Get the most out of your data

Abstract

What is the best way to examine complex datasets involving large samples or patient groups with multiple confounding factors, many of which are inter-dependent? Traditional approaches have used multivariate regression that allows the investigator control over these confounding factors. Nonetheless, regression is less effective at using these inter-dependent interactions as part of the analysis. Latent class analysis (LCA) is a statistical approach to parse samples or patient subjects into class membership. The analysis can incorporate either categorical or continuous data. If one has a user-defined number of classes that are already anticipated, LCA can test that model. Alternatively, the best fit of the data to a derived number of class assignments and the characteristics (phenotypes) of these classes can be determined. In this volume of The Journal, Hendryx et al have used LCA to examine patterns of clinical, socioeconomic, behavioral, and psychosocial variables associated with low birthweight infants and preterm birth. They used data from the Australian Longitudinal Study on Women's Health, a large longitudinal study. Almost 10 000 live singleton births from almost 4000 women were linked to robust perinatal records and birth outcome data. The LCA revealed 5 groups of infants characterized by low education, recent drug use, stress/anxiety/depression, smoking, drinking, low education, multi-risk and a low risk control group. There were associations between individual classes with preterm delivery and others with low birth weight. Although logistic regression identified main effects not captured by LCA, the latent class analysis identified variable combinations not captured by the main effects analysis. Latent class analysis is a robust approach to complex datasets with the potential for great improvement in insights, interpretation, and implications of results. Hendryx et al provide us an example using a clinical problem with pediatric and perinatal relevance. Article page 42▸