Changing population characteristics, effect-measure modification, and cancer risk factor identification

Abstract

Epidemiologic studies have identified a number of lifestyle factors, e.g. diet, obesity, and use of certain medications, which affect risk of colon cancer. However, the magnitude and significance of risk factor-disease associations differ among studies. We propose that population trends of changing prevalence of risk factors explains some of the variability between studies when factors that change prevalence also modify the effect of other risk factors. We used data collected from population-based control who were selected as study participants for two time periods, 1991-1994 and 1997-2000, along with data from the literature, to examine changes in the population prevalence of aspirin and non-steroidal anti-inflammatory medication (NSAID) use, obesity, and hormone replacement therapy (HRT) over time. Data from a population-based colon cancer case-control study were used to estimate effect-measurement modification among these factors. Sizeable changes in aspirin use, HRT use, and the proportion of the population that is obese were observed between the 1980s and 2000. Use of NSAIDs interacted with BMI and HRT; HRT use interacted with body mass index (BMI). We estimate that as the prevalence of NSAIDs use changed from 10% to almost 50%, the colon cancer relative risk associated with BMI >30 would change from 1.3 to 1.9 because of the modifying effect of NSAIDs. Similarly, the relative risk estimated for BMI would increase as the prevalence of use of HRT among post-menopausal women increased. In conclusion, as population characteristics change over time, these changes may have an influence on relative risk estimates for colon cancer for other exposures because of effect-measure modification. The impact of population changes on comparability between epidemiologic studies can be kept to a minimum if investigators assess exposure-disease associations within strata of other exposures, and present results in a manner that allows comparisons across studies. Effect-measure modification is an important component of data analysis that should be evaluated to obtain a complete understanding of disease etiology.

Figure 1

Changes in obesity and aspirin use between 1970 and 2000 in U. S. populations. Obesity data prior to 1991 from NHANES [25], for 1991–94 from colon cancer case-control study [17], and for 1999–2000 from rectal cancer study [42]. Aspirin data from 1970s and 1980 from references [26, 27] and for 1991–94 from colon cancer case-control study [17], and for 1999–2000 from rectal cancer study [42].

Figure 2

Predicted ORs for colon cancer associated with obesity (BMI ≥ 30 kg/m² versus lowest quartile, BMI ≤ 23 kg/m²) where obesity is held constant but with varying prevalence of NSAID or HRT use. ORs calculated from multiple logistic regression models for stratum-specific ORs for obesity in users and non-users of aspirin, and users and non-users of HRT among post-menopausal women. Adjusted for age, sex (NSAIDS/aspirin use only), energy intake, physical activity level, dietary calcium and fiber, NSAIDs/aspirin (HRT model only).