Identifying social determinants of health subgroups and their associations with health outcomes in a prospective US adult COVID-19 cohort: an analysis of the INSPIRE registry

Objective: To identify subgroups with similar social determinants of health (SDOH) characteristics using latent class analysis (LCA) and examine their associations with physical and mental health, cognitive function and missed workdays at 3 and 6 months post-SARS-CoV-2 infection. We hypothesised that intersecting SDOH factors would differentially influence COVID-19-related health outcomes across subgroups. Design: Prospective cohort study from the Innovative Support for Patients with SARS-CoV-2 Infections Registry (INSPIRE), with longitudinal data collection and cross-sectional analyses at baseline, 3-month and 6-month follow-ups. Setting: Multicentre registry across eight US academic medical centres (Chicago, Dallas, Houston, Los Angeles, New Haven, Philadelphia, San Francisco and Seattle). Participants: Adults aged ≥18 years, fluent in English or Spanish, with self-reported acute COVID-19 symptoms and a confirmed positive SARS-CoV-2 test within 42 days before enrolment (9 December 2020 to 12 August 2022), and access to an internet-connected device. Exclusions included incarceration, inability to provide informed consent, lack of confirmed SARS-CoV-2 infection or no internet access. Of 3791 eligible participants with complete baseline data, 2897 (76.4%) completed the 3-month follow-up and 2666 (70.3%) completed the 6-month follow-up; most were aged 18–49 years (74–75%), female (66–67%), white (86.6–87.5%) and non-Hispanic (86.6–87.5%). Outcome measures: Prespecified primary outcomes were physical and mental health (Patient-Reported Outcomes Measurement Information System (PROMIS)-29 V.2.1 T-scores for depression, anxiety, fatigue, sleep disturbance, pain interference, physical function and social participation), cognitive function (PROMIS Cognitive Function Short Form 8 T-scores) and missed workdays due to illness (binary: >1 week vs ≤1 week, from a single-item survey). All measures were self-reported and collected at baseline, 3 months and 6 months; no changes from protocol. Results: LCA identified a 4-class model as optimal (lowest Bayesian Information Criterion (BIC) after evaluating 1–7 class models; significant demographic differences (χ² p<0.001 for gender, age, race, ethnicity, education and income distributions)). Sensitivity analysis using four age groups (18–29, 30–49, 50–64, ≥65 years) yielded higher BIC (45 430.8) than three groups (18–49, 50–64, ≥65 years; BIC=42 150.9), confirming the primary model. Class 3 (middle-aged, high-income non-Hispanic; n=599 (15.8%)) was the reference group. Compared with Class 3, Class 1 (lower-income, predominantly Hispanic young to middle-aged adults; n=499 (13.2%)) and Class 2 (lower-income, older, predominantly Black non-Hispanic; n=1828 (48.2%)) exhibited significantly worse outcomes across physical and mental health domains (eg, Class 1 3-month anxiety β=4.41 (95% CI 3.25 to 5.56, p<0.001); Class 2 3-month depression β=3.58 (95% CI 2.53 to 4.64, p<0.001)). Classes 1 and 2 also reported significantly worse cognitive function at both time points (eg, Class 1 3-month β=−3.29 (95% CI −4.77 to −1.82, p<0.001)) and Class 2 had significantly higher odds of missed workdays at 6 months (OR=1.853 (95% CI 1.192 to 2.880, p=0.006)). In contrast, Class 4 (young to middle-aged, highly educated, high-income non-Hispanic; n=865 (22.8%)) consistently reported the most favourable outcomes, including better physical function (3-month β=2.04, p<0.001) and lower pain interference compared with the reference group. Conclusions: In this US prospective cohort, SDOH-based subgroups showed persistent disparities in health outcomes post-SARS-CoV-2 infection. Findings highlight the urgent need for intersectional approaches to address systemic inequities in post-COVID-19 recovery. Trial registration number: NCT04610515.