11.07.2015 Views

Obesity Epidemiology

Obesity Epidemiology

Obesity Epidemiology

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

INTERPRETING EPIDEMIOLOGIC EVIDENCE AND CAUSAL INFERENCE 45(SES). However, one can argue that biological processes are more likely to be the samein populations with similar demographic characteristics.Although greater ethnic and SES diversity in study populations may increase generalizability,this approach typically increases the heterogeneity of the population, andthus, the chance for unmeasured or uncontrolled confounding. For example, it would benecessary to include smokers and those who are ill in a representative sample to estimatenational obesity prevalence and trends, but inclusion of smokers and those who are ill inthe analyses of obesity and mortality can lead to serious confounding by smoking andreverse causation by existing diseases, and thus, diminish the validity of the estimatedassociations between obesity and mortality. On the other hand, a more homogeneouscohort with respect to residence, education, or occupation would not represent a randomsample of U.S. men and women. As a result, the distribution of dietary and otherlifestyle characteristics may not reflect the general population. However, this does notmean that the identified associations do not apply to other populations. In fact, mostcohort studies do not rely on national samples but draw on participants with similareducational, occupational, or geographic backgrounds. Compared with the general population,more homogeneous cohorts have relatively less unmeasured or uncontrolled confoundingby SES and other variables, and therefore, enhanced internal validity. But ifthere were effect modification by a variable defining a cohort (e.g., educational levels,ethnicity), it would not possible to detect it, and thus, it would affect the generalizabilityof the results. On the other hand, although epidemiologic studies of obesity and mortalitybased on U.S. national datasets, such as the National Health and Nutrition ExaminationSurveys (NHANES), are considered more generalizable, the analyses are more likely tobe confounded and are typically not sufficiently powered to characterize potential effectmodifications by ethnicity, SES, or other covariates. Nonetheless, epidemiologic studydesigns still need to balance validity and generalizability, and improving generalizabilitywithout compromising validity is a challenge for epidemiologists.Calculation and Interpretation of the PopulationAttributable RiskThe PAR is the fraction of disease in the population that is attributable to the exposure,and thus, the percent of cases that would be prevented if the exposure were tobe removed. 38 Widely used in epidemiology and public health, the PAR is sometimesreferred to as population attributable fraction, etiological fraction, or preventive fraction.In that, the PAR provides information about public health significance and burden of anexposure on disease, it can be useful in setting public health priorities. 38 The PAR hasbeen used in obesity epidemiology to estimate the number of deaths attributable to obesity39-41 and the fraction of obesity cases that could be prevented by adopting a healthylifestyle, such as one that includes exercise. 24 In the absence of confounding, to calculatethe PAR, one needs to estimate the association between a dichotomized exposure anddisease (RR) (typically from a cohort study), and obtain the prevalence of exposure (P e)in the population (typically from the specific cohort or population-based surveys, suchas NHANES):PAR = P e(RR − 1)∕(1 + P e(RR − 1))This equation can be generalized to an exposure variable with multiple categories,but could lead to a biased estimate in the presence of confounding and effect modification.42 Benichou 43 discussed several methods for calculating the PAR in a multivariate

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!