Ожирение и риск эпидемических подстрекательских заболеваний: исследование реальных тенденций с использованием электронных медицинских записей.

Introduction

Obesity is a growing global health challenge, characterized by excessive adiposity and its association with numerous chronic diseases 1 , 2 . Beyond its well-established link to cardiometabolic disorders, obesity is increasingly recognized as a state of chronic low-grade inflammation, characterized by infiltration of pro-inflammatory M1-polarized macrophages into adipose tissue, elevated secretion of leptin and pro-inflammatory cytokines, and reduced levels of immunoregulatory adipokine adiponectin 3 – 5 . These alterations disrupt immune homeostasis, promote autoantibody production, and impair the function of regulatory T and B cells—mechanisms implicated in the initiation and propagation of autoimmunity 3 – 5 .

Epidemiologic evidence supports an association between obesity and an increased risk of several immune-mediated inflammatory diseases (IMIDs), such as rheumatoid arthritis, psoriasis, autoimmune thyroiditis, type 1 diabetes mellitus, and other related conditions 6 – 9 . However, for other IMIDs, reported associations have been inconsistent, null, or even inverse. For example, obesity was associated with an increased risk of systemic lupus erythematosus among younger women in the Nurses’ Health Study II, but this association was not observed in older cohorts, and a meta-analysis reported a non-significant 49% overall increase in risk 10 . A Danish cohort study found no association between obesity and systemic lupus erythematosus risk 8 , while the Black Women’s Health Study indicated that only adolescent obesity was associated with increased risk among Black women 11 . No significant associations have been reported between obesity and Sjögren’s syndrome 8 , 12 or ulcerative colitis 6 , 13 . In contrast, population-based studies have suggested inverse associations with certain conditions, including Graves’ disease 14 , celiac disease 8 , and Raynaud’s phenomenon 8 . These heterogeneous patterns highlight the importance of distinguishing shared from disease-specific immunopathogenic mechanisms.

Most previous studies have focused on a limited subset of common IMIDs and have rarely examined rarer conditions such as systemic sclerosis, dermatomyositis, or systemic vasculitis. Furthermore, few investigations have utilized large-scale real-world data with rigorous control for confounding through matched cohort designs. The potential modifying effects of demographic factors, such as age, sex, and race, also remain insufficiently explored.

To address these gaps, we conducted a large-scale real-world cohort study using the TriNetX global health research network to evaluate the association between obesity and the risk of incident IMIDs. Our study employed stringent cohort definitions, comprehensive adjustment for confounders, control outcome validation, and stratified and dose–response analyses, aiming to provide a robust and comprehensive assessment of these associations.

Methods

Study design and data source

We conducted a retrospective, propensity score-matched cohort study using data from TriNetX, a global federated health research network encompassing de-identified electronic health records from over 300 million patients across more than 120 healthcare organizations worldwide. The database includes comprehensive clinical information such as demographics, diagnoses (coded using the International Classification of Diseases, Tenth Revision, Clinical Modification [ICD-10-CM]), procedures, medications, laboratory results, and healthcare utilization 15 . All methods were carried out in accordance with relevant guidelines and regulations. This study utilized de-identified electronic health record data obtained from the TriNetX research network. The use of such anonymized data has been determined by the Western Institutional Review Board to be exempt from Institutional Review Board approval, and the need to obtain informed consent was waived, as it does not involve research on human subjects as defined by federal regulations and institutional policies. This study was granted exemption from full ethical review and waiver of informed consent by the Institutional Review Board of Chi Mei Medical Center (Applicant No.: 11403-E02) on March 17, 2025. Data were accessed on June 10, 2025, and included retrospective records from January 1, 2015, to December 31, 2024.

Study population

Adult patients aged ≥ 18 years with at least 3 clinical encounters and a recorded body mass index (BMI) ≥ 18.5 kg/m 2 between January 1, 2015, and December 31, 2019, were included. To reduce potential misclassification from pregnancy-related weight fluctuations, individuals with relevant pregnancy-related diagnosis codes within 9 months before or after the index date were excluded. Patients with prior diagnoses of any IMIDs, as defined below, were also excluded. Individuals with documented death prior to the index date were excluded. The first qualifying BMI measurement was designated as the index date.

Exposure definition

Patients were classified as obese (BMI ≥ 30 kg/m 2 ) or non-obese (BMI 18.5–29.9 kg/m 2 ) based on their index date measurement. To ensure classification accuracy and account for weight stability, a confirmatory BMI within the same category was required within 1 to 60 days following the index date.

Covariates and propensity score matching

To minimize confounding, 1:1 propensity score matching was conducted within the TriNetX platform using greedy nearest-neighbor matching with a fixed caliper of 0.1, which is predefined by the system and not modifiable by the user. Covariates were selected based on their potential as confounders or established risk factors for IMIDs, including demographic factors (age, sex, and race), common comorbidities (e.g., hypertension, dyslipidemia), and healthcare utilization metrics to minimize surveillance bias. Baseline covariates were evaluated during the 12 months prior to the index date. All covariates were identified using ICD-10 diagnosis codes unless otherwise specified. The following covariates were included:

Demographics: Age at index, sex, race/ethnicity, and indicators of adverse socioeconomic status (based on ICD-10 codes related to housing, employment, and education)

Lifestyle proxies: Tobacco use and nicotine dependence (as proxies for smoking), and alcoholic liver disease (as a proxy for alcohol consumption)

Comorbidities: Hypertension, type 2 diabetes mellitus, hyperlipidemia, chronic kidney disease, asthma, depression, sleep disorders, psychoactive substance use, vitamin D deficiency, neoplasms, chronic obstructive pulmonary disease, chronic viral infections (hepatitis B virus, hepatitis C virus, and human immunodeficiency virus), anxiety, non-alcoholic fatty liver disease, and chronic ischemic heart disease

Healthcare utilization: ICD-10 codes indicating factors influencing health status and contact with health services

In the TriNetX platform, participants with missing categorical covariates (e.g., sex or race/ethnicity) were retained by assigning an explicit ‘unknown’ category in the propensity score model to minimize selection bias and preserve representativeness in real-world electronic health record data. In contrast, patients missing continuous variables required for exposure classification or matching (e.g., BMI) were excluded from the matched analysis. No imputation is performed by the platform. A full list of diagnostic codes is provided in Supplementary Table 1.

Outcomes

A total of 15 IMIDs were selected as primary outcomes. Baseline exclusions were applied to ensure the identification of incident cases. IMIDs were ascertained using ICD-10-CM codes according to our pre-specified definitions (Supplementary Table 1). Comparable ICD-10-based definitions within the TriNetX platform have been employed in prior studies investigating chronic inflammatory and autoimmune conditions 16 , 17 . Each IMID was analyzed as a distinct incident outcome using a separate Cox proportional hazards model. For each outcome, participants with a diagnosis of the specific IMID prior to the index date were excluded. Participants could therefore contribute to more than one outcome analysis if they were free of the respective IMIDs at baseline. For analysis, IMIDs were classified into four categories:

Systemic: Rheumatoid arthritis, systemic lupus erythematosus, Sjögren’s syndrome, systemic sclerosis, dermatopolymyositis, sarcoidosis, and systemic vasculitis

Endocrine: Graves’ disease, autoimmune thyroiditis, and type 1 diabetes mellitus

Cutaneous and musculoskeletal: Psoriasis, psoriatic arthritis, and ankylosing spondylitis

Neurological: Multiple sclerosis and myasthenia gravis

To assess internal validity and detect residual confounding, we included control outcomes:

Positive controls (conditions known to be associated with obesity): Gastroesophageal reflux disease, sleep apnea, and knee osteoarthritis

Negative controls (conditions not known to be associated with obesity): Allergic contact dermatitis, acute appendicitis, and hemorrhoids

To minimize detection bias and reduce the risk of reverse causality, events occurring within 90 days of the index date were excluded. Follow-up began on day 91 post-index and continued until the first occurrence of the outcome, death, loss to follow-up, or the end of the study period (maximum of 5 years). A full list of diagnostic codes is provided in Supplementary Table 1.

Statistical analyses

Baseline characteristics were compared using Student’s t -tests for continuous variables and Pearson’s chi-squared tests for categorical variables. Covariate balance after matching was evaluated using standardized mean differences (SMDs), with values < 0.1 indicating adequate balance.

Incidence rates were reported as events per 1,000 person-years. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). The proportional hazards assumption was tested using the generalized Schoenfeld approach built in the TriNetX platform. For outcomes violating the assumption ( p  < 0.05), a piecewise Cox regression was applied to account for time-varying effects.

To reduce reverse causality, outcome events occurring within 90 days after the index date were excluded. For outcomes requiring piecewise modeling, the remaining follow-up time was divided into three prespecified intervals (day 91–365, day 366–1095, and day 1096–1825), and interval-specific hazard ratios were estimated to evaluate temporal stability of associations.

Separately, sensitivity analyses were conducted by censoring follow-up at 1, 3, and 5 years, irrespective of proportional hazards assumption status, to assess the robustness of associations to different maximum follow-up durations. Additional analyses excluded events occurring within 6 and 12 months after the index date to further reduce potential reverse causality, particularly for conditions with insidious onset, such as systemic lupus erythematosus.

Stratified analyses were conducted to assess potential effect modification by age, sex, and race. Heterogeneity of HRs across strata was tested using Cochran’s Q test, with standard errors calculated from the reported 95% CIs. These analyses assumed independent, similarly adjusted stratum-specific estimates and were performed in Python 3.11.

Dose–response analyses were conducted across BMI categories: overweight (25–29.9 kg/m 2 ), obesity class 1 (30–34.9 kg/m 2 ), class 2 (35–39.9 kg/m 2 ), and class 3 (≥ 40 kg/m 2 ), using normal BMI (18.5–24.9 kg/m 2 ) as the reference. A test for linear trend was performed using weighted least squares regression of the natural logarithm of HRs (log[HR]) on the mean BMI for each category, modeled as a continuous variable. Inverse-variance weights (1/SE 2 of log[HR]) were applied to account for the precision of each estimate.

All statistical tests were two-sided, with P values < 0.05 considered statistically significant. Positive and negative control outcomes were analyzed using the same modeling strategies to evaluate the specificity and robustness of observed associations.

Definitions for all exposures, outcomes, covariates, and exclusion criteria were based on ICD-10-CM coding algorithms where available. The complete list of codes is provided in Supplementary Table 1.

Results

Study cohort and patient selection

A total of 5,160,431 adult individuals with at least 3 clinical encounters and a BMI ≥ 18.5 kg/m 2 were identified during the study period. After applying eligibility criteria and exclusions, including absence of confirmatory BMI, pregnancy-related diagnoses, pre-existing IMIDs, and pre-index death, 720,065 patients were classified in the obese cohort and 1,244,761 in the non-obese cohort. Propensity score matching yielded 698,624 matched pairs (1:1 ratio), which comprised the final analytic population (Fig.  1 ).

Cohort selection flowchart. Abbreviations BMI, body mass index; IMID, immune-mediated inflammatory disease.

Baseline characteristics

Table 1 presents the baseline characteristics of the obese and non-obese cohorts before and after propensity score matching. After matching, all covariates achieved excellent balance, with SMDs < 0.1 across demographic, socioeconomic, lifestyle, and comorbidity variables. The mean age was 51.7 years in the obese cohort and 52.9 years in the non-obese cohort, while the proportion of males was 43.7% and 45.2%, respectively. Mean BMI values were 36.1 kg/m 2 for the obese cohort and 25.1 kg/m 2 for the non-obese cohort. Full pre- and post-matching covariate distributions and SMDs are detailed in Table 1 .

Table 1 Baseline characteristics of study subjects before and after propensity score matching Variables Before matching, No (%) After matching, No (%) Obese Non-obese Obese Non-obese BMI, mean ± SD, kg/m 2 36.3 ± 6.1 24.8 ± 3.1 SMD 36.1 ± 5.9 25.1 ± 3.1 SMD (n = 720,065) (n = 1,244,761) (n = 698,624) (n = 698,624) Age at index, mean ± SD, years 51.9 ± 16.7 50.0 ± 21.0 0.099 51.7 ± 16.8 52.9 ± 17.7 0.068 Sex  Male 313,360 (43.5) 570,665 (45.8) 0.047 305,633 (43.7) 315,812 (45.2) 0.029  Female 406,705 (56.5) 674,096 (54.2) 0.047 392,991 (56.3) 382,812 (54.8) 0.029 Race  White 500,550 (69.5) 858,357 (69.0) 0.012 489,184 (70.0) 490,999 (70.3) 0.006  Black or African American 106,627 (14.8) 116,261 (9.3) 0.168 98,032 (14.0) 94,808 (13.6) 0.013  Hispanic or Latino 65,502 (9.1) 101,967 (8.2) 0.032 63,905 (9.1) 66,834 (9.6) 0.014  Asian 10,288 (1.4) 64,305 (5.2) 0.210 10,283 (1.5) 10,693 (1.5) 0.005 Adverse socioeconomic status  Housing/Economic circumstances 544 (0.1) 1,427 (0.1) 0.013 518 (0.1) 525 (0.1)  < 0.001  Employment/Unemployment problems 255 (0.0) 462 (0.0) 0.001 233 (0.0) 210 (0.0) 0.002  Education and literacy problems 162 (0.0) 392 (0.0) 0.005 145 (0.0) 177 (0.0) 0.003 Lifestyle  Tobacco use (smoking) 1,952 (0.3) 3,384 (0.3)  < 0.001 1,774 (0.3) 1,660 (0.2) 0.003  Nicotine dependence (smoking) 16,263 (2.3) 29,662 (2.4) 0.008 15,056 (2.2) 15,009 (2.1)  < 0.001  Alcoholic liver disease (alcohol use) 908 (0.1) 2,068 (0.2) 0.010 872 (0.1) 898 (0.1) 0.001 Comorbidities  Hypertension 86,672 (12.0) 94,667 (7.6) 0.149 71,478 (10.2) 70,931 (10.2) 0.003  Type 2 diabetes mellitus 44,398 (6.2) 33,191 (2.7) 0.171 30,865 (4.4) 29,773 (4.3) 0.008  Hyperlipidemia 65,149 (9.0) 80,137 (6.4) 0.098 54,644 (7.8) 54,441 (7.8) 0.001  Chronic kidney disease 13,116 (1.8) 18,854 (1.5) 0.024 11,456 (1.6) 11,289 (1.6) 0.002  Asthma 18,749 (2.6) 20,895 (1.7) 0.064 15,079 (2.2) 14,165 (2.0) 0.009  Depression 23,393 (3.2) 29,948 (2.4) 0.051 19,817 (2.8) 19,950 (2.9) 0.001  Sleep disorder 32,049 (4.5) 23,893 (1.9) 0.145 22,283 (3.2) 22,235 (3.2)  < 0.001  Psychoactive substance use 21,689 (3.0) 42,211 (3.4) 0.022 20,276 (2.9) 20,055 (2.9) 0.002  Vitamin D deficiency 17,186 (2.4) 20,924 (1.7) 0.050 14,173 (2.0) 14,087 (2.0) 0.001  Neoplasms 62,533 (8.7) 113,937 (9.2) 0.016 59,262 (8.5) 60,744 (8.7) 0.008  Chronic obstructive pulmonary disease 9,947 (1.4) 14,371 (1.2) 0.020 8,603 (1.2) 8,251 (1.2) 0.005  Chronic hepatitis B 155 (0.0) 615 (0.0) 0.015 144 (0.0) 201 (0.0) 0.005  Chronic hepatitis C 1,412 (0.2) 3,049 (0.2) 0.010 1,332 (0.2) 1,280 (0.2) 0.002  Human immunodeficiency virus disease 1,222 (0.2) 3,537 (0.3) 0.024 1,197 (0.2) 1,073 (0.2) 0.004  Anxiety 21,356 (3.0) 34,473 (2.8) 0.012 19,348 (2.8) 19,544 (2.8) 0.002  Non-alcoholic fatty liver disease 4,726 (0.7) 2,913 (0.2) 0.063 2,786 (0.4) 2,670 (0.4) 0.003  Chronic ischemic heart disease 20,622 (2.9) 30,984 (2.5) 0.023 18,445 (2.6) 18,517 (2.7) 0.001 Contact with health services 163,252 (22.7) 247,654 (21.5) 0.028 150,208 (21.5) 147,534 (21.1) 0.009 Bold font represents a SMD > 0.1. BMI, body mass index; SD, standard deviation; SMD, standardized mean difference

Baseline characteristics of study subjects before and after propensity score matching

Bold font represents a SMD > 0.1.

BMI, body mass index; SD, standard deviation; SMD, standardized mean difference

Association between obesity and immune-mediated inflammatory diseases

Table 2 summarizes the incidence rates and HRs for each IMID and control outcomes. Obesity was significantly associated with an increased risk for multiple IMIDs. The strongest associations were observed for psoriatic arthritis (HR 1.74, 95% CI 1.59–1.89) and type 1 diabetes mellitus (HR 1.41, 95% CI 1.36–1.47). Among systemic IMIDs, risks of rheumatoid arthritis (HR 1.20, 95% CI 1.16–1.25), systemic lupus erythematosus (HR 1.15, 95% CI 1.07–1.22), and sarcoidosis (HR 1.33, 95% CI 1.22–1.45) were significantly elevated in the obese group.

Table 2 Incidence rates and hazard ratios for immune-mediated inflammatory diseases and control outcomes. Outcomes Obese Non-obese Events Incidence rate Events Incidence rate HR (95% CI) (/1,000 PY, 95% CI) (/1,000 PY, 95% CI) Systemic IMIDs  Rheumatoid arthritis 7,143 2.72 (2.66–2.78) 5,782 2.26 (2.20–2.32) 1.20 (1.16–1.25)  Systemic lupus erythematosus 2,001 0.76 (0.73–0.79) 1,699 0.66 (0.63–0.69) 1.15 (1.07–1.22)  Sjögren’s syndrome 2,470 0.94 (0.90–0.97) 2,804 1.09 (1.05–1.13) 0.86 (0.81–0.90)  Systemic sclerosis 323 0.12 (0.11–0.14) 420 0.16 (0.15–0.18) 0.75 (0.65–0.87)  Dermatopolymyositis 231 0.09 (0.08–0.10) 242 0.09 (0.08–0.11) 0.93 (0.78–1.11)  Sarcoidosis 1,265 0.48 (0.45–0.51) 927 0.36 (0.34–0.38) 1.33 (1.22–1.45)  Systemic vasculitis 1,227 0.47 (0.44–0.49) 1,174 0.46 (0.43–0.48) 1.02 (0.94–1.10) Endocrine IMIDs  Graves’ disease 1,344 0.51 (0.48–0.54) 1,349 0.53 (0.50–0.55) 0.97 (0.90–1.05)  Autoimmune thyroiditis 4,191 1.59 (1.54–1.64) 3,733 1.46 (1.41–1.50) 1.09 (1.05–1.14) *  Type 1 diabetes mellitus 5,548 2.11 (2.05–2.17) 3,834 1.50 (1.45–1.55) 1.41 (1.36–1.47) * Cutaneous & musculoskeletal IMIDs  Psoriasis 5,724 2.18 (2.12–2.23) 4,042 1.58 (1.53–1.63) 1.38 (1.33–1.44)  Psoriatic arthritis 1,442 0.55 (0.52–0.58) 807 0.31 (0.29–0.34) 1.74 (1.59–1.89)  Ankylosing spondylitis 781 0.30 (0.28–0.32) 653 0.25 (0.24–0.27) 1.16 (1.05–1.29) Neurological IMIDs  Multiple sclerosis 1,056 0.40 (0.38–0.43) 1,030 0.40 (0.38–0.43) 1.00 (0.92–1.09)  Myasthenia gravis 422 0.16 (0.14–0.18) 376 0.15 (0.13–0.16) 1.09 (0.95–1.25) Positive control outcomes  Gastroesophageal reflux disease 72,218 32.68 (32.44–32.92) 58,871 26.51 (26.30–26.72) 1.26 (1.25–1.27) *  Sleep apnea syndrome 59,349 25.55 (25.34–25.75) 20,374 8.25 (8.14–8.36) 3.25 (3.19–3.30) *  Knee osteoarthritis 43,256 17.48 (17.31–17.64) 24,562 9.90 (9.78–10.03) 1.81 (1.78–1.83) * Negative control outcomes  Allergic contact dermatitis 5,213 1.98 (1.93–2.04) 4,960 1.94 (1.89–1.99) 1.02 (0.98–1.06)  Acute appendicitis 1,753 0.67 (0.64–0.70) 1,719 0.67 (0.64–0.71) 0.99 (0.93–1.06)  Hemorrhoids 10,114 3.88 (3.81–3.96) 9,855 3.90 (3.82–3.97) 1.00 (0.97–1.02) Bold font indicates statistically significant results ( P  < 0.05). IMID, immune-mediated inflammatory disease; HR, hazard ratio; CI, confidence interval; PY, person-years. *Proportionality ( P  < 0.05).

Incidence rates and hazard ratios for immune-mediated inflammatory diseases and control outcomes.

Bold font indicates statistically significant results ( P  < 0.05).

IMID, immune-mediated inflammatory disease; HR, hazard ratio; CI, confidence interval; PY, person-years.

*Proportionality ( P  < 0.05).

Conversely, obesity was inversely associated with Sjögren’s syndrome (HR 0.86, 95% CI 0.81–0.90) and systemic sclerosis (HR 0.75, 95% CI 0.65–0.87). No significant associations were observed for dermatopolymyositis (HR 0.93, 95% CI 0.78–1.11), systemic vasculitis (HR 1.02, 95% CI 0.94–1.10), and neurological IMIDs such as multiple sclerosis (HR 1.00, 95% CI 0.92–1.09) and myasthenia gravis (HR 1.09, 95% CI 0.95–1.25).

Positive control outcomes confirmed expected associations with obesity, including gastroesophageal reflux disease (HR 1.26, 95% CI 1.25–1.27), sleep apnea (HR 3.25, 95% CI 3.19–3.30), and knee osteoarthritis (HR 1.81, 95% CI 1.78–1.83). Conversely, negative control outcomes, such as allergic contact dermatitis (HR 1.02, 95% CI 0.98–1.06), appendicitis (HR 0.99, 95% CI 0.93–1.06), and hemorrhoids (HR 1.00, 95% CI 0.97–1.02), showed no significant differences between groups. These findings support the specificity of the observed IMID risk associations and affirm the internal validity of our analytic design.

Two IMID outcomes (autoimmune thyroiditis and type 1 diabetes mellitus) and three positive control outcomes (gastroesophageal reflux disease, sleep apnea syndrome, and knee osteoarthritis) violated the proportional hazards assumption ( p  < 0.05). For these outcomes, HRs were estimated separately for three follow-up intervals (day 91–365, 366–1095, and 1096–1825). Results were consistent across intervals, supporting the robustness of the associations (Supplementary Table 2).

Sensitivity analyses

Associations between obesity and IMIDs remained consistent across follow-up durations of 1, 3, and 5 years, supporting the robustness of the findings (Supplementary Table 3). For example, the HR for rheumatoid arthritis remained stable over time (1.22 at 1 year, 1.20 at 3 years, and 1.20 at 5 years).

Additional sensitivity analyses applying varying post-index exclusion windows (90, 180, and 365 days) yielded consistent results across IMIDs, with only minor variations in effect size (Supplementary Table 4). Key positive associations remained stable, negative controls showed null effects, and positive controls confirmed expected risks, supporting the robustness and internal validity of our findings.

Stratified analyses

Stratified analyses for 10 IMIDs with significant primary associations revealed demographic variations in obesity-related risk (Fig.  2 – 4 ; Supplementary Tables 5–7).

Age-stratified hazard ratios for incident immune-mediated inflammatory diseases.

Race-stratified hazard ratios for incident immune-mediated inflammatory diseases.

By age Strongest associations occurred in middle-aged adults (41–64 years), notably rheumatoid arthritis (HR 1.32, 95% CI 1.25–1.40) and psoriatic arthritis (HR 1.82, 95% CI 1.61–2.06). Older adults (≥ 65 years) showed attenuated risks for systemic IMIDs, while younger adults (18–40 years) exhibited prominent risks for endocrine IMIDs, including type 1 diabetes (HR 1.47, 95% CI 1.26–1.71) and autoimmune thyroiditis (HR 1.23, 95% CI 1.09–1.38) (P for interaction < 0.05) (Fig.  2 ; Supplementary Table 5).

By sex Females had stronger associations for rheumatoid arthritis (HR 1.23, 95% CI 1.18–1.28), psoriatic arthritis (HR 1.86, 95% CI 1.67–2.08), and type 1 diabetes (HR 1.56, 95% CI 1.47–1.66), whereas males exhibited modestly lower risks for these conditions (P for interaction < 0.05). Inverse associations for Sjögren’s syndrome and systemic sclerosis were similar across sexes (Fig.  3 ; Supplementary Table 6).

Sex-stratified hazard ratios for incident immune-mediated inflammatory diseases.

By race Significant heterogeneity was observed for systemic lupus erythematosus, Sjögren’s syndrome, sarcoidosis, autoimmune thyroiditis, and type 1 diabetes (P for interaction < 0.05). White patients showed consistently elevated risks, especially for type 1 diabetes (HR 1.52, 95% CI 1.44–1.60), psoriatic arthritis (HR 1.76, 95% CI 1.60–1.93), and sarcoidosis (HR 1.43, 95% CI 1.28–1.60). Black/African American and Latino patients had elevated risks for type 1 diabetes, psoriasis, and psoriatic arthritis, but more variable associations for systemic IMIDs (Fig.  4 ; Supplementary Table 7).

Dose–response analyses and trend test

Among the 10 IMIDs with significant associations in the primary analysis, rheumatoid arthritis and psoriasis demonstrated clear dose–response relationships, with risk increasing steadily across BMI categories (P for trend < 0.05) (Supplementary Table 8). Other outcomes showed stepwise or plateaued patterns, suggesting possible dose-related effects without a strictly linear trend.

Discussion

In this large-scale cohort study of over 1.3 million propensity score-matched adults, obesity was significantly associated with increased risks of several IMIDs, particularly rheumatoid arthritis, psoriatic arthritis, psoriasis, and type 1 diabetes mellitus. These associations remained robust across sensitivity analyses with varying follow-up durations and exclusion windows. The expected positive control associations and null findings for negative controls support the internal validity of our results. A clear dose–response gradient across BMI categories further strengthens the observed relationships.

While statistically significant linear trends were seen only for rheumatoid arthritis and psoriasis, most IMIDs showed progressive or plateaued HR increases across BMI strata. Using weighted regression of log(HR) on mean BMI allowed more precise dose–response estimation, though non-linear or threshold effects may not be fully captured.

These findings align with mechanistic evidence that excess adiposity promotes a pro-inflammatory immune environment. Obesity is associated with infiltration of pro-inflammatory M1-polarized macrophages, suppression of regulatory T cells, and a shift toward Th17-mediated immune responses 3 – 5 . Elevated leptin levels enhance dendritic cell activation and pro-inflammatory cytokine secretion 18 , whereas reduced adiponectin diminishes immunoregulatory effects. Additionally, dietary factors such as the Western diet may contribute to gut dysbiosis, further disrupting immune homeostasis 19 .

Not all IMIDs showed positive associations with obesity. Inverse or null associations were observed for Sjögren’s syndrome, systemic sclerosis, dermatomyositis, systemic vasculitis, multiple sclerosis, and myasthenia gravis. These divergent patterns likely reflect a combination of disease-specific immunopathology and non-causal factors. Mechanistically, excess adiposity promotes a pro-inflammatory milieu characterized by increased leptin, reduced adiponectin, macrophage M1 polarization, and Th17 skewing that may increase risk for Th17/IL-23-driven diseases (e.g., psoriasis, psoriatic arthritis) and contribute to beta-cell autoimmunity in type 1 diabetes 3 – 5 . Conversely, some autoimmune conditions may be less influenced by adiposity or may involve altered immune tolerance associated with obesity. Importantly, the persistence of inverse associations after sensitivity analyses excluding outcomes occurring within 6 and 12 months after the index date suggests that acute reverse causation due to preclinical disease-related weight loss is unlikely to be the sole explanation for these findings, raising the possibility of disease-specific immunologic heterogeneity in the relationship between adiposity and autoimmunity. Nonetheless, alternative explanations include diagnostic and ascertainment biases (e.g., delayed recognition of autoimmune phenotypes in individuals with obesity), differential healthcare utilization, and residual confounding by unmeasured factors such as smoking, occupational exposures, or genetic predisposition. Taken together, these factors suggest that inverse or null associations should be interpreted with caution, as they do not necessarily imply a protective effect of adiposity. Future research incorporating more granular measures of adiposity, prospective cohorts with pre-diagnostic weight trajectories, and mechanistic studies of adipokines and immune cell phenotypes will be needed to clarify these relationships.

Stratified analyses showed demographic variation in obesity-associated IMID risk. Younger adults (18–40 years) had stronger links to type 1 diabetes, likely due to early obesity-driven beta-cell stress. Females showed higher risks for rheumatoid arthritis, psoriatic arthritis, and type 1 diabetes, consistent with the higher baseline prevalence of autoimmune diseases in women, potentially mediated by the immunomodulatory effects of sex hormones. Racial differences were seen, particularly for systemic lupus erythematosus, Sjögren’s syndrome, sarcoidosis, autoimmune thyroiditis, and type 1 diabetes, reflecting variations in adiposity patterns (e.g., Black individuals generally have less visceral but more subcutaneous fat than White individuals at the same BMI), immune genetics, metabolic profiles, and healthcare access 20 . Similar racial disparities in IMID prevalence have been documented in a large electronic health record-based study 21 . These findings highlight the need to consider demographic factors in mechanistic research and targeted prevention strategies.

Consistent associations for psoriatic arthritis, psoriasis, rheumatoid arthritis, and type 1 diabetes suggest that obese individuals with relevant symptoms may benefit from earlier rheumatologic, dermatologic, or endocrine evaluation. Incorporating weight management and more precise adiposity measures (e.g., waist-to-hip ratio, visceral fat imaging) into routine care could improve risk stratification and targeted prevention.

While BMI remains the most commonly used metric for defining obesity in epidemiologic research, it has well-recognized limitations. BMI does not differentiate between lean and fat mass, nor does it reflect visceral adiposity, which is more metabolically and immunologically active 22 . Moreover, BMI cannot capture heterogeneity in obesity phenotypes, such as metabolically healthy obesity or sarcopenic obesity 23 . Age, sex, and ethnic differences in body composition also affect its accuracy 21 . Future studies using more precise measures of adiposity, such as waist-to-hip ratio, visceral fat imaging, or dual-energy X-ray absorptiometry, are needed to refine risk stratification and strengthen causal inference 22 .

Our findings align with a growing body of epidemiologic evidence indicating that obesity is associated with an increased risk of selected IMIDs, particularly rheumatoid arthritis, psoriasis, and type 1 diabetes mellitus, supporting the role of adiposity-driven immune activation in these conditions 6 – 9 , 16 . In contrast, evidence for systemic autoimmune diseases such as Sjögren’s syndrome, systemic sclerosis, and certain neurologic autoimmune conditions has been less consistent across prior studies, with reported associations ranging from positive to null depending on study design, population characteristics, and outcome definitions 6 – 8 , 10 – 12 , 16 . Our observation of inverse or null associations for several of these conditions therefore adds to the existing heterogeneity in the literature and highlights the possibility of disease-specific immunopathogenic mechanisms. In this context, our large-scale, globally representative electronic health record-based analysis extends existing literature by simultaneously evaluating a broad spectrum of IMIDs using uniform definitions, rigorous propensity score matching, and multiple validation strategies.

A recent TriNetX-based study by Mousavi et al. also reported associations between obesity and several chronic inflammatory diseases, although results for specific IMIDs differed 16 . Their study reported null associations for Sjögren’s syndrome and systemic sclerosis, but observed elevated risks for dermatomyositis, multiple sclerosis, and myasthenia gravis, findings that differ from ours 16 . Differences in study design may partly explain these variations. Mousavi et al. defined obesity using ICD-10 codes, adjusted for fewer covariates, and did not exclude pre-existing outcomes before matching. In contrast, our study used measured BMI with confirmation, adjusted for a broader set of covariates, incorporated both positive and negative controls, and conducted dose–response analyses, which may enhance exposure classification, confounding control, and temporal clarity.

Strengths and limitations

This study has several strengths, including the use of a large, diverse, and well-validated electronic health record dataset with longitudinal follow-up. Rigorous propensity score matching and multiple validation strategies, including control outcomes and sensitivity analyses, enhance the internal validity of our findings.

Nonetheless, certain limitations should be acknowledged. Outcome misclassification is possible because diagnoses were identified using single ICD-10-CM codes, which may reduce specificity for some autoimmune diseases. This limitation is particularly relevant for type 1 diabetes mellitus, for which differentiation from type 2 diabetes may be imperfect in ICD-based definitions. However, outcome definitions were applied uniformly across exposure groups, and any resulting misclassification is likely to be non-differential, potentially biasing estimates toward the null. Future studies incorporating validated multi-code algorithms or laboratory-based criteria may further refine disease ascertainment. Obesity was defined using objectively measured BMI values with confirmatory recordings within a short time window; however, this approach does not capture the duration or chronicity of obesity exposure, which may be relevant for IMID development. Requiring sustained obesity over a longer pre-index period (e.g., multiple years) could theoretically strengthen causal inference but would have resulted in substantial sample attrition due to the intermittent nature of BMI recording in routine clinical practice. Accordingly, this study prioritized exposure validity and statistical power, a trade-off commonly encountered in large-scale real-world evidence studies. In addition, BMI measured at a single time point limits assessment of weight changes over time. The index date in this study reflects the timing of BMI measurement rather than the onset of obesity. Accordingly, follow-up time should be interpreted as time since BMI assessment, not cumulative exposure duration. This limitation constrains inference regarding the effects of long-term obesity exposure and should be considered when interpreting analyses involving follow-up time. Unmeasured confounders such as diet, physical activity, and genetic factors could influence results. As with all observational studies, residual confounding and reverse causality cannot be fully excluded.

Conclusions

In this large real-world cohort study, obesity was associated with increased risk of several IMIDs, particularly psoriatic diseases, rheumatoid arthritis, and type 1 diabetes mellitus. Demographic heterogeneity highlights the need for personalized risk assessment and targeted screening strategies in high-risk groups. Future prospective studies and interventional trials are warranted to determine whether weight reduction can modify IMID risk and improve patient outcomes.

Supplementary Information

Below is the link to the electronic supplementary material.

Supplementary Material 1

Author contributions

Yeong-Jang Lin: study conception and design, investigation, data collection and assembly, data analysis and interpretation, manuscript drafting, and final approval of the version for publication. Wan-Hsuan Hsu: study conception and design, investigation, data collection and assembly, data analysis and interpretation, and final approval of the version for publication. Chih-Cheng Lai: study conception and design, data collection and assembly, data analysis and interpretation, and final approval of the version for publication. Hung-An Chen: investigation, data analysis and interpretation, and final approval of the version for publication. Chao-Yu Chen: investigation, data analysis and interpretation, and final approval of the version for publication.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

The data that support the findings of this study are derived from the TriNetX global health research network. Due to data use agreements and privacy policies, the raw data are not publicly available. Aggregate-level results can be shared upon reasonable request. Please direct all data requests to the corresponding author, Dr. Yeong-Jang Lin, at linyeongjang@gmail.com.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

All methods were carried out in accordance with relevant guidelines and regulations. This study utilized de-identified electronic health record data obtained from the TriNetX platform. The use of such anonymized data has been determined by the Western Institutional Review Board to be exempt from Institutional Review Board approval, and the need to obtain informed consent was waived, as it does not involve research on human subjects as defined by federal regulations and institutional policies. This study was granted exemption from full ethical review and waiver of informed consent by the Chi Mei Medical Center Institutional Review Board (Applicant No.: 11403-E02) on March 17, 2025.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

Associated Data

Supplementary Materials

Supplementary Material 1

Data Availability Statement

The data that support the findings of this study are derived from the TriNetX global health research network. Due to data use agreements and privacy policies, the raw data are not publicly available. Aggregate-level results can be shared upon reasonable request. Please direct all data requests to the corresponding author, Dr. Yeong-Jang Lin, at linyeongjang@gmail.com.