Factors Associated with Korean Medicine Utilization during the COVID-19 Pandemic Period (2019–2023): A Panel Analysis of the Korea Health Panel

Article information

J Korean Med. 2026;47(3):211-235
Publication date (electronic) : 2026 September 1
doi : https://doi.org/10.13048/jkm.26045
1Ph.D. candidate, Graduate School of Public Administration, Seoul National University
2Research professor, Research Institute for Korean Medicine, Pusan National University
*Correspondence to: Changyon Han, 1 Gwanak-ro, Gwanak-gu, Seoul, 08826, Republic of Korea, Graduate School of Public Administration, Seoul National University, 49, Busan University Road, Mulgeum-eup, Yangsan-si, Gyeongnam, Pusan National University, Tel: +82-51-510-8449, Fax: +82-51-510-8450, E-mail: darae0@snu.ac.kr; cyhan@pusan.ac.kr
Received 2026 July 2; Revised 2026 July 30; Accepted 2026 August 12.

Abstract

Objectives

This study investigates factors associated with Korean Medicine (KM) outpatient utilization in 2019–2023, a period that includes the COVID-19 pandemic, using the Andersen behavioral model. It analyzes individual-level panel data to distinguish factors associated with entry (whether KM is used) from those associated with intensity (how much is used), across three dimensions.

Methods

This study estimated three outcomes using household-member random-effects panel models in accordance with Andersen’s behavioral model: KM outpatient utilization (panel logit), annual visit days (random-effects negative binomial, users only), and annual expenditure (gamma generalized linear model (GLM) with a log link, users only). Unweighted random-effects models were used for the main analysis, and complex-survey design-based estimates using the longitudinal weight were provided as a supplementary check.

Results

Predisposing, enabling, and need factors were all significantly associated with KM utilization, although the associated dimensions differed by factor group. Predisposing factors—older age, female sex, and marriage—were associated with entry and visit days, but not with expenditure. Enabling factors were associated with all three dimensions, most prominently with expenditure, which increased with household income quintile, whereas private insurance was associated with entry. Need factors—the presence of chronic disease (rather than comorbidity severity, CCI) and poorer self-rated health—were consistently associated with higher visit days and expenditure, and health-conscious individuals used KM more.

Conclusions

Within the two-part framework, KM outpatient use in 2019–2023 showed a dual pattern: entry was associated with socioeconomic factors, and intensity with health need, consistent with KM’s role as a complementary medical service.

(1) g (E[Yit∣Pit, Eit, Nit, ui])=β0+β1 Pit+β2 Eit+β3Nit+λt+ui,         ui~N(0,σu2)
Fig. 1

Age-specific Korean Medicine utilization rate by year, 2019, 2021, and 2023

Note. Weighted logistic regression with restricted cubic spline in age (5 knots). Separate models were fitted for each survey year (2019, 2021, and 2023), weighted by the cross-sectional individual sampling weight (I_WSC). Knots were placed at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles of the age distribution (the default placement in Stata's command). Ages 18–85; denominator includes both users and non-users. Analyses are restricted to adults aged 18–85 years; those aged over 85 years were excluded owing to sparse observations. Estimates are based on the Korea Health Panel Survey (2019, 2021, and 2023).

Fig. 3

Forest plot of significant factors associated with Korean Medicine expenditure (Gamma GLM, users only, 2019–2023)

Note. Effect estimates from an unweighted random-effects gamma generalized linear model (log link) for annual Korean Medicine outpatient expenditure among users (Expenditure > 0), 2019–2023. Values are exponentiated coefficients (orange diamonds), interpretable as multiplicative cost ratios; the dashed vertical line marks the null (ratio = 1). For example, a ratio of 1.20 implies 20% higher expenditure. Markers are point estimates with 95% confidence intervals. The same seven covariates are shown; the log-transformed continuous variables (Age, Income) and the reference categories for the categorical covariates are defined in the note to Figure 2. Expenditure is modeled on the original KRW scale via the log link (not log-transformed), avoiding retransformation bias.

Fig. 2

Forest plot of significant factors associated with Korean Medicine utilization and visit days (Random-effects panel models, 2019–2023)

Note. OR (Odds Ratio). IRR (Incidence Rate Ratio). Effect estimates from unweighted random-effects panel models (2019–2023). Utilization is shown as odds ratios (OR, navy circles) from a panel logit model; Days as incidence-rate ratios (IRR, red squares) from a random-effects negative binomial model among users (visits > 0). Markers are point estimates and horizontal bars are 95% confidence intervals; the dashed vertical line marks the null value (OR/IRR = 1). The figure displays seven selected covariates. Age and income are entered as natural logs. For the categorical covariates, estimates are relative to the following reference groups: never-married for Married; National Health Insurance for National Medical Aid; having a chronic disease for No chronic; very poor self-rated health for Good health; and smoker for Non-smoker. Because the two outcomes use different link functions, OR and IRR are not directly comparable in magnitude.

Descriptive Statistics of the Study Population by Year, Korea Health Panel 2019–2023 (Unweighted)

Korean Medicine Outpatient Utilization, Days, and Expenditure by Year: Unweighted and Cross-Sectionally Weighted Estimates, Korea Health Panel 2019–2023

Factors Associated with Korean Medicine Utilization, Days, and Expenditure: Panel Random-Effects and Survey-Weighted Estimates, 2019–2023

Summary of Hypothesis Tests: Joint Wald Tests by Andersen Factor Group

References

1. Lim B. M.. 2013;Korean medicine coverage in the national health insurance in Korea: Present situation and critical issues. Integrative Medicine Research 2(3):81–88.
2. Park J., Yi E., Yi J.. 2021;The provision and utilization of traditional Korean medicine in South Korea: Implications on integration of traditional medicine in a developed country. Healthcare 9(10):1379.
3. Ministry of Health and Welfare. 2024 Korean medicine utilization and herbal medicine consumption survey [Internet] Bureau of Korean Medicine; 2025. [cited 2026 Apr 10]. Available from: URL:https://www.mohw.go.kr/board.es?mid=a10503010100&bid=0027&tag=&act=view&list_no=1485130&cg_code=.
5. Han C. Y.. 2025;Blinder-Oaxaca decomposition of changes in Korean medicine utilization: Evidence from the Korea Health Panel Survey 2009 and 2018. Journal of Korean Medicine 46(3):71–85.
6. Moynihan R., Sanders S., Michaleff Z. A., Scott A. M., Clark J., To E. J., et al. 2021;Impact of COVID-19 pandemic on utilisation of healthcare services: A systematic review. BMJ Open 11(3):e045343.
7. Andersen R. M.. 1995;Revisiting the behavioral model and access to medical care: Does it matter? Journal of Health and Social Behavior 36(1):1–10.
8. Park M. J., Kwon S. M.. 2014;Socioeconomic determinants of Korean medicine ambulatory services: Comparing panel fixed effect model with pooled ordinary least squares. Health Policy and Management 24(1):47–55.
9. Park M. J., Ko S. G.. 2015;The effect of private health insurance coverage reduction on the Korean medicine utilization. Journal of Society of Preventive Korean Medicine 19(3):57–66.
10. Seo S. K., Park B. H., Choi S. J.. 2016;A study on the sociodemographic characteristics of adult users of Korean traditional medicine. Journal of Korean Public Health Nursing 30(1):136–148.
11. Choi B. H., Son J. H., Lim B. M.. 2014;The effect of Korean medicine utilization on Western medicine use for musculoskeletal diseases: Using the Korea Health Panel data. Journal of Korean Medicine 35(3):22–31.
12. Shin B. Y.. 2025. Analysis of age-period-cohort effects on the use of Korean medicine ambulatory service. Master’s thesis Seoul National University Graduate School;
13. Oh H. J.. 2021. The influence of income level on herbal medicine utilization in Korean medical clinic. Master’s thesis Seoul National University Graduate School;
14. Yoon L. S., Lim B.. 2021;The determinants and behavioral intentions of Korean medicine utilization in youth aged 19–39 years: Based on the microdata of national survey on Korean medicine utilization and herbal medicine consumption, 2017. Journal of Society of Preventive Korean Medicine 25(2):85–98.
15. Lim J. W., Lee K. J.. 2021;Influencing factors of using Korean medicine services: Focusing on the 2017 Korean medicine utilization survey. Journal of Korean Medicine 42(1):12–25.
16. Park J. S.. 2023;Factors affecting the utilization of Korean medicine in the last 1 year: Analysis of the national survey on Korean medicine utilization, 2020. Journal of Society of Preventive Korean Medicine 27(2):49–59.
17. Sung A. R., Park H., Kim H., Lee S. D.. 2017;The patterns of Korean medicine usage with coverage by private health insurance: Based on usage and consumption of Korean medicine report in 2014. Journal of Society of Preventive Korean Medicine 21(1):67–81.
18. Yoo J. H., Kim Y. J., Koo B. S., Lee S. W.. 2011;Differences in utilization behaviors of Western and Oriental medical care according to the Korea National Health and Nutrition Examination Survey. Journal of Physiology & Pathology in Korean Medicine 25(3):582–588.
19. Jung B., Bae S., Kim S.. 2017;Use of Western medicine and traditional Korean medicine for joint disorders: A retrospective comparative analysis based on Korean nationwide insurance data. Evidence-. Based Complementary and Alternative Medicine 2017(1):2038095.
20. Yoon H. C.. 2024;The comparative study on change of utilization in Korean medicine before and after COVID-19. Journal of Society of Preventive Korean Medicine 28(2):67–83.
21. Jang S., Kim D., Yi E., Choi G., Song M., Lee E. K.. 2021;Telemedicine and the use of Korean medicine for patients with COVID-19 in South Korea: Observational study. JMIR Public Health and Surveillance 7(1):e20236..
22. Yu C. H., Kim K. I., Jung H. J., Lee B. J.. 2023;Effects of Korean medicine treatment at public health center for post-acute COVID-19 patients: A retrospective case-series of 11 patients. Journal of Korean Medicine 44(2):132–148.
23. Quan H., Sundararajan V., Halfon P., Fong A., Burnand B., Luthi J. C., et al. 2005;Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Medical Care 43(11):1130–1139.
24. Manning W. G., Mullahy J.. 2001;Estimating log models: To transform or not to transform? Journal of Health Economics 20(4):461–494.
25. Solon G., Haider S. J., Wooldridge J. M.. 2015;What are we weighting for? Journal of Human Resources 50(2):301–316.

Article information Continued

Fig. 1

Age-specific Korean Medicine utilization rate by year, 2019, 2021, and 2023

Note. Weighted logistic regression with restricted cubic spline in age (5 knots). Separate models were fitted for each survey year (2019, 2021, and 2023), weighted by the cross-sectional individual sampling weight (I_WSC). Knots were placed at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles of the age distribution (the default placement in Stata's command). Ages 18–85; denominator includes both users and non-users. Analyses are restricted to adults aged 18–85 years; those aged over 85 years were excluded owing to sparse observations. Estimates are based on the Korea Health Panel Survey (2019, 2021, and 2023).

Fig. 2

Forest plot of significant factors associated with Korean Medicine utilization and visit days (Random-effects panel models, 2019–2023)

Note. OR (Odds Ratio). IRR (Incidence Rate Ratio). Effect estimates from unweighted random-effects panel models (2019–2023). Utilization is shown as odds ratios (OR, navy circles) from a panel logit model; Days as incidence-rate ratios (IRR, red squares) from a random-effects negative binomial model among users (visits > 0). Markers are point estimates and horizontal bars are 95% confidence intervals; the dashed vertical line marks the null value (OR/IRR = 1). The figure displays seven selected covariates. Age and income are entered as natural logs. For the categorical covariates, estimates are relative to the following reference groups: never-married for Married; National Health Insurance for National Medical Aid; having a chronic disease for No chronic; very poor self-rated health for Good health; and smoker for Non-smoker. Because the two outcomes use different link functions, OR and IRR are not directly comparable in magnitude.

Fig. 3

Forest plot of significant factors associated with Korean Medicine expenditure (Gamma GLM, users only, 2019–2023)

Note. Effect estimates from an unweighted random-effects gamma generalized linear model (log link) for annual Korean Medicine outpatient expenditure among users (Expenditure > 0), 2019–2023. Values are exponentiated coefficients (orange diamonds), interpretable as multiplicative cost ratios; the dashed vertical line marks the null (ratio = 1). For example, a ratio of 1.20 implies 20% higher expenditure. Markers are point estimates with 95% confidence intervals. The same seven covariates are shown; the log-transformed continuous variables (Age, Income) and the reference categories for the categorical covariates are defined in the note to Figure 2. Expenditure is modeled on the original KRW scale via the log link (not log-transformed), avoiding retransformation bias.

Fig. 2

Forest plot of significant factors associated with Korean Medicine utilization and visit days (Random-effects panel models, 2019–2023)

Note. OR (Odds Ratio). IRR (Incidence Rate Ratio). Effect estimates from unweighted random-effects panel models (2019–2023). Utilization is shown as odds ratios (OR, navy circles) from a panel logit model; Days as incidence-rate ratios (IRR, red squares) from a random-effects negative binomial model among users (visits > 0). Markers are point estimates and horizontal bars are 95% confidence intervals; the dashed vertical line marks the null value (OR/IRR = 1). The figure displays seven selected covariates. Age and income are entered as natural logs. For the categorical covariates, estimates are relative to the following reference groups: never-married for Married; National Health Insurance for National Medical Aid; having a chronic disease for No chronic; very poor self-rated health for Good health; and smoker for Non-smoker. Because the two outcomes use different link functions, OR and IRR are not directly comparable in magnitude.

Table 1

Descriptive Statistics of the Study Population by Year, Korea Health Panel 2019–2023 (Unweighted)

Year

2019 2020 2021 2022 2023 Total Test
N 7,995 (22.1%) 7,167 (19.8%) 7,319 (20.3%) 6,933 (19.2%) 6,719 (18.6%) 36,133 (100.0%)
 User 380 (4.8%) 370 (5.2%) 422 (5.8%) 392 (5.7%) 389 (5.8%) 1,953 (5.4%) 0.016
Sex (ref: male)
 Female 4,549 (56.9%) 4,031 (56.2%) 4,173 (57.0%) 3,970 (57.3%) 3,852 (57.3%) 20,575 (56.9%) 0.708
Age (18 years old or over) 52.408 (14.790) 53.419 (14.870) 54.288 (15.065) 55.288 (15.122) 56.190 (15.274) 54.245 (15.075) <0.001
Age (ref: 18–64 years old) <0.001
 65 years old or over 1,889 (23.6%) 1,922 (26.8%) 2,186 (29.9%) 2,302 (33.2%) 2,439 (36.3%) 10,738 (29.7%)
Education (ref: High school) 0.996
 College and university 2,861 (35.8%) 2,573 (35.9%) 2,630 (35.9%) 2,493 (36.0%) 2,415 (35.9%) 12,972 (35.9%)
 Graduate school 291 (3.6%) 272 (3.8%) 276 (3.8%) 273 (3.9%) 262 (3.9%) 1,374 (3.8%)
Marital status (ref: Never) 0.024
 Married 5,927 (74.1%) 5,271 (73.5%) 5,304 (72.5%) 5,009 (72.2%) 4,841 (72.0%) 26,352 (72.9%)
 Married but not now 1,074 (13.4%) 1,002 (14.0%) 1,056 (14.4%) 1,040 (15.0%) 1,035 (15.4%) 5,207 (14.4%)
Region (ref: Metropolitan)
 Non-metropolitan 6,689 (83.7%) 6,087 (84.9%) 6,232 (85.1%) 5,904 (85.2%) 5,745 (85.5%) 30,657 (84.8%) 0.017
Income per capita (log) 7.299 (0.713) 7.387 (0.667) 7.529 (0.542) 7.591 (0.542) 7.606 (0.606) 7.476 (0.632) <0.001
Income quintile per household (ref: 1st: lowest) 0.267
 2nd 1,589 (19.9%) 1,418 (19.8%) 1,543 (21.1%) 1,486 (21.4%) 1,378 (20.5%) 7,414 (20.5%)
 3rd 1,765 (22.1%) 1,593 (22.2%) 1,558 (21.3%) 1,459 (21.0%) 1,497 (22.3%) 7,872 (21.8%)
 4th 1,773 (22.2%) 1,574 (22.0%) 1,559 (21.3%) 1,444 (20.8%) 1,424 (21.2%) 7,774 (21.5%)
 5th (Highest) 1,781 (22.3%) 1,635 (22.8%) 1,645 (22.5%) 1,565 (22.6%) 1,496 (22.3%) 8,122 (22.5%)
Public insurance (ref: National Health Insurance) 0.889
 National Medical Aid 147 (1.8%) 139 (1.9%) 144 (2.0%) 123 (1.8%) 121 (1.8%) 674 (1.9%)
Private insurance (ref: Yes) <0.001
 No 2,773 (34.7%) 2,418 (33.7%) 2,338 (31.9%) 2,129 (30.7%) 1,970 (29.3%) 11,628 (32.2%)
Working (ref: Yes) 0.004
 No 2,390 (29.9%) 2,083 (29.1%) 2,045 (27.9%) 1,941 (28.0%) 1,835 (27.3%) 10,294 (28.5%)
Chronic diseases (ref: Yes) <0.001
 No 4,527 (56.6%) 3,803 (53.1%) 3,650 (49.9%) 3,140 (45.3%) 2,881 (42.9%) 18,001 (49.8%)
Self-rated health (ref: 1 Very bad) <0.001
 Bad 998 (12.5%) 939 (13.1%) 1,043 (14.3%) 999 (14.4%) 964 (14.3%) 4,943 (13.7%)
 Normal 3,792 (47.4%) 3,497 (48.8%) 3,305 (45.2%) 3,228 (46.6%) 3,105 (46.2%) 16,927 (46.8%)
 Good 2,800 (35.0%) 2,409 (33.6%) 2,664 (36.4%) 2,423 (34.9%) 2,365 (35.2%) 12,661 (35.0%)
 Very good 338 (4.2%) 277 (3.9%) 254 (3.5%) 217 (3.1%) 234 (3.5%) 1,320 (3.7%)
Disability (ref: Yes) 0.272
 No 7,629 (95.4%) 6,824 (95.2%) 6,946 (94.9%) 6,574 (94.8%) 6,366 (94.7%) 34,339 (95.0%)
Charlson comorbidity index score (0–7) 0.221 (0.553) 0.243 (0.580) 0.267 (0.614) 0.297 (0.647) 0.319 (0.671) 0.267 (0.613) <0.001
Smoking (ref: Yes) 0.309
 No 5,265 (65.9%) 4,611 (64.3%) 4,734 (64.7%) 4,518 (65.2%) 4,398 (65.5%) 23,526 (65.1%)
Drinking (ref: Yes) <0.001
 No 3,746 (46.9%) 3,341 (46.6%) 3,673 (50.2%) 3,480 (50.2%) 3,153 (46.9%) 17,393 (48.1%)
Regular exercise (ref: Yes) <0.001
 No 3,774 (47.2%) 3,508 (48.9%) 3,579 (48.9%) 3,128 (45.1%) 2,873 (42.8%) 16,862 (46.7%)
Unmet medical need (ref: Yes) <0.001
 No 6,272 (87.5%) 6,359 (86.9%) 6,124 (88.3%) 6,088 (90.6%) 24,843 (88.3%)

Note. Values are n (%) for categorical variables and mean (SD) for continuous variables (Age, Income per capita(log), Charlson comorbidity index score). The analytic sample is the estimation sample of the main utilization model (N=36,133; adults aged 18+, unweighted). The Test column reports differences across years: Pearson’s chi-squared test for categorical variables and one-way ANOVA (regression-based) for continuous variables. Because of the large sample size, even small differences may reach statistical significance; the magnitude of the year-to-year change should be considered alongside the p-value. Unmet medical need was surveyed only in 2020–2023 and is therefore shown for those years (2019 not available; N=28,138). Expenditure denotes annual out-of-pocket payments for Korean Medicine outpatient care, measured in Korean won (KRW) and presented in 1,000 KRW in the tables. Days denote annual KM outpatient visit days (days per year; users only).

Table 2

Korean Medicine Outpatient Utilization, Days, and Expenditure by Year: Unweighted and Cross-Sectionally Weighted Estimates, Korea Health Panel 2019–2023

Year N(user) Unweighted Weighted

Utilization Rate Days Expenditure Utilization Rate Days Expenditure
2019 7995(380) 0.048 15.268 208.360 0.046 13.823 239.152
2020 7167(370) 0.052 14.457 229.412 0.047 12.138 266.489
2021 7319(422) 0.058 17.346 232.792 0.051 14.186 244.001
2022 6933(392) 0.057 17.301 270.179 0.050 15.056 319.510
2023 6719(389) 0.058 16.645 293.921 0.053 13.967 355.530

Note. Utilization is the proportion of individuals with at least one Korean Medicine outpatient visit in the year; Days and Expenditure are means among users only (visits > 0), with expenditure expressed in 1,000 KRW. Unweighted columns report raw sample estimates; Weighted columns apply the cross-sectional individual sampling weight (I_WSC) to yield year-specific population-representative estimates. The longitudinal weight (I_WSL) is not used here, as it is unavailable for 2019 and applies only to the tracked panel sample; it is reserved for the panel regression analysis. Expenditure denotes annual out-of-pocket payments for Korean Medicine outpatient care, measured in Korean won (KRW) and presented in 1,000 KRW in the tables. Visit days denote annual KM outpatient visit days (days per year; users only). Among users in the unweighted estimation sample, visit days and expenditure were right-skewed (skewness = 3.100 and 4.450; kurtosis = 16.475 and 31.243, where 3 indicates a normal distribution; n = 1,953 and 1,904, respectively). No outlier trimming was applied, and these skewed distributions were accommodated by the negative binomial and gamma (log-link) specifications in the regression analysis. All values are rounded to three decimals.

Table 3

Factors Associated with Korean Medicine Utilization, Days, and Expenditure: Panel Random-Effects and Survey-Weighted Estimates, 2019–2023

Variables Panel Random-Effects Survey-Weighted Estimates

(1) Utilization (Panel logit) (2) Days (Panel NB) (3) Expenditure (Gamma GLM) (4) Utilization (Survey logit) (5) Days (Survey NB) (6) Expenditure (Survey gamma GLM)
Sex (ref: male)
 Female 0.229** (0.098) −0.203** (0.085) 0.006 (0.119) 0.148 (0.121) −0.175 (0.141) 0.147 (0.178)
Age(log) 0.642*** (0.193) 1.083*** (0.168) −0.266 (0.233) 0.325 (0.258) 1.364*** (0.313) 0.304 (0.325)
Education (ref: High school)
 College and university 0.046 (0.085) 0.035 (0.074) −0.008 (0.102) 0.013 (0.103) 0.196* (0.119) 0.485*** (0.147)
 Graduate school 0.103 (0.171) 0.161 (0.147) 0.256 (0.205) 0.054 (0.193) 0.275 (0.248) 0.577** (0.268)
Marital status (ref: Never)
 Married 0.571*** (0.150) −0.026 (0.140) 0.313 (0.191) 0.590*** (0.190) −0.284 (0.251) 0.324 (0.243)
 Married but not now 0.541*** (0.172) 0.084 (0.156) 0.455** (0.214) 0.501** (0.208) 0.043 (0.282) 0.531* (0.277)
Region (ref: Metropolitan)
 Non-metropolitan −0.118 (0.086) −0.195*** (0.071) −0.127 (0.100) 0.030 (0.103) 0.033 (0.109) 0.314*** (0.119)
Income per capita (log) 0.149** (0.075) 0.028 (0.062) 0.089 (0.083) 0.289*** (0.102) −0.092 (0.132) 0.141 (0.134)
Income quintile per household (ref: 1st: lowest)
 2nd 0.015 (0.112) 0.050 (0.094) 0.217* (0.131) −0.016 (0.153) 0.332* (0.177) 0.264 (0.207)
 3rd −0.143 (0.134) 0.106 (0.111) 0.530*** (0.155) −0.344* (0.184) 0.298 (0.210) 0.516** (0.254)
 4th −0.156 (0.151) 0.049 (0.126) 0.502*** (0.174) −0.354* (0.201) 0.306 (0.249) 0.272 (0.274)
 5th (Highest) −0.027 (0.173) 0.075 (0.144) 0.563*** (0.197) −0.336 (0.237) 0.332 (0.283) 0.505 (0.323)
Public insurance (ref: National Health Insurance)
 National Medical Aid −0.051 (0.240) 0.377* (0.203) −0.754** (0.305) −0.401 (0.269) 0.643 (0.426) 0.053 (0.646)
Private insurance (ref: Yes)
 No −0.153** (0.070) 0.046 (0.058) −0.080 (0.081) −0.073 (0.091) −0.008 (0.101) −0.188 (0.141)
Working (ref: Yes)
 No 0.019 (0.069) 0.202*** (0.057) 0.238*** (0.079) 0.041 (0.088) 0.124 (0.095) 0.082 (0.117)
Chronic diseases (ref: Yes)
 No 0.153** (0.077) −0.221*** (0.064) −0.250*** (0.089) −0.003 (0.104) −0.355*** (0.112) −0.264* (0.146)
Self-rated health (ref: 1 Very bad)
 Bad 0.171 (0.301) −0.231 (0.219) −0.315 (0.304) 0.077 (0.428) −0.749 (0.467) −0.890 (0.608)
 Normal −0.038 (0.301) −0.422* (0.221) −0.568* (0.306) −0.012 (0.425) −0.879* (0.459) −1.335** (0.598)
 Good −0.052 (0.303) −0.481** (0.223) −0.685** (0.309) −0.063 (0.430) −1.043** (0.460) −1.358** (0.612)
 Very good −0.378 (0.342) −0.589** (0.265) −0.771** (0.368) −0.325 (0.480) −1.232** (0.523) −1.721*** (0.665)
Disability (ref: Yes)
 No −0.071 (0.143) −0.218* (0.115) 0.057 (0.167) −0.090 (0.175) −0.193 (0.181) 0.465** (0.194)
Charlson comorbidity index score (0–7) −0.085 (0.055) 0.012 (0.043) −0.093 (0.061) −0.121* (0.073) −0.001 (0.077) −0.079 (0.093)
Smoking (ref: Yes)
 No 0.267*** (0.100) 0.271*** (0.087) 0.198 (0.121) 0.258** (0.127) 0.376*** (0.142) 0.115 (0.189)
Drinking (ref: Yes)
 No 0.141** (0.065) 0.057 (0.056) 0.072 (0.077) 0.149* (0.079) 0.092 (0.098) 0.174 (0.123)
Regular exercise (ref: Yes)
 No −0.280*** (0.057) −0.111** (0.049) −0.139** (0.066) −0.139* (0.077) −0.232*** (0.086) −0.054 (0.112)
Year (ref: 2019, weighted: 2020)
 2020 0.081 (0.081) −0.090 (0.067) 0.061 (0.091)
 2021 0.178** (0.080) −0.012 (0.067) 0.105 (0.092) 0.039 (0.095) 0.075 (0.106) −0.050 (0.146)
 2022 0.118 (0.083) −0.026 (0.070) 0.193** (0.096) −0.019 (0.097) 0.175 (0.119) 0.094 (0.149)
 2023 0.142* (0.084) −0.106 (0.071) 0.064 (0.097) 0.060 (0.101) 0.058 (0.111) 0.235 (0.159)
Constant −7.963*** (0.864) −1.553** (0.738) 11.910*** (1.021) −6.795*** (1.109) −1.444 (1.363) 9.769*** (1.549)
Overdispersion, ln(α) −1.001*** (0.067) 0.314*** (0.083)

Random-effect variance, var(u) 0.557*** (0.041) 1.271*** (0.082)
Observations (Group/Population) 36133 (9584) 1953 (1488) 1904 (1457) 28482 (30122.01) 1565 (1504.51) 1531 (1471.34)
Intraclass correlation, ρ 0.363 — — — — —
Model Wald χ² F-tests/design-based F 208.99*** 343.58*** 107.67*** 3.60*** 7.06*** 3.88***
LR test (RE/overdispersion) 477.8*** 346.5*** 479.4*** — — —

Note. Models (1)–(3): unweighted household random-effects panel models (2019–2023); Models (4)–(6): complex-survey design-based estimates with the longitudinal weight (2020–2023), household member. Utilization = panel logit; Days = random-effects negative binomial(NB); Expenditure = gamma GLM (log link) on users. Weighted Days/Expenditure use the user subpopulation (n = 1,565/1,531); variance estimation uses the full design sample. ρ = between-individual share of variance (utilization). Likelihood Ratio(LR) tests assess the random effect (utilization: ρ=0; expenditure: vs. gamma) and overdispersion (days: vs. NB); all p < 0.001. Overall significance of weighted models is assessed by design-based F tests (Models (4)–(5)); for Model (6) the design-based Wald F is obtained by post-estimation joint test. Standard errors in parentheses.

*

p < 0.10,

**

p < 0.05,

***

p < 0.01.

Table 4

Summary of Hypothesis Tests: Joint Wald Tests by Andersen Factor Group

Groups Factors χ2 Degree of freedom p-value
Utilization Predisposing factors 75.066 7.000 <0.001
Enabling factors 17.697 8.000 0.024
Need factors 53.354 10.000 <0.001

Days Predisposing factors 83.266 7.000 <0.001
Enabling factors 18.691 8.000 0.017
Need factors 55.424 10.000 <0.001

Expenditure Predisposing factors 8.519 7.000 0.289
Enabling factors 52.965 8.000 <0.001
Need factors 35.360 10.000 <0.001

Note. Utilization (panel logit, entry); Days (random-effects negative binomial, users only); Expenditure (gamma GLM with log link, users only). Each row reports the joint Wald test that all coefficients of the factor group equal zero in the corresponding model. Hypotheses 1–3 are evaluated primarily on the entry model, with the intensity models informing the exploratory research question on stage-specific associations.