gms | German Medical Science

Background: The collection of cross-sectional data is usually less resource-intensive as compared to longitudinal data. Therefore, valid methods for the prediction of longitudinal outcomes on the basis of repeated independent cross-sections would be desirable. Repeated independent cross-sectional data, i.e., data collected at different time points (T0, T1) among independent samples could be obtained by real world data provided by routine data collection systems. In the current application, our aim is to use such a pseudo-panel of independent cross-sectional data (i.e. data of T0 and T1) to predict the longitudinal health trajectory of refugees (T0-T1). We will illustrate the proposed methods by the example of studying contextual effects (e.g. remoteness of refugee accommodation) on health (e.g. low/high self-rated general health) among refugees by calculating probabilities for the transitions from one state (e.g. “rural-healthy”) to another (e.g. “rural-sick”) with associated variances.

Methods: Following the post-migration trajectory, two large-scale cross-sectional health surveys among randomly selected refugee samples in reception centers (T0) and accommodation centers (T1) located in Baden-Württemberg were conducted as part of the RESPOND study [1]. Self-reported measures of physical and mental health, health-related quality of life, health care access, and unmet medical needs of 560 refugees were collected. Missing data was handled by multiple imputations based on the method of Fully Conditional Specification [2] and the Predictive Mean Matching method [3], [4]. For each imputed data set transition probabilities were calculated based on (i) Probabilistic Discrete Event Systems with Moore-Penrose generalized inverse matrix method [5], and (ii) Propensity Score Matching [6]. By application of sampling approaches, exploiting the fact that status membership is multinomially distributed, results of both methods were pooled by Rubin's Rule [7], accounting for within and between imputation variance. Application of, e.g., Bland-Altman plots [8] allowed to quantify and investigate the agreement between both methods.

Results: The results of most of the analyzed estimates of the transition probabilities and associated variances are comparable between both methods. However, it seems that they handle the occurrence of sparse cells (i.e. the occurrence of states with small numbers of members) differently: either assigning an average value for the point estimate of the transition probability for all states with high certainty (i), or assigning a more extreme value for the point estimate of the transition probability, but a large value for the variance estimate (ii).

Conclusion: Further research on the potential to extrapolate the results of cross-sectional data to predict longitudinal outcomes is needed. To inform and advise future studies, the results of this analysis will be compared to the results of a prospective natural experiment study with longitudinal data collection of contextual and individual factors.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.