AbstractIntroductionThe perceived absence of human implicit and explicit biases, scalability, and the potential for rapid improvement with algorithmic decision-making systems make compelling arguments for the widespread use of this technology. Unfortunately, real-world performance of some algorithmic decision-making systems demonstrates the reinforcement of discriminatory human biases in a way that is hidden from the human user. This study aims to retrospectively investigate if the widely used HOSPITAL score and LACE index used to predict hospital readmissions exhibit bias on the basis of sex.Materials and MethodsAll adult medical patients discharged from the SIU-School of Medicine (SIU-SOM) Hospitalist service from Memorial Medical Center from January 1, 2015, to January 1, 2017, were studied retrospectively to determine if patient sex had an influence on the ability of the HOSPTIAL score and LACE index to predict the likelihood of any cause hospital readmission within 30 days. Receiver operating characteristic (ROC) curves were constructed comparing risk prediction tool performance by sex by measuring the area under the curve (AUC).ResultsThe analysis includes data for the 1781 discharges for 1410 individual patients that met inclusion criteria. Of these discharges, 456 (27%) were readmitted to the same hospital within 30 days. The overall study population was 47% women, had an average age of 63 years and spent an average of 7.9 days in the hospital. Comparison of the performance of the LACE index in women and men showed no differences between AUCs (0.565 and 0.578, p = 0.613) and an ABROCA of 0.013. Sensitivity (67% and 70%), specificity (46% and 46%), PPV (30% and 31%), NPV (80% and 82%) and accuracy (51% and 52%) for the LACE index are very similar for women and men.Comparison of the performance of the HOSPITAL in women and men showed no differences between AUCs (0.56 and 0.58, p = 0.407) and an ABROCA of 0.008 indicating highly similar performance. Sensitivity (16% and 21%), specificity (96% and 95%), PPV (59% and 57%), NPV (77% and 78%) and accuracy (76% and 76%) for the HOSPITAL score are very similar for women and men.Discussion and ConclusionsThe performance of the HOSPITAL and LACE readmission risk prediction tools appears to have equivalent performance when used for women or men in this small, single-center, retrospective study. Further research is needed to explore the potential of bias and discrimination on risk prediction tools used in healthcare.