Application of life course trajectory methods to public health data: A comparison of sequence analysis and group-based multi-trajectory modeling for modelling childhood adversity trajectories

Abstract

There is increasing awareness of the importance of modelling life course trajectories to unravel how social, economic and health factors relate to health over time. Different methods have been developed and applied in public health to classify individuals into groups based on characteristics of their life course. However, the application and results of different methods are rarely compared. We compared the application and results of two methods to classify life course trajectories of individuals, i.e. sequence analysis and group-based multi-trajectory modeling (GBTM), using public health data. We used high-resolution Danish nationwide register data on 926,160 individuals born between 1987 and 2001, including information on the yearly occurrence of 7 childhood adversities in 2 dimensions (i.e. family poverty and family dynamics). We constructed childhood adversity trajectories from 0 to 15 years by applying (1) sequence analysis using optimal matching and cluster analysis using Ward's method and (2) GBTM using logistic and zero-inflated Poisson regressions. We identified 2 to 8 cluster solutions using both methods and determined the optimal solution for both methods. Both methods generated a low adversity, a poverty, and a consistent or high adversity cluster. The 5-cluster solution using sequence analysis additionally included a household psychiatric illness and a late adversity cluster. The 4-group solution using GBTM additionally included a moderate adversity cluster. Compared with the solution obtained through sequence analysis, the solution obtained through GBTM contained fewer individuals in the low adversity cluster and more in the other clusters. We find that the two methods generate qualitatively similar solutions, but the quantitative distributions of children over the groups are different. The method of choice depends on the type of data available and the research question of interest. We provide a comprehensive overview of important considerations and benefits and drawbacks of both methods.