Study design and population
We conducted a population-based retrospective cohort study using data from the National Health Insurance Services (NHIS) database of Korea, which covers approximately 97% of the Korean population. Demographic, socioeconomic, and clinical data, including sex, age, height, weight, BMI, smoking status, alcohol consumption, exercise habits, income level, blood glucose, cholesterol levels, blood pressure, and eGFR, were obtained from the NHIS database. Height and weight were measured during routine health screenings conducted by trained medical personnel following standardized protocols, and BMI was calculated as weight (kg) divided by height squared (m²). Our study focused on individuals aged 50 years and above who participated in health screenings during 200933.
From an initial population of 4,234,412 individuals, we refined our cohort through a multi-step selection process (Fig. S1). Exclusion criteria included: age below 50 years, pre-existing knee OA or related interventions, insufficient follow-up data, and limited claims (1 or 2 within the year prior to screening) for atopic diseases. The final study cohort consisted of 880,300 individuals.
We followed up each participant starting one year after their health screening date in 2009 until they received a knee OA diagnosis, died, or until December 31, 2020, whichever came first.
Data sources and ethical considerations
This study utilized the National Health Insurance Services (NHIS) database of Korea, a comprehensive resource covering approximately 97% of the Korean population. Established by the Korean government, the NHIS database provides extensive healthcare service claims and screening data. The reliability of NHIS cohorts has been validated in previous studies, ensuring a robust foundation for our research34,35.
We conducted the study in accordance with the Declaration of Helsinki, and the Institutional Review Board (IRB) of the Catholic University of Korea approved it (protocol number VC24ZISI0188). Given the retrospective nature of the study and the use of de-identified data, the IRB of the Catholic University of Korea waived the requirement for individual informed consent. This approach ensured ethical rigor while facilitating access to a large-scale, representative dataset for population-based analysis.
Definition of atopic diseases
We defined atopic diseases in this study as asthma, atopic dermatitis, and allergic rhinitis, using the International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes: asthma (J45-46), atopic dermatitis (L20), and allergic rhinitis (J301-304)33. We classified participants as having an atopic disease if they had at least one confirmed diagnosis of any component of the atopic triad. To ensure diagnostic accuracy, we required a minimum of three documented clinical visits per year for each condition. This threshold has been previously applied in NHIS-based studies to identify clinically significant atopic disease cases36,37. Individuals without any atopic triad diagnosis were categorized as nonatopic.
Primary outcome
The primary endpoint was the new onset of knee OA. We identified knee OA using ICD-10 codes specific to knee OA (M17) or general OA (M15 for polyarthrosis, M19 for other forms of arthrosis), along with a procedure code for knee X-ray within the same medical claim. This methodology is consistent with validated approaches from previous research38. The follow-up period commenced one year after the initial health screening and continued until knee OA diagnosis, death, or December 31, 2020, whichever occurred first.
Assessment of health behaviors and comorbidities
We assessed lifestyle factors and the presence of comorbid conditions through a comprehensive review of patient-reported outcomes and clinical data. Lifestyle behaviors were self-reported via standardized questionnaires. Socioeconomic status was determined through income tiers, with the lowest tier representing the bottom 25% of the population by income. Smoking habits were categorized into three distinct groups: non-smokers, former smokers, and active smokers. Alcohol consumption was classified based on daily intake: non-drinkers, moderate consumption (less than 30 g per day), and significant consumption (30 g or more per day). Physical activity levels were delineated based on type and frequency: non-exercisers, moderate exercisers (over 30 min of moderate activity at least once a week), and regular exercisers (over 30 min of moderate activity at least five times a week or over 20 min of vigorous activity at least three times a week).
Comorbid health conditions such as hypertension, diabetes mellitus, and dyslipidemia were identified using a combination of ICD-10-CM diagnostic codes, prescribed medication records, and biometric measures from health examinations. The criteria for these conditions were consistent with previously established and validated protocols39. In our health screenings, fasting blood tests were conducted to measure serum glucose and lipid levels, following at least eight hours of fasting, to ensure the accuracy of these diagnostic markers. Further information on the operational definitions used comorbid conditions can be found in the supplementary material of this study, delineated in table S1.
Statistical evaluation methods
Our data analysis commenced with the delineation of baseline characteristics, where we reported continuous variables as mean values with their corresponding standard deviations and categorized variables in frequencies and proportions. Comparative analysis of continuous data was executed via the application of the t test or non-parametric alternatives when appropriate. Chi-square testing facilitated the comparison of categorized variables.
We determined the incidence rates of knee OA by tallying new cases and dividing by the accumulated person-years, expressing the result as the number of events per 1,000 person-years. The risk estimation for the occurrence of knee OA was performed using the Cox proportional hazards regression model, yielding hazard ratios and 95% confidence intervals40. Our model adjustments progressed sequentially: the primary model accounted for age and gender, the secondary model additionally considered lifestyle factors and comorbidities, and the tertiary model incorporated further adjustments for chronic inflammatory conditions.
The strength of association was quantified using Cohen’s d, calculated from the natural logarithm of the hazard ratio and standardized41. We conducted a retrospective power calculation to validate the robustness of our findings. Subgroup analyses were stratified based on demographics, lifestyle factors, and comorbidity status, with the Bonferroni method employed to adjust for multiple testing.
The Kaplan-Meier estimator was used to plot survival curves, with the log-rank test comparing the curves. The Cox model also supported the survival analysis, and interactions were tested to identify significant differences between subgroups. We established significance for all tests at a p-value of less than 0.05 and conducted two-tailed analyses.
All statistical computations were performed with SAS (version 9.4, SAS Institute, Inc, Cary, NC, 2013) and R program (version 3.2.4, R Core Team, Vienna, Austria, 2017), ensuring rigorous data handling and analysis integrity.
link