Future demand for hospitalization is typically projected by applying the rate of inpatient care in the overall population to the estimated future population [15]. However, this method is unsuitable for projecting long-term hospital bed demand because of the small fluctuations in the admission and discharge of long-term inpatients. Therefore, we adopted the cohort-component method, which is widely used in demography [16] with the aim to analyze trends in the number of long-term inpatients in psychiatric care beds in Japan. Long-term hospitalization was defined as a hospital stay lasting 1 year or more. Our projection method is shown in Figure 1.
The total number of long-term inpatients in 2017 and the number of long-term inpatients by age group were taken from the 2017 Patient Survey [17]. The Patient Survey is an official statistical survey conducted once every 3 years by the MHLW that randomly samples patients from medical facilities throughout Japan to create a snapshot of the injuries and illnesses of these patients. The survey is conducted on a certain date in mid-October that differs for each facility. According to the 2014 Patient Survey, the sampling rate was approximately 38%. In this study, we independently aggregated data from the Patient Survey that was provided to us by the MHLW in accordance with Article 33 of the Japanese Statistics Act, because we wanted to incorporate data in addition to that which is publicly available.
We used the 630 Survey to determine the number of long-term discharges and new long-term admissions to be used for calculations of discharge-admission differences [2]. The 630 Survey is a snapshot of patients using psychiatric hospitals, psychiatric clinics, and home-visit nursing services on June 30 (i.e., 6/30) every year. It is a statistical survey conducted to obtain information to promote mental health and welfare policies. The response rate exceeds 95%, and thus the 630 Survey is highly exhaustive.
For both the Patient Survey and the 630 Survey, we divided mental illnesses into four categories: schizophrenia, dementia, mood disorder, and “other”. These mental illnesses were defined as the following International Statistical Classification of Disease and Related Health Problems (10th revision) codes: F20–29 for schizophrenia; F00, F01, F03, G30, and G31 for dementia; F30–39 for mood disorder; and all other F codes and G40 for “other.” For age groups in the Patient Survey, ages 0−89 years were divided into 5-year age groups, while all ages ≥90 years were treated as a single group. For age groups in the 630 Survey, we used the classifications in the original survey, although they are approximations: <20 years, 20−39 years, 40−64 years, 65−74 years, and ≥75 years.
General mortality rates by age group were determined with the 2017 Report on Vital Statistics. Reports on Vital Statistics are published annually by the MHLW to determine the demographics of the Japanese population, including foreigners. General survival rates by age group were determined by subtracting the general mortality rates by age group from 1.
Calculations were made in three stages. First, we estimated deaths by age group based on long-term inpatient numbers in the 2017 Patient Survey and general mortality rates by age group listed in the 2017 Report on Vital Statistics. Long-term inpatient numbers were classified by illness and age group. For general mortality rates by age group, based on the assumption that current trends will continue, we used the same numbers to project from 2017 onwards. Associated statistics were multiplied by the previously described general mortality rates by age group to estimate future deaths for each illness and age group. Deaths of patients who are long-term inpatients as of 2017 were calculated in five-year intervals. These numbers were then added together, and the results were divided into those for ages ≤74 years and ≥75 years.
Next, for discharge-admission differences, we subtracted the “number of new long-term admissions” from the “number of long-term inpatients discharged for reasons other than death” in the 2012−2016 630 Survey; future differences were estimated as the average difference for the past 5 years. Thus, the 2017 difference was estimated as the average difference for 2012−2016, the 2018−2020 difference was estimated as the average difference for 2013−2017, and the 2021−2025 difference was estimated as the average difference for 2016−2020. For 2026−2030, 2031−2035, and 2036−2040, we used the average difference for 2021−2025. Considering that deaths will also occur among new long-term admissions, we classified new long-term admission numbers by age group and multiplied them by general survival rates by age group (determined as described earlier) to determine the number of survivors among new long-term admissions. We then subtracted the number of survivors among new long-term admissions from the number of long-term discharges to determine discharge-admission differences.
Lastly, we subtracted the number of deaths and discharge-admission differences (calculated as described above) from long-term inpatients (as of 2017) to estimate the number of demands for long-term hospitalization by illness.