Download This PaperAging of Measurement in the Association between Anthropometry and Mortality in Children
52 Pages Posted: 11 Dec 2022 Last revised: 31 Dec 2022
Date Written: December 30, 2022
Abstract
Background. Millions of children under 60 months old are undernourished globally. Olofin et al. (2013) investigates the risks associated with suboptimal growth in such children using older data from prospective cohort studies in 10 countries. The Cox proportional hazards model is fit to each data set, and the results pooled meta-analytically.
Methods. To engage critically with Olofin, the study is replicated. 5 of the 10 data sets are obtained, as well as one not used in Olofin. The original results are approximately replicated. A methodological issue is exposed: the Cox model’s assumption that hazard ratios are fixed over time is violated by aging of measurement. Hazard ratios are higher in data sets with more frequent follow-up. Two alternative models are applied: a probit model for death within one year of measurement, and a stochastic process (“MIG”) model.
Results. The two new methods yield one-year mortality RRs that are lower than some corresponding HRs in Olofin. Where Olofin reports an HR for weight-for-height z (WHZ) < –3 vs. WHZ > –1 of 11.63, the probit and MIG RRs for WHZ = –3.5 vs. –0.5 are 3.75 (2.22, 6.31) and 5.61 (2.58, 12.18).
Conclusions. For coherence, pooled analyses of historical data with various follow-up frequencies should employ methods that are less sensitive to timing effects and should state results with reference to specified follow-up periods. HRs and RRs are not constant.
Keywords: Nutrition, child mortality, Cox proportional hazards model, aging of measurement, stochastic process survival modeling
JEL Classification: I18
Suggested Citation: Suggested Citation