Multicenter Study of Outcomes Among Persons With HIV Who Presented to US Emergency Departments With Suspected SARS-CoV-2

Christopher L. Bennett, MD, MA; Emmanuel Ogele, MD; Nicholas R. Pettit, DO, PhD; Jason J. Bischof, MD; Tong Meng, MPP; Prasanthi Govindarajan, MBBS, MAS; Carlos A. Camargo, Jr, MD, DrPH; Kristen Nordenholz, MD, MSC; Jeffrey A. Kline, MD


J Acquir Immune Defic Syndr. 2021;88(4):406-413. 

In This Article


This study was a planned secondary analysis of a multicenter registry of patients from 45 medical centers, with 116 EDs in 27 US states, the REgistry of suspected COVID-19 in EmeRgency care (RECOVER) Network.[15,16] Our December 2020 data cut of the registry included 25,721 unique patients with a qualifying emergency department (ED) visit at participating medical centers for suspected SARS-CoV-2 infection and an accompanying polymerase chain reaction based test for SARS-CoV-2. Patients with ED visits lacking a reasonable probability of being related to SARS-CoV-2 (eg, trauma, alcohol or drug intoxication, or testing performed purely for admission policy) were excluded from enrollment in the registry; all entries in the registry had a minimum of 30-day follow-up information.[15,16]

The registry used a REDCap platform ( with 204 questions among 7 domains (visit information, demographics, symptoms and risk factors, vital signs, medical history, current medications, test results, and outcomes). As described previously, the platform used programming to force data entry for critical fields, perform error checking, and ensure sensible alpha-numeric content and ranges for numeric data.[15,16] Data were obtained by abstractors using the local electronic health record at each study site and supported by an administrative core and a steering committee. Training of data abstractors was conducted through teleconference with the principal investigator and supplemented by a program manager and extensive guidance documentation; this included guidance documentation embedded in the REDCap.[15,16]

The protocol for the RECOVER Network was reviewed by the institutional review board at all participating sites. Detailed methodology (including study design, setting, registry development, patient selection, and other characteristics) for the RECOVER Network has been described elsewhere.[15,16] The population of interest for this study was patients in the registry with a documented HIV infection or AIDS. Patient data was stratified by SARS-CoV-2 and HIV test results—positive and negative.

Values were summarized and presented with descriptive statistics including medians with interquartile range and proportions. We used χ2 tests and unpaired Student t tests to compare clinical characteristics and outcomes between the stratified groups and to compare patients with HIV stratified by SARS-CoV-2 status. To explore factors associated with our primary and secondary outcomes, which include death, intubation, and hospital length of stay, we first calculated unadjusted odds ratios (ORs) and corresponding 95% confidence intervals (CIs) to determine whether the odds of experiencing the outcomes varied by factor; factors explored included age, sex, race, smoking status, obesity (defined by a body mass index greater than 35 kg/m2), insured status/type, and presence of a do not resuscitate order. Factors were selected based on previous studies and clinical judgment.[15–18] We then constructed multivariable logistic regression models to further identify associations between the aforementioned factors and outcomes. Given that death was our primary outcome of interest, we also constructed Kaplan–Meier curves comparing survival probabilities for patients with HIV (stratified by SARS-CoV-2–positive and SARS-CoV-2–negative status) and survival probabilities for patients with SARS-CoV-2 (stratified by HIV-positive and HIV-negative status). Log-rank tests were used to compare survival distributions. Per our a priori protocol, categorical data that were not charted were considered absent and not imputed.[15,16] Missing (>0.1%) continuous data (ie, age, vital signs, and body mass index) were analyzed for monotonicity and replaced using multiple imputation.[15,16] Furthermore, we tested for collinearity but did not identify any variables that were colinear; we did not test for interaction. We were unable to include information about HIV treatment status and CD4 count in our analysis or modeling given a high degree of missingness (46% were missing); these data were not imputed. An α level of <0.01 was considered significant, and all analyses were completed using SAS software (SAS Institute Inc., Cary, NC).[19]