Selection Bias, Confounding, or Information Bias?
To the Editor,
As a public health advocate against tobacco use, I find the recent article by Jatoi et al1 interesting and quite informative. As an epidemiologist, I also have a few concerns surrounding the methodologic issues of such study findings.
One of the most essential components of a comprehensive tobacco control program is to disseminate the correct health communication message that is based on sound scientific evidence. Because of the nature of this current study,1 the findings are to be interpreted with caution. Although nonrandomized observational studies are never free from methodologic limitations, it is important that the investigators of this study1 were explicit in spelling out such limitations at the outset.
First, the recruitment procedures for study subjects are not clear. It appears as if the patients did attend a specialized “hypertension” clinic in a general hospital. What intrigues me is the status of such patients being left “untreated”! Does this suggest a “healthy volunteer” effect,2 or a systematic sampling error? Also, a selected cohort is never representative of a general population. Taken together, such issues could have introduced selection bias, a serious methodologic limitation indeed.
Second, if the patients are recruited from a specialized hypertension clinic, the data recording system should be understandably comprehensive and should also not lose information on time-varying exposures or on risk factors potentially associated with hypertension or cardiovascular events. For example, the number of cigarettes smoked over time is an important consideration. This is particularly relevant when the authors argue that smoking even 1 cigarette “immediately” raises arterial stiffness.1,3 It might occur so that the long-term quitters were “heavy” smokers, whereas the recent quitters were “light” smokers.
Apart from residual confounding because of unmeasured confounders, other important measurable but modifiable risk factors that are potential confounders for the exposure–outcome relation studied, including physical activity levels and dietary habits, have not been addressed adequately.4 Such effects were indirectly reflected in Table 1, where cholesterol, triglycerides and hip:waist ratios, though found to be significant, were not factored into the model (Table 2). If such factors were adjusted for, the mean estimates of pulse wave velocity (a surrogate measure for arterial stiffness, which in itself is a limitation for outcome measure5) for former smokers (10.4) might have been similar (or even lower) to those reported for the never smokers (10.3), and the 95% CI also overlapped (Table 2).
Because of an apparent association between smoking and adiposity, a better indicator for adiposity, namely, hip:waist ratio, should have been factored into the model instead of body mass index. Not surprisingly, the adjusted analysis did not substantially change the crude estimates (Table 2). Also, it is not clear whether autocorrelation between the independent variables was accounted for.
In addition to selection bias and confounding, information bias because of inadequate information on exposure levels clearly undermines the scientific rigor of a nonrandomized observational study. In short, a greater transparency in methodologic approaches was warranted from the investigators before drawing an apparently strong conclusion. Nonetheless, the study findings should stimulate better designed clinicoepidemiological studies in the near future.
Z.K. worked briefly as a research fellow at the Department of Pharmacology and Therapeutics, Trinity College, Dublin, Ireland.
Jatoi NA, Jerrard-Dunne P, Feely J, Mahmud A. Impact of smoking and smoking cessation on arterial stiffness and aortic wave reflection in hypertension. Hypertension. 2007; 49: 981–985.
Pinsky P, Miller A, Kramer B, Church T, Reding D, Prorok P, Gelmann E, Schoen R, Buys S, Hayes R, Berg C. Evidence of a healthy volunteer effect in the prostate, lung, colorectal, and ovarian cancer screening trial. Am J Epidemiol. 2007; 165: 874–881.
Mahmud A, Feely J. Effect of smoking on arterial stiffness and pulse pressure amplification. Hypertension. 2003; 41: 183–187.
Tanaka H, Safar ME. Influence of lifestyle modification on arterial stiffness and wave reflections. Am J Hypertens. 2005; 18: 137–144.
Franks PW. Comparing the roles of physical activity and fitness in arterial stiffness: how important is exposure measurement error? Hypertension. 2005; 45: e1.