This preprint used 11 years of US Medicaid and Medicare claims to describe mortality patterns and identify conditions that predict death among adults with Down syndrome. The investigators employed a case-control design with risk-set sampling to align control selection with the distribution of incident Alzheimer’s disease time points in the cohort. A machine-learning model—gradient boosted trees—was trained to determine which claims-based conditions most strongly predicted mortality both in the overall sample and within age strata.
Researchers assembled a cohort of 137,293 adults identified as having Down syndrome in Centers for Medicare and Medicaid Services data. Death was identified using the CMS-reported date of death. Health conditions were captured through ICD-9 and ICD-10 diagnosis codes appearing in claims. The study period covered 11 years of claims data; the preprint does not specify the exact start and end calendar years in the abstract.
The primary outcome was death, as recorded by CMS. To support prediction analyses and comparison with incident Alzheimer’s disease timing, the team used a case-control framework with risk-set sampling so that controls reflected the temporal distribution of incident Alzheimer’s disease in the cohort. Details of eligibility criteria, matching variables, and specific sampling ratios beyond this description are not reported in the abstract.
Investigators trained gradient-boosted tree models on claims-derived diagnoses to rank predictors of mortality. The approach sought to identify the most influential claims-based conditions contributing to death risk across the full cohort and within age-defined subgroups. Exact model parameters, training-validation splits, and performance metrics (for example, area under the curve or calibration) are not provided in the abstract and therefore were not reported here.
Among the 137,293 adults with Down syndrome, 30,894 (22.5%) died during the study period. The mean age at death for decedents was 55 years (SD 10). When stratified by claims for Alzheimer’s disease, mean age at death was 59 years (SD 7) in those with an Alzheimer’s claim and 52 years (SD 12) in those without such a claim.
The machine-learning ranking identified several top predictors drawn from claims data. The most influential included any claim for dementia (including Alzheimer’s disease), any claim for pneumonia, recurring claims for cardiovascular disease in the three years before index death, and claims for heart failure and epilepsy. These predictors were reported as the leading contributors to mortality risk in the cohort.
The authors indicate they evaluated predictors in age-stratified analyses, but the abstract provides only summary-level age findings for mean age at death with and without Alzheimer’s claims. Specific age-stratified predictor rankings, differences in predictor importance by age band, and subgroup effect sizes are not described in the abstract and therefore were not reported here.
The results are consistent with previous clinical literature indicating elevated mortality in people with Down syndrome and a substantial contribution from neurodegenerative disease. The prominence of claims for dementia, pneumonia, cardiovascular disease, heart failure, and epilepsy as predictors highlights potential target areas for prevention, surveillance, and clinical management to reduce mortality in this population. The authors emphasize intervenable contributors to death risk, suggesting opportunities for improved infection prevention, cardiovascular care, seizure management, and dementia-related supports.
This work is a preprint and has not undergone peer review; the authors caution against using the findings to guide clinical practice without further validation. The abstract notes that health conditions were ascertained via ICD-9 and ICD-10 claims, which may be subject to coding limitations. Specific methodological details—such as exact calendar years covered, full inclusion/exclusion criteria, model tuning and performance metrics, and age-stratified predictor details—are not reported in the abstract. The study was funded by the National Institute on Aging (R01AG073179). The Boston University Medical Campus Institutional Review Board considered the project exempted research. Competing interests declared by authors are reported in the preprint.
Overall, this claims-based analysis of 137,293 adults with Down syndrome identified a high cohort mortality (22.5% during the study period), with dementia, pneumonia, recurrent cardiovascular disease, heart failure, and epilepsy emerging as the top predictors of death in gradient-boosted tree models. Further peer review and detailed methodological reporting will be needed to assess model performance and to guide translation into clinical practice.