Incorporating zip code-level social determinants of health variables into an already high-performing clinical risk model for in-hospital STEMI death did not meaningfully enhance prediction, indicating that short-term, in-hospital survival is primarily driven by immediate clinical factors rather than upstream socioenvironmental risks.
Key Findings
Results
Both the clinical model and the full SDOH model demonstrated excellent and equivalent discrimination for predicting in-hospital STEMI death.
The clinical model achieved a concordance statistic of 0.95 and an area under the receiver operating characteristic curve (AUC) of 0.85.
The full SDOH model also achieved a concordance statistic of 0.95 and AUC of 0.85.
There was 'no meaningful difference in overall predictive performance' between the two models.
DeLong's test was used to evaluate incremental value between models.
Methods
The study analyzed a large imputed cohort of 749,410 STEMI admissions from the GWTG-CAD registry between 2018 and 2020.
The cohort had a median age of 62 years.
Approximately 72% of the cohort were men.
Data came from the American Heart Association's Get With The Guidelines-Coronary Artery Disease (GWTG-CAD) registry.
Missing data were addressed with multiple imputation by chained equations.
Methods
The clinical risk model included demographic, clinical, and medical history variables and served as the baseline comparator.
The full SDOH model additionally incorporated six zip code-level SDOH variables: college graduation, high school graduation, per-capita income, poverty, foreign-born status, and median household income.
Logistic regression was used to construct both models.
Discrimination was assessed using the concordance statistic and area under the receiver operating characteristic curve.
Results
In exploratory analyses, five of the six individual SDOH variables each produced a modest statistically significant improvement in model discrimination.
High school graduation, college graduation, per-capita income, poverty, and foreign-born status each had 'a modest statistical improvement in model discrimination' when added individually.
Median household income did not improve model discrimination.
Despite statistical significance, these improvements were characterized as 'modest' and not clinically meaningful.
Conclusions
The study concludes that short-term, in-hospital STEMI survival is primarily driven by immediate clinical factors rather than upstream socioenvironmental risks.
Incorporating SDOH variables into the already high-performing clinical risk model 'did not meaningfully enhance prediction.'
The authors note that SDOH are 'increasingly recognized as contributors to disparities in cardiovascular outcomes' but their incremental predictive value for short-term STEMI death appears limited.
The findings suggest that while SDOH may influence longer-term or population-level outcomes, they add little to in-hospital risk stratification beyond clinical variables.
What This Means
This research examined whether adding neighborhood-level social and economic factors — such as income, education levels, poverty rates, and the proportion of foreign-born residents — to existing clinical risk models could better predict which patients hospitalized for a heart attack (specifically STEMI, a severe type) would die in the hospital. The study used a very large dataset of over 749,000 heart attack hospitalizations from a national registry covering 2018 to 2020. Researchers compared a model based solely on clinical information (like age, sex, and medical history) to one that also included zip code-level social determinants of health.
The study found that both models performed equally well, with high accuracy in predicting in-hospital death (AUC of 0.85 in both cases). Adding the social and economic variables did not meaningfully improve the model's ability to identify high-risk patients beyond what clinical information alone could provide. While five of the six social variables tested showed small, statistically detectable improvements in some analyses, these differences were not considered clinically meaningful.
This research suggests that for predicting whether a STEMI patient will survive their hospitalization, the most important factors are immediate clinical ones — such as the severity of the heart attack and the patient's existing health conditions — rather than the social and economic environment they come from. This does not mean social determinants are unimportant to heart disease overall; they may play a larger role in longer-term outcomes or in determining who gets a heart attack in the first place. However, for the specific purpose of short-term, in-hospital risk prediction, clinicians and hospital quality programs may not gain added value from incorporating neighborhood-level social data into their risk models.
Pliakos E, Shultz K, Reddy K, Eberly L, Khatana S, Fanaroff A, et al.. (2026). Role of Social Determinants of Health in Risk Prediction Models for ST-Segment-Elevation Myocardial Infarction Death: A Registry-Based Study.. Journal of the American Heart Association. https://doi.org/10.1161/JAHA.125.047813