Work

Evaluating social influences as risk factors of type 2 diabetes subtypes and long COVID comorbidities: An application of machine learning in a retrospective outpatient cohort

Year: 2026

Type: article

Abstract: Background: Underlying type 2 diabetes mellitus (T2D) is a risk factor for coronavirus disease 2019 (COVID-19) severity, including hospitalization and invasive mechanical ventilation. A critical research gap is determining the specific influences of T2D on long COVID risk, accounting for extensively heterogeneous clinical manifestations in diabetes and key social factors. Among participants with T2D, T1D, impaired fasting glucose (IFG), or impaired glucose tolerance (IGT) in the US, our study objectives are to: 1) elucidate the impacts of T2D on the probability of long COVID, including considering social factors (individually, index) as effect modifiers; and 2) assess the impacts of social factors (individually, index) on the likelihood of T2D complication subtypes. Methods: In this retrospective cohort study, our inclusion criteria were any individual with available data in the OCHIN database (2012-2023) and ≥ 1 diagnosis or treatment-related ICD code for IFG, IFT, T1D, or T2D. The OCHIN database includes electronic health records of outpatient visits at community-based health centers. We excluded participants with incomplete data for key independent variables (T2D indicators and complications), dependent variables (long COVID during 2019-2023), and covariates. Due to evolving definitions of long COVID throughout the study duration, we consider four long COVID definitions based on at-risk periods (population-, individual-level) and ICD-9 and -10 codes (e.g., U09.9). We will define T2D complication subtypes with unsupervised clustering (e.g., k-means) approaches. Social influences will be proxied by individual-level (federal poverty level, self-reported lack of housing) and geospatial- (residential zip code-based environmental justice indices, GINI coefficient [income inequality]) variables. We will randomly split the analytic dataset into discovery (70%) and final hold-out validation (30%) cohorts. We will use binomial or Poisson multivariable regressions to evaluate associations between type 2 diabetes (overall, subtypes, relative to no diabetes, and long COVID risk, adjusting for key covariates (e.g., biological sex, race-ethnicity, age, vaccinations). Discussion: Our analysis is in progress, however anticipated findings are two-fold. First, we will identify which specific T2D subtypes affect long COVID risk, including overall and specific symptoms. These results could provide tailored insights that improve the detection of long COVID, and inform future therapeutic studies. Second, we will identify specific social factors affect T2D and comorbidities, particularly long COVID. Our expected findings could provide key insights for policymakers, particularly regarding major social needs that should be prioritized, in light of the additional benefits of reducing adverse consequences due to T2D and long COVID, respectively.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.19456125

Authors: Yu, Elaine A., Jackman, Rachael, Musaad, Salma

Institutions: University of California, San Francisco, Vitalant Research Institute

Language: English

Topic: Long-Term Effects of COVID-19

Subfield: Neurology

Field: Medicine

Domain: Health Sciences

SDGs: Good health and well-being, Reduced inequalities

Open Access status: green

Funder: National Institutes of Health