An Interpretable, Artificial Intelligence-Empowered Mortality Risk Score for Inborn Errors of Immunity

One foundational goal of global health and a shared inspiration for all humankind is to reduce premature mortality from preventable and treatable conditions. This imperative becomes markedly more urgent for the uniquely vulnerable population of patients with inborn errors of immunity (IEI), among whom mortality rates exceed the global average by a staggering 27-fold margin over the past 47 years. To reduce this profound mortality burden requires accurate individualized mortality risk prediction and identification of underlying risk factors, yet such tools remain lacking in the IEI field. Here, we developed an interpretable, artificial intelligence (AI)-empowered mortality risk score for IEI patients (IEIMRS) using machine learning and explainable AI techniques on large-scale, real-world US national Electronic Health Records (68,408 patients included with 5,685 deceased and 62,723 censored). IEIMRS outperformed existing intensive care unit (ICU), hospital, and all-cause mortality risk scores in individualized risk prediction (area under the receiver operating characteristic curve [AUROC] of 0.96, 95% confidence interval [CI] = [0.95, 0.97], and Delong test p < 0.001) with rigorous calibration (Brier score = 0.03), demonstrating stable and consistently superior performance across patient clinical and demographic subgroups (Delong test p < 0.05 for ethnicity, sex, age, and certain IEI subtypes and race). To facilitate broader implementation in diverse clinical settings, a simplified version of IEIMRS with comparable performance using merely 30 features was introduced as well (AUROC of 0.95, 95% CI = [0.94, 0.96], and Delong test p > 0.05). Critically, IEIMRS moved beyond outstanding prediction alone by offering transparent contributing risk factors to interpret each risk estimate. Analysis of these risk factors verified well-established mortality risk factors and identified novel risk factors enriched in IEI patients. Interestingly, at the subtype level, IEIMRS revealed heterogeneity in both mortality risk pattern and risk factor profile across IEI subtypes (F[1, 42] = 41.54, p < 0.001). As the first mortality risk score specifically designed for IEI populations, IEIMRS enables clinicians to confidently identify high-risk patients through accurate, interpretable predictions. Beyond clinical applications, the identified risk factors and risk patterns at both cohort and subtype levels offer insight into IEI mortality mechanisms and may inform fundamental immunology research on how specific pathophysiology and manifestations progress to fatal outcomes. Figure 1.