Area under the receiver operating characteristic (ROC) curve
An area under the receiver operating characteristic curve analysis was performed for both the BOOST score and LACE index. The resulting c-statistics were 0.667 (955 CI 0.559 – 0.775, p = 0.005) for the BOOST score, and 0.685 (95% CI 0.579 – 0.792, p = 0.002) for the LACE index, demonstrating both as significant predictors of readmission (Figure 1 ). We believe the similarity in these c-statistics to demonstrate equivalent accuracy in predicting readmissions by either model.
Cut-off scores for optimal sensitivity and specificity in each model as determined by the maximum Youden’s index (YI) were found to be a BOOST score of 4 (YI = 0.28) and a LACE index of 11 (YI = 0.29). Patients with a BOOST score of 4 or more would be classified as ‘high-risk’ and have a sensitivity of 69.4% and specificity of 58.1% for predicting readmission. The high-risk category for the LACE index was determined to be 11 or more, with a sensitivity of 77.8% and a specificity of 51.4% for readmission.
Cross tabulation demonstrated a readmission rate of 20.4% in those with a low-risk BOOST score, and a 44.6% readmission rate in the high-risk category. Pearson chi-squared testing showed those who had a high-risk BOOST score were significantly more likely to be readmitted (χ2 = 7.356, N = 110, P = 0.007). Those in the low-risk LACE category had a readmission rate of 17.4%, and in the high-risk category 43.8%, thus making it a poor rule-out test, similar to BOOST at 20.4%. Those with a high-risk LACE index were significantly more likely to be readmitted (χ2 = 8.446, N = 110, P = 0.004). High-risk BOOST patients had odds of reattendance 2.5 times higher than low-risk patients, and high-risk LACE patients had odds of reattendance 3.0 times higher than low-risk patients.