Examining Hacking Practices in Published Clinical Prediction Models: A Protocol for a Cross-Sectional Observational Study

This research project will examine statistical hacking practices in recently published high-ranking clinical prediction model articles. We will compare the distribution of reported area under the curve performance values, sensitivity and specificity values to expected distributions as well as compare reported area under the curve values against extracted area under the curve values from receiver operating characteristic curves. Previous research has identified questionable research practices in reporting area under the curve values at arbitrary thresholds, however, questions still remain if these practices are occurring in recently published models, other performance values and how these hacking practices are occurring. This research will answer the following questions: 1) Is there evidence of AUC-hacking in recently published, high ranking journal articles? 2) Is there evidence of hacking practices in other prediction model statistics? 3) Is there evidence researchers are altering reported AUC values or does AUC-hacking occur during the analysis of data?