Study design, setting, data sources, and period

In this retrospective cohort study, we used data from the DeSC database from April 2014 to December 2022. The DeSC database is a large commercial medical claims database provided by DeSC Healthcare Co., Ltd., in Japan. It includes data from multiple insurers covering the North Kanto, Kinki, Tokai, and Shikoku regions, which encompass both urban and rural areas (Figure S1). Japan’s universal healthcare system is primarily based on occupational health insurance schemes16. However, individuals aged ≥ 75 years are enrolled in the regional Long-Life Medical Care System (LLMCS). Using data from the DeSC database, we focused on the LLMCS, which covers nearly the entire elderly population aged ≥ 75 years in the region, excluding only those receiving welfare assistance, due to the universal enrollment of this age group in Japan’s healthcare system.

Study population

The study participants comprised patients aged 75–89 years at the start of the observation period, who had at least two chronic conditions and had available data on their status (alive or deceased) at the end of the follow-up period. To enhance cohort homogeneity, we excluded individuals aged ≥ 90 years, a group known to have distinct clinical and functional characteristics17. Chronic conditions were defined based on the Fortin list, and conditions were considered chronic if they had been continuously recorded for at least 6 months7,9,18. The full list of conditions is provided in Table S1. We excluded individuals who were hospitalized at the start of the observation period to ensure that all participants began follow-up in the outpatient setting. (Fiugre S2) This allowed us to assess the association between outpatient visit patterns and subsequent hospitalizations. To ensure sufficient observation time for outpatient visit patterns, we excluded individuals with a follow-up period of less than six months after cohort entry19. The index date (cohort entry) was defined as the date of initial inclusion in the database for those who were already aged ≥ 75 years at the start of data collection, and as the date they turned 75 years for those who became eligible during the observation period. (Figure S2)

Outcomes

The primary outcome was all-cause mortality, which was derived from mortality data recorded in the claims database. Secondary outcomes included all-cause hospitalization and hospitalization due to ambulatory care-sensitive conditions (ACSC), which are conditions that may be prevented from progressing to severe stages requiring hospitalization through appropriate primary care interventions. ACSC-related hospitalizations were included as a secondary outcome because they are widely recognized indicators of primary care quality and continuity, and are known to increase with poor care coordination and fragmentation20,21. We hypothesized that patients with either no regular visits or excessively high numbers of visited facilities may have higher risks of ACSC-related hospitalizations due to poor care continuity or fragmentation, while those with a moderate number of regularly visited facilities would experience better coordination and fewer preventable hospitalizations. ACSCs were defined according to established literature, with hospitalizations classified as ACSC if the primary diagnosis during hospitalization matched the ACSC list22,23. Additionally, outpatient medical costs within the first year of observation were calculated to assess the financial burden associated with polydoctoring. Rather than serving as a direct measure of utilization, outpatient costs were used as a proxy for healthcare resource input in the outpatient setting, reflecting both the frequency and intensity of care. This approach allowed us to examine the association between the level of resource allocation and downstream outcomes.

Study variables

The main exposure variable was the number of regularly visited facilities (RVFs), which served as a proxy indicator for polydoctoring7,8,9. RVFs were defined as the number of healthcare facilities visited at least thrice a year and continuously for 6 months, calculated based on the 1 st year from the start of follow-up. Previous studies have shown that RVFs are associated with fragmented care in patients with multimorbidity and with polypharmacy and outpatient medical costs7. Covariates included age, sex, and the Charlson Comorbidity Index (CCI), which was calculated using ICD-10 codes, according to previous literature24. Given that previous studies have demonstrated regional differences in ACSC hospitalizations, stratification was conducted based on geographic region25,26. The dataset used for the analysis had no missing data for any of the key variables used in the study, including covariates such as age, sex, and geographic location. Therefore, all participants included in the analysis had complete information.

Statistical analysis

As this was an exploratory observational study, a convenience sample was used, and no prior sample size calculation was conducted. Age was treated as a continuous variable, while sex, RVF, CCI, and geographic region were treated as categorical variables. RVF and CCI values above the 95th percentile were grouped into a single category. RVF was categorized into six groups (0, 1, 2, 3, 4, and ≥ 5) based on facility visit frequency, and CCI was categorized similarly (0, 1, 2, 3, 4, and ≥ 5) according to the Charlson comorbidity score distribution.

Kaplan–Meier curves were plotted to illustrate the time to all-cause mortality, first hospitalization, and ACSC hospitalization. Kaplan–Meier curves stratified by CCI were reviewed to assess whether the CCI was appropriately adjusted for the risk of mortality associated with multimorbidity.

Multivariable Cox proportional hazard models were used to estimate adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for both primary and secondary outcomes. In each model, the dependent variable was the time to event (either all-cause mortality, all-cause hospitalization, or ACSC-related hospitalization), and the independent variables included RVF, age, sex, CCI, and geographic region. The time to event was measured from the start of the observation period to the occurrence of the specified outcome or the end of follow-up. Proportional hazard assumptions were assessed using log-log plots; this variable was treated as a stratification factor rather than a covariate, given that the assumption did not hold for geographic regions. (Figure S3)

For outpatient medical costs, adjusted cost ratios were calculated using a log-linear regression model after adding a constant of 1 to the actual cost. RVFs, age, sex, and CCI were included as covariates in the adjusted models. Due to the correlation between RVFs and CCI, an interaction term between these two variables was included as a covariate in all multivariable models (Figure S4). All tests were two-sided, and a p-value of < 0.05 was considered statistically significant. All analyses were conducted using R version 4.3.1 in RStudio version 2023.06.1.

Ethical considerations

The Ethics Committee of Keio University School of Medicine approved this study (ID 20231056). This retrospective study utilized anonymized claims data provided by DeSC Healthcare Co., Ltd., which were collected based on agreements with each insurer. Before being shared with the researchers, DeSC fully anonymized all data, and no personal identifiers were accessible to the investigators. Given this, the Ethics Committee of Keio University School of Medicine waived the requirement for individual informed consent. This study was conducted in accordance with the Declaration of Helsinki.