In praise of replication studies and null results

CORR Insights®: Machine Learning on Medicare Claims Poorly Predicts the Individual Risk of 30-day Unplanned Readmission After Total Joint Arthroplasty, Yet Uncovers Interesting Population-level Associations With Annual Procedure Volumes

“…As part of the response to the “replication crisis,” many journals have moved toward accepting more papers with null findings [12], but the bar remains higher for authors to demonstrate why their null findings are worthy of readers’ precious time. Null findings serve to dispel misconceptions, steer the scientific apparatus toward more fertile ground, and help prevent us from making the same mistakes multiple times [13]. In this paper, Kunze et al [9] show that data characteristics prevail over sheer quantity, reinforcing previous findings suggesting that predicting 30-day hospital readmissions is a complex task that likely requires incorporation of other clinical features not typically found in large administrative databases.…”