Study design and participants
This national cross-sectional survey was conducted between December 2021 and January 2022, during the COVID-19 pandemic, a period characterized by heightened psychological stress, social distancing measures, and limited in-person interactions in Korea. A proportionately stratified sampling method was used to ensure representative sampling of 2.5% of the Korean population. The study included participants aged 18–85 years and was conducted across all 17 provinces in South Korea. Participants were recruited through a professional survey agency using stratified sampling based on national demographic distributions of age, sex, and region. Recruitment was conducted primarily through an existing population panel managed by the agency. Among those contacted for this study, approximately 80% agreed to participate. Because the panel itself was constructed using probability-based sampling to represent the Korean population, the characteristics of respondents and non-respondents are expected to be comparable. A total of 2,699 individuals completed the survey, exceeding the initial target of 2,500. Each participant received a 5,000 KRW (approximately 5 USD) voucher as compensation. As this was a nationally representative survey, no additional exclusion criteria were applied beyond the initial sampling frame. The study was approved by the Institutional Review Board of the Samsung Medical Center (IRB number: 2021-03-005).
Measurement
The Korean version of the Aggression Questionnaire (AQ) was used to assess perpetration of aggression [22]. The AQ comprises subscales of physical aggression, verbal aggression, anger, and hostility. These subscales evaluate the behavioral aspects of physical and verbal aggression, cognitive dimensions through measures of hostility, and emotional dimensions through the measurement of anger. The questionnaire comprises 27 items rated on a 5-point Likert scale ranging from 1 (extremely uncharacteristic of me) to 5 (extremely characteristic of me), reflecting how well each statement described the respondent [22]. The Cronbach’s alpha values for Korean version of AQ was 0.86 [22].
Based on the GAM, which posits that aggressive behavior results from interactions between situational inputs and individual traits [11], we selected health-related variables that are known to influence emotional regulation and behavioral responses. Impaired physical function and chronic pain have been linked to heightened irritability and frustration, which may reduce an individual’s threshold for aggressive reactions. Similarly, psychological factors such as anxiety and depression are associated with impaired impulse control and heightened threat perception, both of which may contribute to reactive aggression. These domains were therefore selected as key predictors to capture the multifactorial underpinnings of aggression in the general population. Sociodemographic characteristics, such as age (≤ 29: Young adults, 30–49: Middle, 50–64: Senior and ≥ 65: Older adults), sex, residential area (Seoul, Gyeonggi-do, and Incheon; Non-metropolitan area: Chungcheong-do, Gangwon-do, Jeolla-do, Gyeongsang-do, and Jeju-do), marital status, employment status, educational level, and annual total household income over the past year, including wages, real estate income, financial instrument income, government subsidies and pensions, allowances from relatives or children, and jobs with heavy manual labor, were gathered through self-reported questionnaires.
Physical factors included physical function, pain interference, comorbidities, and physical disability. Comorbidities were assessed by using medication-related questionnaires. Comorbidities were defined as the presence of physician-diagnosed physical health conditions including hypertension, diabetes, hyperlipidemia, cancer, consistent with prior studies. Physical function was assessed using the Korean version of the Patient-Reported Outcomes Measurement Information System (PROMIS) Physical Function short form v2.0 (10a), which measures perceived limitations in performing physical tasks. Example items include: “Does your health now limit you in vigorous activities?”, “Does your health now limit you in hiking a couple of miles?”, and “Does your health now limit you in participating in active sports?”. Each item was rated on a 5-point Likert scale ranging from 1 (unable to do) to 5 (without any difficulty), with higher scores indicating better physical function [23]. To measnure pain, we used the PROMIS Pain Interference Item Bank v1.1, which includes 56 items [23]. The PROMIS(pain interference assessment determines how pain affects physical, mental, and social activities, while also evaluating pain levels. Example items include: “How much did pain interfere with your relationships with other people?” and “How much did pain interfere with your ability to work (including work at home)?” Each item is rated on a 5-point Likert scale, with higher scores indicating greater pain interference. The range of Cronbach’s alpha values for Korean version of physical function and pain were 0.87–0.91 and 0.95–0.97, respectively, in various population [24,25,26]. For physical function, scores ≤ 40 were considered moderate to severe impairment. For pain interference, scores ≥ 60 were also defined as moderate to severe.
Psychological factors assessed included anxiety, depression, and use of psychiatric medication. The use of psychiatric medication within the last 30 days as prescribed by a doctor was repoted using the questionnaire. Anxiety was assessed using the PROMIS-Anxiety questionnaire, which features 29 items that evaluate cognitive, emotional, and physical anxiety symptoms. The response options ranged from 1 (never) to 5 (always), with higher scores indicating higher anxiety levels. Depression was evaluated utilized the PROMIS-Depression 28-item questionnaire, which encompasses diverse depressive aspects on a Likert scale. The scores reflect various depressive experiences, including sadness, loss of interest, hopelessness, low energy, and sleep disturbances [27]. The Cronbach’s alpha coefficients of depression, and anxiety were 0.97, and 0.96, respetvely [28]. For anxiety and depression, scores of 55–59 were classified as mild, and ≥ 60 as moderate to severe. These thresholds are consistent with both U.S.-based scoring guidelines and prior validation studies of the Korean version of PROMIS.
Statistical analysis
We performed a post hoc power calculation using the observed proportion of participants classified in the high aggression group (10%) and the distribution of key exposure variables. With a total sample size of 2,699 and a two-sided α of 0.05, the study had over 80% power to detect an odds ratio of 1.6 or greater for the association between aggression and primary exposures, such as moderate-to-severe depression, using multivariable logistic regression.
To calculate the aggression score, raw AQ scores were linearly transformed to yield domain scores within the range of 27–135, where a higher score indicated a stronger tendency towards aggression [22]. For the analysis, participants who responded in the highest 10 percentiles were grouped into the “higher aggression” category while participants responding below the highest 10 percentiles were grouped into the “normal” category.
The PROMIS scores, which measure physical function, pain interference, depression, and anxiety, were converted into t-scores via the PROMIS Assessment Center. The PROMIS T-scores are based on the US population, with an average score of 50 and a standard deviation of 10 points. Higher pain interference scores indicate greater pain interference. Despite potential differences in local item parameters, previous research has demonstrated that when country-specific parameters are rescaled to the U.S. metric, the resulting PROMIS scores exhibit negligible mean differences while maintaining international interpretability [29].
We conducted multivariable logistic regression analyses to identify the demographic, physical, psychological, and social relational factors associated with perpetration of aggression (prevalence ratio (PR) and 95% confidence interval (CI) were estimated). We adjusted for age, sex, physical function, pain interference, comorbidities, physical disability, marital status, psychological characteristics, anxiety, depression, and the use of psychiatric medication. To evaluate the fit of the final multivariable logistic regression model, we calculated the residual deviance and the Akaike Information Criterion (AIC). These metrics assess how well the model fits the observed data while accounting for model complexity.
Differences in score and distribution between the two groups were identified using the standardized mean difference (SMD). Statistical analyses were performed using the R 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria).