Data
We utilized a database of healthcare claims data from the Fukuoka Prefecture Wide-Area Association of the Latter-Stage Elderly Healthcare Insurance (LSEHI) and long-term care claims data from the Fukuoka Prefecture Wide-Area Association of the LTCI, from April 2014 to March 2019. These databases contain cost information for each beneficiary.
The LSEHI is an insurer for all individuals aged ≥75 years and those between 65 and 74 years who have a certain level of disorder. Each prefecture administrates this insurance, and the Fukuoka prefecture had 613,952 beneficiaries as of March 2015 [12]. The LSEHI database includes monthly healthcare data per person, such as disease diagnosis, healthcare procedures, medication, and healthcare costs.
The LTCI includes all individuals aged ≥40 years in Japan, who can receive services once approved should they need care for any reason when they are aged 65 years or older. As mentioned, there are seven care levels, with higher levels meaning people require more help. These levels are first determined by a computer, based on the examination of the application and the report from the patient’s doctor, after which the committee for certification reviews and confirms the right level of care. The limitation of long-term care services increases as the long-term care level rises. In Fukuoka prefecture, the number of approved individuals was 39,499 as of March 2019, with 33.0% requiring help and 32.0% with care level ≥ 3 [13]. The LTCI database includes monthly long-term care data per person, such as long-term care level, care details, and total long-term care costs. The database used in this study contains anonymized individual numbers from the LSEHI; thus, we linked these two databases using these anonymized numbers.
The names, residential addresses, and individual numbers of all beneficiaries were deidentified by constructing specific databases using a secured workstation (i.e., not connected to any network and within a locked room).
Study design and participants
This retrospective cohort study was conducted in Fukuoka, Japan. Patients included were those newly diagnosed with PD in the 2014 fiscal year (April 1, 2014–March 31, 2015). The participants were aged ≥75 years with a confirmed diagnosis using the ICD-10 code (G20: PD) and current prescription of antiparkinsonian agents using the therapeutic category code (116: antiparkinsonian agents). Those aged 65 to 74 were excluded to avoid selection bias because the insurance scheme and its coverage are conditional; only those with a specific disability are eligible to join the insurance. We gathered data from the first day of their PD diagnosis up to March 31, 2019. Data were also gathered from patients who lost their insurance eligibility, due to and up to their death.
Statistical analysis
The distributions of sex, age category, residential facility, long-term care level, and comorbidity (malignancy, ischemic heart disease, cerebrovascular disease, dyslipidemia, diabetes mellitus, and dementia) by physician visit frequency for PD treatment were examined using chi-square tests.
Physician visit frequency for PD treatment was divided into two groups: a higher frequency of physician visits and a lower frequency of physician visits. According to a 2014 report by the Ministry of Health, Labour and Welfare of Japan, that calculated the number of days that antiparkinsonian agents were prescribed for per one physician visit, approximately 70% of patients were prescribed antiparkinsonian agents for ≤30 days, and 90% were prescribed for ≤60 days [14]. In Japan, because medication can only be obtained for the specific amount prescribed, most patients were assumed to need to visit a physician at least once every 2 months. As the guidelines for PD in Japan do not specify the number of days for which medication should be prescribed for and healthcare claims data are compiled monthly, physician visit frequency was defined as a patient’s number of months that PD treatment was claimed for divided by the number of follow-up months: a higher frequency of physician visits was defined if the number of months of PD treatment divided by the number of follow-up months was ≥0.5, and lower frequency of physician visits if < 0.5. The age categories were divided into four groups: 75–79, 80–84, 85–89, and ≥ 90 years, reflecting the onset of PD. Residential facilities were categorized into home, retirement home, long-term care facility, and healthcare institute. Patients who received inpatient care for > 28 days per month were assumed to have lived in a healthcare institute. The long-term care levels were categorized into four levels: none, requiring help that can prevent long-term care needs, long-term care levels 1–2, and long-term care levels 3–5. Comorbidity refers to an identified prevalent disease in older adults, and we extracted the records of each disease without suspicion from the onset of PD to the end of this study period. A diagnosis of dementia was understood to include Lewy body dementia.
Survival analysis was conducted to evaluate the mortality of the study participants using the Kaplan-Meier method and generalized Wilcoxon test to compare higher and lower frequency of physician visits, based on survival months. Censoring was present due to participants becoming ineligible for insurance, such as moving outside of Fukuoka. Unadjusted and adjusted restricted mean survival time (RMST) were calculated to examine the effect of physician visit frequency and variables on RMST, and RMST differences, ratios, and 95% confidence intervals were also calculated. The RMST is a summary measure of the survival time distribution μ, defined as the area under the curve of the survival function up to a truncation time point τ (≤60), where S(t) is the survival function for time t for integration dt [15]. The function is as follows:
$$\mu ={\int}_0^{\tau }S(t) dt$$
The covariates were sex, age category, residential facility, long-term care level, and comorbidities. The reference measures were lower frequency of physician visits, male, aged 75–79 years, home resident, no long-term care level, and no comorbidity.
Generalized linear models (GLMs) with gamma distribution were constructed to evaluate healthcare services, healthcare, and long-term care costs among each variable. We calculated the number of inpatient and outpatient days, costs of healthcare services, and long-term care costs per month, using records from the database during this study period and the number of follow-up months. The independent variables were physician visit frequency, sex, age category, residential facility, long-term care level, mortality, and comorbidity. The reference measures were a lower frequency of physician visits, male, aged 75–79, home resident, no long-term care level, no morbidity, and no comorbidity. Following these analyses, we calculated the marginal means of days (each separately: inpatient and outpatient days for all diseases, inpatient and outpatient days for PD) and marginal means of cost (each separately: total of healthcare and long-term care costs, healthcare and inpatient costs for all diseases, healthcare and inpatient costs for PD, and long-term care costs, USD 1 = JPY 110) per month, per participant. Inpatient and outpatient days were assumed, except for participants with healthcare institutes for resident facilities.
We used Microsoft SQL server Management Studio 18 software to extract the data and Stata BE 17.0 (StataCorp LLC, College Station, TX, USA) for the analyses.
This study was approved by the Institutional Review Board of Kyushu University (Clinical Bioethics Committee of the Graduate School of Healthcare Sciences, Kyushu University).
