A total of 23 hospitals participated in this study, including eight specialized hospitals for alcohol treatment (Dasarang Central Hospital, Aju Pyeonhan Hospital, Jin Hospital, W Jin Hospital, Dasarang Hospital, Hansarang Hospital, Yesarang Hospital and Onsarang Hospital), six general hospitals (Hanyang University Hospital, Konyang University Hospital, Hallym University Chuncheon Sacred Heart Hospital, Yangsan Pusan National University Hospital, Jesus Hospital, and Daejeon Eulji University Hospital), and nine psychiatric hospitals (Bugok National Hospital, Jeonbuk Maumsarang Hospital, Incheon Chamsarang Hospital, Yonkang Hospital, With Hospital, Hyeongju Hospital, Cheonju St. John Hospital, Haenam Hyemin Hospital, and Daedong Hospital). All participating hospitals offered the 4 levels of care except for level 0.5. Patients with level 0.5 were referred to community center for treatments. All patients began treatment at one of the 23 hospitals and each of the 23 hospitals treated at least one patient. Individuals with alcohol problems who visited these hospitals from March 20th, 2020, to December 31, 2021, and were newly entering treatment for an AUD were invited to participate in this study. People who agreed to participate were screened using the self-reported Alcohol Use Disorder Identification Test-Korean (AUDIT-K) [10]. Males with score of 10 or more and females with score of 6 or more were eligible for enrollment. Participants had to be ≥ 18 years of age with no confirmed history of brain damage, organic mental disorder, or suspected intellectual disability. All subjects were informed of the study purpose, contents, and potential risks and provided written consent. The study was approved by the institutional review boards (IRBs) of all participating institutes (IRB NO. HYUH 2020-01-032, IRB NO. 2019-02-010, IRB NO. 2020-03-010, IRB NO. 206-82-07306, IRB NO. 05–2021-007).

Enrolled participants underwent evaluations through self-report questionnaires and computer-facilitated intake assessments based on the KPPC at baseline and 1-month follow up. The questionnaires included questions about demographic characteristics (sex, education, occupation, marital status, socioeconomic status (SES)), AUDIT-K, AUDIT-Consumption (AUDIT-C), the Center for Epidemiologic Studies-Depression Scale (CES-D), Hanil Alcohol Insight Scale (HAIS), Readiness to Change Questionnaire (RCQ), and the Clinical Institute Withdrawal Assessment of Alcohol Scale, Revised (CIWA-Ar). Physical health condition, risk of relapse, and environmental risks were accessed through structured clinical interviews by case managers.

Each participant’s LOC was determined at every assessment, ranging from level 0.5 to 4 (0.5, 1, 2, 3, 4) [18]. These respective levels indicate early intervention (0.5), basic outpatient (1), intensive outpatient (2), basic inpatient (3), and intensive inpatient (4) treatment. Based on the LOC, participants were guided to receive interventions that could include medications (including naltrexone, acamprosate, antidepressants and anxiolytics), motivational enhancement therapy, cognitive behavioral therapy, 12-step programs, relapse prevention education, disease education, family therapy, hospital-based case management, community center-based case management, therapeutic communities, or Alcoholics Anonymous (Fig. 2). Some patients also utilized the Addiction Support Center, a resource within the South Korean substance use care system which offers assistance with diagnosis, access to economic resources, Alcoholics Anonymous meetings, and rehabilitation services. Case managers provided one to four counseling sessions per month to assist and engage subjects in treatment.

Interventions guideline for level of care. Programs for level 0.5 include sobriety program and community center-based case management. Medication include acamprosate, naltrexone, antidepressants and anxiolytics. Programs for level 1,2,3,4 include motivational enhancement therapy, cognitive behavioral therapy, 12-step programs, relapse prevention education, disease education, family therapy, hospital-based case management, community center-based case management, therapeutic communities, and Alcoholics Anonymous

Participants freely chose to receive the recommended LOC or not at every assessment. We categorized individuals who received treatment at the recommended LOC or higher as KPPC-matched and those who received lower level treatment as KPPC-mismatched. We compared patients regarding duration of alcohol abstinence and number of treatment completions according to the number of KPPC-matched treatments they received: 0, 1, or 2.

The outcomes were assessed at 1-month and 3-months follow up. Duration of alcohol abstinence was determined by self-reported AUDIT-C score of 0, indicating one month of no alcohol. We assessed the AUDIT-C score at one- and three-month follow-up and summed up the number of months with no alcohol for each participant—0, 1, or 2. Individuals who missed the second AUDIT-C assessment at the 3-month follow up were considered to be not alcohol abstinent for the month. Treatment program completion was defined as successful conclusion of a 30-day treatment program. The number of months with successful treatment completion was recorded—0, 1, 2, or 3 out of three-months. We also conducted separate analyses to determine if the effect of adhering to KPPC recommendations varied based on whether the first referral was for inpatient or outpatient care.

We used the KPPC to decide patients’ LOC which includes multiple scales and questionnaires in six dimensions: (1) Intoxication and Withdrawal (2) Biomedical Conditions and Complications, (3) Emotional or Behavioral Conditions, (4) Treatment Readiness, (5) Potential for Relapse, and (6) Environmental Conditions (Fig. 1, [18]). The scales included were AUDIT-K, AUDIT-C, CES-D, RCQ, HAIS, and CIWA-Ar. Besides the scales, the assessment included participants’ subjective report on self-health, diagnosis and treatment history, presence of suicidal ideation, use of community addiction management center, drug compliance, subjective report of craving, and environmental factors such as family, occupation, friends, religion, and financial status. KPPC was evaluated twice—at baseline and 1-month follow-up. We compared the participants by the number of times they received KPPC-matched treatments—0, 1 and 2.

The control variables included biological sex, employment status, and previous hospitalization for AUD. These factors were controlled because male gender is associated with a higher risk of lifetime AUD [7], and being employed is linked to better abstinence outcomes in individuals with AUD [5]. Hospitalizations are known to promote behavior change by initiating medication for AUD [15], which might have positive effect on the outcomes.

Multiple contingency tables were formed using the number of KPPC-matched treatments and each of the other variables. Subsequently, frequency analysis followed by the Chi-square or Mantel–Haenszel Chi-square test was performed to examine the baseline characteristics of study participants and to gauge the effects of the KPPC-matched treatments on primary outcomes. These variables were also provided with Cramer's V as the effect size. For the primary outcome, a post-hoc power analysis was conducted using GPower software [3]. The effect size used in G*Power version 3.1 [4] was w, calculated as w = Cramer’s V × √(r-1), where r represents the number of categories in the smaller variable of the contingency table [3].

We conducted ordinal logistic regression analyses with an unadjusted model and an adjusted model that added potential confounders to the unadjusted model [1]. The predictor variable of primary interest was the number of KPPC-matched treatments, and the confounders were biological sex, employment status, and prior hospitalization for AUD.

For each outcome variable which was measured on an ordinal scale, the satisfaction of the proportional odds assumption was checked before attempting to interpret the results of the two ordinal logistic regression models. Additionally, to identify the better model, the likelihood ratio test based on -2LLs from the two models was performed while the model fit statistics, AIC and SC were also compared between the models [19].

For each model, odds ratios (ORs) and 95% confidence intervals (CIs) were examined to judge substantive significance and p values were reviewed for statistical significance. Quasi-complete separation [23] was also checked to ensure the integrity of the analysis results.

Meanwhile, the same ordinal logistic regression models were separately fitted for subgroups based on inpatient and outpatient care to differentiate between referrals to inpatient and outpatient settings.