Introduction: Social determinants of health (SDOH) are known to impact mortality risk for critically ill children. It is unknown whether SDOH confer additional risk above a patient’s predicted mortality at admission. The goal of this study was to evaluate if SDOH variables including the Child Opportunity Index (COI), insurance status, and preferred language impact the relationship between predicted mortality at admission and actual mortality in the PICU. Methods: This single center retrospective cohort study included all patients admitted to an academic tertiary care PICU over a two-year period (Nov ‘22-Nov ‘24). Data were extracted from the Virtual Pediatric Systems (VPS) database and SDOH variables were obtained from the electronic health record (EHR). COI was assigned based on home residence zip code. Language preference was classified as English or non-English. Insurance status was divided into self-pay, public, private, and Tricare. Predicted mortality was determined using the PRISM3 and PIM2 scoring systems calculated upon admission to the PICU. Only one PICU encounter was collected from each patient. Four logistic regression models were fit with mortality as the outcome: two with PRISM3 and PIM2 scores alone as covariates, and two expanded models with additional SDOH covariates. The models were compared to assess the effect of SDOH in mortality prediction. Results: 4100 unique patients were identified from VPS and data extraction from the EHR identified 3178 patients. 3.8% (154/4100) of patients died. 17.6% (713/4055) of patients were classified as very low COI and 29.8% (1208/4055) as low COI. 89.9% (3344/3718) of families preferred English. 2.9% (109/3713) were uninsured and 58.2% (2160/3713) had public insurance. The models including only PIM 2 and PRISM 3 scores showed good performance (AUC 0.908 and 0.899, respectively). The expanded models had similar performance with no improved prediction of mortality. Conclusions: In this single center study, the addition of SDOH variables did not result in an improved model for mortality prediction. Future studies should assess the impact of SDOH on other outcomes such as functional status at discharge and/or parental missed work. Larger datasets may offer more power to detect meaningful differences that can inform targeted efforts to improve outcomes.
Carr et al. (Sun,) studied this question.