cigarettes a day,32 then the amount of nicotine to which the average ETS-exposed nonsmoker is exposed is roughly equivalent to smoking one-eighth of a cigarette a day.
There are significant uncertainties in using cigarette equivalents to try to quantify ETS cancer risk. Estimates of ETS exposure using cigarette equivalents vary enormously depending on the compound chosen. Researchers
---28 Testimony presented by Dr. Michael Guerin, Oak Ridge National Laboratory, at the July 21 ETS hearing.
---29 The formula for cigarette equivalents: amount from ETS exposure/amount from smoking one cigarette.
---30 Assumes an average nicotine concentration of 18 g/m3. Exposures longer than 8 hours would lead to proportionately higher cigarette equivalents, as would higher breathing rates resulting from physical exertion at work. Based on calculations presented in Hammond et al., 1995.
---31 Recent newspaper advertisements by R.J. Reynolds Tobacco Company stated that nonsmokers are exposed to only slightly more than one "cigarette equivalent" a month in the workplace. However, this statement is misleading as it refers to nicotine cigarette equivalents and therefore underestimates exposure to many other toxic and carcinogenic compounds in ETS.
---32 U.S. Centers for Disease Control. Morbidity and Mortality Weekly Report, Vol. 41, 1992. p. 354.
CRS - 18
do not know how the levels of these individual compounds relate to overall ETS exposure, or exposure to those ETS constituents that may be linked to lung cancer. Indeed, they do not know which ETS constituents are responsible for lung cancer and other health effects attributed to ETS exposure. Although 4-ABP is a bladder carcinogen, it does not appear to be associated with lung cancer. Finally, the contrasting breathing patterns of active and passive smokers may strongly influence the degree of exposure and uptake of various tobacco smoke constituents in the lungs of smokers and nonsmokers.
In order to estimate ETS lung cancer risk using cigarette equivalents researchers assume that there is a linear relationship between exposure (number of cigarettes smoked a day) and cancer risk that extends from the relatively intense exposures typical of active smoking down to the much lower exposures associated with passive smoking. EPA uses this type of straight-line extrapolation from high exposures down to zero exposure in all its cancer-risk assessments but researchers do not know the actual shape of the exposure-risk relationship for passive smoking and lung cancer.
CRS - 19
ETS AND LUNG CANCER - EPIDEMIOLOGY
INTRODUCTION
This chapter presents a review of the epidemiology evidence for the possible relationship between ETS and lung cancer, based on results for spousal exposure. The review will particularly address the dose-response relationship between ETS exposure and lung cancer risk reported in many of these studies. Results of these studies will be presented first, followed by a discussion of the uncertainties associated with the analyses. The section will conclude with a discussion of the principal sources of possible alternative explanations of the results given in the studies. Attention is given to confounders and misclassifications errors.
BACKGROUND
The chemical similarities between mainstream and sidestream smoke and the association of active smoking with lung cancer are reasons for a possible relationship between ETS and lung cancer. But, they do not prove the relationship, since ETS is substantially diluted and aged compared to even low levels of active smoking. It is possible that ETS exposures are too small to be the cause of lung cancer in any meaningful sense; it is possible that some exposures are large enough to have an effect and others are not; and, it is possible that even a very limited exposure could cause some disease.
Epidemiologic studies are statistical studies of actual populations that are aimed at testing those hypotheses. By and large, these studies use as a measure of exposure to ETS, marriage to a smoker.
With only a few exceptions, these studies are of the "case-control" type. A group of non-smoking women ill with lung cancer (cases) are questioned as to the smoking status of their husbands and a comparable group from the population at large (controls) are also questioned. If a larger fraction of the cases have been exposed than of the controls, the risk of ETS is positive. The risk is usually expressed as a relative risk ratio (or odds ratio), which is the ratio of exposed to unexposed among the cases, divided by the ratio of exposed to unexposed among the controls. If the risk ratio is, for example, 1.2, that means that exposure to ETS increases the risk of lung cancer by twenty percent. (Such a risk would be quite small in absolute terms, however, because lung cancer among nonsmokers is quite rare).
An alternative but rarely used approach is a cohort study, where a large group in the population is followed and the exposure levels of those who develop the disease and those who don't are compared. Cohort studies are superior in theory to case control studies, but because lung cancer is extremely rare among nonsmokers thus requiring a large group, and because of the lengthy period of time required, these studies tend to be rare.
CRS - 20
Some studies have also asked questions regarding the degree of exposure, by asking subjects how long and/or how much their husbands smoked. If there is an effect of ETS on lung cancer, it should be greater with greater exposure measured by either intensity or duration. As statistical studies, the interpretation of the findings in these studies are subject to many limitations of statistical inference, and these limitations have been the subject of considerable controversy in the debate on ETS and lung cancer.
First, only a sample of the population is studied, and it is possible that any relationships observed are due to chance. Statistical results are always qualified by their degree of statistical significance, which is merely another way of measuring the probability that the results hold for the entire population and not just the particular sample under study. This measure is often expressed as a confidence interval (CI), which is centered on the actual measure of risk. For example if a 95 percent confidence interval is given, it means that there is a 95 percent chance that the truth lies between the two limits. There is a 5 percent chance that the answer falls outside the interval: 2 and 1/2 percent that it is larger and 2 and 1/2 percent that it is smaller. If the entire confidence interval falls in the positive risk range (the lower limit is at or above one), then the study would be interpreted as showing a positive risk at the 95 percent level, and we would normally accept the hypothesis (were there no other problems) that ETS poses a risk.
For large samples, the confidence interval will be narrow; for small samples it will be wide. Thus, in a small sample, the measured risk would have to be very high to achieve statistical significance. Indeed, researchers also sometimes refer to the power of a study to detect a small risk -- small studies have less power than large ones. The limited ability of small studies to accurately inform us of the true risk is important to keep in mind in evaluating the results. For example, seven of the eleven U.S. studies reviewed by EPA had only about a 20 percent chance of detecting a statistically significant risk of 50 percent (i.e., risk ratio -- 1.5) using a 95 percent confidence interval.
Over time, certain conventions for the level of statistical significance have developed; 95 percent is common. Statisticians are faced with two types of potential error: type I, accepting the hypothesis when it is not true, and type II, failing to accept the hypothesis when it is true. Any convention that is adopted balances between these errors -- the more you minimize one error, the greater the likelihood of the other error. If a standard convention for statistical significance is chosen, then small studies are more likely to be subject to type II error. However, there is no objective standard for determining what level of significance is necessary to accept a hypothesis; one is always dealing with some degree of probability.33
---33 There has been some criticism about the standard used by the EPA, which was a 90 percent confidence interval rather than a 95 percent interval. Critics have complained that standard was atypically chosen to ensure statistical significance in the over all weighted average of the EPA's combined studies. The EPA has responded with a justification for their choice. This issue is a procedural matter, and not one that relates directly to the evidence.
CRS - 21
In addition to considering sampling error in determining whether the results of a study are valid, there are other potential problems. Questions of statistical significance and statistical power relate only to the issue of sampling from a population. There are other potential problems with interpreting the results of studies, which primarily have to do with two issues: (1) are there other factors independently associated with both the development of lung cancer and exposure to ETS that could account for the relationship? and (2) are subjects properly identified into the correct groups -- for example, are all exposed cases truly ill with primary lung cancer, truly nonsmokers, are they all truly exposed to ETS, or have they correctly reported their exposure level? Some studies make considerable efforts to control for other factors and to verify the classification of subjects into the proper categories; others do little in that regard. Even the best of studies, however, face practical limitations on their abilities to verify and control.
Some critics have also suggested that there is a publication bias -- a tendency for studies that yield positive results -- those which support the hypothesis -- to be published.34 This behavior does not necessarily mean a deliberate bias on the part of editors and researchers. For example, in some cases a researcher might study many potential cancer-causing factors and simply not mention those that do not support the hypothesis being tested. If this tendency occurs, then published studies will be biased in favor of positive results. For that reason, large studies that are aimed at the beginning towards studying ETS may be more reliable.
Given the limitations of statistical analysis, what standards are used to evaluate the results, even when results are statistically significant? In 1964, a group of experts was brought together by the Surgeon General to define a set of criteria for causal inference. These criteria, which are often referred to as the Bradford Hill criteria, are widely used by epidemiologists today and are summarized in the box below.35
Epidemiologists typically await the results of several studies before weighing all the available evidence for a causal relationship. The first criterion is the strength of the association. How large is the relative risk? Hill argued that a strong association -- usually taken to mean a risk ratio of at least three -- is more likely to be causal than a weaker association because if it was due to confounding or some other bias, this effect would have to be large enough that it would presumably be evident. On the other hand, weak associations are more likely to be explained by undetected biases. The fact that an association is weak, however, does not rule out a cause-effect relationship. The strength of an association is not a biologically consistent feature but rather a characteristic that depends on the relative prevalence of other causes.
---34 LeVois, M.E. and Layard, M., Regulatory Toxicology and Pharmacology, Vol.21, 1995, p. 184-191.
---35 For a more detailed discussion of the Hill criteria, see Rothman, K.J., Modern Epidemiology. Little, Brown and Co., Boston, Massachusetts, 1986.
CRS - 22
Criteria for Causal Inference Strength of association: How big is the relative risk?
Consistency of association: Do similar studies by other researchers yield similar results?
Dose-response relationship: Does the risk increase with increasing exposure?
Temporal relationship: Does exposure precede the onset of illness?
Biological plausibility: Does the association make sense in light of biological knowledge?
Coherence: Is the association consistent with existing knowledge about the natural history of the disease?
Specificity of association: Is exposure linked to a single disease?
The second criterion is consistency of association; whether similar studies by other researchers yield similar results. If the relative risk is small, then the evidence of a dose-response relationship -- the third criterion -- becomes very important in attempting to determine causation. Does the risk increase with increasing ETS exposure?
The remaining criteria have more to do with the underpinnings of the basic theory rather than statistical matters, and are addressed elsewhere in the paper.
OVERALL EFFECTS AND PREVIOUS STUDIES
In performing its assessment of the possible contribution of ETS exposure to lung cancer in non-smokers, EPA relied on 31 epidemiology studies published over the period 1981-1992.36 These studies, which were carried out in several countries in addition to the United States, examined the possible lung cancer-ETS linkage using predominantly case-control methods to measure the relative risk of developing lung cancer due to exposure to ETS. In all cases, the primary objects of the study were non-smoking women subjected to ETS from a smoking spouse. The studies relied primarily on questionnaires to the case and control group members, or their surrogates, to determine ETS exposure and other information pertinent to the studies. All of the studies reported an average relative risk for the entire case group and several reported relative risk as a function of the dose of ETS reported to have been received by the case group members. In addition, 95 percent confidence intervals for the relative risk values were generally provided.
---36 EPA Report, p.114.
CRS - 23
Nearly all of the debate about the possible health effects of ETS, to date, has focused on overall relative risk. The EPA considered 31 studies -- including 11 from the U.S. -- in its analysis of ETS and lung cancer risk. Using a method of combining studies, called meta-analysis, it concluded that there is an overall relative risk of 1.19 for developing lung cancer for female non-smokers in the U.S. with a 90 percent confidence interval of (1.04, 1.35). In a 1986 report assessing the health effects of ETS, the National Research Council estimated a relative risk of 1.32, with 95 percent confidence limits of (1.16,1.52), for female non-smokers in this country? Both the NRC and the EPA concluded after further analyses of these results that a causal relationship existed between ETS and lung cancer in non-smokers. The earlier NRC study, however, had available a much smaller number of studies (9 overall and 3 from the United States).
The EPA report then used this result to calculate overall risk (annual deaths) due to exposure to ETS, assuming the risk was uniform among nonsmokers.
For a variety of reasons, EPA's conclusions have been controversial. While many in the scientific community have accepted the EPA conclusions, other have criticized them. First, the findings in the studies were mixed, and of the 30 studies examined by EPA (one Japanese study could not be used because of the presentation of data), 24 found an increased risk, though only five were statistically significant at the 95 percent level, and six actually found a negative risk (with one statistically significant). Of the eleven U.S. studies, eight found a positive risk and three found a negative risk, though none was statistically significant.
These studies originally considered by the EPA and their confidence intervals are shown in figure i (next page), ordered by increasing level of risk. Note that large studies have narrow confidence intervals and small studies have very wide ones. They incorporate a downward correction for a certain type of bias -- smoker misclassification -- that has been of some concern in evaluating the results of these studies. Note also that the EPA examined studies and ranked them in tiers with respect to their usefulness in four tiers; the fourth tier studies were deemed too poor to use in the analysis. (These studies are Lui, Wu-Williams, Geng, and Inoue; none was in the U.S.).
Figure i also includes four U.S. studies38 that appeared after the EPA cutoff, one of them the final version of the Fontham, et.al., (hereafter Fontham) study, which is the expanded and refined version of the original Fontham study included in the EPA report. (Thus, the original Fontham study should be subsumed by the new one and the final study should not be viewed as wholly new evidence). The risk estimate in the final Fontham study is similar to the original one included in the EPA study, but attains statistical significance
---37 NRC Study, p.231.
---38 Brownson, R.C., et.al.; Fontham, E.T.H., et.al.; Stockwell, H.G., et. al., and Kabat, G.C., et.al.
CRS - 24
Figure 1: Residential Epidemiologic Studies of Passive Smoking and Lung Cancer
[GRAPHIC MISSING]
Means plus 95 percent confidence intervals. Data from Tables 5-2 and 5-5, U.S. EPA, 1992.
* U.S. studies.
because of its larger numbers of observations. The other three new studies show, in one case, no effect (the Brownson, et.al,. hereafter Brownson study) in the other cases a positive effect that is not statistically significant (Stockwell,et.al., hereafter Stockwell, Kabat, et.al., here after, Kabat), and in the case of the Kabat study, very small.
None of these new studies was adjusted for smoker misclassification and their risk ratios would presumably be smaller if the standard EPA adjustment were made. The Brownson results would probably show a negative risk overall, the Stockwell results a smaller positive risk, which would remain statistically insignificant, and the Kabat result might disappear or even be negative). The EPA did adjust the original Fontham study, but only by a small amount because of the care taken in testing for misclassification in that study. In the final
CRS - 25
Fontham study, a small adjustment could render the overall Fontham results statistically insignificant at the 95 percent level.
Simply comparing results of different studies is of limited value, since, as noted above, small studies provide limited information because of sampling error. For that reason, the EPA combined the studies (through a meta-analysis) to yield the overall estimate of a risk of 1.19 percent. The rationale behind combining studies is simple: if there are a lot of small studies that each do not obtain statistical significance, but each have a positive effect, then if they could have been studied as one group of observations, the test would have been more powerful. Combining the studies takes into account the probabilities associated with the whole body of studies.
Although this approach is valid, and is superior to just counting up the studies, it still does not entirely clarify the risk. Even when overall risk is considered, it is a very small risk and is not statistically significant at a conventional 95 percent level. Moreover, problems of bias and confounding that are mentioned above (and will be discussed subsequently) still occur in most studies; they probably occur to some extent, but with different degrees of seriousness, in all of the studies. Some studies were much more careful in controlling for other factors that might influence the study's results.
The new studies, including the very large Brownson study, did not clarify the existence of a risk. Indeed, they complicated the interpretation of the evidence, since the two largest U.S. studies -- Fontham and Brownson - found in one case a positive risk that was barely statistically significant and the other no risk at all.
For these and other reasons, the conclusions in the EPA study have generated considerable controversy. While receiving support from a segment of the scientific community, others have registered criticism focusing on the uncertainty inherent in such low risk values and argued that there were potentially other explanations for these results if indeed they were not due to chance alone.39
Missing from most of these analyses was any emphasis on the dose-response relationships observed in many of the studies, traditionally an issue that is considered in establishing causality. In many studies, respondents were questioned as to the degree of exposure, either in number of cigarettes per day the husband smoked, number of years the husband smoked, or a multiple (pack years). If there is a risk from ETS, it would be expected to rise with exposure.
---39 See for example Review of the U.S. Environmental Protection Agency's Tobacco and Smoke Study, hearing before the Subcommittee on Specialty Crops and Natural Resources, Committee on Agriculture, U.S House of Representatives, July 21, 1993, Washington, DC; and Smith, Cart J., et.al.
CRS - 26
Of the 31 studies reviewed by EPA, 17 presented data on the variation of relative risk as a function of ETS exposure levels.40 EPA carried out an analysis of these studies including the calculation of pooled risk estimates, confidence intervals and trend tests.
EPA also looked at high exposure levels to see if there was a significant effect. EPA went on to say that "It appears that relatively high exposure levels are necessary to observe an effect in the United States, .... "in its assessment of dose response trends. As noted above, positive trends were viewed as evidence of an effect, but no further consideration of dose-response relationships was given in the EPA analysis.
In particular, EPA did not use dose-response relationships in its estimates of population risk. If risk does vary by exposure level, then this assumption may not give a true picture of the risk distribution of developing lung cancer from ETS.
Attention to the dose-response trends is particularly important because of the possibility that much of the risk may be concentrated at the largest, integrated, ETS exposure levels (daily ETS exposure times its duration). If so, such an observation could have substantial consequences for possible mitigation actions. In addition, dose-response analyses can be used as an additional test of the possible role of confounders and misclassification biases in explaining reported ETS health risk.
Also, most analyses of other potential environmental hazards consider the effect of dose levels when assessing the possibility of public health dangers and policy response. Regulations to protect the public from such hazards usually have exposure limitations rather than banning exposure altogether. Given the potential importance of dose-response relationships for ETS and the extensive comments that have already been made on the EPA analysis of the average relative risk, this analysis has chosen to concentrate on the dose-response issue.
Turning to more specific measures of exposure does, however, introduce a potential new form of bias -- recall bias. The more specific the question about exposure, the more precise the measure, but the less accurate the recall. That is, there is likely to be a very small error rate in reporting marriage to a smoker, but there could be a significant error in reporting actual amounts of exposure, such as numbers of cigarettes smoked by a spouse, particularly in the past.
In reviewing the dose-response analysis of ETS, the 17 studies listed in the EPA report which reported dose-response data along with three other studies, not considered by EPA, which also examined the dose-response relationship, were examined.41 These three, Brownson, Stockwell, and Kabat, appeared after
---40 EPA Report, p. 144.
---41 See Appendix B for list of studies used in the table.
CRS - 27
the EPA report was published. The Fontham study was only partially completed when included in the EPA analysis.
All but two of the studies used a case-control method. The others were cohort studies. Cases were selected from various sources of lung cancer patients or those who recently died of lung cancer, and the controls were chosen using various random selection techniques. Only individuals who stated that they had never smoked, or, in some cases, had quit several years prior to the study, were selected as participants in the case and control groups.
The participants were interviewed directly if possible -- a number of the cases in nearly all of the studies required surrogates -- to determine exposure to ETS and other information relevant to the studies. For example, data on age, educational level obtained, occupation, and other factors were obtained to permit matching of controls and cases and to eliminate as many factors as possible that may compromise the results. In addition, many of the studies attempted to obtain data on dietary habits to control for these potential confounders in calculating the relative risk values. More will be presented on this issue below.
Finally in all of these studies, the cases and controls who stated they had been exposed to ETS were asked for information about the extent of exposure. In most of the studies, this information was provided separately for number of cigarettes per day and for duration of exposure. In a few of the studies, an integrated exposure level, packs of cigarettes per day times years exposed at that daily level, was provided.
There are two final caveats to interpreting these data. First, unlike the overall results presented earlier, these measures have no downward correction for smoker misclassification. Second, by segmenting the observations in the study, the numbers become smaller and the tests less powerful (less able to detect a statistically significant risk).
RESULTS
The results are summarized in tables 3, 4 and 5 on the following two pages. Each study used standard statistical methods to carry out the analyses. Relative risk values (odds ratios) -- labeled RR in the tables -- and 95 percent confidence levels (in most studies) -- labeled CI in the tables -- were calculated using logistic regression techniques or related methods. Those confidence intervals marked with an * are at the 90 percent level. Average relative risk values and confidence intervals were measured along with those at various exposure levels. Only the latter are reported in the tables.
The tables are organized by exposure measures. Table 3 is cigarettes per day, table 4 is pack-years (packs per day times years at that level), and table 5 is smoker years. All but one of the studies in the last category also reported results in terms of one of the other two exposure measures. In the table, exposure levels were adjusted from the reported levels when possible to keep the
CRS - 28
studies as comparable as possible. The key to the numbers in the column marked study is in Appendix B.
-----------------------------------------------------------------
Table 3 -- ETS Dose-Response Observations -- (Cigarettes per Day)
Study Exposure RR 95% CI Study Exposure RR 95% CI --- 1 1-20 1.93 (1,29,2.88) 2 1-20 1.95 (1.13,3.36) _>21 2.07 (1.07,4.01) _>21 2.55 (1.31,4.93) --- 3 1-19 1.27 (0.85,1.89) 4 1-19 1.41 (1.03,1.94) _>20 1.10 (0.77,1.61) _>20 1.93 (1.35,2.74) --- 5 1-10 0.82 (0.42,1.61) 6 1-19 1.3 (0.7,2.3)* _>11 1.06 (0.49,2.30) _>20 1.7 (0.9,3.2)* --- 7 1-19 1.12 (0.7,1.8) 8 1-9 1.40 (1.1,1.8) _>20 2.11 (1.1,4.0) 10-19 1.97 (1.4,2.7) _>20 2.76 (1.9,4.1) --- 9 1-20 1.8 (0.6,5.6)* 10 1-20 1.54 (0.8,3.0) _>21 1.2 (0.3,5.2)* _>21 1.71 (0.9,3.4) --- 11 1-20 1.76 (1.0,3.2) 12 1-15 1.02 (0.6,1.8) _>21 1.19 (0.5,3.0) _> 16 (1.0,9.5) --- 19 5-19 1.58 (0.4,5.7)* _>20 3.09 (1.0,11.8)*
Table 4 - ETS Dose-Response Observations -- (Pack-Years)
Study Exposure RR 95% CI Study Exposure RR 95% CI --- 14 1-40 1.18 (0.44,3.20) 15 1-39.9 1.02 (0.82,1.26) _>41 3.52 (1.45,8.59) _>40 1.43 (1.07,1.91) --- 16 1-40 0.70 (0.52,1.18) 17 1-24 0.71 (0.37,1.35) ->41 1.30 (1.0,1.7) 25-49 0.98 (0.47,2.05) _>50 1.10 (0.47,2.56)-----------------------------------------------------------------
Most of the studies report a small but positive effect which increases as exposure level increases. Three of the studies show effects of less than 10 percent excess risk at the highest exposure levels, and four of the studies show no indication of a trend of increasing risk with increasing exposure. In addition, two studies which reported more than one measure of exposure, showed conflicting results. In one case a trend was indicated while using the other measure, it was not. Only 10 of the studies showed any results which are
CRS - 29 statistically significant at the 95 percent level, and for four of those studies, only the highest exposure levels yielded statistically significant results. Three of the latter group reported its results in terms of pack-years. One of that group, however, the study by Fontham did not show any statistically significant results when exposure was expressed in terms of smoker years.42-----------------------------------------------------------------
Table 5 - ETS Dose-Response Observations -- (Smoker-Years)
Study Exposure RR 95% CI Study Exposure RR 95% CI 18 1-21 1.6 (0.8,3.2) 6 1-19 21 (1.0,4.3)* 22-39 1.4 (0.7,2.9) 20-39 1.5 (0.8,2.7)* >_40 2.4 (1.1,5.3) >_40 1.3 (0.7,2.5)* --- 8 _<19 1.49 (1.15,1.94) 10 _<19 1.26 (0.56,2.87) 20-39 2.23 (1.54,3.22) 20-39 1.62 (0.82,3.19) >40 3.32 (2.11,5.22) >40 1.88 (0.82,4.33) --- 13 1-30 12 15 1-15 1.10 (0.83,1.46) >_31 2.0 16-30 1.33 (0.98,1.80) >_31 1.23 (0.91,1.66) --- 20 20-29 1.1 (0.7,1.8) 30-39 1.3 (0.8,2.1) >_40 1.7 (1.0,2.9)-----------------------------------------------------------------
Only eight of the studies which tested for trend found it to be statistically significant at the 95 percent level. Included in this group are two tier 4 studies;43 without these studies, and with the 95 percent standard, only six would be significant. All of the trend analyses include zero exposure. If the trend was linear down to zero exposure, then including that level in the trend analysis would yield the same results as when excluded. If there was a threshold effect, then a trend test which included the zero exposure level might show a trend even if an analysis which included only exposures above zero did not show such a trend. In other words a sharp rise at some exposure level above zero could incorrectly be interpreted as a dose response trend over all exposure levels.
As mentioned above, EPA calculated an overall relative risk from the relative risk values at the highest exposure levels even though these studies did not all use the same measure of exposure level. For the seven U.S. studies giving such information, a combined relative risk of 1.38 with a 90 percent
---42 It should be noted that when reporting relative risk for non-smoking females against smoker years of exposure, Fontham included all sources of exposure at home while the results measured against pack-years included only spousal exposure.
---43 In assessing the utility of the various epi studies for evaluating a linkage between ETS and lung cancer, EPA established a ranking system of four tiers, the lowest of which is tier 4. Studies falling in tier 4 were excluded by EPA from its analysis of ETS and lung cancer.
CRS - 30
confidence interval of (1.13,1.70) was calculated.44 The EPA also performed a trend test for the combined U.S. studies and found it to be statistically significant at the 99 percent level.
It is also worth examining the reported risk values at the lower exposure levels. Based on the distribution of controls in these studies, a much higher fraction of the non-smoking population in the United States which is exposed to ETS, is exposed to the lower levels. Therefore, if there is a real effect at these lower levels, most of the risk would reside there. If there is a threshold exposure, however, it may be that most of the exposed non-smoking population would be at no risk from ETS. The studies reporting their results as function of cigarettes per day and smoker years which show a trend, give no indication of a threshold, i.e., a level below which the measured effect is negligible. For those studies presenting their results in terms of pack-years, however, all of them show negligible risks below some level, in the range of 40 pack-years. One study in this group showed no effect at any level.
ANALYSIS
Risk and Exposure Measurement
The results presented by these studies indicate that if there is any risk of developing lung cancer from exposure to ETS, it increases as the exposure level increases. As mentioned, however, both the size of the effects measured and the lack of consistent, statistically significant data lead to considerable uncertainty.
An additional problem in trying to extract any conclusions from these 20 studies is the different measures of exposure levels used, cigarettes per day, smoker years and pack-years. Pack-years -- an integrated exposure of daily intensity summed over time -- is probably a better way to measure exposure levels than cigarettes per day. This measure, however, is probably the least precise of the three measures because it is most subject to recall error. Evidence from studies linking direct smoking with lung cancer indicates that the risk increases in proportion to the number of years smoked at a given level. One might suspect that any lung cancer risk from ETS would behave similarly.
Only if there is perfect correlation between cigarettes per day and number of years of smoking would these measures serve as well as the pack-year measure. If that correlation is imperfect, the other dose measures are inferior to pack years, although the overall direction is likely to be the same.
At the same time, each of these measures require less recall. It is likely, however, that recall errors are more serious for number of cigarettes per day than for number of years, especially if smoking occurred in the past. That is, it is probably easier to remember how many years someone smoked than how much they smoked. If so, years might be the best measure of exposure if recall bias is severe.
---44 EPA Report, p. 144.
CRS - 31
One implication of the potential disparity between the different types of exposure measurements is that combining risk assessments of several studies at the highest exposure levels probably yields misleading results.
All of the twelve studies using cigarettes per day as a measure of exposure show elevated risk at the highest exposure level although only about half are statistically significant -- not surprising given that most studies are small. Not all show a consistent trend, however. All four of the pack-year studies also show elevated risk at high exposures, with three out of four statistically significant. (Again, the largest studies show a statistically significant risk.) Of the six studies using years, all involve positive results but only two are identified as statistically significant.
The pack-year studies also offer evidence that non-smokers exposed to lower levels of ETS -- below 40 pack-years -- have little or no relative risk of developing lung cancer from ETS. The two largest case-control studies in terms of sample size -- Brownson and Fontham -- show this threshold behavior. Neither study, however, claims to be able to demonstrate a threshold effect because they lack the statistical power to make such precise measurements at such small levels of relative risk.45 Indeed, as pointed out above, most epidemiologists state that it is virtually impossible to measure a relative risk below 1.1 using currently available epidemiology techniques. When considering the confidence intervals for the various exposure levels for these two studies, several different curves could be drawn, including a straight line, to represent the variation of relative risk as a function of exposure. Nevertheless, the possibility cannot be ruled out that a threshold level does exist if there is a real effect from ETS.
Confounding
Critics of studies which assert that ETS is associated with an increased risk of lung cancer claim that these studies have not adequately accounted for potential confounders. They argue that the small values of the relative risk found in these studies (usually less than 2) makes the probability relatively high that confounders are the cause. Potential confounders are behavioral patterns or biological conditions which may be a risk factor for the disease under investigation. To be an actual confounder, however, these patterns and/or conditions must be associated with the exposure under study in that study. This pattern and/or condition also must be present in sufficient strength to be a plausible source of the excess risk in the situation under study. A third test of a candidate confounder can be made using dose-response observations.46 Any confounder that is to explain that risk likely would have to become stronger if and as the integrated ETS exposure increases.
---45 Dr. Michael Alavanja, personal communication, June 12, 1995.
---46 Noel S. Weiss,, et.al., American Journal of Epidemiology, Vol. 113, No. 5, May 1981, p.487-
CRS - 32
Critics argue that association of potential confounders with ETS exposure is likely to be met in ETS studies because the health habits of non-smoking spouses of smokers are similar to their smoking spouses and are, therefore, inferior to non-smoking spouses of non-smokers. Several studies have investigated this assertion. One group has examined the differences between exposed and unexposed non-smokers in terms of several dietary and related factors without attempting to measure relative risk, while the other group includes several studies which measured the relative risk of these factors in conjunction with that of ETS.
Two recent studies examined the dietary habits of a large populations of individuals who are exposed to ETS either at home or in the workplace.47 48 The two studies attempted to measure consumption of dietary nutrients suspected of being associated with cancer risk, often as an inhibitor to developing cancer. Neither study attempted to measure the differences in dietary behavior as a function of level of ETS exposure. Both studies showed a difference in diets between non-smokers exposed to ETS and those not exposed for most of the nutrients tested.
In one of the studies, however, only a few of the differences for the nutrients were statistically significant, and then only at the highest intake differences. The other study found that the differences investigated were all statistically significant, but that the dietary differences between exposed and unexposed non-smokers was much less than the corresponding differences between smokers and non-smokers. That study also concluded that the nutrient consumption by both exposed and unexposed non-smokers generally exceeded the recommended daily allowance. The authors speculated, however, that ETS and nutrients may interact in a way that would increase any nutrient requirements as a cancer inhibitor compared to when no ETS was present. The only disagreement between the two studies was dietary fat where Emmons, et.al., found that those exposed to ETS consumed a higher percentage of calories from fat than those unexposed, while Matanoski, et.al., found no difference in intake of fatty acids between the two classes of exposure.
In another study which investigated both the effect of ETS exposure and diet on lung cancer risk, only small differences were found between cases and controls for all foods included in the study except fruit.49 The study found fruit intake generated a statistically significant relative risk for lung cancer of less than one; i.e., it acted as an inhibitor. Controlling for each of these factors showed them to be independent of one another in affecting lung cancer relative risk measurements.
---47 Matanoski, et.al., American Journal of Epidemiology, Vol. 142, No.2
---48 Emmons, E.M.,, et.al., European Journal of Clinical Nutrition, Vol. 49, 1995, p.336-345.
---49 Kalandidi, A., et.al., Cancer Causes and Controls, Vol. 1, 1990, p.15-21.
CRS - 33
A study focusing on beta carotene intake for non-smokers exposed to ETS compared to those not exposed found a statistically significantly lower amount in the former compared to the latter.50 The authors estimated that such differences could act to reduce the measured relative risk -- total relative risk 'due to ETS by about 10 percent. No relationship between beta carotene intake and duration of exposure to ETS was found.
A 1991 study examined specific dietary habits of individuals exposed to ETS compared to those not exposed to ETS.51 The study was confined to factors for which there has been evidence of an association with an increased risk of lung cancer, diets low in beta carotene, and high in cholesterol and total fat. Results showed an inverse correlation between ETS exposure levels and consumption of beta carotene, cholesterol and total fat among non-smokers. Exposure levels were measured by cotinine levels and, therefore, only measured current exposure. On the basis of risk values relating a low beta carotene diet to the risk of lung cancer, the researchers calculated corrections to the ETS risk values in order to determine the adjustment that may be needed because of reduced beta carotene consumption. He found corrections to the measured ETS risk values of about 11.5 to 12 percent. For cholesterol and total fat, however, since consumption decreased with increasing ETS exposure, any confounding correction would tend to raise the measured ETS risk value. No numerical corrections were presented in the paper.
Another study examined the possible contribution of a large number of food types as well as ETS to lung cancer risk among non-smoking women.52 The study measured the relative risk of developing lung cancer as a function of the food dosage consumed. The only dietary components to have statistically significant relative risk factors were saturated fat, citrus fruits and juice, and beans and peas. The last food reduced the risk as its consumption increased. No effect due to beta carotene was observed. Furthermore, the authors reported that no interaction between ETS and the various dietary components could be measured. The most important contributor to increased lung cancer relative risk was saturated fat. Women who consumed the highest amounts of saturated fat -- a mean value of 20 percent of their daily calories -- had a lung cancer risk value of over 6. The paper reported that a biological link between saturated fat consumption and lung cancer was still speculative although preliminary experimental evidence of such a connection existed. The authors, however, were not able to offer any explanation for the connection between citrus fruit consumption and lung cancer risk.
Analysis of other potential confounders is not as extensive as for dietary factors but some work has been completed. One study explored the relationship
---50 Sidney, S., et.al., American Journal of Epidemiology, Vol. 129, No.6, June 1989, p. 1305-1309.
---51 Loic Le Marchand, et.al., Cancer Causes and Control, Vol. 2, p.11-16.
---52 Alavanja, M.C.R., et.al., Journal of the National Cancer Institute,Vol.85,No.23, Dec. l, 1993, p. 1906-1916.
CRS - 34
between pre-existing lung disease (asthma, pneumonia, emphysema, bronchitis and tuberculosis) and lung cancer risk.53 The authors measured a risk value of about 1.4 for never smoking women. From these results, the authors concluded that about 13 percent of all lung cancer deaths in never smoking women were due to a pre-existing lung disease. The research did not find any interaction between ETS exposure and pre-existing lung disease.
A 1983 study examined various factors including alcohol and marijuana consumption, and exposure to workplace hazards by a sample of the subscriber population at a Kaiser-Permanente Medical Care Center.54 They found that these three factors were correlated with ETS exposure and, further, increased as exposure to ETS, as measured in hours per week, increased. The percentage of those exposed to ETS who also used alcohol and/or marijuana on a weekly basis was quite small, 7 percent or less, and included both males and females. The percentage exposed to workplace hazards ranged from 30 percent at no ETS exposure to 37 percent at the highest ETS exposure. The rate of increase in exposure to occupational hazards with ETS exposure reported in the study was modest. The number of survey participants who reported exposure to occupational hazards increased 7.5 percent as ETS exposure increased over 800 percent on average. The connection, if any, between rate of increase in exposure to occupational hazards and increased lung cancer risk was not given.
Finally, a study just published reviewed the lung cancer risk of a variety of potential confounders? This paper reviewed interactions between the various suspected contributors to lung cancer in non-smoking women. The authors determined that about 48 percent of all those lung cancers could be explained by the seven factors they covered. The largest contributor measured in the study was saturated fat (22 percent) followed by former smoking (17.5 percent), pre-existing non-malignant lung disease (10 percent), ETS (6 percent), occupation (5 percent), family history of lung cancer (4 percent) and domestic radon (1.5 percent). All but the ETS and radon measurements were statistically significant. When only lifetime non-smokers were considered, however, the ETS contribution increased and became statistically significant. Among this group of non-smokers, ETS was measured to have accounted for 7.5 percent of all lung cancer deaths. This contribution was still exceeded by previous lung disease and saturated fat. In making these calculations, the authors controlled for all items except the particular factor being considered. No interactions between any of the items was found.
The evidence from these studies appears inconclusive about whether confounders may be responsible for the measured ETS risk values, particularly
---53 Alavanja, M.C.R., et.al., American Journal of Epidemiology, Vol. 136. No.6, Sept. 15, 1992, p.623-632
---54 Gary D. Friedman, et.al., American Journal of Public Health, Vol. 73, No. 4, April 1983, p.401-405.
---55 Alavanja, M.C.R., et.al., Cancer Causes and Control, Vol. 6, 1995, p.209.
CRS - 35
those at the most extensive ETS exposure levels. While it is fairly clear there are differences between exposed and unexposed non smokers for many of these potential confounders, it is uncertain whether that difference will be of consequence in developing lung cancer. There are several reasons for this. First, with few exceptions the measured relative risks of these potential confounders are about the same as those measured for ETS exposure and are at least as uncertain as the ETS values. As a result, in order to account for much or all of the measured risk value, a confounder or combination of confounders would have to be present at levels intense enough to affect the etiology of lung cancer in many or all of the cases for which ETS induced lung cancer is suspected. Second, the potential confounder has to be either a likely cause or inhibitor of lung cancer. For example, alcohol consumption, which has been shown to be greater in exposed than unexposed non-smokers, and which is a suspected cause of some cancers, has not been shown to be connected by itself with lung cancer. There are indications, however, that excessive alcohol consumption in conjunction with smoking can increase the lung cancer risk.
Furthermore, the uncertainties exhibited in the measurements of the risk of most of the potential confounders, as expressed by the absence of statistical significance or conflicting results, suggests that none of them can be considered a clear cause or inhibitor. For example, there is considerable uncertainty about the role of beta carotene -- long thought to be a cancer inhibitor -- in affecting the risk of lung cancer. Beta carotene is often mentioned as a confounder because non smokers exposed to ETS appear to consume less than unexposed non-smokers. A recent study found that beta carotene not only did not inhibit the development of lung cancer, but may actually enhance the risk.56
A third reason is that there is disagreement, as reported above, about whether there are consumption differences between exposed and unexposed nonsmokers for the potential confounder with the largest measured risk for lung cancer -- saturated fat. Fourth, studies which have attempted to control for these potential confounders -- in particular those by Fontham and Brownson 'do not find that they contribute any confounding to the measured ETS induced risk in those studies.
Fifth, evidence of potential confounders being correlated with increasing ETS exposure so as to offer a possible explanation for ETS dose response observations, is mixed. Examples of such confounder tracking has been reported, but for many of these confounders there is a question about whether they are a lung cancer risk factor. The cholesterol and total fat observations may mean that some confounders could raise the measured ETS risk values. Trend data showing the relationship between the levels of potential confounders and ETS exposure, are limited, however, so this possibility is speculative at this time.
---56 The Alpha-Tocopherol, Beta Carotene Cancer Prevention Study Group, New England Journal of Medicine, Vol.330, No.15, April 14, 1994, p.1029-1035.
CRS - 36
Misclassification Bias
Bias is generated from errors in the design, conduct, or analysis of an epidemiology study which result in a false measure of an association. There are several types of bias encountered in epi studies, including smoker misclassification, exposure misclassification and recall bias. Smoker misclassification would result from incorrectly assigning lifetime non smoker status to someone who actually smokes or who was a former smoker. Exposure, or random, misclassification would be the result of assigning someone to the exposed category when they actually had not been exposed to ETS. Recall bias occurs when someone reports an incorrect level of exposure to ETS because they are unable to recall the correct levels. Included is the situation of not recalling that one's spouse actually smoked. Some of these errors may be systematic in that they are a result of events or behavior which could be predicted to push the error in one direction. An example would be if some case group members provide incorrect information about their smoking or exposure status because of their disease status. Control group members, who do not have lung cancer, would have no reason to provide such incorrect information. Random errors cannot be predicted by events or behavior. Such errors are just as likely to occur in the case as control groups.
In this analysis, the consequences of each of the three types of misclassification will be examined using a mathematical model developed by EPA to calculate the downward correction to the observed relative risk values to account for smoker misclassification bias.57 The model has been expanded to examine the effect of exposure misclassification and recall bias. In addition, modifications were made to allow for differential misclassification.
Smoker Misclassification
Smoker misclassification has drawn the most attention in the ETS studies to date. Surveys have indicated that a fraction of self-reported nonsmokers are actually current or former smokers. Because the relative risk of developing lung cancer from direct smoking is so high compared to any of the measured ETS risk values, it is possible that only a small percentage of smokers would need to be misclassified as nonsmokers to account for a large part of the measured ETS risk. Furthermore, while such misrepresentation can occur for both exposed and unexposed non-smokers (both cases and controls), it may be more likely to occur to the former because smokers tend to be married to smokers. This situation would create a bias resulting in an overestimation of the risk value because it would increase disproportionally the observations in the exposed cases.
The EPA model used to assess the consequences of smoker misclassification is dependent on a number of parameters including the misclassification rates of current regular female smokers (although they may have just recently quit), former female smokers, occasional female smokers, and the risk of developing
---57 EPA Report, p.311-335.
CRS - 37
lung cancer from smoking for each group.58 In addition, the prevalence rate of never smokers for the group under study is required. There are several other parameters needed for the model which must be derived from experimental observations, but those listed above are the most critical.
Table 6 - Smoker Misclassification Consequences ---------------------------------------------------------------- Misclassification Condition Required Rates and Adjusted RR Values Rate - % (RR=1.0) Rate-% (CI_<1.0) Non-differential 10.1 2.8
To see the consequences of smoker misclassification, the model is used to calculate the misclassification rate of current (or recently quit) female smokers that would be needed to reduce a measured ETS risk value to 1.0.(59) A second test is to determine the current smoker misclassification rate needed to cause the measured risk to no longer be statistically significant at the 95 percent level. For this test, the model is used to find the rate when the lower limit of the 95 percent confidence interval drops below 1.0. The results are shown in table 6. The Fontham study is used for this analysis because it provides most of the data needed for the model and the rest of the data is available from the EPA study.60 The calculations are carried out on the measured risk value at the highest exposure level, 79.9 pack-years or more. For example, if the smoker misclassification rate were 10.1 percent, the measured risk of 1.87 for that exposure group would actually be 1.0, indicating no risk from ETS. All of the measured risk would be due to a group of smokers who had been incorrectly identified as non-smokers.
---58 The most common method of determining smoker misclassification is through measurements of cotinine levels in the blood. Because of the time sensitivity of this measurement, it will not pick up individuals who have recently quit (within a several month period prior to the measurement). Such people still retain the high risk associated with smoking, however, and should be included in any complete accounting of the current smoker misclassification rate.
---59 The EPA model assumes that the former smokers have not been smoking for at least 10 years. This condition led EPA to use a value for excess risk for former smokers which was about 9 percent of the value of the excess risk of current smokers. This assumption results in a total relative risk for those in the former smoker category only slightly higher than measured values of ETS relative risk. Therefore, smokers who had quit more recently and had a higher excess risk, must be included in the current smokers category. For occasional smokers, EPA assumes that their relative risk is 16 percent of that of current smokers based on cotinine measurements which showed levels of cotinine in occasional smokers to be on average 16 percent of that of current smokers.
---60 EPA Report, p.327. It is important to note that Fontham undertook extensive efforts to minimize the effect of smoker misclassification (see below). The use of the Fontham data for these misclassification rate calculations does not imply that such rates are necessarily likely for their study.
CRS - 38
To simplify the analysis, misclassification rates for long ago former smokers and occasional smokers are arbitrarily set to zero.61 The result of this analysis is shown in table 6 above for the case of non-differential misclassification (equal rates in both the control and case populations), the standard assumption made in smoker misclassification corrections.
Exposure Misclassification
Next, the model is used to examine the consequences of exposure misclassification. When a case and/or control group member identifies herself as having a smoking spouse but is actually unexposed to ETS, the participant is incorrectly counted as exposed. Adjusting for such exposure misclassification would increase the measured relative risk. Table 7 shows the effect of this misclassification on the measured values of risk as a function of exposure level for the Fontham study. Two misclassification rates are chosen for illustration -- 10 percent and 20 percent. For example, at the highest exposure level -- above 80 pack years -- if 10 percent of those cases or controls who state their spouses smoke actually are not exposed to ETS, the measured risk rate of 1.87 would actually be 1.89. If that exposure misclassification rate were 20 percent, the actual risk would be 1.90. No studies have been done to date attempting to measure exposure misclassification rates. In order to carry out this illustrative calculation, it has been assumed that the misclassification rate is the same for both controls and cases. Further, the misclassified individuals were distributed among the various exposure levels in proportion to the number of cases and controls in that level.
Table 7 - Relative Risk -- Exposure Misclassification Exposure Level Misclassification (pack-years) 0 10% 20% <15.0 1.02 1.04 1.06 15.1-39.9 1.02 1.03 1.05 40.0-79.9 1.34 1.35 1.38 >_80.0 1.87 1.89 1.90 All Levels 1.12 1.13 1.15
Recall Bias
Recall bias is simulated in the model by assuming that a fraction of the exposed members of the case and control groups have either overestimated or underestimated their exposure level. To see the effect of recall bias, a few illustrations are presented. The data from the Fontham paper are used for
---61 Including them at levels used in the EPA analysis (11.7 percent for long ago, former
smokers and 24.2 percent for occasional smokers) and with the same assumed lung cancer risk rates used would result in a decrease -- about 20 percent -- in the regular smoker misclassification rates needed to drive the relative risk to zero or to make the measured risk no longer statistically significant at the 95 percent level. If either or both of the relative risk values for occasional and long ago former smokers is increased above those assumed by EPA the contributions of these two categories to smoker misclassification bias will grow.
CRS - 39
these illustrative cases.62 First, recall bias rates are calculated which would be required to reduce the relative risk for each exposure level to the average relative risk for all levels; i.e., to eliminate the dose response trend. In this first illustration, it is assumed that only the cases are subject to recall bias. The effect of recall bias in the controls will be discussed below. For the data being used -- Table 3 in the Fontham paper -- the average relative risk is 1.12. The results, shown in table 8, give the required recall bias rates for each level to reach that value. A positive rate means that some participants at that level overestimated their exposure levels and actually belong at lower exposure levels. Negative values indicate how much these exposure levels should grow in order to flatten the dose-response curve. For example, at the highest exposure level -- above 80 pack-years -- if 40.1 percent of the case members in that group had over estimated their exposure and it actually ranged between 40 to 79.9 pack-years, the actual risk of the above 80 pack year group would drop to 1.12 from the measured value of 1.87. For the exposure level below 15 pack-years, if 9.3 percent had underestimated their exposure and actually belong in the next highest group -- 15.1 to 39.9 pack-years -- the actual risk for the lowest exposure level would rise to 1.12 from the measured value of 1.02. The final column gives the net number of people shifted to each level corresponding to the recall bias rate. A negative number, of course, means that participants are lost from that level. None of the adjusted relative risk values are statistically significant.
Table 8 - Effects of Recall Bias Exposure RR Bias Adj RR Cases Level Rate( % ) Shifted _<15.0 1.02 -9.3 1.12 14 15.1-39.9 1.02 -10.0 1.12 9 40.0-79.9 1.34 16.2 1.12 -13 >_80.0 1.87 40.1 1.12 -10
Another indication of the effect of recall bias can be seen by calculating the change in smoker misclassification rate needed to push the relative risk at the highest exposure level -- 80 pack-years and above -- to 1.0 (no risk) for a given recall bias rate. Again, Fontham data were used. For a recall bias rate of 0, a smoker misclassification rate is 10.1 percent would be required to cause this reduction. If the recall bias rate at the highest exposure level increases to 10 percent, the smoker misclassification rate required for the actual risk to be 1.0 drops to 9.4 percent. A third test shows that with a smoker misclassification rate of zero, a recall bias rate of 4.5 percent in the highest exposure level will push the lower limit of the 95 percent confidence interval to below 1.0. These calculations were all done assuming an exposure misclassification rate of zero.
While a recall bias which overestimates ETS exposure in the cases reduces the upper level relative risk, the same type of recall bias in controls would raise
---62 Fontham, et.al., p.1754.
CRS - 40
it. With no smoker or exposure misclassification, a 10 percent recall bias rate in the controls would cause the relative risk value to increase from 1.87 to 2.08. Finally, if the recall bias in the cases is in the other direction -- i.e., non-smokers underestimate their ETS exposure, the effect is to raise the relative risk. A 10 percent recall bias in the highest level of exposed cases would increase the relative risk from 1.87 to 2.06. Of course, a similar type of recall bias in the controls would act to lower the relative risk.
Discussion
The calculations presented above are just a sample of the very large number of misclassification rate combinations possible in these ETS studies. It seems clear from those results, however, that possible combinations of small rates -below 10 percent -- could drive ETS relative risks in the highest exposure groups to values no longer distinct from 1.0, even in a study that produces relatively high risks. While these results were obtained from the Fontham study, similar results are likely from the Brownson study.63 Even smaller values of these rates -- below 3 percent -- could be combined to reduce the lower bounds of the 95 percent confidence intervals well below 1.0 for these studies. On the other hand, it appears possible to construct combinations of relatively small misclassification rates -- again less than 10 percent -- which would increase the measured relative risk. The major problem with assessing the likelihood of any of these paths is the absence of data. While there exist some spotty data on smoker misclassification, there is very little information to provide guidance about values for the other two rates -- exposure misclassification and recall bias. The rest of this discussion focuses on each of the three error rates.
Smoker Misclassification - Discussion.
Few studies have been done to measure smoker misclassification rates results to date. EPA used a rate of 1.09 percent for current smokers which was determined by measuring cotinine levels in self-reported female non-smokers. There has been criticism of its choice of that value.64 More recent unpublished results by Roger Jenkins of the Oak Ridge National Laboratory indicate that the rate may range from 2.5 to 4.6 percent depending on how one classifies former or current smokers according to chemical markers? Given the potential influence of former smokers (not accounted for in this illustration), whose cancer risk could be higher than that assumed by the EPA, and the sampling variability of misclassified smokers in different samples, smoker misclassification
---63 Brownson, et.al., p.1528. A complete set of parameters necessary for carrying out these calculations is lacking for the Brownson, et.al., study. It is unlikely, however, that those parameters will differ sufficiently from the Fontham, et.al., case to change this conclusion.
---64 Dr. Maxwell W. Layard, personal communication.
---65 Roger A. Jenkins, Addendum to comments on Proposed Rulemaking Occupational Safety and Health Administration 29CFR parts 1910, 1915, 1926, and 1928 Indoor Air Quality; Proposed Rule, Oak Ridge National Laboratory, December 22, 1994, p.30.
CRS - 41
could explain all the measured risk even at high exposure levels even for studies such as Fontham and Brownson.
A major question about smoker misclassification is the degree to which an investigator would be able to find out whether a case -- or control -- participant is actually a non-smoker. The probability that the truth could be determined seems good but not certain. It is difficult to believe that the medical records of a smoker who developed lung cancer would not indicate that person's smoking status. The records may not have been complete or accurate in all cases, however, and, for various reasons, the records were not always reviewed. The Fontham study, for example, made a substantial effort to control for this factor. For example, it used cotinine screening to test participants and eliminate them from the study if the concentrations exceed a pre-determined threshold. It also used extensive follow-up questionaires and physician interviews to check on the smoking status of the case and control members. The cotinine screening, of course, will only determine current smokers. Smokers who quit upon developing lung cancer and then denied that they ever smoked when answering the questionnaire would not be discovered by this screening. Also, the follow-up questionaires and interviews are still subject to incomplete or false information.
There is also an issue as to the incidence of former smokers reporting themselves as non smokers. The only studies that exist currently are those which identify the rate based on discordant answers -- instances where different answers were given on different questionnaires, and these data are limited. That evidence, although skimpy and mixed, led to a misclassification rate of 11.7 percent. There is no information on individuals who consistently misrepresent their former smoking status.66 Although this misclassification rate was high relative to the current smoker rate, it did not loom very large in the EPA adjustment because they also assigned a very low cancer risk rate to former smoking, under the assumption that these were long term ex-smokers. But, if the cancer risk rate is larger (because these individuals quit in the past few years) or if there is more misclassification, these effects could be much larger.
Another issue is whether the smoker misclassification rate could be differential, i.e., higher for either cases or controls. In the non-differential situation, the misclassification rates for cases and controls are equal. Because the relative risk for lung cancer from smoking is so much greater than any of the estimates from ETS, however, the effective rate for cases would be much greater than for controls because the former is weighted by the smoker lung cancer risk value. If the misclassification rate for controls was greater than that for cases, part or all of this weighting would be offset depending on the size of the differential. The difference would have to be substantial, however, to offset the downward correction. If the rate was higher for the cases, the necessary rate to reduce the measured relative risk to 1.0 would drop. Little data exist about non-differential misclassification. In the Fontham study, cotinine
---66 A possible reason for such misrepresentation relates to life insurance policies which differentiate premiums. In some instances such policy can become null regardless of what the individual died from if he lied on his application.
CRS - 42
measurements to eliminate possible smokers did find a higher rate among the controls than the cases.67 Because of the limitations of cotinine as a means of determining smoker misclassification, this result may not be indicative of any differential smoker misclassification more generally. For instance, if lung cancer victims quit smoking upon diagnosis and denied ever smoking, they could be misclassified as non-smokers but would not be discovered by cotinine methods.
The test to determine the number of case group members that would need to be misclassified from current to never smokers in order for the calculated risk value to lose its statistical significance in the above illustration, shows much lower misclassification rates. For the Fontham data, a non-differential rate of less than 3 percent would cause the measured risk value at the highest exposure levels -o 80 pack-years or more -- to lose its statistical significance. The implication of this result is that misclassification rates well within those measured could render all measured ETS risk values no longer statistically significant. Whether that is sufficient to conclude there is no lung cancer risk from ETS depends on how one interprets statistical significance as discussed earlier in this report.
Exposure Misclassification - Discussion.
The results of the simulation show that exposure misclassification has little effect on the measured relative risk. For misclassification rates up to 20 percent, the actual risk value would increase by 4 percent or less from the measured value in all examples considered. One source of exposure misclassification is those cases and controls whose spouse smoked, but did not do so in the presence of the non-smoking spouse. This situation is likely to be confined to the lowest exposure categories since it would seem very difficult to avoid exposure in those cases where the spouse smoked heavily. Accounting for this possibility in the model shows less than a percent change in the corrected relative risks. It is also interesting to note that the inclusion of exposure misclassification actually reduces the smoker misclassification rates needed to drive the highest exposure relative risk to 1.0 or to reduce the lower bound of the 95 percent confidence interval below 1.0, although the changes are small. Since correcting for exposure misclassification removes cases and controls from the exposed category, apparently a smaller percentage of misclassified smokers would be needed in that category to account for the measured risk.
Actual measurements of exposure misclassification appear to be beyond current study techniques because of the detailed recall which would be required by the cases and controls. Measurements of non-smoker's exposure to ETS using monitors has been attempted (see section on exposure), but these measurements are only feasible over short periods, and no attempt has been made so far to look at the variation in actual exposure for similar levels of spousal smoking. In any event, the corrections are quite small and do little to account for the uncertainties associated with ETS and lung cancer risk.
---67 Fontham, et.al.,p.1757.
CRS - 43Recall Bias -- Discussion
Simulations of recall bias show substantial rates would be necessary, by themselves, in order to reduce the highest measured relative risk values to those essentially no different than 1.0 for the Fontham study. The direction of the simulated bias is for those at the highest exposures to overestimate their exposure by enough to put them into a lower exposure level. The same effects could occur if controls underestimated their exposure on average. Such a differential might simply occur because cases are more focused on exposure due to their disease status and provide more accurate answers, while casual answers to exposure questions by controls tend to recollect less exposure than actually occurred. Much lower rates of recall error, of course, would be needed to render relative risk values no longer statistically significant.
While there are no direct data available about recall bias rates, there are data which might be used to evaluate the likelihood of the recall bias. One study examined reliability of responses to questionnaires about exposure to ETS by repeating the questionnaire six months after first given.68 The authors found that reliability was quite high in terms of whether the participants were exposed, but much lower about the level of that exposure. The results indicated that second interviews reported lower duration of exposure and that reliability of reporting duration was poor. A second study done on smokers compared recall of their smoking habits with inf