Influence of the Social Deprivation Index on COVID-19 Hospital Outcomes

The inclusion criteria were: (1) adults (≥ 18 years old) (2) COVID-19 confirmed by positive RT-PCR test or ICD-10 diagnosis (3) admission to the emergency room or hospital between March 1, 2020 and February 8, 2021. Patients living in a nursing home before their index was submitted were excluded because postcodes in the EHR may not represent their residence. The resulting cohort included 30,016 unique patients with confirmed COVID-19.

Exposure: social deprivation index

We linked clinical data using the patient’s residential postcode with social data from the postcode tabulation area (ZCTA) to calculate the social deprivation index (SDI)⁸ for 2020 using publicly available sources⁹. The SDI is a composite of six socioeconomic characteristics (income, education, employment, housing, household characteristics and transportation) determined at the ZCTA level. We mapped the residential postal codes of the patients on the ZCTAs. We categorized all ZCTAs into quintiles based on the SDI score.

Results

The two outcomes of interest were in-hospital intubation and in-hospital mortality. Intubation was defined as mechanical ventilation during the hospital stay based on the presence of relevant orders and procedure codes. In-hospital mortality was defined as deaths occurring during hospitalization recorded in the hospital EHR or reflected in the diagnosis-related group.

Patient characteristics

We reviewed demographics, baseline comorbidities, and vital signs at admission. Demographics included age, gender, race (white or non-white), and ethnicity (Hispanic or non-Hispanic). Diagnosis codes set^ten were used to identify baseline comorbidities including hypertension, diabetes, coronary artery disease, heart failure, chronic obstructive pulmonary disease, asthma, cancer, obesity, and hyperlipidemia. Vital signs that were securely captured by participating health systems included systolic and diastolic blood pressure and body mass index (BMI) at admission.

statistical analyzes

To predict the two binary outcomes (intubation and mortality), we considered two methods: logistic regression and random forests (RF). First, we constructed a sequence of models with logistic regression to assess the incremental effect of each group of predictors. Model 1 included demographics, comorbidity, and vital signs; we added the SDI quintile to build model 2, added the time since the start of the pandemic to build model 3, and finally added an SDI interaction by time to build model 4.

Similar models (except model 4) were built with Random Forests, a machine learning algorithm that automatically models the complex interactions between predictors. The performance of the model was estimated by the AUROC curve (Area under the receiver operating characteristic) using a fivefold cross-validation and its 95% confidence interval. Missing data in predictors (range 3.6–12%) were imputed with random forest¹¹ to produce an imputed data set.

Ethical information

Weill Cornell IRB (#20-04021948) has approved this study and has determined that this study meets the waiver requirements of HHS 45 CFR 46.104(d). All data management and analysis was performed in a HIPAA-compliant manner.