Identifying co-morbidity patterns of health conditions via cluster analysis of pairwise concordance statistics

Abstract

Identification of co-morbidity patterns of health conditions is critical for evidence- based practice to improve the prevention, treatment, and health care of relevant diseases. Existing approaches focus mainly on either using descriptive measures of co-morbidity in terms of the prevalence of co-existing conditions, or addressing the prevalence of co-morbidity based on a particular disease (e.g. psychosis) or a specific population (e.g. hospital patients). As coincidental co-morbidity by chance increases with the prevalence rates of the conditions, which in turn depend heavily on the population under study, research findings on co-morbidity patterns using those approaches may provide unreliable results. In this paper, we propose an asymmetric version of Somers' D statistic to provide a quantitative measure of co-morbidity that accounts for co-occurrence of conditions by chance, and develop a unified clustering algorithm to identify co-morbidity patterns with adjustment for multiple testing and control for the false discovery rate. We assess the applicability of the proposed co-morbidity measure and investigate the performance of the proposed procedure for the adjustment of multiple testing by conducting a comparative study and a sensitivity analysis, respectively. The proposed method is illustrated using a national survey data set of mental health and wellbeing and a national health survey data set in Australia. Keywords: asymmetric Somers' D statistic; co-morbidity; concordance statistic; multiplicity problem; national survey data; overlapping clusters.