Predicting Hospice Use Among American Indian/Alaska Native Persons with End-Stage Kidney Disease

BackgroundAmerican Indian/Alaska Native individuals are disproportionately affected by end-stage kidney disease. Once diagnosed, treatment options include receiving a kidney transplant or starting dialysis. Disparities in treatment options are linked to social determinants of health, and in rural areas, dialysis is often the preferred treatment due to limited access to transplant. A recent nationwide analysis of survival differences showed a survival advantage after starting dialysis but also found that patients in these populations are less likely to receive hospice care before death. They face early diagnosis, leading to more years on dialysis and reduced quality of life. Using a predictive model can help identify factors that affect hospice use among patients. This study proposes a predictive model to estimate the likelihood of hospice use before death among patients who received a kidney transplant. MethodsUsing the 2022 USRDS Standard Analysis Files, adults who fit inclusion criteria were retrospectively identified. Area Deprivation Index (ADI) data, used for this analysis, includes 17 variables from US Census data, such as income, education, housing security, employment, and healthcare access, to generate the standardized ranking. This study employed regression tree, random forest, boosting, and support vector machine techniques to predict hospice use among participants. The covariates used in this study are race, IHS region, age, sex, mean ADI, comorbidities, and transplant status. Models are assessed based on their predictive performance using accuracy, sensitivity, and specificity. ResultsThe random forest model outperforms others in accuracy, boosting offers the best sensitivity, and support vector machine and random forest excel in specificity. Overall, random forest and boosting are top models, with logistic regression as second-best. Logistic regression provides more interpretable results. The results, however, suggest a nonlinear relationship between covariates and the response that logistic regression might not capture. Based on both metrics, IHS region, age, and ADI are the most important features. ConclusionsIncluding measures like the ADI in predictive models highlights geographic disparities that are often overlooked and can guide interventions toward communities that have been historically underserved. Researchers and healthcare systems should improve access to hospice and palliative care for those who face these disparities.