Deeper Look at Weight Loss Program Effectiveness: Beyond Prevalence Rate Reduction

Recent studies showing a decrease in the obesity prevalence rate have been cited as progress in fighting the obesity epidemic. But is the prevalence rate a sufficient statistic to monitor changes over time? A population measure should reflect the relationship between obesity and health outcomes and costs to accurately reflect the population risks associated with obesity. In the case that health risks and economic costs were the same for all obese individuals, the prevalence rate would be sufficient. Instead, the risk of mortality and developing cancer, type II diabetes, and heart disease, increase with body mass index (BMI) level. Moreover, the medical and economic costs associated with obesity show the same upward trend. Obese employees with higher BMI levels on average exhibit higher absenteeism, and more frequent presenteeism.

Thus while health risks and cost are increasing in BMI, the prevalence rates are sensitive only to changes in the number of obese people, and not the severity of their condition. Development economics can provide a complimentary measure, known as the severity, which is the normalized squared deviations from the obese threshold. It is sensitive to the depth of an individual’s obesity (i.e., how much higher BMI is relative to the obese threshold) and the distribution of weight in the obese population. Thus it can detect changes in population obesity that directly relate to population health risk.

This paper uses the severity measure to show how it can provide additional insight when analyzing weight loss intervention effects.  Baseline and 6-month follow-up data from a 2-group, cluster-randomized controlled trial of 28 worksites were used to compare changes in prevalence and severity of obesity following an internet-based weight loss intervention with and without monetary incentives. The obesity prevalence rate declined significantly in the incentive-based program (-3.56%, p=0.002) but showed no significant change in the comparison intervention (0.46%, p=0.79). Similarly, the incentive-based program also demonstrated a significant reduction in severity (-4.03%, p=0.009) while the comparison program did not (-2.67%, p=0.16).  Subgroup analysis revealed an upward trend in African American severity even in the presence of a decline in prevalence for the comparison program. Given the importance of addressing obesity in high-risk subgroups, our results confirm that the two measures provide necessary complimentary statistics that ensures health risks mitigation is directly incorporated into weight loss treatment assessment.

To determine the effect of the intervention on the severity of obesity we had to take into consideration the irregular distribution of severity, which shows a large point mass over the value zero, as well as the clustering at the worksite level. For our preliminary model we utilized a generalized linear model with a log link, gamma family and cluster robust standard errors and controlled for comorbidities (diabetes, heart disease, high blood pressure, and arthritis) and demographics (age, age squared, gender, race, marriage, income, tenure and occupation). The treatment exhibited no significant effect, however, we plan to now move to a generalize estimation equation model to take in account both correlation within clusters and across time.