Healthcare utilization among Japanese older adults during later stage of prolonged pandemic
Our primary data source is the Medical Claims Data with Income Tax Information for the Oldest-Old in Japan (MCD-Tx), a comprehensive dataset of Japan’s Latter-Stage Elderly Healthcare System (LSEH) collected by the Ministry of Health, Labour and Welfare (MHLW). The LSEH is Japan’s universal healthcare system for the 75+ population, and as of September 2022, it covered 18.52 million individuals, comprising 98.6% of Japan’s 75 + population.
The MCD-Tx, spanning from November 2021 to November 2022, captures all LSEH enrollees, regardless of their healthcare service utilization. For each individual, the dataset provides detailed monthly healthcare utilization and cost information from medical claims, linked with individual demographic and socioeconomic data including income information. This dataset represents the first instance in Japan where income information is integrated with medical claims data. Informed consent was waived by the Ethics Review Committee for Research Involving Human Subjects at Waseda University (Approval No.: 2022-HN038). As this study used anonymized secondary data, the requirement for informed consent was waived by the ethics committee. Furthermore, all methods used in this study were conducted in accordance with relevant guidelines and regulations regarding using data of human participants.
To complement the MCD-Tx, we incorporated three additional datasets. The first tracks daily new COVID-19 cases at the secondary medical region (SMR) level—administrative units organized around specialized healthcare facilities that provide more granular geographical resolution than prefectures—from the onset of Japan’s first reported case until September 2022. The second dataset provides biweekly records of hospital beds designated for COVID-19 patients and actual admissions at the SMR level, from December 2021 to April 2023. The third dataset covers the daily implementation status of SoE and SoPE at the prefecture level, from March 2020 to September 2022.
In constructing our sample, we first convert these three supplementary datasets to a monthly format. We then link them with the MCD-Tx claim records based on the SMR or prefecture of residence of the oldest-old (detailed linkage procedures are described in Appendix A). The final sample covers an 11-month period from November 2021 to September 2022, capturing the later phase of the pandemic (Waves VI and VII). This sample includes 1,769,537 individuals aged 75+ and 189,841,257 associated health insurance claim records.
Measurements
Our analysis examines both extensive and intensive margins of healthcare utilization. For the extensive margin, we construct four binary indicators associated with different healthcare services: overall healthcare utilization, hospital admission, outpatient visits, and dental care. Each indicator takes a value of one if an individual use the respective service within a given month, and zero otherwise. As shown in Table 1 and 84.2% of our sample access some form of healthcare during the study period, with outpatient visits being the most common (78.5%), followed by dental services (20%), while hospital admissions are less frequent (5.4%).
For the intensive margin, we analyze four variables representing monthly medical costs: total costs, inpatient costs, outpatient costs, and dental costs (measured in 10,000 Japanese Yen, JPY). Among those who utilize services, average monthly inpatient care costs (651,320 JPY) are substantially higher than outpatient care (43,370 JPY) and dental care costs (14,630 JPY).
To measure pandemic conditions, we construct two variables. The first captures pandemic severity through the monthly aggregated number of new COVID-19 cases per million people within each SMR, averaging 0.013 during our study period. The second indicates the presence of SoPE measures, taking a value of one for months with active measures in the resident’s prefecture (present in 26% of our study period). Notably, no prefecture implemented SoE measures during our study timeframe.
Our models include several control variables to account for individual characteristics and healthcare capacity. Individual-level controls comprise age quintiles (ranging from mean age 75.56 years in Q1 to 92.25 years in Q5), income quintiles (from mean 37.42 million JPY in Q1 to 554.22 million JPY in Q5), and gender (39.8% male). We also include indicators for non-COVID-19 main diagnoses in both inpatient and outpatient care, with cardiac conditions being most prevalent (22.1% of hospitalizations and 26.7% of outpatient visits). Healthcare capacity is measured through COVID-19 hospital bed occupancy quintiles, ranging from 5.0 to 68.6%, reflecting regional variations in healthcare system strain during the pandemic.
Analytical approach
We use two complementary models to examine healthcare utilization patterns. Our first model analyzes the association between SoPE measures and healthcare utilization:
$${Y}_{ijt}={\beta}_{0}+{\beta}_{1}SoP{E}_{jt}+{\beta}_{2}f\left(COVI{D}_{jt}\right)+{X}_{ijt}^{{\prime}}\gamma+{\delta}_{t}+{\eta}_{j}+{\tau}_{j}t+{\varepsilon}_{ijt}.$$
(1)
Here, \({Y}_{ijt}\) represents healthcare utilization and costs for individual \(i\) in SMR \(j\) at time \(t\). \(SoP{E}_{jt}\) is the indicator of SoPE measures, our primary variable of interest. We anticipate healthcare utilization to be negatively associated with SoPE implementation, consistent with documented healthcare avoidance patterns. The relationship between medical costs and SoPE measures may be more complex, potentially showing either negative associations due to fewer visits or positive associations if delayed care leads to more intensive treatment needs. \(f\left(COVI{D}_{jt}\right)\) represents a quadratic polynomial function of new COVID-19 cases. The quadratic specification captures potential non-linear patterns in the relationship between case numbers and healthcare utilization, allowing for diminishing marginal correlations at higher case levels. \({X}_{ijt}\) represents individual characteristics and healthcare capacity controls. The model incorporates year-month fixed effects \({\delta}_{t}\), geographic fixed effects at the SMR level \({\eta}_{j}\), and geographic linear trends \({\tau}_{j}t\) to control for unobserved temporal and regional variations. \({\varepsilon}_{ijt}\) denotes the error term, with standard errors clustered at the SMR level.
Our second model examines how the association between healthcare utilization and pandemic severity varies with public health measures:
$$\begin{aligned}{Y}_{ijt}&={\theta}_{0}+{\theta}_{1}SoP{E}_{jt}+{\theta}_{2}f\left(COVI{D}_{jt}\right)+{\theta}_{3}\left[SoP{E}_{jt}\times f\left(COVI{D}_{jt}\right)\right]+{X}_{ijt}^{{\prime}}\gamma+{\delta}_{t}+{\eta}_{j}\\ &\quad +{\tau}_{j}t+{\varepsilon}_{ijt}.\end{aligned}$$
(2)
This model fully interacts the SoPE indicator with the quadratic function of case numbers, allowing us to examine whether the relationship between pandemic severity and healthcare utilization differs during periods with and without public health measures. To interpret these relationships, we calculate average marginal effects of changes in case numbers at different pandemic severity levels, ranging from low (below 0.04 cases per million) to high (above 0.14 cases per million).
To assess the robustness of our approaches, we conduct additional analyses (see Appendix B). We estimate models with a linear specification of pandemic severity as an alternative to our quadratic specification. We also perform subgroup analyses by age groups (75–84 and 85+) and gender to examine potential heterogeneity in patterns. These analyses reveal patterns consistent with our main findings, suggesting that the observed relationships between healthcare utilization, public health measures, and pandemic severity are robust across specifications and subpopulations.
link
