Abstract
From its initial applications in the social sciences to its subsequent adoption in medical science, meta-analysis is typically associated with evaluations of studies of specific interventions for a particular clinical condition or phenomenon. It is an excellent, systematic method to evaluate evidence resulting from an array of small and under-powered studies to obtain an objective index of the magnitude of an effect while improving overall statistical power. As illustrated in the work by Kraft and Dorstyn (this issue) meta-analysis can also be used to determine the magnitude of the relationships that may exist between specific variables across quantitative studies of a particular research problem. Meta-analysis relies heavily on the quality of the studies under investigation. Most recommended guidelines for and criticisms of meta-analytic procedures emphasize the need for methodological rigor and reporting. Consistent with best practice, Kraft and Dorstyn present clear inclusion and exclusion criteria and report important details to help us interpret their findings. The quality of the studies evaluated varied considerably and twenty studies were found with overlapping samples, necessitating an adjustment by the authors to minimize a potentially disproportionate influence in calculating effect sizes. As the authors note a degree of overlap is expected in studies that emanate from the Model Systems database and it is likely present in other ongoing, archived databases. The degree to which samples overlap on any measured variable is valuable information that will be germane to future meta-analytic studies involving data routinely collected by the Model Systems. Although the Model Systems maintains a repository of published papers at National Spinal Cord Injury Database website (https://www.nscisc.uab.edu/nscisc-database.aspx), it may be worthwhile to consider additional details about the samples used in each study to assist future inquiries about overlap. Meta-analysis of specific variables is best applied when guided by theoretical models of relationships that may exist between variables. Ideally, theoretical models guide a priori tests of predicted relationships and they are essential for isolating potential mechanisms that may be targeted in clinical interventions. They are essential in interpreting relationships. For example, depression is often comorbid with anxiety and it compromises quality of life. 1 Consequently, the magnitude of the associations between depression and life satisfaction, worry and anxiety are not particularly compelling. The strong associations found between depression and ‘…affective feelings, and thoughts and beliefs specific to SCI’ are consistent with cognitive-behavioral models of adjustment, generally, and provide further support for their use in conceptualizing adjustment and developing interventions. Theoretical models also guide advanced techniques like path analytic and structural equation modeling (SEM) and latent growth mixture modeling to test hypothesized relationships between and among variables. With the limitations of their approach Kraft and Dorstyn excluded studies that featured SEM. This should not discourage the use of these sophisticated procedures: Indeed, more studies using SEM (and other contemporary techniques, such as hierarchical linear modeling) are recommended to further our understanding of the contextual relationships between and among variables that occur over time in predicting outcomes in a manner that informs our policy and practice. 2 One of the most important, contemporary studies of psychosocial adjustment following spinal cord injury nicely demonstrates how theory can make meaningful predictions for behavior and outcomes with practical implications. In an a priori test of a popular model of resilience (and one that was excluded from the Kraft and Dorstyn research) using latent growth mixture
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have