Session Type: Abstract Submissions (ACR)
Prognostic and explanatory models are frequently utilized in clinical medicine to assist in judgments about diagnosis, prognosis, and treatment options. I surveyed predictive and explanatory models in rheumatology by conducting a systematic review using a PubMed (Medicine) search of a two year period, starting January 1, 2010, and ending December 31, 2011. This identified 564 articles using the key words: “Rheumatology,” and “prediction,” “models,” “stepwise,” “regression,” “multivariate,” “survival,” “prognosis,” “propensity scores,” “cost-effectiveness,” “decision analysis,” and “diagnosis.”
Out of the 564, I identified 76 articles, which I reviewed by abstract, and of these 42 articles were selected for review of their full text for adherence to methodologic standards.
Studies addressed a variety of issues including diagnosis, prognosis, identification of risk factors, biomarker studies, and disease activity. Data sets included institutional, multi-institutional, and national registries. Prospectively acquired, retrospectively acquired, and cross-sectional data was used. Regression techniques included logistic, linear, generalized estimating equation (GEE) and generalized linear model (GLM). Two studies used classification and regression trees (CART) analysis, 2 used a Bayesian approach, and one used neural network. Only one study explicitly described a power calculation. Internal validation by bootstrapping, split sample, or similar techniques was used in 13 of 42 studies, but only 4 used an external validation step. Discrimination (ROC Curves) or clinical utility (sensitivity and specificity) was described in 12 studies, but only 3 were subjected to calibration.
A minority of published prognostic and explanatory models adhered to established methodologic standards. Very few studies contained a validation study and those that did were small and thus subject to over fitting. Several studies provided information on discrimination; however, few of these were calibrated using data from the validation study. A consensus scoring system of prognostic and explanatory models similar to that which has been utilized for systemic reviews needs to be constructed and adopted by the rheumatologic clinical epidemiology and clinical trial community.
D. A. Albert,
« Back to 2012 ACR/ARHP Annual Meeting
ACR Meeting Abstracts - https://acrabstracts.org/abstract/systematic-review-of-explanatory-and-predictive-models-in-rheumatology-lack-of-adhearance-to-methodologic-standards/