Unveiling the BMI Risk Threshold for Osteoarthritis: Multi‐Database Causal and Nonlinear Evidence Authors: Shaohong Yu, Li Zeng, Lei Yang, Yangjiang Ou, Chenglin Yang, Xiaoru Lin, Jingyi Shao, Qing Guan, Guobo Li, Xiuquan Lin Published: 2026 ABSTRACT ABSTRACTObjectiveTo characterize the nonlinear relationship between BMI and osteoarthritis (OA), and to identify BMI thresholds that inform precise prevention strategies.MethodsThis multi‐database study integrated Global burden of disease 2021, National Health and Nutrition Examination Survey 2007–2018, and Genome‐Wide Association Studies. A generalized additive model was performed to visualize the BMI‐OA relationship, adjusting for multiple confounders. We applied segmented logistic regression models to identify potential threshold effects and used Mendelian randomization to estimate the causal effects of BMI on OA subtypes.ResultsFrom 1990 to 2021, the age‐standardized prevalence and years lived with disability rates for OA were highest in regions with high SDI. OA prevalence rose nonlinearly with BMI, with breakpoints at 24.00 and 41.58 kg/m2. Each unit increase in BMI was associated with higher odds of OA between 24.00 and 41.58 kg/m2 (OR = 1.022, 95% CI: 1.003–1.041) and above 41.58 kg/m2 (OR = 1.055, 95% CI: 1.022–1.090). Women and individuals aged ≥ 45 years exhibited a higher susceptibility to knee osteoarthritis. BMI was causally associated with knee osteoarthritis (OR = 1.63, 95% CI 1.50–1.77) and hip osteoarthritis (OR = 1.54, 95% CI 1.40–1.70).ConclusionsThese findings suggest that OA risk awareness and weight‐management strategies should begin before BMI reaches the high range, particularly among individuals with BMI exceeding 24.00 kg/m2. Abbreviations body mass index Centers for Disease Control and Prevention generalized additive model Global Burden of Disease Genome‐Wide Association Studies hospital event statistics hip osteoarthritis Inverse Variance Weighting knee osteoarthritis Mobile Examination Center Mendelian Randomization National Center for Health Statistics National Health and Nutrition Examination Survey osteoarthritis population attributable fraction Socio‐demographic Index Summary Exposure Value World Health Organization years lived with disability Introduction Osteoarthritis (OA) is a chronic joint disease characterized by progressive degeneration of articular cartilage, hyperosteogeny, and alterations in joint structure [1]. The disease commonly affects weight‐bearing joints, such as the knee and hip, leading to pain and functional impairments. These symptoms significantly impair patients' daily functioning and quality of life and impose a substantial socioeconomic burden [2]. In the United States, for example, direct medical expenditures related to OA reached approximately $101.8 billion in 2019, and the overall economic burden exceeded $303.5 billion [3]. The World Health Organization (WHO) has classified OA as one of the fastest‐growing major public health issues globally and the second leading cause of disability [4]. Therefore, identifying modifiable key risk factors for OA is crucial for slowing disease progression and alleviating its societal burden [5]. Global Burden of Disease (GBD) 2021 identifies body mass index (BMI) as the only clearly quantifiable and modifiable risk factor for OA. In recent years, global obesity prevalence and mean BMI have persistently increased; according to pooled analyses by the Non‐Communicable Disease (NCD) Risk Factor Collaboration, these trends extend through 2022 [6]. A meta‐analysis has shown that for every 5 kg/m2 increase in BMI, the risk of knee osteoarthritis (KOA) increases by 35% [7]. This association may be mediated by increased mechanical load due to weight gain and obesity‐related metabolic disturbances, such as insulin resistance and chronic inflammation, contributing to cartilage degradation [8]. Therefore, in the context of the global obesity epidemic, further investigating whether a causal and non‐linear dose–response relationship exists between BMI and OA is crucial for identifying high‐risk populations, characterizing potential threshold effects, and developing more targeted prevention strategies. Although previous studies have consistently demonstrated a nonlinear positive association between elevated BMI and the risk of osteoarthritis [9, 10, 11, 12, 13], the dose–response pattern and potential intervention thresholds of this relationship remain insufficiently characterized. Furthermore, many studies have been restricted to single databases, lacking cross‐validation across multiple data sources. Therefore, this study integrates the GBD database, the National Health and Nutrition Examination Survey (NHANES) database, and the Genome‐Wide Association Studies (GWAS) database to conduct a comprehensive analysis of the relationship between BMI and OA. The aim is to validate the impact of BMI on OA in adults and explore potential threshold effects within this relationship. Additionally, by utilizing the latest GWAS database, this study provides genetic evidence from a larger sample to assess the causal effect of BMI on OA and its sub‐types. The findings are expected to provide high‐quality epidemiological evidence for precise prevention and control strategies for OA, thereby supporting global efforts to address the burden of OA. Materials and Methods Data Collection and Study Population Global Burden of Disease (GBD) Database The GBD database collects detailed data on 370 diseases and injuries across 204 countries and regions from 1990 to 2021, integrating demographic data at both national and regional levels. The GBD study employs a systematic analytical approach that synthesizes various epidemiological data to calculate the relative risk and population attributable fraction (PAF) for diseases associated with BMI and provides data standardization along with uncertainty interval estimation. In this study, the prevalence and years lived with disability (YLDs) data for OA and its subtypes were sourced from the Global Health Data Exchange online platform (https://vizhub.healthdata.org/gbd‐results/). The study was approved by the Institutional Review Board of the University of Washington, and as the study used anonymized publicly available data, informed consent was waived. National Health and Nutrition Examination Survey (NHANES) Database The NHANES is a nationwide survey project led by the National Center for Health Statistics (NCHS), designed to systematically assess the health and nutrition status of the U.S. population across all age groups. The survey gathers detailed individual‐level data through face‐to‐face interviews, physical examinations, and laboratory tests. This study utilized data from the 2007–2018 NHANES cycles (https://www.cdc.gov/nchs/nhanes/). The participant selection process for the NHANES database is outlined in the flowchart (Figure 3). Of the initial 59 842 participants, the following individuals were excluded: those with missing BMI or OA history data; individuals < 20 years old; those with a BMI > 60 kg/m2. Missing covariate data were addressed using multiple imputation. The final NHANES study population consisted of 32 783 individuals. BMI data was obtained from the NHANES body measures examination variable BMXBMI, which was calculated from measured weight and height collected in the Mobile Examination Center (MEC). The NHANES protocol has been approved by the Centers for Disease Control and Prevention (CDC). Before participation in the NHANES study, all participants provided informed consent, with parental or guardian consent required for minors. NHGRI‐EBI GWAS Catalogue Database The GWAS Catalogue (https://www.ebi.ac.uk/gwas) is a public database that archives large‐scale genetic studies, providing support for disease mechanism research and precision medicine by identifying genetic loci associated with diseases and phenotypes through GWAS. The GBD 2021 report identifies high BMI as a risk factor for KOA and hip osteoarthritis (HOA). In this study, the causal relationship between BMI and OA was further validated using Mendelian Randomization (MR). According to the retrieved datasets [14, 15], the GWAS summary data for BMI (GCST008025), KOA (GCST90566800), and HOA (GCST90566798) include 49 335, 1 316 500 (172 256 cases, 1 144 244 controls), and 1 152 707 participants (97 328 cases, 1 055 379 controls), respectively. All data used in this study were obtained from public sources, and the original research teams have received appropriate ethical approval. OA Diagnosis In the GBD 2021 database, OA diagnosis is based on radiographic confirmation, using the Kellgren‐Lawrence grading scale, and includes only symptomatic OA cases with grades 2–4 [16, 17]. HOA corresponds to ICD‐10 code M16, while KOA corresponds to ICD‐10 code M17. The ICD‐10 terminology replaces the ICD‐9 term ‘osteoarthrosis’ with ‘osteoarthritis’ to emphasize the inflammatory component of the disease. In NHANES, OA diagnosis is based on self‐reported questionnaires. Participants aged 20 and older were classified as having OA if they answered ‘Yes’ to the question ‘Doctor ever said you had arthritis’ and further specified ‘Osteoarthritis or degenerative arthritis’ when asked ‘Which type of arthritis was it?’ [18]. In GWAS, OA cases are selected based on hospital event statistics (HES) using ICD‐10 codes for KOA and HOA to ensure diagnostic accuracy [19]. Although case definitions differed across databases, each data source contributed complementary evidence at a distinct level of inference: GBD provided radiographically anchored estimates of OA burden, NHANES captured physician‐diagnosed OA as reported in a nationally representative community sample, and GWAS identified hospital‐coded OA in clinically ascertained populations. Therefore, the BMI threshold derived from NHANES should be interpreted as a population‐level risk‐acceleration point rather than a universal clinical cutoff, while the consistency of findings across GBD, NHANES, and GWAS strengthens the overall robustness of the observed association between BMI and OA. Variable Description The primary exposure variable in this study was BMI. In the NHANES database, height and weight were measured through standardized physical examination procedures conducted at the Mobile Examination Center. BMI was calculated as weight in kilograms divided by height in meters squared. In our analysis, BMI was treated as a continuous variable. Established risk factors for OA confirmed by previous studies served as the covariates in this study, including demographic information (sex, age, race, education level, marital status, ratio of family income to poverty), physical examination data (waist circumference), and questionnaire data (alcohol status, smoking status, physical activity, hypertension, hyperlipidemia, diabetes, cardiovascular diseases) [20, 21, 22, 23]. All covariates were collected through standardized household interviews and in‐person measurements. The outcome variable was OA status, defined according to self‐reported physician diagnosis (yes/no). To ensure data quality, a comprehensive assessment of missing data was conducted, and the proportions of missing data for key variables in the overall NHANES population and stratified by osteoarthritis status are presented in eTable 1. Missing covariates were handled using multiple imputation. In the revised analysis, the imputation model explicitly included osteoarthritis status, BMI, and the covariates used in the multivariable models. Statistical Analysis For GBD 2021, this study explores the trends in the association of high BMI with OA prevalence and YLDs rates across different geographic regions and time periods. All prevalence and YLDs rates were age and sex standardized using the GBD world standard population weights to account for demographic differences. To further investigate the relationship between the Socio‐demographic Index (SDI) and the Summary Exposure Value (SEV) of high BMI, Pearson's correlation coefficient was used to investigate the correlation between the SEV of high BMI and SDI. Stratified analyses were conducted by age and sex to examine the distribution of OA. For NHANES, continuous variables were expressed as mean ± standard deviation, whereas categorical variables were presented as numbers and percentages. Between group comparisons were performed using Pearson's chi squared test and Welch's two sample t‐test. To assess the nonlinear relationship between BMI and OA prevalence, we first fitted a generalized additive model (GAM) with adjustment for covariates [24]. Subsequently, a segmented logistic regression model was applied to further identify potential BMI‐related breakpoints. BMI was modelled as a piecewise linear variable with two unknown breakpoints, and the breakpoint locations and segment specific regression coefficients were estimated simultaneously using an iterative maximum likelihood procedure. The effects within different BMI intervals were expressed as odds ratios and 95% confidence intervals (CI) for each 1 kg/m2 increase in BMI [25, 26]. This study explores the causal relationship of BMI with KOA and HOA using two‐sample Mendelian randomization. Instrumental variables (IVs) were derived from large‐scale GWAS and selected based on their genome‐wide significant association with BMI (p < 5 × 10−8) and independence from each other (r 2 < 0.001). The F‐statistics of the included SNPs ranged from 29.13 to 133.77, all exceeding the conventional threshold of 10, indicating adequate instrument strength. The exact F‐statistics and R 2 values for all included SNPs are provided in eTable 2. The primary analysis was conducted using inverse‐variance weighted (IVW) estimation, supplemented by sensitivity analyses using weighted median, MR‐Egger, weighted mode, and simple mode estimations [27]. Additionally, a leave‐one‐out analysis was performed to assess the robustness of the results. After cross‐referencing with summary statistics for KOA and HOA, 11 IVs were extracted. All statistical tests were two‐tailed, with a significance level set at p < 0.05. All statistical analyses and graphical representations were performed using R software (version 4.4.2). Results As shown in Table 1, in both 1990 and 2021, high‐SDI regions consistently ranked highest for YLDs and prevalence across the global and five SDI regions. In 1990, the age‐standardized prevalence and age‐standardized YLDs rate in high‐SDI regions reached their highest values, at 7371.33 per 100 000 (95% UI: 6609.68–8130.17) and 262.74 per 100 000 (95% UI: 126.00–529.15), respectively. Notably, the United States had the highest absolute number of cases and YLDs, with 25 007 735.07 (95% UI: 22 549 194.96–27 580 854.86) and 900 534.50 (95% UI: 432 395.48–1 819 200.41), and its age‐standardized prevalence and YLDs rates ranked among the top five globally. By 2021, this pattern had not changed substantially: the age‐standardized prevalence in high‐SDI regions remained at 7897.27 per 100 000 (95% UI: 7067.13–8689.88) and the age‐standardized YLDs rate at 283.13 per 100 000 (95% UI: 136.04–570.53), still at the highest levels, with the United States continuing to rank among the top. The results for each country are visualized in map form eFigure 1. TABLE 1The changes in age‐standardized prevalence and age‐standardized YLDs of OA globally and in five SDI regions in 1990 and 2021.Age‐standardized Prevalence rate per 100 000Age‐standardized YLDs rate per 100 00019902021Percentage change 1990–202119902021Percentage change 1990–2021Global6393.12 (5683.20, 7059.53)6967.29 (6180.70, 7686.06)0.09 (0.08, 0.10)222.80 (106.65, 450.30)244.501 (117.061, 493.106)0.10 (0.09, 0.10)Low SDI5080.55 (4507.60, 5674.54)5605.58 (4967.54, 6230.60)0.10 (0.09, 0.12)170.90 (82.56, 345.11)190.93 (91.63, 384.26)0.12 (0.11, 0.13)Low‐middle SDI5326.73 (4722.18, 5924.73)6106.25 (5419.32, 6763.20)0.15 (0.13, 0.16)180.18 (86.93, 364.19)209.35 (100.40, 422.62)0.16 (0.15, 0.18)Middle SDI6066.13 (5375.33, 6744.29)6903.80 (6123.00, 7643.11)0.14 (0.12, 0.16)208.27 (100.47, 419.68)240.40 (115.09, 483.99)0.15 (0.13, 0.17)High‐middle SDI6557.50 (5802.43, 7265.76)7120.38 (6297.96, 7879.76)0.09 (0.07, 0.10)228.99 (109.20, 462.13)250.58 (119.78, 503.69)0.09 (0.08, 0.11)High SDI7371.33 (6609.68, 8130.17)7897.27 (7067.13, 8689.88)0.07 (0.07, 0.08)262.74 (126.00, 529.15)283.13 (136.04, 570.53)0.08 (0.07, 0.09) The changes in age‐standardized prevalence and age‐standardized YLDs of OA globally and in five SDI regions in 1990 and 2021. The results of the comparative analysis across time periods (1990 vs. 2021) indicated that (Figure 1) among the five different SDI regions, SEV for high BMI in high‐SDI regions consistently remained the highest. Furthermore, in all SDI regions, the age‐standardized exposure rate for high BMI in 2021 showed a significant increase compared to 1990. Age‐standardized summary exposure value for high body mass index in five SDI regions in 1990 and 2021. From 1990 to 2021, there were significant differences in the age‐standardized YLDs rates for OA across global and five SDI regions (eFigure 2). Overall, age‐standardized YLDs for KOA accounted for a larger proportion of OA‐related disability than HOA. Over time, the age‐standardized YLDs rates for OA have shown an upward trend, but the increase was relatively gradual in high and middle‐high SDI regions. Among these, the rise in OA was primarily driven by KOA, while the trend for HOA remained relatively flat. Age‐stratified analyses from 1990 to 2021 revealed that the increases in the prevalence and YLDs rates of OA, HOA, and KOA were predominantly concentrated in individuals aged 45 and older, with a significant upward trend in disease burden as age increased (Figure 2). Specifically, HOA was more prevalent in those aged 65 years and older, while KOA had the highest prevalence in the 45–64 age group; the distribution of YLDs across age groups mirrored the prevalence patterns. Gender‐stratified analyses showed that females generally had higher age‐standardized prevalence estimates than males for overall OA, with a similar pattern observed for KOA; in contrast, male and female prevalence estimates for HOA were broadly comparable, with substantial overlap in their confidence intervals. For age‐standardized YLD rates, females also tended to have higher estimates than males for overall OA and KOA, whereas the male and female trajectories for HOA were similar over time, with largely overlapping uncertainty intervals, suggesting a less pronounced sex disparity (eFigure 3). The YLDs and prevalence rates of OA, HOA and KOA in three age groups from 1990 to 2021. The study population included a total of 32 783 participants, of whom 15 881 (48%) were male and 16 902 (52%) were female (Figure 3). Table 2 summarizes these characteristics. OA patients were predominantly characterized by the following traits: female sex, older age, higher education level, better socioeconomic status, non‐Hispanic White, and higher waist circumference index. A higher prevalence of OA was also observed among former smokers, former drinkers, and individuals with no physical activity. Additionally, individuals with hypertension, hyperlipidemia, diabetes, and cardiovascular disease were more likely to develop OA. NHANES data showed significant differences in all variables between individuals with and without OA (p < 0.05). Flow chart of the study population selection process. TABLE 2Baseline characteristics of study participants (N = 32 783).VariableOverall N = 32 783 a Non‐OA N = 29 367 a OA N = 3416 a p b GenderMale15 881 (48%)14 655 (50%)1226 (36%)< 0.001Female16 902 (52%)14 712 (50%)2190 (64%)Age (year)49.67 ± 17.7048.06 ± 17.4463.52 ± 13.31< 0.001Age category≥ 65 year7801 (24%)6045 (21%)1756 (51%)< 0.00120‐44 year13 745 (42%)13 406 (46%)339 (9.9%)45–64 year11 237 (34%)9916 (34%)1321 (39%)RaceNon‐Hispanic White13 262 (40%)11 198 (38%)2064 (60%)< 0.001Non‐Hispanic Black7063 (22%)6521 (22%)542 (16%)Mexican American4916 (15%)4640 (16%)276 (8.1%)Other race7542 (23%)7008 (24%)534 (16%)Education levelHigh school7475 (23%)6710 (23%)765 (22%)< 0.001>High school17 177 (52%)15 233 (52%)1944 (57%)