Body Weight, Weight Change, and Mortality Risk

Abstract

Many studies find that overweight people have lower mortality than people with clinically normal body weight, despite higher rates of disease onset. One popular hypothesis to explain this is "reverse causation": the idea that while weight affects health, health also affects weight. Specifically, the hypothesis states that death risk typically increases with body weight, all else being equal, but that sickness-induced weight loss creates a spurious association between low (though clinically normal) body weight and high mortality. The hypothesis remains largely untested, however. This dissertation assesses its plausibility.

Chapter 1 tests whether past weight loss can account for elevated death rates among people with clinically normal body mass index (BMI). Data come from five longitudinal surveys, each based in a different country: Australia, Costa Rica, England, Taiwan, and the USA. The study uses logistic regression with P-splines and finds that, in all five surveys, lowest mortality occurs among the overweight, adjusting for age (range: 50-79), age squared, sex, and smoking. However, people losing weight have higher mortality than weight-gainers or weight-maintainers. Adjusting the BMI--mortality curve for weight change thus attenuates the overweight longevity advantage. This provides modest support for the reverse causation hypothesis.

Chapter 2 explores possible reasons for high mortality among weight-losers, testing whether diagnosed disease predicts subsequent or concurrent weight loss in two population-based surveys: one in Costa Rica and one in England. The study uses logistic regression to estimate associations of two weight loss outcomes with six (self-reported) diagnoses -- cancer, diabetes, heart attack, stroke, arthritis, and lung disease -- adjusting for age (range: 52-79), sex, smoking, and initial BMI. Associations between disease and weight loss are more apparent in England than in Costa Rica. This suggests the impact of reverse causation could vary across populations.

Finally, Chapter 3 uses simulation to demonstrate that any of three competing hypotheses could explain a longevity advantage among the overweight. Where death risk increases with BMI, even modest sickness-induced weight loss is sufficient to produce the phenomenon. However, the simplest models of reverse causation are still surprisingly complex. Together, the three chapters provide tentative support for the reverse causation hypothesis.