Impact of the mental health law revision restricting hospitalization on healthcare utilization in South Korea using interrupted time series analysis

Study population and data

Data analyzed during this study were acquired from the Korean National Health Insurance Sharing Service (NHISS)^9,10,11. The National Health Insurance Service (NHIS) routinely collects all claims data and provides a database for academic investigation and policymaking. The NHIS is the only insurer and covers approximately 97% of the total population of Korea. The claims data of patients receiving medical aid because of low income (approximately 3% of the total population) were also included in the database. We used diagnostic information and inpatient and outpatient utilization data from this database.

To explore the effect of the law revision on healthcare utilization by patients with psychiatric disorders, we targeted patients with severe symptoms that usually result in compulsory psychiatric department admission. We chose the psychiatric population by using the diagnostic codes defined by the International Classification of Disease, 10th revision (ICD-10). As psychotic and mood disorders constitute the majority of involuntary admissions, we selected patients with these conditions as our inclusion criteria for the analysis. From the dataset spanning from 2010 to 2021, we excluded data from 2010 to eliminate episodes involving individuals already hospitalized before the study period. Additionally, to avoid the confounding effects of the coronavirus disease 2019 pandemic on the evaluation of the impact of the law, we excluded data from 2020. Consequently, we analyzed 39,842,254 episodes, including 592,953 admission episodes and 39,249,301 outpatient episodes, from 2011 to 2019.

Variable of interest

In this study, the variable of interest was revision of the Mental Health and Welfare Law. The Mental Health and Welfare Law was revised on May 29, 2016, and enacted on May 30, 2017. Our analysis included monthly intervals; thus, we designated June 2017 as the time of implementation of the revision to the Mental Health and Welfare Law. Consequently, until May 2017, the intervention variable was coded as “0,” and from June 2017 onward, the intervention variable was coded as “1.”

Dependent variables and covariates

The dependent variables included metrics of healthcare utilization by the study population, such as number of admissions, length of hospitalization, and emergency department (ED) visits. The number of ED visits was defined as the monthly occurrences of psychiatric patients being admitted through the ED to reflect crisis situations. Additionally, the proportion of total hospital admissions that occurred through the ED was used in the analysis.

The independent variables in this study were age group (0–19 years, 20–29 years, 30–39 years, 40–49 years, 50–59 years, 60–69 years, or 70 years or older), sex (male or female), social security status (insurance or medical aid), income level (low, middle, or high), region (metropolitan, city, or rural), disability status (present or absent), and Charlson comorbidity index (low, middle, or high). Disability status in the data indicates whether an individual is registered as having a disability in the national welfare system. To assess patient comorbidities, the Charlson comorbidity index was used, a tool commonly applied in longitudinal studies using administrative data. The index was calculated by weighting scores of 1–6 for 19 comorbid diseases¹².

Statistical analysis

The chi-squared test was performed to evaluate differences in the characteristics of the study population before and after the law revision. To evaluate the impact of the law revision, we performed a single interrupted time-series analysis with segmented regression at individual levels^13,14,15,16. The following equation for interrupted time-series using a generalized estimating equation was used for the individual-level analysis^17,18,19.

In the aforementioned regression equation, “time” represents time on monthly basis, “intervention” is the dummy variable (which assigns a value of 1 if the time was after the law revision), “TAI” indicates the time after intervention (which is the time variable started at the intervention point), “X” indicates the covariate, and “\(\:\epsilon\:_it\)” represents the residual error. Regarding the regression coefficients in this model, intercept “\(\:\beta\:_0\)” estimates the baseline level of the outcome, “\(\:\beta\:_1\)” estimates the preintervention trend of the outcome, “\(\:\beta\:_2\)” estimates the level change after the intervention, thus indicating the immediate effect size of the intervention, and “\(\:\beta\:_3\)” estimates the slope change after the intervention.

The results are presented in terms of β coefficients, standard errors, and p-values, with significance considered at p < 0.05. For hospital admissions, ED visits, and readmissions for which a logit link function was used, the β coefficients were presented as exponentiated values. We categorized hospitals into two categories: general hospitals and psychiatric hospitals/clinics. While both types can perform involuntary admissions, psychiatric departments in general hospitals typically handle more cases involving acute conditions and shorter length of hospital stay, whereas psychiatric hospitals/clinics tend to manage more cases involving chronic conditions and longer length of hospital stay. To investigate whether the differences between these hospital types responded distinctively to the legislative amendment, we conducted a subgroup analysis using this classification. To verify the robustness of our results, we conducted a comparative interrupted time-series analysis. Here, the control group included patients with neurosis (ICD-10 F4x), as they were expected to be relatively less affected by the revised law.