SDOH Improves Performance of Heart Failure Mortality Predictive Model
Researchers have found that machine-learning models that incorporate social determinants of health data perform better than traditional methods of predicting heart failure deaths among Black patients.
Source: Getty Images
By - A new study published in JAMA Cardiology shows how machine learning (ML)–based models that incorporate social determinants of health (SDOH) data improve the prediction of in-hospital mortality among patients with heart failure, particularly Black patients.
The Centers for Disease Control and Prevention (CDC) report that about 6.2 million adults in the United States have heart failure and that the condition was mentioned on 13.4 percent of death certificates in 2018. Heart failure cost the US an estimated $30.7 billion in 2012.
According to a 2020 study published in Circulation: Heart Failure, racial disparities in heart failure outcomes persist despite some improvement over the last decade. Age-adjusted heart failure-related death rates are higher for Black patients, with young Black men and women having death rates that are 2.6-and 2.97-fold higher than their White counterparts. Similarly, the hospitalization rate for Black patients is 2.5-fold higher than that of White patients, with significantly higher costs in the year following hospitalizations.
With these disparities in mind, researchers who published the JAMA Cardiology study aimed to examine how incorporating SDOH can improve patient outcomes. Traditional methods for predicting in-hospital mortality for heart failure patients do not take SDOH into account, which may perpetuate disparities and negatively impact health outcomes, the study authors noted.
To develop an SDOH-incorporated model, the researchers used retrospective data from the Get With The Guidelines–Heart Failure (GWTG-HF) registry to identify heart failure hospitalizations between Jan. 1, 2010, and Dec. 31, 2020. Using this data, race-specific and race-agnostic ML models were developed and trained to predict in-hospital mortality. Patients were analyzed in two categories: Black, which comprised 19 percent of the dataset, and non-Black, which comprised the rest. The non-Black category included 2.1 percent Asian patients, 91 percent White, and 6.9 percent labeled “other race and ethnicity.”
The ML models demonstrated high performance in the internal testing subset, which comprised 82,420 patients, and the external validation set, which contained 3,469 patients from the Atherosclerosis Risk in Communities (ARIC) study between 2005 and 2014. The ML models also outperformed standard logistic regression models typically used in clinical settings.
The performance of the ML models was identical when using the race-specific and race-agnostic approaches in the GWTG-HF and external validation cohorts. In the GWTG-HF cohort, the addition of zip code–level SDOH parameters to the ML model with clinical covariates only was associated with higher performance in Black patients but not in non-Black patients.
The researchers stated that these findings indicate how ML models incorporating SDOH data may improve risk prediction of in-hospital mortality after hospitalization for heart failure, particularly in Black adults. However, further research is needed to ensure that ML models incorporating SDOH data improve health outcomes without perpetuating biases or additional disparities in a clinical setting.
