Extraction of unadjusted estimates of prognostic association for meta-analysis: simulation methods as good alternatives to trend and direct method estimation

• Published in: Journal of clinical epidemiology Vol. 99; pp. 153 - 163
• Main Authors: Pérez, T., McLellan, J., Perera, R.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.07.2018; Elsevier Limited
• Subjects: Data Analysis; Data extraction; Epidemiology; Estimates; Generalized least squares for trend estimation; Least-Squares Analysis; Medical prognosis; Meta-analysis; Meta-Analysis as Topic; Methods; Prognosis; Prognostic marker; Simulated data; Simulation; Statistical analysis; Statistical methods; Studies; Systematic review; Systematic reviews; Systematic Reviews as Topic; Womens health; Generalized least squares for trend estimation; Systematic reviews; Prognostic marker; Simulated data; Meta-analysis; Data extraction
• ISSN: 0895-4356; 1878-5921
• DOI: 10.1016/j.jclinepi.2017.12.017

Systematic reviews and meta-analysis are the standard methods to assess the association between prognostic markers and major events/conditions. However, the summary measures reported are not always explicitly presented and therefore different indirect methods of extracting estimates have been proposed. The aim of this study is to present two new alternative methods for obtaining summary statistics to be included in a meta-analysis of prognostic studies based on simulating individual patient data and to compare them with the already known generalized least squares for trend (glst) estimation method and direct method. We have checked the performance of these methods using a between study comparison, including 122 studies, and a within study comparison, based on data from one of the studies. The results obtained in this study show that glst estimation method appears to overestimate the effect size when reported information is incomplete. For the within-study comparison, the closest approximation to the direct estimates was obtained using the approach based on simulating individual patient data. The proposed simulation methods are a good alternative when other well-known indirect methods cannot be used.