Comparing the Accuracy of ABCD-10 and SCORTEN in Predicting the In-Hospital Mortality of Stevens-Johnson Syndrome/Toxic Epidermal Necrolysis: A Multi-Institutional Study from Central China.

BACKGROUND

The newly described ABCD-10 (age, bicarbonate, cancer, dialysis, 10% body surface area [BSA]) is a 5-item mortality prediction model for patients with Stevens-Johnson syndrome (SJS)/toxic epidermal necrolysis (TEN). It was developed in the United States, has at present been externally tested only in the United States, Spain, and Singapore, and remains to be validated in resource-restricted settings. We sought to compare the accuracy of ABCD-10 and Score of Toxic Epidermal Necrolysis (SCORTEN) in predicting in-hospital mortality in a cohort from central China. Due to disease progression affecting the accuracy of the prediction model during hospitalization, for example, higher predictive accuracy of SCORTEN based on parameters collected on day 3 of hospitalization, we also assessed the overall predictive value of ABCD-10 on days 1 and 3, respectively.

METHODS

A retrospective study was performed over a 10-year period (2010-2020) from 3 medical institutions in Wuhan. The performance of predictive models was assessed by both discrimination and calibration. Receiver-operating characteristic (ROC) curves, Hosmer-Lemeshow goodness-of-fit tests and calibration plots were used to evaluate the model discrimination and calibration.

RESULTS

Of 84 included patients, 11 (13.1%) did not survive. The discrimination power of ABCD-10 was not significantly different from that of SCORTEN (area under the curve: day 1, p > 0.05; day 3, p > 0.05). Although the calibration of ABCD-10 was good, it was inferior to SCORTEN as it underestimated total mortality (Hosmer-Lemeshow goodness-of-fit test: day 1, p = 0.17 vs. p = 0.63; day 3, p = 0.35 vs. p = 0.93). Besides, the performance of ABCD-10 was slightly better on day 3 relative to day 1. During hospitalization, bacteremia developed in 21 (25.0%) patients, which was associated with a higher risk of death in our cohort (odds ratio, 22.88; 95% CI, 4.38-119.40; p < 0.001).

CONCLUSION

ABCD-10 showed acceptable overall performance, but revealed mortality underestimation and was inferior to the performance of SCORTEN. In consistence with SCORTEN, ABCD-10 was a better model when using values collected at day 3 of hospitalization relative to day 1.