A Hierarchical Nonlinear Modelling Framework of Uncertainty Surveillance of Aggregated Global Cholesterol

Abstract

Cholesterol is a key risk factor that can be altered and has become a primary factor in cardiovascular disease. Still, trend analyses at the global level often use aggregated estimates as error-free, which may exaggerate accuracy. The objective of the study was to measure long-term trends in the means of total cholesterol across countries and had the advantage of clearly modelling demographic structure, in addition to accounting for measurement error. The aims were to analyse changes over time by sex and age group, to examine nonlinear dynamics in age-period models, and to compare conventional and uncertainty-aware modelling. Multidecade-long country-year-sex-age group data on harmonised data used multilevel growth models, generalised additive mixed models and uncertainty-aware weighted multilevel models in which inverse-variance weights were based on reported 95% uncertainty intervals. Findings revealed that the average total cholesterol decreased significantly worldwide (weighted annual change: -0.0027 mmol/L, p = 0.046), with women having higher baseline levels than men (\(\beta\) = 0.09 mmol/L difference, p < 0.001). The temporal declines were steeper in older age, but the interaction between years and test age ranged from -0.0008 mmol/L/year to -0.0084 mmol/L/year (p < 0.001). Uncertainty-aware estimates were consistently smaller than those from unweighted models, indicating less overconfidence. The nonlinear age-period interaction was significant (p < 0.001). Its novelty lies in propagating reported uncertainty intervals directly into hierarchical models, which is the primary statistical constraint in previous studies. Among the actionable recommendations, it is possible to mention implementing models of uncertainty-aware global surveillance and focusing on age- and sex-specific approaches to lipid control to achieve faster progress toward cardiovascular disease reduction goals.