Mortality risk estimation for autistic older adults: comparing novel machine-learning derived weights versus standard weights for the Charlson Comorbidity Index.
L’essentiel
Aim: We aimed to compare Quan and colleagues (2011) established weights for the Charlson Comorbidity Index (CCI) conditions to autism-specific weights for predicting mortality risk in autistic older adults. Materials & methods: We used inpatient healthcare claims from autistic older adults (aged 65+; n = 2829) using the Medicare Standard Analytic Files from 2021 to 2023. We used a machine learning technique called stochastic hill climbing to assign weights to the 12 CCI conditions to maximize predictive ability for 30-day and 1-year mortality. We then compared the resulting area under the curve (AUC) against the established weights. Results: The established weights had poor predictive ability for 30-day (AUC: 0.68; 95% CI: 0.62-0.74) and 1-year mortality (AUC: 0.67; 95% CI: 0.63-0.72). The autism-specific weights also had poor predictive ability for 30-day (AUC: 0.67; 95% CI: 0.61-0.73) and 1-year mortality (AUC: 0.67; 95% CI: 0.62-0.71). Conclusion: The established and autism-specific CCI weights performed similarly in predicting mortality among autistic older adults. Findings may suggest adjusting CCI weights alone is insufficient to accurately predict mortality risk in autistic older adults, and additional health conditions not currently captured by the CCI may need to be added to better predict mortality in this population. Future studies on developing an autism-specific mortality risk index are warranted.
Synthèse détaillée
Résumé original
Aim: We aimed to compare Quan and colleagues (2011) established weights for the Charlson Comorbidity Index (CCI) conditions to autism-specific weights for predicting mortality risk in autistic older adults. Materials & methods: We used inpatient healthcare claims from autistic older adults (aged 65+; n = 2829) using the Medicare Standard Analytic Files from 2021 to 2023. We used a machine learning technique called stochastic hill climbing to assign weights to the 12 CCI conditions to maximize predictive ability for 30-day and 1-year mortality. We then compared the resulting area under the curve (AUC) against the established weights. Results: The established weights had poor predictive ability for 30-day (AUC: 0.68; 95% CI: 0.62-0.74) and 1-year mortality (AUC: 0.67; 95% CI: 0.63-0.72). The autism-specific weights also had poor predictive ability for 30-day (AUC: 0.67; 95% CI: 0.61-0.73) and 1-year mortality (AUC: 0.67; 95% CI: 0.62-0.71). Conclusion: The established and autism-specific CCI weights performed similarly in predicting mortality among autistic older adults. Findings may suggest adjusting CCI weights alone is insufficient to accurately predict mortality risk in autistic older adults, and additional health conditions not currently captured by the CCI may need to be added to better predict mortality in this population. Future studies on developing an autism-specific mortality risk index are warranted.