Causality, correlation, and cardiac disease: Does smoking cause cardiac hypertrophy and diastolic dysfunction?

The dangers of cigarette smoking are not disputed and yet it remains a multi-billion dollar industry.¹ Despite their first hand experience with complications of cigarette smoking, even many physicians continue to smoke.² Furthermore, alternatives to traditional cigarette smoking, such as e-cigarette use, are on the rise.³ This illustrates the on-going public appetite for smoking and reinforces the strong public interest in continuing to better understand relationships between smoking and disease.

Cigarette smoking can cause disease directly, but smoking is also associated with a suite of health behaviors and socioeconomic determinants of health.⁴ If we knew definitively that cigarettes caused heart failure, a public health campaign to reduce heart failure could have a precise and targeted mandate. On the other hand, if we knew definitively that cigarette smoking was merely a marker of a man or woman’s exposure to outdoor air pollution, access to health care, a healthy diet, or medication adherence then the public health approach might be much different. This is not meant to diminish the importance of smoking cessation. Smoking is known to cause a number of diseases and cessation remains a vitally important pillar of public health, but it is important to try and understand where it fits into a complex, multi-faceted, and nuanced understanding of disease causation.¹

The importance of trying to understand causality is a core principle in epidemiology. At a time when the world was convinced that improved life expectancy was a direct result of germ theory, antibiotics, and medical science, the microbiologist René Dubos demonstrated that mortality had been falling for several decades before these medical advances. He noted, “No medical discovery made during recent decades can compare in practical importance with the introduction of social and economic decency in the life of the average man. The greatest advances in the health of the people were probably the indirect results of better housing and working conditions, the general availability of soap, of linen for underclothing, of glass for windows… and at least one square meal a day.”⁵

In this issue of Circulation: Cardiovascular Imaging, Nadruz and colleagues try to understand some nuances in the relationship between cigarette smoking and heart failure by looking at relationships between cigarette smoking and cardiac morphology in the aging population of the Atherosclerosis Risk in Communities (ARIC) Study.⁶ The authors report that current smokers of otherwise similar body size have greater left ventricular mass than non-smokers. Current smokers also had worse diastolic function when compared to non-smokers. These differences were not readily explained by differences in diabetes, hypertension, heart rate, alcohol consumption, or carotid-femoral pulse wave velocity.

The authors thoughtfully considered their population. Specifically, they were concerned that elderly participants who smoked might be unique. An individual who continues to smoke into their mid-seventies without succumbing to health related complications might have “better protoplasm” then an individual who is smoking in their forties when health problems have not asserted themselves and discouraged on-going smoking. These differences can make groups less comparable, impose bias, and skew the results of observational studies. Approaches to address pervasive selection bias is vital in elderly populations.⁷ Ideally, one might begin obtaining serial cardiac imaging before age-related selective pressure exists and then demonstrate increasing left ventricular mass in smokers and stable or decreasing cardiac mass among non-smokers. This approach is time consuming and expensive. Instead, the authors used a more practical, but still established approach. They attempted to restore balance to their cohort by giving greater analytic weight in sensitivity analyses to participants who looked most like participants who did not attend the fifth ARIC study visit.⁸

The authors were careful to state that their results cannot be assumed to be causal. This is a standard limitation of observational research; however, it is extraordinarily unlikely and would be unethical for a randomized study of sustained cigarette smoking to ever be performed.⁹ Observational evidence is therefore one of the best sources of insight into the relationship between sustained cigarette smoking and cardiac morphology in men and women.

When there is no potential for randomized evidence, it is useful to have a rubric to evaluate the strength of observational results. While it is rare for observational evidence to prove causality beyond the shadow of a doubt, the goal of observational research is often to infer causality with varying degrees of confidence. One guideline that has been used for over sixty years to help assess the strength of causal inference in observational study is the Bradford Hill criteria, which include evaluating:¹⁰

1. Strength of the Association (strong associations are more likely to be causal as they are less likely to be explained by an unrecognized confounding factor)

2. Consistency (consistent findings by different investigators in different populations are more likely to be causal)

3. Specificity of the association (another explanation is not more likely)

4. Temporality (the cause must precede the effect in a causal relationship)

5. Biologic Gradient (although not all causal relationships have a dose-response, identifying a dose-response reinforces the impression of causality)

6. Plausibility (although this biases against paradigm shifting observations, a relationship is more likely to be causal when we can explain it with extant knowledge of mechanism)

7. Coherence (a relationship is more likely to be causal when there is agreement between all streams of evidence, e.g. epidemiologic and laboratory evaluation)

8. Experiment (perhaps the strongest evidence of causality is the ability to show that changing the exposure under controlled conditions changes the outcome)

9. Analogy (again this biases against paradigm shifting observations, but when the proposed cause and effect resemble other relationships that we already accept there is a greater impression of causality)

The Bradford Hill criteria are somewhat subjective; however, for the study in this issue of Circulation: Cardiovascular Imaging, one might speculate that: the effect is statistically significant, but the magnitude of the effect is relatively modest and accounts for only a five to ten percent difference in smokers relative to non-smokers (strength of the association). The findings are generally comparable with some, but not all other studies. Most notably results agree quite well with the Multi-Ethnic Study of Atherosclerosis (consistency).¹¹ The association with cardiac hypertrophy is not particularly specific and may plausibly be explained through confounding (e.g., health determinants associated with cigarette smoking, such as air pollution, activity, or diet) (specificity).¹² While this is functionally a cross-sectional study of a prospective cohort, smoking is well remembered and we can be reasonably sure that smoking preceded a participant’s cardiac imaging (temporality). Admirably, the authors show evidence of a dose-response both with pack-years of smoking and total years of smoking, which is one of the strongest votes for a causal relationship in this paper (biologic gradient). Finally the link between smoking and cardiac hypertrophy is both plausible and coherent. While the authors argue against cardiac hypertrophy mediated by increased resting blood pressure or vascular stiffness, there are other potential mechanisms that have been established in pre-clinical models. For instance, smoking-related mediators, such as prostaglandin E₂ and tissue factor, may act directly on cardiac myocytes to cause hypertrophy.^{13, 14}

Taken together, the current analysis offers a moderately robust case for a causal relationship between cigarette smoking and cardiac hypertrophy. The key concern is that the magnitude of the effect is modest enough that, while it may be causal, it may also be explained by unrecognized confounding. Most notably, smoking can be a marker of low socioeconomic status, which in turn is associated with a number of plausible mediators of cardiac hypertrophy. This study highlights the fact that many important questions rely on observational data to provide insight in man. In observational studies, high quality epidemiology may minimize bias, account for confounding, and contextualize different streams of evidence.

In sum, Nadruz and colleague provide an important, but incremental step toward understanding the nuanced relationship between smoking and heart failure.⁶ There is an epidemic of diastolic dysfunction and heart failure with preserved ejection fraction, but a paucity of decisively disease modifying therapy.¹⁵ The implication that smoking may lead to diastolic dysfunction in the absence of differences in arterial stiffness is intriguing, hypothesis generating, and may ultimately highlight high value mechanisms in diseases complicated by diastolic dysfunction.