Data source
We utilized data from the Korea National Health Insurance Service-Elderly Sample Cohort (NHIS-ESC), supported by Korea National Health Insurance (NHI). The NHI is a single insurer offering universal healthcare coverage to all Korean citizens and involving all healthcare providers. The NHIS-ESC is a nationwide, representative database that has been prospectively collecting data across South Korea from January 1, 2002, to December 31, 2019, except for instances of disqualification due to emigration or death [19]. This cohort includes 545,831 participants, representing approximately 8% of the adult population aged 60 years and older in Korea as of 2008. The NHIS-ESC database encompasses a wide array of data, including sociodemographic variables, mortality records, diagnostic records, results of medical checkups, and prescribed medications.
Study design and participants
A nested case-control study involving 250,654 participants diagnosed with geriatric depression was conducted. To ensure accurate diagnoses, we excluded individuals potentially misdiagnosed and those with illegal proxy prescriptions, limiting our study to patients who had visited outpatient clinics at least three times for geriatric depression (Fig. 1). Additionally, we excluded patients who lacked psychiatric specialist visits one year before or after their depression diagnosis. The stringent inclusion criteria are based on local legislation and enforcement rules in Korea, where physicians, other than psychiatric specialists, are required to refer patients with depression to psychiatric specialists within 60 days [20].
A two-year washout period was employed to identify newly diagnosed cases of geriatric depression. Participants who had previously benefited from the Medical Aid Program before their diagnosis of depression were also excluded from the study. Eligible participants were tracked from the date of their initial diagnosis with geriatric depression until the earliest occurrence of one of the following events: diagnosis of dementia, disqualification from the NHI, death, or the conclusion of the observation period on December 31, 2019.
Identification of cases and controls
Patients were diagnosed with forms of dementia were identified as cases and selected based on the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10). Participants who had visited an outpatient clinic more than three times or were admitted more than once because of dementia were classified as cases. The index date was set as the date of dementia diagnosis, and the follow-up period extended from the date of geriatric depression diagnosis to the date of dementia diagnosis. Patients with a dementia diagnosis before the diagnosis of depression and follow-up duration of less than one year was excluded from previous research [21, 22].
Eligible controls were defined as participants were at risk of dementia at the time of case occurrence. The matching of controls was conducted based on several attributes corresponding to the cases, including sex (male or female), age (within ± one year), the duration of follow-up, and the date of the first dementia diagnosis (within ± 30 days). Each case was matched with controls selected with replacement, allowing one matched control to serve as a control for multiple cases. Additionally, each case served as their own control. The controls shared the same index date as matched cases.
We further implemented additional exclusion criteria: (1) presence of missing data and (2) physical activity data collected at least once every three years (due to a national health checkup, which collects self-reported physical activity data every other year and to exclude participants who showed poor adherence to the national checkup [23]).
After applying exclusion criteria, controls were randomly selected at a ratio of 1:5 from the matched control group. Controls without corresponding matched cases were excluded. Ultimately, the study included 2,121 patients and 9,901 controls. Among the 2,121 cases, 51 were matched at a 1:1 ratio, 73 at a 1:2 ratio, 90 at a 1:3 ratio, and 101 at a 1:4 ratio.
Exposure
Exposure consisted of self-reported physical activity data categorized by the intensity level. Physical activity data were obtained using structured questionnaires similar to the short form of the International Physical Activity Questionnaire [24]. Physical activity intensity was divided into two levels: vigorous-intensity aerobic physical activity (VIPA), defined as intense activities that made participants breathe much harder than usual, such as running, aerobics, and fast-paced cycling, lasting at least 20 min; and, moderate-intensity aerobic physical activity (MIPA), defined as moderate activities that made participants breathe slightly harder than usual, such as brisk walking, playing double tennis, cycling at a moderate pace, and lasting at least 30 min. Participants were asked how many days they engaged in VIPA or MIPA over the last seven days. In 2018 and 2019, they were additionally asked to specify the exact duration (in minutes) and frequency of VIPA and MIPA activities per week.
The long-term physical activity was measured from the date of depression diagnosis to the index date. Long-term physical activity during the follow-up period was approached from two perspectives: the average fulfillment rate of the World Health Organization (WHO) recommendations and the absolute amount of physical activity measured in metabolic equivalents of task (MET) min per week. According to the WHO recommendations for physical activity in older adults aged ≥ 65 years, the guidelines are as follows: 1) engagement in a minimum of 150 min of MIPA weekly, 2) engagement in at least 75 min of VIPA weekly, or 3) a combination of MIPA and VIPA that is equivalent [25]. Therefore, engaging in VIPA four times or more per week, MIPA five times or more per week, or equivalent combinations was considered to adhere to the recommendations. Physical activity that did not meet the recommendations was defined as the achievement of half of the recommended goals. The average WHO recommendation fulfillment rate during the follow-up period was calculated, ranging from 0% for the inactive group to 100% for the “always followed recommendations” group. Recommendation fulfillment rate ranging from 62.5–100% were categorized as the “Generally followed recommendations” group, 37.5–62.5% as the “Sometimes followed recommendations” group, and 0–37.5% as the “Never followed recommendations” group.
Additionally, for the absolute level of physical activity as a continuous variable, the average MET-min/week during the follow-up period was calculated and categorized into quartiles (7.0 MET-min/week for VIPA; 4.0 MET-min/week MIPA) [26].
Covariates
Based on the index date, three socioeconomic characteristics were involved: economic status (subdivided by the premium amount of NHI: low, middle, and high): area of residence (capital area, metropolitan area, and province), and insurance type (employee, dependents of an employee, self-employed, and dependents of a self-employed).
To account for geriatric depression severity and type, adjustments were made for each specific antidepressant subclass intake (selective serotonin reuptake inhibitors, selective norepinephrine reuptake inhibitors, tricyclic antidepressants, mirtazapine, trazodone, bupropion, and tianeptine sodium) and the cumulative dose of the seven subclasses of antidepressants during the follow-up period. Based on the cumulative defined daily doses (cDDDs) established by the WHO, the cumulative dose was calculated (less than 14, 14–30, 30–60, 60–180, 180–365, 365–730, and > 730 cDDDs). We defined antidepressant non-users as those with fewer than 14 cDDDs, according to the antidepressant usage guidelines in Korea and the United States [27, 28]
To adjust for participants’ health status, physical activity levels before geriatric depression were included. The physical activity data collected between two and six years prior to the depression diagnosis date were utilized. This approach was taken to exclude the potential influence of prodromal symptoms of depression manifesting before the official diagnosis. Alcohol status (social drinker, consuming alcohol less than once a week; and current drinker, consuming alcohol more than once a week), smoking status (non-, past, and current smokers), systolic blood pressure, fasting glucose level, body mass index, total cholesterol level, outpatient visits per year, and number of admissions during follow-up period were also included as health status covariates. Comorbidity such as chronic kidney disease, congestive heart failure, hypertension, atrial fibrillation, chronic liver disease, diabetes mellitus, cancer, stroke, and ischemic heart disease were included to adjust. Several medications prescribed for a minimum of 90 days during the follow-up period, encompassing both outpatient visits and hospital admissions, such as antihypertensive medications, antihyperglycemic medications, statins, antiplatelet agents, and first- and second-generation antipsychotics were also included as covariates. All ICD-10 diagnostic and major ingredient codes for the medications used in the analyses are provided in Supplementary Table 1.
Statistical analysis
A conditional logistic regression model was employed to investigate the association between physical activity and the risk of dementia. As the patients were treated as self-referent controls, unbiased estimations of the hazard ratio (HR) were obtained by employing odds ratios derived from the logistic regression analysis [29–31]. Crude HRs with 95% confidence intervals (CI) and adjusted HRs (adjusted for all covariates) with 95% CI were calculated. Trend analyses (p for trend) were conducted by treating either the recommendation fulfillment rate or the MET-min/week as ordinal variables to investigate the potential dose-response association in a logistic model [32]. Subgroup analyses were performed based on age (individuals below their 50s, those in their 60s and 70s, and those aged ≥ 80 years) and sex. Although these variables were not matched, subgroup analyses were performed based on physical activity before depression onset and antidepressant use. Sensitivity analyses were conducted using alternative recommendation fulfillment rate criteria (0%, 0 < rate ≤ 25%, 25 < rate < 75%, 75 ≤ rate < 100%, 100%) and 600 MET-min/week during follow-up period, adopting different dementia diagnoses criteria (prescription of cognitive enhancers for at least 30 days after dementia diagnosis; donepezil, galantamine, rivastigmine, or memantine), and varying exclusion criteria for the minimum follow-up duration (no washout period; follow-up duration of less than three years). For all analyses, SAS Version 9.4 was utilized, and a two-tailed P-value of ≤ 0.05 was deemed statistically significant.