Abstract 13406: Surveillance Bias: Inpatient Outcomes of Heart Failure in Substance Users as an Example

Abstract

Introduction: Substance use (SU), including opioid, methamphetamine, and cocaine use, is not infrequent among those with heart failure (HF). We aimed to characterize substance use among patients with heart failure and its association with inpatient outcomes. Methods: We used the Nationwide Readmission Database (NRD) for 2019 for our retrospective study. First, we extracted all cases older than 18 years that included a primary diagnosis of HF. Appropriate survey and domain analyses were applied to obtain national estimates using SAS 9.4. Results: We identified 1,318,006 discharges with a primary diagnosis of HF. SU was present in 5% of the study cohort (n= 57,395 cases). Patients with SU were younger than those without SU (mean age 55 years vs. 72 years, p<0.001) and fewer women (29% vs. 49%, p<0.001). Patients with SU had a higher prevalence of underlying hypertension, acquired immunodeficiency syndrome, and chronic lung and liver disease, but a lower prevalence of malignancy, dementia, end-stage renal disease, diabetes, and cerebrovascular disease (p<0.001). Substance users had significantly lower inpatient mortality (1.4% vs. 2.9%, p<0.001), but higher leaving against medical advice (AMA) (8% vs. 1%, p<0.001), length of stay (mean 9 days vs. 8 days, p<0.001), and all-cause 30-day readmission rates (58% vs. 46%, p<0.001). Even after adjusting for age, gender, and comorbidities, the presence of SU continued to be associated with lower inpatient mortality compared to those without SU (OR 0.81 [0.73-0.90]). In this case, mortality rates are calculated based on discharge-level records rather than patient-level records. This leads to a dilution of the inpatient mortality rate because drug users who have multiple readmissions contribute more to the denominator. After excluding patients who left AMA or had readmission in the same year, SU was not associated with any favorable outcome (OR 1 [0.90-1.12]). Conclusions: While administrative data offers opportunities to study clinical outcomes, understanding how data is created and analyzed is key. Because the NRD database is built based on discharge-level rather than patient-level records, this is an example of how these calculations can introduce bias.