© 2007 American Thoracic Society doi: 10.1164/rccm.200708-1275ED
Of Contagion and Inherited Susceptibility: An Epidemiologic Tribute to George W. Comstock
International Union against Tuberculosis and Lung Disease In this issue of the Journal (pp. 1281–1288), van der Eijk and colleagues venture into the past with modern analytic tools (1). Taking prior analyses a step further, it is also a tribute to the late George W. Comstock, one of the most preeminent epidemiologists of the 20th century in general and tuberculosis specifically (2). The Prophit survey was an exceptionally well-designed twin study, identifying cases with a twin sibling among a tuberculosis population. Among the 101 twin pairs meeting the inclusion criteria, 39 of the co-twins also had tuberculosis. The influence of zygosity on the risk of tuberculosis in the co-twin, and thus heredity, was the research question in the original survey, the reanalysis by Comstock (3), and the current analysis (1). The original and both reanalyses found a higher concordance of tuberculosis among monozygotic than dizygotic twins. The original study explained this finding with a greater exposure and a higher incidence of sputum smear–positive tuberculosis among the former compared with the latter. Comstock used a then-emerging multivariate analysis technique allowing adjustment for potential confounders and found that monozygotes had an approximately threefold elevated risk of developing tuberculosis compared with dizygotes even after this adjustment. van der Eijk and colleagues obtained identical results as Comstock using a standard multivariate logistic regression model. However, accounting for effect modifiers, zygosity vanished as a risk factor: all apparent genetic susceptibility turns out to be just a masked difference in exposure risk. That a different analytic strategy gives the opposite results begs for an explanation. Bias or confounding factors distort study findings and may suggest a spurious association if in fact there is none or if they mask one that exists. The very rationale for Comstock to reanalyze the data was his acute awareness of this potential pitfall. For example, if sex emerges as a risk factor for a condition, the risk for which also increases continuously with age, sex as a putative risk factor might be solely attributable to an age difference between the two sexes. Finding a crude risk difference by sex in this setting is then said to be confounded by age. Confounders are addressed by either scrutinizing sex differences within appropriately sized age groups (stratified analysis) or, as is common now, using standard logistic regression analysis for adjustment. Standard logistic regression averages out the strata effects of a confounder. It will thus be inappropriate if sex differences go in one direction in some age groups but in the opposite in others, age being here an effect modifier. To prevent missing such an age effect requires adding so-called interaction terms in the logistic model. Several interactions with zygosity were identified by van der Eijk and colleagues. These include age, sputum smear status in the index case, and other factors that potentially influence intensity of exposure. Among twins with characteristics that expose each to the co-twin more intensely, the index twin rather than the parent or another sibling might have been the source of infection, most conspicuously if the index twin was sputum smear positive, but the putative source (parent or a nontwin sibling) was not. It is intuitively appealing to approach tuberculosis epidemiology by taking guidance from the classification system of the American Thoracic Society (4). Exposure is the conditio sine qua non for acquisition of latent tuberculosis infection, and the latter for the development of clinically active tuberculosis. Exposure is, however, not amenable to quantitative measurement, and information on latent infection was not available in the Prophit survey. All three analyses thus required assumptions about an unmeasurable extent of exposure leading through an unmeasured extent of infection prevalence to measured tuberculosis. If the risk of tuberculosis were dependent solely on the prevalence of latent infection, then morbidity would be directly proportional to intensity of exposure. However, all evidence points to fundamental differences between the risk of becoming infected and the risk of developing tuberculosis after infection. For instance, it is common that young adult males have a higher risk of becoming infected than females of the same age, but once infected, females in this age range are at higher risk of tuberculosis than their age-matched male counterparts (5). Comstock has likened the pathogenesis of tuberculosis to the two-hit theory espoused for cancer, with Mycobacterium tuberculosis as the necessary, and some reduction in the body's defenses as the sufficient, cause (6). The first hit, acquisition of infection, is by and large exogenous; the second, progression to tuberculosis, is by and large endogenous in nature. The increased risk of progression from latent infection to tuberculosis in the presence of chemical or biological immunosuppressants testifies to the large role of innate, genetically determined immune functions in suppressing metabolic activity of M. tuberculosis. The authors point to failed attempts to unambiguously identify genes that specifically modify the risk of progression from infection to disease. Comstock and colleagues (7) and Vynnycky and Pine (8) estimated the baseline lifetime risk of tuberculosis after infection during late childhood to be on the order of 10 to 15%. The real research challenge is thus not to identify additional factors that increase the risk of tuberculosis but rather factors that determine the probability of escaping it. It is axiomatic that genetic makeup protects the majority of latently infected humans from ever developing tuberculosis. The current reanalysis of the Prophit survey data clarifies that the most parsimonious explanation for the survey findings is extent of exposure and resulting infection with M. tuberculosis. Because it is unknown who was infected and escaped tuberculosis, no analysis of the Prophit survey data, however sophisticated, can determine the role of genetics in protection from tuberculosis. This reanalysis is a fitting tribute to George Comstock's contribution to our understanding of the epidemiology of tuberculosis. It updates our knowledge and shows the appropriate use of analytic tools that have become available. Appropriately, this study raises new questions. The Prophit survey confirms the overwhelming role of the necessary cause for tuberculosis, but lack of information on infection prevalence leaves us short of addressing the issue of the sufficient cause. FOOTNOTES Conflict of Interest Statement: H.L.R. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. REFERENCES
Related articles in AJRCCM:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
