Time trends and forecasts of body mass index from repeated cross‐sectional data: a different approach

Abstract

In this paper, we report a case study on a technical generalization of the Lee-Carter model, originally developed to project mortality, to forecast body mass index (BMI, kg/m2). We present the method on an annually repeated cross-sectional data set, the Dutch Health Survey, covering years between 1981 and 2008. We applied generalized additive models for location, scale and shape semi-parametric regression models to estimate the probability distribution of BMI for each combination of age, gender and year assuming that BMI follows a Box-Cox power exponential distribution. We modelled and extrapolated the distribution parameters as a function of age and calendar time using the Lee-Carter model. The projected parameters defined future BMI distributions from which we derived the prevalence of normal weight, overweight and obesity. Our analysis showed that important changes occurred not only in the location and scale of the BMI distribution but also in the shape of it. The BMI distribution became flatter and more shifted to the right. Assuming that past trends in the distribution of BMI will continue in the future, we predicted a stable or slow increase in the prevalence of overweight until 2020 among men and women. We conclude that our adaptation of the Lee-Carter model provides an insightful and flexible way of forecasting BMI and that ignoring changes in the shape of the BMI distribution would likely result in biased forecasts.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 14 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.