SGR 2004 – Inside Cover

[Surgeon General’s Smoking Report,  health reports Continued from Part  92 ]

2004 The Health Consequences of Smoking: A Report of the Surgeon General.  Office of the Surgeon General (US); Office on Smoking and Health (US). Atlanta (GA): Centers for Disease Control and Prevention (US);  Forty years of reports. Cancers of stomach, uterine cervix, pancreas, kidney, acute myeloid leukemia, pneumonia, aneurysm, cataract, periodontitis. Conclusion “smoking generally diminishes the  health of smokers.”

https://www.ncbi.nlm.nih.gov/books/NBK44698/

[Part 93 The Health Consequences of Smoking: A Report of the Surgeon General. 2004. Introduction, causal inference. Smoking is the single greatest cause of avoidable morbidity and mortality in the United States. ]

The Health Consequences of Smoking: A Report of the Surgeon General.

Office of the Surgeon General (US); Office on Smoking and Health (US).

Atlanta (GA): Centers for Disease Control and Prevention (US); 2004.

• Contents

Top of Form

1 Introduction and Approach to Causal Inference

Introduction

This report of the Surgeon General on the health effects of smoking returns to the topic of active smoking and disease, the focus of the first Surgeon General’s report published in 1964 (U.S. Department of Health, Education, and Welfare [USDHEW] 1964). The first report established a model of comprehensive evidence evaluation for the 27 reports that have followed: for those on the adverse health effects of smoking, the evidence has been evaluated using guidelines for assessing causality of smoking with disease. Using this model, every report on health has found that smoking causes many diseases and other adverse effects. Repeatedly, the reports have concluded that smoking is the single greatest cause of avoidable morbidity and mortality in the United States.

Of the Surgeon General’s reports published since 1964, only a few have comprehensively documented and updated the evidence on active smoking and disease. The 1979 report (USDHEW 1979) provided a broad array of information, and the 1990 report on smoking cessation (U.S. Department of Health and Human Services [USDHHS] 1990) also investigated major diseases caused by smoking. Other volumes published during the 1980s focused on specific groups of diseases caused by smoking (USDHHS 1982, 1983, 1984), and the 2001 report was devoted to women and smoking (USDHHS 2001). Because there has not been a recent systematic review of the full sweep of the evidence, the topic of active smoking and health was considered an appropriate focus for this latest report. Researchers have continued to identify new adverse effects of active smoking in their ongoing efforts to investigate the health effects of smoking. Lengthy follow-ups are now available for thousands of participants in long-term cohort (follow-up) studies (National Cancer Institute [NCI] 1997).

This report also updates the methodology for evaluating evidence that the 1964 report initiated. Although that model has proved to be effective, this report establishes a uniformity of language concerning causality of associations so as to bring greater specificity to the findings of the report. The following section of this chapter describes the approach and its rationale. Beginning with this report, conclusions concerning causality of association will be placed into one of four categories with regard to strength of the evidence: (1) sufficient to infer a causal relationship, (2) suggestive but not sufficient to infer a causal relationship, (3) inadequate to infer the presence or absence of a causal relationship, or (4) suggestive of no causal relationship.

This approach separates the classification of the evidence concerning causality from the implications of that determination. In particular, the magnitude of the effect in the population, the attributable risk, is considered under “implications” of the causal determination. For example, there might be sufficient evidence to classify smoking as a cause of two diseases but the number of attributable cases would depend on the frequency of the disease in the population and the effects of other causal factors.

This report covers active smoking only. Passive smoking was the focus of the 1986 Surgeon General’s report and subsequent reports by other entities (USDHHS 1986; U.S. Environmental Protection Agency [EPA] 1992; California EPA 1997; International Agency for Research on Cancer [IARC] 2002). The health effects of pipes and cigars, also not within the scope of this report, are covered in another report (NCI 1998).

In preparing this report, the literature review approach was necessarily selective. For conditions for which a causal conclusion had been previously reached, there was no attempt to cover all relevant literature, but rather to review the conclusions from previous Surgeon General’s reports and focus on important new studies for that topic. The enormous scope of the evidence precludes such detailed reviews. For conditions for which a causal conclusion had not been previously reached, a comprehensive search strategy was developed. Search strategies included reviewing previous Surgeon General’s reports on smoking, publications originating from the largest observational studies, and reference lists from important publications; consulting with content experts; and conducting focused literature searches on specific topics. For this report, studies through 2000 were reviewed.

In addition, conclusions from prior reports concerning smoking as a cause of a particular disease have been updated and are presented in this new format based on the evidence evaluated in this report (Table 1.1). Remarkably, this report identifies a substantial number of diseases found to be caused by smoking that were not previously causally associated with smoking: cancers of the stomach, uterine cervix, pancreas, and kidney; acute myeloid leukemia; pneumonia; abdominal aortic aneurysm; cataract; and periodontitis. The report also concludes that smoking generally diminishes the health of smokers.

Table 1.1

Diseases and other adverse health effects for which smoking is identified as a cause in the current Surgeon General’s report. 

Despite the many prior reports on the topic and the high level of public knowledge in the United States of the adverse effects of smoking in general, tobacco use remains the leading preventable cause of disease and death in the United States, causing approximately 440,000 deaths each year and costing approximately $157 billion in annual health-related economic losses (see Chapter 7, “The Disease Impact of Cigarette Smoking and Benefits of Reducing Smoking”). Nationally, smoking results in more than 5.6 million years of potential life lost each year. Although the rates of smoking continue to decline, an estimated 46.2 million adults in the United States still smoked cigarettes in 2001 (Centers for Disease Control and Prevention [CDC] 2003). In 2000, 70 percent of those who smoked wanted to quit (CDC 2002a). An increasingly disturbing picture of widespread organ damage in active smokers is emerging, likely reflecting the systemic distribution of tobacco smoke components and their high level of toxicity. Thus, active smokers are at higher risk for cataract, cancer of the cervix, pneumonia, and reduced health status generally.

This new information should be an impetus for even more vigorous programs to reduce and prevent smoking. Smokers need to be aware that smoking carries far greater risks than the most widely known hazards. Health care providers should also use the new evidence to counsel their patients. For example, ophthalmologists may want to warn patients about the increased risk of cataract in smokers, and geriatricians should counsel their patients who smoke, even the oldest, to quit. This report shows that smokers who quit can lower their risk for smoking-caused diseases and improve their health status generally. Those who never start can avoid the predictable burden of disease and lost life expectancy that results from a lifetime of smoking.

Preparation of the Report

This report of the Surgeon General was prepared by the Office on Smoking and Health, National Center for Chronic Disease Prevention and Health Promotion, CDC, USDHHS. Initial chapters were written by 19 experts who were selected because of their expertise and familiarity with the topics covered in this report. Their various contributions were summarized into six major chapters that were then reviewed by more than 60 peer reviewers. The entire manuscript was then sent to more than 20 scientists and experts, who reviewed it for its scientific integrity. After each review cycle was completed, the drafts were revised by the editors on the basis of the experts’ comments. Subsequently, the report was reviewed by various institutes and agencies within USDHHS.

Publication lags, even short ones, prevent an up-to-the-minute inclusion of all recently published articles and data. Therefore, by the time the public reads this report, there may be additional published studies or data. To provide published information as current as possible, this report includes an appendix of more recent studies that represent major additions to the literature.

This report is also accompanied by a companion database of key evidence that is accessible through the Internet (see http://www.cdc.gov/tobacco). The database includes a uniform description of the studies and results on the risks of smoking that were presented in a format compatible with abstraction into standardized tables. Readers of the report may access these data for additional analyses, tables, or figures. The Office on Smoking and Health at CDCintends to maintain this database and will periodically update its contents as new reports are published.

Organization of the Report

This report covers major groups of the many diseases associated with smoking: cancers, cardiovascular diseases, respiratory diseases, reproductive effects, and other adverse health consequences. This chapter (Chapter 1) includes a discussion of the concept of causation and introduces new concepts of causality that are used throughout this report. Chapter 2 discusses each of the main sites of cancer and their relationship to smoking. Cardiovascular diseases, including atherosclerosis, coronary heart disease, stroke, and abdominal aortic aneurysm are the focus of Chapter 3, which begins with an extensive review of newer findings on the mechanisms by which smoking causes this group of very common diseases. Chapter 4 includes both acute respiratory diseases associated with smoking and the chronic respiratory diseases long known to be caused by smoking, including accelerated loss of lung function with aging. The full scope of adverse reproductive effects caused by smoking in both men and women is covered in Chapter 5. Chapter 6 discusses other specific effects of smoking on the eyes, the bones, and oral health, along with evidence on more general adverse effects related to health status overall. Chapter 7 updates prior estimates of the burden of diseases caused by smoking. Finally, Chapter 8 discusses “A Vision for the Future” outlining broad strategies and courses of action for tobacco control in the future.

Smoking: Issues in Statistical and Causal Inference

The U.S. Surgeon General’s reports on the health effects of smoking have long had a central role in the translation of scientific evidence into policies for tobacco control. A critical and essential aspect of this role has been the judgment that smoking is a cause of specific diseases or health conditions. The statement that an exposure “causes” a disease in humans represents a serious claim, but one that carries with it the possibility of prevention. Causal determinations may also carry substantial economic implications for society and for those who might be held responsible for the exposure or for achieving its prevention. The qualitative judgment that an exposure causes a particular disease signifies that in the absence of exposure some fraction of cases or deaths would not occur or would occur at a later age (USDHEW 1964; Rothman and Greenland 1998). Given these implications, the grounds for making the causal designation must be well founded and clear.

The need for guidelines for causal determination was recognized by the committee that authored the first Surgeon General’s report, and by the scientists whose work served as the foundation for that report (Cornfield et al. 1959). The difficulty of attempting to both adjudicate causal relationships and choose the language to describe them was apparent then (USDHEW 1964). In a section titled “Criteria for Judgment” in the 1964 report, the committee wrote that after “vigorous discussions,” they could neither precisely define nor replace the word “cause,” a reflection of the same problem that philosophers have confronted over the centuries. The main approach is summarized below:

When a relationship or an association between smoking…and some condition in the host was noted, the significance of the association was assessed.

The characterization of the assessment called for a specific term…. The word cause is the one in general usage in connection with matters considered in this study, and it is capable of conveying the notion of a significant, effectual relationship between an agent and an associated disorder or disease in the host.

No member was so naive as to insist upon mono-etiology in pathological processes or in vital phenomena. All were thoroughly aware…that the end results are the net effect of many actions and counteractions.

Granted that these complexities were recognized, it is to be noted clearly that the Committee’s considered decision to use the words “a cause,” or “a major cause,” or “a significant cause,” or “a causal association” in certain conclusions about smoking and health affirms their conviction (USDHEW 1964, p. 21).

The key descriptors in the above passage include “effectual,” “significant,” and “major.” Reading these phrases now, it is unclear whether the committee intended to describe the underlying causal relationship itself, the size of an estimated effect, the degree of statistical evidence for that estimated effect, the strength of the causal claim, or some combination of these elements of the evidence. The report further described the criteria for determining a causal relationship. These criteria, which were just emerging into public health, have since become widely accepted and used in epidemiology and public health: that any alleged association should demonstrate consistency, strength, specificity, temporality, and coherence. This report has served as a lasting model for the comprehensive evaluation of scientific evidence.

However, at that time strict terminology was not in place for describing the status of the evidence. Thus, in the 1964 and subsequent Surgeon General’s reports, as well as in other reports, the language used to characterize conclusions about relationships between smoking and disease varied. Table 1.2 contains examples of these variations used in every Surgeon General’s report published between 1964 and 1990. For example, for atherosclerosis outcomes there is the following sequence of terms: “likely risk factor” (USDHEW 1971, p. 9), “major risk factor” (USDHEW 1973, p. 23), “strong associations” (USDHEW 1974, p. 19), “major risk factor” (USDHEW 1979, p. 1–14), “major, independent risk factor” (USDHHS 1980, p. 7), “the most powerful risk factor” (USDHHS 1983, p. 8), and finally, “a cause of and the most powerful risk factor” (USDHHS 1989, p. 63). For pancreatic cancer the sequence proceeds in a similar manner: “significant association” (USDHEW 1972, p. 75), “data confirm the association” (USDHEW 1974, p. 59), “a dose-response relationship” (USDHEW 1979, p. 1–17), and in 1982 “a contributory factor” that “by no means excludes the possibility of a causal role…” (USDHHS 1982, p. 7). For some other outcomes, statements on causality were more qualified, such as “for the purposes of preventive medicine, it can be concluded that smoking is causally related to coronary heart disease…” (USDHEW 1979, p. 1–15).

Table 1.2

Variations in terminology from previous Surgeon General’s reports concerning smoking as a cause of the listed diseases. 

One would not expect that conclusive language in these earlier reports would be identical, as each committee analyzed successively larger bodies of evidence, often with different cumulative support for causal claims. But without standardized terminology, authors contributing to the reports sometimes introduced their own phrasing to convey the extent of the evidence and attendant uncertainty. The intent of this chapter is to establish a more structured framework for reporting conclusions for this report and for those that follow.

Twenty-seven Surgeon General’s reports on the health effects of smoking and related issues have been published since 1964. They contain the full range of information available on smoking and health for the purpose of evaluating the evidence. This evidence has come from studies of the composition of tobacco smoke, toxicologic investigation of smoke and of particular smoke components in experimental systems, and observational or epidemiologic studies of associations of smoking with diseases or other adverse health consequences. The observational evidence has also extended to mortality statistics, cancer incidence data, and disease prevalence figures, all of which capture the occurrence of diseases possibly caused by smoking. Changes in disease patterns across the twentieth century were a substantial impetus for hypotheses proposing that smoking causes disease. The epidemiologic evidence, now abundant for many diseases caused by smoking, has been given substantial weight in identifying smoking as a cause of disease. The observational data have been complemented by experimental data from the laboratory, which support the plausibility of causation and give an ever-deepening understanding of the mechanisms by which tobacco smoking causes disease.

Since the earliest reports of the Surgeon General, evidence has become available on the benefits of smoking cessation, primarily from observations of smokers who have stopped and from observations of patterns of disease occurrence over time.

Across these 27 reports the strength of evidence has mounted, new conclusions have been added, and older conclusions have been strengthened and expanded. Since the 1964 report, there has never been any reason to reverse earlier conclusions of causality.

This chapter returns to the topic of causality, including causal inference and terminology for characterizing the strength of evidence for causality. This topic has not been addressed comprehensively since the 1964 report. In view of the continued importance and public health relevance of causal conclusions, updating the 1964 report was considered necessary.

Terminology of Conclusions and Causal Claims

The first step in introducing this revised approach is to outline the language that will be used for summary conclusions regarding causality, which follows hierarchical language used by Institute of Medicine committees (Institute of Medicine 1999) to couch causal conclusions, and by IARC to classify carcinogenic substances (IARC 1986). These entities use a four-level hierarchy for classifying the strength of causal inferences based on available evidence as follows:

1. Evidence is sufficient to infer a causal relationship.
2. Evidence is suggestive but not sufficient to infer a causal relationship.
3. Evidence is inadequate to infer the presence or absence of a causal relationship (which encompasses evidence that is sparse, of poor quality, or conflicting).
4. Evidence is suggestive of no causal relationship.

For this report, the summary conclusions regarding causality are expressed in this four-level classification. Use of these classifications should not constrain the process of causal inference, but rather bring consistency across chapters and reports, and greater clarity as to what the final conclusions are actually saying. As shown in Table 1.1, without a uniform classification the precise nature of the final judgment may not always be obvious, particularly when the judgment is that the evidence falls below the “sufficient” category. Experience has shown that the “suggestive” category is often an uncomfortable one for scientists, since scientific culture is such that any evidence that falls short of causal proof is typically deemed inadequate to make a causal determination. However, it is very useful to distinguish between evidence that is truly inadequate versus that which just falls short of sufficiency.

There is no category beyond “suggestive of no causal relationship” as it is extraordinarily difficult to prove the complete absence of a causal association. At best, “negative” evidence is suggestive, either strongly or weakly. In instances where this category is used, the strength of evidence for no relationship will be indicated in the body of the text.

In this new framework, conclusions regarding causality will be followed by a section on implications. This section will separate the issue of causal inference from recommendations for research, policies, or other actions that might arise from the causal conclusions. This section will assume a public health perspective, focusing on the population consequences of using or not using tobacco and also a scientific perspective, proposing further research directions. The proportion of cases in the population as a result of exposure (the population attributable risk), along with the total prevalence and seriousness of a disease, are more relevant for deciding on actions than the relative risk estimates typically used for etiologic determinations. In past reports, the failure to sharply separate issues of inference from policy issues resulted in inferential statements that were sometimes qualified with terms for action. For example, based on the evidence available in 1964, the first Surgeon General’s report on smoking and health contained the following statement about the relationship between cardiovascular diseases and smoking:

It is established that male cigarette smokers have a higher death rate from coronary artery disease than non-smoking males. Although the causative role of cigarette smoking in deaths from coronary disease is not proven, the Committee considers it more prudent from the public health viewpoint to assume that the established association has causative meaning, than to suspend judgment until no uncertainty remains (USDHEW 1964, p. 32).

Using this framework, this conclusion would now be expressed differently, probably placing it in the “suggestive” category and making it clear that although it falls short of proving causation, this evidence still makes causation more likely than not. The original statement makes it clear that the 1964 committee judged that the evidence fell short of proving causality but was sufficient to justify public health action. In this report, the rationale and recommendations for action will be placed in the implications section, separate from the causal conclusions. This separation of inferential from action-related statements clarifies the degree to which policy recommendations are driven by the strength of the evidence and by the public health consequences acting to reduce exposure. In addition, this separation appropriately reflects the differences between the processes and goals of causal inference and decision making.

Implications of a Causal Conclusion

The judgment that smoking causes a particular disease has immediate implications for prevention of the disease. Having reached a causal conclusion, one of the immediate and appropriate next steps is to estimate the burden of disease that might be avoided through prevention and cessation of smoking. This estimation is made with the population attributable risk, a measure first proposed by Levin (1953) to calculate the proportion of lung cancer caused by smoking. Levin’s attributable risk is central to the estimates made by the Smoking-Attributable Mortality, Morbidity, and Economic Costs (SAMMEC) application developed by CDC (2002b).

The burden of avoidable disease in a population depends on the strength of smoking as a factor causing the disease and the prevalence of smoking in the population of interest. The attributable risk could vary across populations that have different patterns of smoking or in the same population over time as smoking changes. The attributable risk may also be influenced by the population’s exposures to other causes of this disease of interest and by whether those other causes modify the effect of smoking.

Because the attributable risk is population dependent, the report separates the causal conclusion from this quantitative assessment of its implications. This assessment is placed in the separate section, “Implications,” immediately following the statement of conclusions.

There are also implications of not reaching a causal conclusion. The attributable risk can still be calculated to estimate how much disease is potentially avoidable, given a causal determination. Additionally, the evidence review may indicate needed areas of research to address remaining gaps and uncertainties that have precluded a causal designation.

Judgment in Causal Inference

A causal conclusion conveys the inference that changing a given factor will actually reduce a population’s burden of disease, either by reducing the overall number of cases or by making disease occur later than it would have (Robins and Greenland 1989). Without the mantle of “causal,” the identification of a “risk factor” does not necessarily carry with it the certainty of disease prevention or delayed onset following exposure reduction or removal. As noted in the 1964 Surgeon General’s report, the characteristics of evidence that merit calling an association causal involve extra-statistical judgments. Because the claim is so central to disease prevention, it is important to review some of the complexities inherent in this concept and the epidemiologic criteria that have been proposed to decide whether the causal designation should be made.

In this report, the definition of cause is based on the notions of a “counterfactual” state, a concept with origins at least as far back as the English philosopher David Hume (1711–1776) (Steinberg 1993). In the twentieth century, this concept was further developed and applied by statisticians, philosophers, and epidemiologists (Bunge 1959; Lewis 1973; Rubin 1974; Robins 1986, 1987; Greenland 1990; Splawa-Neyman 1990; Greenland et al. 1999; Pearl 2000; Parascandola and Weed 2001). A counterfactual definition holds that something is a cause of a given outcome if, when the same person is observed with and without a purported cause and without changing any other characteristic, a different outcome would be observed. For example, the counterfactual state for a smoker is the same individual never having smoked. The word “counterfactual” comes from the fact that no person can actually be observed under exactly the same conditions twice. For example, it is not possible to actually observe the same human being under identical conditions (including being the same age) except for smoking status. The situation that cannot be observed is called the counterfactual state; literally, counter to the observed facts. The unobservability of the counterfactual state is what makes causal relationships based on observational data subject to uncertainty and questioning.

Properly designed studies provide a scientific basis for inferring what the outcome of the counter-factual state would be, and permit related uncertainty to be properly quantified. In a laboratory, scientists are able to predict, fairly confidently, the outcome in this counterfactual state by repeating an experimental procedure with every important factor tightly controlled, varying only the factor of interest. But in observational studies of humans, scientists must try to infer what the outcome would be in a counterfactual state by studying another group of persons who, at least on average, are substantively different in only one relevant variable, the exposure under study. The outcome of this second group is used to represent what would have occurred in the original group if it had been observed with a different exposure, as in its counter-factual state (Greenland 1990). In the case of smoking and disease, this comparison is between disease risk in smokers and nonsmokers. Because experiments cannot be ethically done that randomize people to smoke or not to smoke, most evidence on smoking and disease is observational.

In the absence of a randomized assignment of exposure, two groups may differ on average in more factors than just the variable of interest. If these other factors affect outcome, then their effects can combine with the causal effect of the factor of interest, biasing the measured effect of that factor. These ancillary causes are called confounders. An example of a confounding factor might be a characteristic associated both with taking a medication and cardiovascular risk, which appears to be the current situation with hormone replacement therapy (HRT) in women. The observational studies showed a clearer cardiovascular benefit from HRT than did a large randomized trial, suggesting that there may be some cardioprotective characteristics or behaviors of women who voluntarily take HRT that are at least partly responsible for the apparent benefit of HRT in the observational studies (Hulley et al. 1998; Blumenthal et al. 2000). In fact, the results of the Women’s Health Initiative Trial of HRT showed increased risk for cardiovascular disease incidence in women randomized to HRT (Pradhan et al. 2002). Confounding by cardioprotective characteristics associated with taking HRT may have obscured this unanticipated consequence of HRT in the observational studies.

If confounders are recognized and their effects measured, these effects can often be statistically minimized or removed by the analysis of a study. However, if a confounder is poorly measured, or its effects poorly characterized, then its effects cannot be controlled for in the analysis phase of a study, resulting in a causal effect that is distorted or confounded by the unwanted factor. The most extreme version of this phenomenon occurs with unmeasured confounding, causal factors that are not measured at all and whose effects are therefore not controllable, which can result in biased estimates and underestimates of uncertainty, because standard analyses implicitly assume an absence of confounding from all unmeasured factors.

One solution to this problem of unmeasured or poorly controlled confounding is to randomize the factor of interest between different groups of people. This solution is obviously not applicable to harmful agents or behaviors such as smoking cigarettes (although randomization to cessation is possible because a benefit is anticipated), but understanding the role of randomization can deepen insights into the interpretation of nonrandomized designs used to study smoking effects. Randomization makes a proposed causal factor independent of potentially confounding factors, and provides a known probability distribution for the potential outcomes in each group under a given mathematic hypothesis (i.e., null) (Greenland 1990). It does not mean that inference from an individual randomized study is free of unmeasured confounding (it is free of unmeasured confounding only on average), but it does mean that measures of uncertainty about causal estimates from randomized studies have an experimental foundation. In the absence of randomization, uncertainty about causal effects depends in part on the confidence that all substantive confounding has been eliminated or controlled either by the study design or by the analysis. Such confidence is ultimately based on scientific judgment.

One way to reduce the uncertainty that occurs with both randomized and observational designs is to repeat the studies. Similar results in a series of randomized studies make it increasingly unlikely that unmeasured confounding is accounting for the findings, since the process of randomization makes the mathematic probability of such confounding progressively smaller as the total sample size or number of studies increases. In observational studies, however, increasing the number of studies may reduce the random component of uncertainty, but not necessarily the systematic component attributable to confounding. Without randomization, there is no mathematic basis to assume that imbalance in unknown confounders will decrease with an increase in the number of studies. For example, many observational studies of HRT use in women have shown a strong cardioprotective effect. If unmeasured cardioprotective characteristics are consistently more common among women who use HRT, then having multiple studies will not necessarily reduce the effect of unmeasured confounding. However, if observational studies are repeated in different settings, with different subjects, different eligibility criteria, and/or different exposure opportunities (e.g., therapeutic HRT use after hysterectomy), each of which might eliminate another source of confounding from consideration, then confidence that unmeasured confounders are not producing the findings is increased. How many studies need to be done, how diverse they need to be, and how relevant they are to the question at hand are matters of scientific judgment.

Confidence that unmeasured confounding is not producing the observed results is further increased by understanding the biologic process by which the exposure might affect the outcome. This understanding allows better identification and measurement of relevant confounders, making it more unlikely that what is unmeasured is of concern. It can also serve as the basis for a judgment that the observed difference could be produced only by an implausible degree of confounder imbalance between exposed and unexposed groups. Thus, causal conclusions from observational studies typically require more and stronger biologic evidence to support plausibility and the absence of confounding than is required for causal inferences based on randomized studies.

Making causal inferences from observational data can be a challenging task, requiring expert judgment as to the likely sources and magnitude of confounding, together with judgments about how well the existing constellation of study designs, results, and analyses addresses this potential threat to inferential validity. To aid this judgment, criteria for the determination of a cause have been proposed by many philosophers and scientists over the centuries. The most widely cited criteria in epidemiology and public health more generally were set forth by Sir Austin Bradford Hill in 1965 (Weed 2000). Five of the nine criteria he listed were also put forward in the 1964 Surgeon General’s report as the criteria for causal judgment: consistency, strength, specificity, temporality, and coherence of an observed association. Hill also listed biologic gradient (dose-response), plausibility, experiment (or natural experiment), and analogy. Many of these criteria have been cited in earlier epidemiologic writings (Lilienfeld 1959; Yerushalmy and Palmer 1959; Sartwell 1960), and Susser has extensively refined them by exploring their justification, merits, and interpretations (Susser 1973, 1977; Kaufman and Poole 2000).

Hill (1965) clearly stated that these criteria were not intended to serve as a checklist:

Here are then nine different viewpoints from all of which we should study association before we cry causation. What I do not believe … is that we can usefully lay down some hard-and-fast rules of evidence that must be obeyed before we accept cause and effect. None of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a sine qua non. What they can do, with greater or less strength, is to help us to make up our minds on the fundamental question—is there any other way of explaining the facts before us, is there any other answer equally, or more, likely than cause and effect? (Hill 1965, p. 299)

All of these criteria were meant to be applied to an already established statistical association; if no association has been observed, then these criteria are not relevant. Hill explained how, if a given criterion were satisfied, it strengthened a causal claim. Each of these nine criteria served one of two purposes: either as evidence against competing noncausal explanations or as evidence supporting causal ones. Noncausal explanations for associations include chance; residual or unmeasured confounding; model misspecification; selection bias; errors in measurement of exposure, confounders, or outcome; and issues regarding missing data (which can also include missing studies, e.g., publication bias). The criteria are briefly discussed below.

Consistency

This criterion refers to the persistent finding of an association between exposure and outcome in multiple studies of adequate power, and in different persons, places, circumstances, and times. Consistency can serve two purposes. The first purpose, which was discussed previously, is to make unmeasured confounding an unlikely alternative explanation for an observed association. Such confounding would have to persist across diverse populations, exposure opportunities, and measurement methods. The confounding is still possible if the exposure (in this case smoking) were very strongly tied to an alternative cause, as was claimed in the form of the “constitutional hypothesis” put forward in the early days of the smoking-disease debate (USDHEW 1964). This hypothesis held that there was a constitutional (i.e., genetic) factor that made people more likely to both smoke and develop cancer. So consistency serves mainly to rule out the hypothesis that the association is produced by an ancillary factor that differs across studies, but not one factor that is common to all or most of them (Rothman and Greenland 1998).

The second purpose of the consistency criterion is to make the hypothesis of a chance effect unlikely by increasing the statistical strength of a finding through the accumulation of a larger body of data. It does not include the qualitative strength of such studies, which Susser subsumes under his subsidiary concept of “survivability,” relating to the rigor and severity of tests of association (Susser 1991).

Strength of Association

This criterion includes two dimensions of strength: the magnitude of the association and its statistical strength. An association strong in both aspects makes the alternative explanations of chance and confounding unlikely. The larger the measured effect, the less likely that an unmeasured or poorly controlled confounder could account for it completely. Associations that have a small magnitude or a weak statistical strength are more likely to reflect chance, modest bias, or unmeasured weak confounding. However, the magnitude of association is reflective of underlying biologic processes and should be consistent with understanding the role of smoking in these processes.

Specificity

Specificity has been interpreted to mean both a single (or few) effect(s) of one cause, or no more than one possible cause for one effect. In addition to specific infectious diseases that are caused by specific infectious agents, some other examples include asbestos exposure and mesothelioma and thalidomide exposure during gestation and the resulting unusual constellation of birth defects. This criterion is rarely used as it was originally proposed, having been derived primarily from the Koch Postulates for infectious causes of disease (Evans 1993). When specificity exists, it can strengthen a causal claim, but its absence does not weaken it (Sartwell 1960). For example, most cancers are known to have multifactorial etiologies, many cancer-causing agents can cause several types of cancer, and these agents can also have noncancerous effects. Similarly, there are multiple causes of cardiovascular disease.

In considering specificity in relation to the smoking-lung cancer association, the 1964 Surgeon General’s report (USDHEW 1964) provides a rich discussion of this criterion. The committee recognized the linkage between this criterion and strength of association and offered a symmetric formulation of specificity in the relationship between exposure and disease; that is, a particular exposure always results in a particular disease and the disease always results from the exposure. The committee acknowledged that smoking does not always result in lung cancer and that lung cancer has other causes. The report notes the extremely high relative risk for lung cancer in smokers and the high attributable risk, and concludes that the association between smoking and lung cancer has “a high degree of specificity.”

Temporality

Temporality refers to the occurrence of a cause before its purported effect. Temporality is the sine qua non of causality, as a cause clearly cannot occur after its purported effect. Failure to establish temporal sequence seriously weakens a causal claim, but establishing temporal precedence is by itself not very strong evidence in favor of causality.

Coherence, Plausibility, and Analogy

Although the original definitions of these criteria were subtly different, in practice they have been treated essentially as one idea: that a proposed causal relationship not violate known scientific principles, and that it be consistent with experimentally demonstrated biologic mechanisms and other relevant data, such as ecologic patterns of disease (Rothman and Greenland 1998). In addition, if biologic understanding can be used to set aside explanations other than a causal association, it offers further support for causality. Together, these criteria can serve both to support a causal claim (by supporting the proposed mechanism) or refute it (by showing that the proposed mechanism is unlikely).

Biologic understanding, of course, is always evolving as scientific advances make possible an ever deeper exploration of disease pathogenesis. For example, in 1964 the Surgeon General’s committee found a causal association of smoking with lung cancer to be biologically plausible. Nearly 40 years later, this association remains biologically plausible, but that determination rests not only on the earlier evidence but on more recent findings that address the genetic and molecular basis of carcinogenesis.

Biologic Gradient (Dose-Response)

The finding of an increment in effect with an increase in the strength of the possible cause provides strong support in favor of a causal hypothesis. This is not just because such an observation is predicted by many cause-effect models and biologic processes, but more importantly, because it makes most noncausal explanations very unlikely. One would have to posit that some unmeasured factor was changing in the same manner as the exposure of interest if that factor, rather than the factor of interest, is to explain the gradient. Except for confounders that are very closely related to a causal factor, it is very difficult for such a pattern to be created by virtually any of the noncausal explanations for an association listed earlier. The finding of a dose-response relationship has long been a mainstay of causal arguments in smoking investigations; virtually all health outcomes causally linked to smoking have shown an increase in risk and/or severity with an increase in the lifetime smoking history, generally number of cigarettes smoked per day, duration of smoking, or a cumulative measure of consumption. This criterion is not based on any specific shape of the dose-response relationship.

Experiment

This criterion refers to situations where natural conditions might plausibly be thought to imitate conditions of a randomized experiment, producing a “natural experiment” whose results might have the force of a true experiment. An experiment is typically a situation in which a scientist controls who is exposed in a way that does not depend on any of the subject’s characteristics. Sometimes nature produces similar exposure patterns. The reduction in risk after smoking cessation serves as one such situation that approximates an experiment; an alternative noncausal explanation would have to posit that an unmeasured causal factor of that health outcome was more frequent among those who did not stop smoking than among those who did. The causal interpretation is further strengthened if risk continues to decline in former smokers with increasing length of time since quitting. Similar to the dose-response criteria, observations of risk reduction after quitting smoking have the dual effects of making most noncausal explanations unlikely, and supporting the biologic model that underlies the causal claim.

Applying the Causal Criteria

The more that an association fulfills the previous criteria, the more difficult it is to offer a more compelling alternative explanation. Which of these criteria may be more important, and whether some can be unfulfilled and still justify the causal claim, is a judgmental issue. Temporality, however, cannot be violated. When there is a still incompletely understood pathogenic mechanism, the causal claim might still be justified by very strong, direct empirical evidence of higher rates in smokers (i.e., strong, consistent associations). Less strong associations (e.g., relative risks between 1 and 2) in only a few studies, without adequate understanding of potential confounders or with weak designs, might result in a suspicion of causal linkage.

The process of applying the criteria extends beyond simply lining the evidence up against each criterion. Rather, the criteria are used to integrate multiple lines of evidence, coming from chemical and toxicologic characterizations of tobacco smoke and its components, epidemiologic approaches, and clinical investigations. Those applying the criteria weigh the totality of the evidence in a decision-making process that synthesizes and, of necessity, involves a multidisciplinary judgment.

The 1964 Surgeon General’s report still stands as one of the finest examples of the power of applying these criteria systematically and comprehensively. Starting with the criterion for consistency, the committee noted that all 29 retrospective (i.e., case-control) and 7 prospective (i.e., cohort) studies at the time reported strong smoking-lung cancer relationships. They further noted that all of the studies comparing smokers with nonsmokers showed very high relative risks for lung cancer (ranging from approximately 5 to 20). Dose-response effects were also observed in almost every study that provided the necessary data. The temporal sequence was reported to be not absolutely certain, but seemed to be very unlikely in the lung cancer-smoking direction, as cancer typically appears many years or decades after the onset of smoking. With regard to coherence of the association with known facts, the studies noted the ecologic increase in lung cancer rates with increased smoking in the population; the gender differential in lung cancer, which at the time was consistent with more smoking by men; an urban-rural difference, which air pollution could not completely explain; socioeconomic differentials in lung cancer for which smoking seemed to be the strongest explanation; and the localization of cancer within the respiratory tract in relation to the type of smoking. The studies also cited the known reduction in risk among former smokers, with greater risk reductions correlated with more time spent not smoking. These observations, in combination with histopathologic evidence, basic biologic observations, and an in-depth discussion of each competing nonsmoking-related explanation (e.g., occupation, constitutional hypothesis, infections, and environmental factors such as pollution), produced a case for causation that was essentially irrefutable.

Statistical Testing and Causal Inference

Hill made a point of commenting on the value, or lack thereof, of statistical testing in the determination of cause: “No formal tests of significance can answer those [causal] questions. Such tests can, and should, remind us of the effects the play of chance can create, and they will instruct us in the likely magnitude of those effects. Beyond that, they contribute nothing to the ‘proof’ of our hypothesis” (Hill 1965, p. 299).

Hill’s warning was in some ways prescient, as the reliance on statistically significant testing as a substitute for judgment in causal inference remains today (Savitz et al. 1994; Holman et al. 2001; Poole 2001). To understand the basis for this warning, it is critical to recognize the difference between inductive inferences about the truth of underlying hypotheses, and deductive statistical calculations that are relevant to those inferences but that are not inductive statements themselves. The latter include p values, confidence intervals, and hypothesis tests (Greenland 1998; Goodman 1999). The dominant approach to statistical inference today, which employs those statistical measures, obscures this important distinction between deductive and inductive inferences (Royall 1997), and has produced the mistaken view that inferences flow directly and inevitably from data. There is no mathematic formula that can transform data into a probabilistic statement about the truth of an association without introducing some formal quantification of external knowledge, such as in Bayesian approaches to inference (Goodman 1993; Howson and Urbach 1993). Significance testing and the complementary estimation of confidence intervals remain useful for characterizing the role of chance in producing the association in hand.

There are many kinds of statements that appear to be, but are not, formal inferences about a hypothesis. For example, consider the statement “the frequency of cirrhosis in smokers is statistically significantly greater than the frequency in nonsmokers.” This statement is based on a deductive mathematic calculation that assumes the truth of the null hypothesis of no association. It is not a knowledge claim of an inductive statement about the likely truth of the cirrhosis-smoking relationship, although it may serve as a foundation for that claim. An inductive inference would be a statement based on this and other evidence, that smokers are likely to have a higher risk of cirrhosis than nonsmokers. Determining whether or not this elevated risk was causally related to smoking would represent a causal judgment.

In this report, language is used to make as clear as possible what kind of statement is being made, and to avoid certain kinds of ambiguities that are widespread in the scientific literature. Certain words imply causal conclusions by suggesting an active effect of smoking on disease (Petitti 1991). For example, the statement that smoking “is associated” with disease could mean that disease frequency is higher in smokers, that it is statistically significantly higher, or that an inferential conclusion about the association has been reached. Depending on the context, words like “effect” or “contributor” can fall into that category, as do statements like smoking “increases risk.” Such language often appears to be a causal conclusion, albeit without consideration of all of the causally relevant evidence.

Another type of claim is that smoking is a “risk factor” for disease, or that the observed association is “real” or “true.” This claim represents an inference, a conclusion that the risk of disease differs in at least an actuarial sense, at different levels; that is, more events overall and at younger ages can be expected in smokers. Such a statistical finding does not yet have the status of a causal claim. In addition, this phrasing does not make it clear whether the factor has predictive value over and above all other known risk and causal factors, which would be indicated by the words “independent risk factor” or “independent contributor.”

Statements like these will be avoided, or at least qualified, to make clear whether they are statements about the data, about statistical significance, or are actual statistical or causal inferences. All causal claims in this report will be clearly identified using the word “cause,” and classified according to the previously outlined criteria.

Conclusions

Inferences, whether about causality or statistical associations, are always uncertain to a degree. The goal of this report, as in all previous ones, is to explain and communicate scientific judgments as to whether observed associations between smoking and disease are likely to be causal, based on the totality of scientific evidence. This report will employ an ordinal scale and standardized language to express the strength of the evidence bearing on causality. This approach will help not only to clarify what the assessment is, but will make it possible for subsequent groups to measure progress or calibrate standards by comparing their summary judgments with those expressed here. This structure also encourages the articulation of the sources of uncertainty in the evidence, which hopefully will stimulate necessary research.

In addition, causal conclusions are separated from public health recommendations. This decoupling is necessary, as decision making in the face of uncertainty involves different issues than those that pertain to the uncertainty itself, and past reports have sometimes combined the two perspectives.

Just as this series of reports has documented progress in understanding the connections between smoking and disease, this report represents progress in how that understanding is assessed and communicated. A debt is owed to the many scientists who have both performed and synthesized smoking-related research in the past. The framework used in this report should assist researchers, the readers, and those who must perform this task in the future to accurately represent what is and what is not known about the impact of smoking on human health.

Major Conclusions

Forty years after the first Surgeon General’s report in 1964, the list of diseases and other adverse effects caused by smoking continues to expand. Epidemiologic studies are providing a comprehensive assessment of the risks faced by smokers who continue to smoke across their life spans. Laboratory research now reveals how smoking causes disease at the molecular and cellular levels. Fortunately for former smokers, studies show that the substantial risks of smoking can be reduced by successfully quitting at any age. The evidence reviewed in this and prior reports of the Surgeon General leads to the following major conclusions:

1. Smoking harms nearly every organ of the body, causing many diseases and reducing the health of smokers in general.
2. Quitting smoking has immediate as well as long-term benefits, reducing risks for diseases caused by smoking and improving health in general.
3. Smoking cigarettes with lower machine-measured yields of tar and nicotine provides no clear benefit to health.
4. The list of diseases caused by smoking has been expanded to include abdominal aortic aneurysm, acute myeloid leukemia, cataract, cervical cancer, kidney cancer, pancreatic cancer, pneumonia, periodontitis, and stomach cancer.

References

1. Blumenthal RS, Zacur HA, Reis SE, Post WS. Beyond the null hypothesis—do the HERS results disprove the estrogen/coronary heart disease hypothesis? American Journal of Cardiology. 2000;85(8):1015–7. [PubMed]
2. Bunge MA.  Causality; The Place of the Causal Principle in Modern Science. Cambridge (MA): Harvard University Press; 1959. 
3. California Environmental Protection Agency.  Health Effects of Exposure to Environmental Tobacco Smoke. Sacramento (CA): California Environmental Protection Agency, Office of Environmental Health Hazard Assessment, Reproductive and Cancer Hazard Assessment Section and Air Toxicology and Epidemiology Section; 1997. 
4. Centers for Disease Control and Prevention. Cigarette smoking among adults—United States, 2000. Morbidity and Mortality Weekly Report. 2002a;51(29):642–5. [PubMed]
5. Centers for Disease Control and Prevention. Smoking Attributable Mortality, Morbidity, and Economic Costs (SAMMEC). Atlanta: Centers for Disease Control and Prevention; 2002b. [accessed: October 14, 2002].  http://apps​.nccd.cdc​.gov/sammec/intro.asp.
6. Centers for Disease Control and Prevention. Cigarette smoking among adults—United States, 2001. Morbidity and Mortality Weekly Report. 2003;52(40):953–6. [PubMed]
7. Cornfield J, Haenszel W, Hammond EC, Lilienfeld AM, Shimkin MB, Wynder EL. Smoking and lung cancer: recent evidence and a discussion of some questions. Journal of the National Cancer Institute. 1959;22(1):173–203. [PubMed]
8. Evans AS.  Causation and Disease: A Chronological Journey. New York: Plenum, Medical Book Company; 1993. 
9. Goodman SN. p values, hypothesis tests, and likelihood: implications for epidemiology of a neglected historical debate. American Journal of Epidemiology. 1993;137(5):485–96. [PubMed]
10. Goodman SN. Towards evidence-based medical statistics. 1: the P value fallacy. Annals of Internal Medicine. 1999;130(12):995–1004.[PubMed]
11. Greenland S. Randomization, statistics, and causal inference. Epidemiology. 1990;1(6):421–9. [PubMed]
12. Greenland S. Probability logic and probabilistic induction. Epidemiology. 1998;9(3):322–32. [PubMed]
13. Greenland S, Robins JM, Pearl J. Confounding and collapsibility in causal inference. Statistical Science. 1999;14(1):29–46.
14. Hill AB. The environment and disease: association or causation? Proceedings of the Royal Society of Medicine. 1965;58(5):295–300.[PMC free article] [PubMed]
15. Holman CD, Arnold-Reed DE, de Klerk N, McComb C, English DR. A psychometric experiment in causal inference to estimate evidential weights used by epidemiologists. Epidemiology. 2001;12(2):246–55.[PubMed]
16. Howson C, Urbach P.  Scientific Reasoning: The Bayesian Approach.2nd ed. Chicago: Open Court Publishing Company; 1993. 
17. Hulley S, Grady D, Bush T, Furberg C, Herrington D, Riggs B, Vittinghoff E. Randomized trial of estrogen plus progestin for secondary prevention of coronary heart disease in postmenopausal women: Heart and Estrogen/progestin Replacement Study (HERS) Research Group. Journal of the American Medical Association. 1998;280(7):605–13. [PubMed]
18. Institute of Medicine.  Veterans and Agent Orange: Update 1998.Washington: National Academy Press; 1999. 
19. International Agency for Research on Cancer.  IARC Monographs on the Evaluation of the Carcinogenic Risk of Chemicals to Man: Tobacco Smoking. Vol. 38. Lyon (France): International Agency for Research on Cancer; 1986. 
20. International Agency for Research on Cancer.  IARC Monographs on the Evaluation of Carcinogenic Risks to Humans: Tobacco Smoke and Involuntary Smoking. Vol. 83. Lyon (France): International Agency for Research on Cancer; 2002. 
21. Kaufman JS, Poole C. Looking back on “causal thinking in the health sciences. Annual Review of Public Health. 2000;21:101–19. [PubMed]
22. Levin ML. The occurrence of lung cancer in man. Acta Unio Internationalis Contra Cancrum. 1953;9:531–41. [PubMed]
23. Lewis DK.  Counterfactuals. Cambridge (MA): Harvard University Press; 1973. 
24. Lilienfeld AM. “On the methodology of investigations of etiologic factors in chronic diseases”—some comments” Journal of Chronic Diseases. 1959;10:41–6. [PubMed]
25. National Cancer Institute.  Smoking and Tobacco Control Monograph No 8. Bethesda (MD): U.S. Department of Health and Human Services, National Institutes of Health, National Cancer Institute; 1997. Changes in Cigarette-Related Disease Risks and Their Implication for Prevention and Control. NIH Publication No. 97–4213.
26. National Cancer Institute.  Smoking and Tobacco Control Monograph No 9. Bethesda (MD): U.S. Department of Health and Human Services, National Institutes of Health, National Cancer Institute; 1998. Cigars: Health Effects and Trends. NIH Publication No. 98–4302.
27. Parascandola M, Weed DL. Causation in epidemiology. Journal of Epidemiology and Community Health. 2001;55(12):905–12. [PMC free article] [PubMed]
28. Pearl J.  Causality: Models, Reasoning, and Inference. Cambridge (MA): Cambridge University Press; 2000. 
29. Petitti DB. Associations are not effects. American Journal of Epidemiology. 1991;133(2):101–2. [PubMed]
30. Poole C. Causal values. Epidemiology. 2001;12(2):139–41. [PubMed]
31. Pradhan AD, Manson JE, Rossouw JE, Siscovick DS, Mouton CP, Rifai N, Wallace RB, Jackson RD, Pettinger MB, Ridker PM. Inflammatory biomarkers, hormone replacement therapy, and incident coronary heart disease: prospective analysis from the Women’s Health Initiative observational study. Journal of the American Medical Association. 2002;288(8):980–7. [PubMed]
32. Robins J. A new approach to causal inference in mortality studies with a sustained exposure period— application to control of the healthy worker survivor effect. Mathematical Modeling. 1986;7(9–12):1393–512.
33. Robins J. A graphical approach to the identification and estimation of causal parameters in mortality studies with sustained exposure periods. Journal of Chronic Diseases. 1987;40(Suppl 2):139S–161S. [PubMed]
34. Robins J, Greenland S. The probability of causation under a stochastic model for individual risk. Biometrics. 1989;45(4):1125–38. [PubMed]
35. Rothman KJ, Greenland S.  Modern Epidemiology. 2nd ed. Philadelphia: Lippincott-Raven; 1998. 
36. Royall RM.  Statistical Evidence: A Likelihood Paradigm. London: Chapman and Hall; 1997. 
37. Rubin DB. Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of Education Psychology. 1974;66(5):688–701.
38. Sartwell PE. “On the methodology of investigations of etiologic factors in chronic diseases”—further comments” Journal of Chronic Diseases. 1960;11:61–3. [PubMed]
39. Savitz DA, Tolo KA, Poole C. Statistical significance testing in the American Journal of Epidemiology, 1970–1990. American Journal of Epidemiology. 1994;139(10):1047–52. [PubMed]
40. Splawa-Neyman J. On the application of probability theory to agricultural experiments: essay on principles. Section 9. Reprinted in Statistical Science. 1990;5(4):465–80.
41. Steinberg E, editor. Hume D: An Enquiry Concerning Human Understanding. 2nd ed. Indianapolis (IN): Hackett Publishing Company; 1993. 
42. Susser M.  Causal Thinking in the Health Sciences; Concepts and Strategies of Epidemiology. New York: Oxford University Press; 1973. 
43. Susser M. Judgment and causal inference: criteria in epidemiologic studies. American Journal of Epidemiology. 1977;105(1):1–15.[PubMed]
44. Susser M. What is a cause and how do we know one? A grammar for pragmatic epidemiology. American Journal of Epidemiology. 1991;133(7):635–48. [PubMed]
45. US Department of Health and Human Services.  The Health Consequences of Smoking for Women A Report of the Surgeon General. Washington: U.S. Department of Health and Human Services, Public Health Service, Office of the Assistant Secretary for Health, Office on Smoking and Health; 1980. 
46. US Department of Health and Human Services.  The Health Consequences of Smoking: Cancer A Report of the Surgeon General.Rockville (MD): U.S. Department of Health and Human Services, Public Health Service, Office on Smoking and Health; 1982. DHHS Publication No. (PHS) 82–50179.
47. US Department of Health and Human Services.  The Health Consequences of Smoking: Cardiovascular Disease A Report of the Surgeon General. Rockville (MD): U.S. Department of Health and Human Services, Public Health Service, Office on Smoking and Health; 1983. DHHS Publication No. (PHS) 84–50204.
48. US Department of Health and Human Services.  The Health Consequences of Smoking: Chronic Obstructive Lung Disease A Report of the Surgeon General. Rockville (MD): U.S. Department of Health and Human Services, Public Health Service, Office on Smoking and Health,; 1984. DHHS Publication No. (PHS) 84–50205.
49. US Department of Health and Human Services.  The Health Consequences of Involuntary Smoking A Report of the Surgeon General. Rockville (MD): U.S. Department of Health and Human Services, Public Health Service, Centers for Disease Control, Center for Health Promotion and Education, Office on Smoking and Health; 1986. DHHS Publication No. (CDC) 87–8398.
50. US Department of Health and Human Services.  Reducing the Health Consequences of Smoking: 25 Years of Progress A Report of the Surgeon General. Rockville (MD): US Department of Health and Human Services, Public Health Service, Centers for Disease Control, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 1989. DHHS Publication No. (CDC) 89–8411.
51. US Department of Health and Human Services.  The Health Benefits of Smoking Cessation A Report of the Surgeon General. Atlanta: US Department of Health and Human Services, Public Health Service, Centers for Disease Control, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 1990. DHHS Publication No. (CDC) 90–8416.
52. US Department of Health and Human Services.  Preventing Tobacco Use Among Young People A Report of the Surgeon General. Atlanta: US Department of Health and Human Services, Public Health Service, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 1994. 
53. US Department of Health and Human Services.  Women and Smoking A Report of the Surgeon General. Rockville (MD): U.S. Department of Health and Human Services, Public Health Service, Office of the Surgeon General; 2001. 
54. US Department of Health, Education, and Welfare.  Smoking and Health: Report of the Advisory Committee to the Surgeon General of the Public Health Service. Washington: U.S. Department of Health, Education, and Welfare, Public Health Service, Center for Disease Control; 1964. PHS Publication No. 1103.
55. US Department of Health, Education, and Welfare.  The Health Consequences of Smoking A Public Health Service Review: 1967.Washington: U.S. Department of Health, Education, and Welfare, Public Health Service, Health Services and Mental Health Administration; 1967. PHS Publication No. 1696.
56. US Department of Health, Education, and Welfare.  The Health Consequences of Smoking 1968 Supplement to the 1967 Public Health Service Review. Washington: U.S. Department of Health, Education, and Welfare, Public Health Service; 1968. DHEW Publication No. 1696 (Supplement)
57. US Department of Health, Education, and Welfare.  The Health Consequences of Smoking 1969 Supplement to the 1967 Public Health Service Review. Washington: U.S. Department of Health, Education, and Welfare, Public Health Service; 1969. DHEW Publication No. 1696–2.
58. US Department of Health, Education, and Welfare.  The Health Consequences of Smoking A Report of the Surgeon General: 1971.Washington: U.S. Department of Health, Education, and Welfare, Public Health Service, Health Services and Mental Health Administration; 1971. DHEW Publication No. (HSM) 71–7513.
59. US Department of Health, Education, and Welfare.  The Health Consequences of Smoking A Report of the Surgeon General: 1972.Washington: U.S. Department of Health, Education, and Welfare, Public Health Service, Health Services and Mental Health Administration; 1972. DHEW Publication No. (HSM) 72–7516.
60. US Department of Health, Education, and Welfare.  The Health Consequences of Smoking A Report of the Surgeon General, 1973.Washington: U.S. Department of Health, Education, and Welfare, Public Health Service, Health Services and Mental Health Administration; 1973. DHEW Publication No. (HSM) 73–8704.
61. US Department of Health, Education, and Welfare.  The Health Consequences of Smoking A Report of the Surgeon General, 1974.Washington: U.S. Department of Health, Education, and Welfare, Public Health Service, Center for Disease Control; 1974. DHEW Publication No. (CDC) 74–8704.
62. US Department of Health, Education, and Welfare.  The Health Consequences of Smoking A Report of the Surgeon General, 1975.Washington: U.S. Department of Health, Education, and Welfare, Public Health Service, Center for Disease Control; 1975. [DHEW Publication No. (CDC) 76–8704]. 
63. US Department of Health, Education, and Welfare.  Smoking and Health A Report of the Surgeon General. Washington: U.S. Department of Health, Education, and Welfare, Public Health Service, Office of the Assistant Secretary for Health, Office on Smoking and Health; 1979. DHEW Publication No. (PHS) 79–50066.
64. U.S. Environmental Protection Agency.  Respiratory Health Effects of Passive Smoking: Lung Cancer and Other Disorders. Washington: Environmental Protection Agency, Office of Research and Development, Office of Air and Radiation; 1992. Publication No. EPA/600/6-90/006F.
65. Weed DL. Epidemiologic evidence and causal inference. Hematology-Oncology Clinics of North America. 2000;14(4):797–807. [PubMed]
66. Yerushalmy J, Palmer CE. On the methodology of investigations of etiologic factors in chronic diseases. Journal of Chronic Diseases. 1959;10:27–40. [PubMed]

Copyright Notice

Bookshelf ID: NBK44

Disclaimer: The author of each article published on this web site owns his or her own words. The opinions, beliefs and viewpoints expressed by the various authors and forum participants on this site do not necessarily reflect the opinions, beliefs and viewpoints of Utah Standard News or official policies of the USN and may actually reflect positions that USN actively opposes. No claim in public domain or fair use.    © Edmunds Tucker. UTopiAH are trade marks of the author. Utopia was written in 1515 by Sir Thomas More, Chancellor of England.

[ 2004 Health Consequences of Smoking: the Surgeon General continues next at Part  94 ]