Changes in rehabilitation service utilisation during COVID-19 bed surges in Japan: a seasonal autoregressive integrated moving average (SARIMA) analysis of care utilisation with 10-year claims data

Abstract

Objective

To explore the association between COVID-19 bed utilisation and changes in rehabilitation service provision in Japan, focusing on cerebrovascular, musculoskeletal and respiratory rehabilitation in the pre-COVID-19, during-COVID-19 and post-COVID-19 periods.

Design

Retrospective study using seasonal autoregressive integrated moving average model to predict expected values, which were compared with actual values to calculate observed-to-expected (OE) ratios.

Setting

Acute care hospitals in Kanagawa Prefecture, which has the second largest population in Japan after Tokyo, covering April 2014 to March 2024.

Participants

Patients aged 0–74 years who were enrolled in the National Health Insurance of Kanagawa Prefecture and underwent the studied types of rehabilitation.

Exposure

COVID-19 pandemic waves and associated bed utilisation rates, with multiple distinct peaks.

Outcome measures

The difference between the predicted and actual values of the volume of rehabilitation services provided and the number of patients per insured person calculated as OE ratios during periods of peak bed utilisation for COVID-19.

Results

The OE ratio of inpatient rehabilitation services for cerebrovascular showed a significant decrease after five waves at −14.3%, and musculoskeletal conditions showed a similar decline during periods of high COVID-19 bed utilisation. Outpatient services experienced sharp declines initially but showed differential recovery patterns. Respiratory rehabilitation displayed unique patterns, with inpatient services increasing up to 62.4% above expected levels until September 2021, before sharply declining. By March 2024, musculoskeletal rehabilitation demonstrated complete recovery, cerebrovascular rehabilitation showed partial recovery, while respiratory rehabilitation exhibited mixed patterns with persistent outpatient deficits.

Conclusions

Rehabilitation services in Japan showed substantial changes associated with the COVID-19 pandemic, with inpatient services for cerebrovascular and musculoskeletal conditions being particularly vulnerable to disruptions during high COVID-19 bed utilisation periods. The differential recovery patterns across rehabilitation types, with some structural changes persisting beyond the acute pandemic phase, indicate the need for flexible healthcare systems to deal with future healthcare crises. These findings underscore the importance of developing strategies to maintain essential rehabilitation services during public health emergencies, especially considering the ageing global population and rising demand for rehabilitation.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Used 72 months of pre-pandemic claims data (April 2014–March 2020) to establish baseline trends through seasonal autoregressive integrated moving average (SARIMA) modelling.

• Analysed 120 months of continuous data spanning pre-, during- and post-COVID-19 periods, enabling assessment of multiple pandemic waves and recovery patterns.

• A SARIMA model using 72 months of pre-COVID data enabled accurate prediction of expected service levels across pre-, during- and post-pandemic periods.

• The study design could not identify the causal relationship between healthcare resource constraints and changes in patient care-seeking behaviours as primary drivers of decreased service utilisation.

• Analysed 120 months of continuous data spanning pre-, during- and post-COVID-19 periods, enabling assessment of multiple pandemic waves and recovery patterns.

Background

The COVID-19 pandemic, declared by the WHO on 11 March 2020, has substantially affected various health services globally.¹ Although the WHO declared the end of the pandemic in May 2023, especially at the initial phase of the pandemic, many countries implemented lockdowns to mitigate the spread of the pandemic. With limited medical resources, non-urgent surgeries were cancelled or delayed, patient numbers were restricted and rehabilitation beds were repurposed for COVID-19, except for urgent rehabilitation cases.² These resource allocation measures involve prioritising limited healthcare resources to those who can potentially benefit from treatment.³

Several studies have shown a decrease in elective surgeries during the COVID-19 pandemic.⁴,⁹ However, studies of the impact of COVID-19 on rehabilitation within healthcare delivery systems are limited. Rehabilitation is crucial for improving patient function and quality of life, with global demand increasing by 63% from 1990 to 2019, driven primarily by population ageing.¹⁰ The WHO estimates that the global population over 60 years will increase from 12% to 22% between 2015 and 2050.¹¹ In Japan, which has one of the highest ageing rates globally, national plans include converting acute care beds into postacute care beds with rehabilitation functions.¹² Rehabilitation is cost-effective¹³ and improves outcomes across multiple conditions, including musculoskeletal dysfunction,^{14 15} respiratory diseases.¹⁶,¹⁸ In stroke patients, where early rehabilitation improves functional outcomes and shortens hospitalisation.^{19 20} However, COVID-19 disrupted rehabilitation services through treatment protocol changes, unit closures and staff shortages,²¹,²³ while respiratory rehabilitation demand increased due to its role in improving COVID-19 sequelae.²⁴

Previous studies had several limitations that must be overcome. First, the status of medical provisions for rehabilitation varies by type, and cross-site studies are limited. Second, most studies focused on the early stages of the pandemic and did not examine long-term trend changes including post-COVID-19. Therefore, we used long-term claims data, including data from the initial phase of COVID-19. In this study, we hypothesised that the volume of rehabilitation services provided and the decrease in patient numbers were associated with increased bed utilisation during the surge of COVID-19 cases. Additionally, we assumed that increased bed constraints due to COVID-19 may have led to certain difficulties in the provision of rehabilitation care.²⁵ This study aimed to examine changes in rehabilitation services associated with COVID-19 bed utilisation in Japan in the pre-COVID-19, during-COVID-19 and post-COVID-19 periods.

Methods

Study design and setting

We conducted a retrospective study using claims data from Kanagawa Prefecture, Japan, covering the time frame from April 2014 to March 2024. Our study was divided into three periods: pre-COVID-19, during COVID-19 and post-COVID-19.

We selected Japan for this study because it is one of the first countries to have an ageing population, and a significant impact of COVID-19 on rehabilitation services was anticipated.²⁶ However, studies on quantitative and long-term effects are limited.²⁷ Kanagawa Prefecture was selected for three primary reasons. First, it has the second largest population in Japan after Tokyo, ensuring high representativeness. Second, it has the lowest number of hospital beds per capita in Japan,²⁸ making it particularly sensitive to bed shortages. Third, previous studies have not fully addressed the impact of the December 2020–January 2021 outbreak and the effects of Delta and Omicron strains on medical resources.²⁷ This study focused on three types of rehabilitation: cerebrovascular, musculoskeletal and respiratory.

Data source and participants

We used the Kokuho Database System (KDB), one of the largest databases in Japan, which contains claims data for Kanagawa Prefecture. The KDB has been used in previous clinical and epidemiological studies.^{29 30} The claims data included age, sex, place of residence, medical service codes, disease codes, quantity and frequency of services, and unique de-identified patient codes.

The participants were patients aged 0–74 years who were enrolled in the National Health Insurance of Kanagawa Prefecture between April 2014 and March 2024 and received rehabilitation (cerebrovascular, musculoskeletal or respiratory). The National Health Insurance is a public insurance programme for private employers, non-workers and retirees, covering approximately 18.0% of the population aged 0–64 years and 68.5% of the population aged 65–74 years.³¹ To examine the impact of hospital bed shortage, we specifically included patients who received treatment at acute care hospitals in Kanagawa Prefecture. We defined acute hospitals as those that claimed a fee for acute hospitals during the study period.

Outcome measures

The primary outcome was the difference between the predicted and actual values of the volume of rehabilitation services provided and the number of patients per insured person after April 2020 following the COVID-19 outbreak. These values were calculated separately for inpatient and outpatient services. In Japan, one rehabilitation unit is claimed for each defined hour of rehabilitation. We defined the monthly volume as the total number of units claimed for each month.

We descriptively observed the relationship between the peak periods of bed utilisation and outcomes during several waves of the pandemic: (first: 2020/4 first state of emergency in Japan, third: 2021/1 third wave, first bed shortage, fifth: 2021/8 Delta strain outbreak, sixth: 2022/2 Omicron outbreak, seventh: 2022/8 and eighth: 2023/1). The data were obtained from the Ministry of Labor and Welfare, and the peak utilisation rate for each pandemic wave was analysed.³²

Statistical analysis

We employed a seasonal autoregressive integrated moving average (SARIMA) model for our analysis. The SARIMA comprises an autoregressive component that predicts the future using historical data, a difference component that removes data trends, a moving average component that predicts the future using historical forecast errors and a seasonal component that accounts for seasonal patterns.

The SARIMA model is represented as (p, d, q)×(P, D, Q, m), where p, d and q are the parameters of the ARIMA model, and P, D, Q and m are the parameters of the seasonal component, with m being the period of seasonality. We estimated trends for 72 months, without missing data, from April 2014 to March 2020 regarding the volume of rehabilitation units provided and the number of patients per million insured persons in each rehabilitation programme. Because the number of insured individuals was publicly available only through FY2022, we used the FY2014–FY2022 average percentage change for FY2023 values (April 2023–March 2024).³³ The model selection procedure involved four steps. First, we used the Dickey–Fuller test as a unit root test to assess stationarity, with a significance level of p<0.05. Second, the parameters of the SARIMA model were set using Stata’s Arimaauto package, which selects the model with the lowest Akaike Information Criterion among the candidate parameters.³⁴ Third, the selected models were subjected to a white noise test using the Ljung–Box χ² test, with a significance level of p>0.05. If the white noise test result was below this value, the parameters were reset, and the procedure was repeated until the result exceeded 0.05. Finally, to evaluate the fit of the final selected model, we calculated the mean absolute percentage error using the values of the forecast model and actual measurements from April 2015 to March 2020. Municipalities with missing data were excluded. Each parameter is indicated in online supplemental table 1.

For most ARIMA models, CIs were calculated using Stata’s stdp option, which provides theoretical standard errors accounting for parameter uncertainty. For complex models where stdp was unavailable (particularly with moving average components), we used residual-based standard deviation (square root of sum of squared residuals divided by df), similar to approaches employed in healthcare utilisation forecasting studies when theoretical standard errors are unavailable. The residual-based method produces wider, more conservative CIs than stdp as it does not account for precision gains from parameter estimation. Both methods constructed 95% CIs as predicted value±1.96×SE, with lower bounds constrained to zero. All statistical analyses were conducted using Stata, MP16.1.

Patient and public involvement

Patients and/or the public were not involved in the design, conduct, reporting or dissemination plans of this research.

Results

After excluding 8 municipalities with missing data, the analysis included data from 48 municipalities to ensure a comprehensive and consistent dataset throughout the study period. Figure 1 illustrates trends in the utilisation of COVID-19 beds in Kanagawa Prefecture. The graph reveals multiple distinct peaks across the study period: January 2021 (third wave) at 47% utilisation, August 2021 (fifth wave) at 72% utilisation, February 2022 (sixth wave) at 65% utilisation, September 2022 (seventh wave) at 88% utilisation and January 2023 (eighth wave) at 78% utilisation. These peaks correspond to major surges in COVID-19 cases and provide context for the changes in the provision of rehabilitation services. The seventh wave in September 2022 showed the highest bed utilisation rate at 88%, indicating the most severe strain on the healthcare system throughout the entire observation period.

Figure 1. COVID-19 bed utilisation trends.

Distribution of rehabilitation services

Table 1 presents the gender distribution across different rehabilitation services during the three study periods. For cerebrovascular rehabilitation, males consistently comprised approximately 61% of inpatient cases and 57%–58% of outpatient cases across all periods, with no statistically significant changes observed (p=0.12 for inpatient, p=0.08 for outpatient). Musculoskeletal rehabilitation showed a contrasting pattern, with females representing the majority in both settings: approximately 59% of inpatient cases and 65% of outpatient cases, remaining stable throughout the study periods (p=0.34 for inpatient, p=0.15 for outpatient). Respiratory rehabilitation demonstrated a significant shift in gender distribution for inpatient services, where the proportion of males decreased from 67.8% before COVID-19 to 65.9% in the post-COVID-19 period (p=0.02), while outpatient services showed fluctuating patterns without statistical significance (p=0.18).

Table 1. Gender distribution by rehabilitation services.

	Before COVID-19 (2014/4–2020/3)	After COVID-19 (2020/4–2023/4)	Post COVID-19 (2023/5–2024/5)	P value*
Cerebrovascular rehabilitation
Hospitalised				0.12
Male	61.6%	61.1%	61.2%
Female	38.4%	38.9%	38.8%
Outpatient				0.08
Male	58.0%	57.2%	57.3%
Female	42.0%	42.8%	42.7%
Musculoskeletal rehabilitation
Hospitalised				0.34
Male	40.8%	41.2%	40.8%
Female	59.2%	58.8%	59.2%
Outpatient				0.15
Male	35.1%	35.2%	34.6%
Female	64.9%	64.8%	65.4%
Respiratory rehabilitation
Hospitalised				0.02
Male	67.8%	67.2%	65.9%
Female	32.2%	32.8%	34.2%
Outpatient				0.18
Male	70.0%	68.2%	73.2%
Female	30.0%	31.8%	26.8%

^*χ2 test.

Table 2 illustrates the age distribution changes across the study periods. Cerebrovascular rehabilitation patients showed a notable trend toward younger ages, with inpatient mean ages decreasing from 63.4 years before COVID-19 to 62.1 years post-COVID-19, and outpatient mean ages declining more substantially from 48.9 to 44.0 years (both p<0.01). Conversely, musculoskeletal rehabilitation patients demonstrated a gradual ageing trend, with inpatient mean ages increasing from 64.5 to 65.5 years and outpatient mean ages rising from 61.7 to 62.9 years (both p<0.01). Respiratory rehabilitation patients maintained the highest mean ages across all categories (66–68 years), with minimal but statistically significant changes over time (both p<0.01).

Table 2. Age distribution by rehabilitation services.

	Mean (SD)						P value*
	Before COVID-19 (2014/4–2020/3)		After COVID-19 (2020/4–2023/4)		Post COVID-19 (2023/5–2024/5)
Cerebrovascular rehabilitation
Hospitalised	63.4	(12.3)	63.0	(12.6)	62.1	(13.0)	<0.01
Outpatient	48.9	(23.7)	45.6	(24.5)	44.0	(24.4)	<0.01
Musculoskeletal rehabilitation
Hospitalised	64.5	(11.9)	65.4	(11.3)	65.5	(11.0)	<0.01
Outpatient	61.7	(14.4)	62.4	(13.9)	62.9	(13.3)	<0.01
Respiratory rehabilitation
Hospitalised	66.2	(11.4)	66.4	(10.7)	66.9	(11.0)	<0.01
Outpatient	67.1	(11.1)	68.2	(8.3)	67.7	(9.2)	<0.01

^*ANOVA (Analysis of Variance) test.

Table 3 presents the number of units and percentage of patients by rehabilitation type during the study period. For cerebrovascular rehabilitation, 95.4% of the units provided inpatient care and 4.6% provided outpatient care; 68.4% of the patients received inpatient care and 31.6% received outpatient care. Musculoskeletal rehabilitation showed a different pattern, where 69.5% of the units provided inpatient services and 30.5% provided outpatient services; however, 31.1% of the patients received inpatient care and 68.9% received outpatient care. Respiratory rehabilitation was predominantly inpatient, with 96.3% of the units and 88.1% of the patients in inpatient settings compared with 3.7% of the units and 11.9% of the patients in outpatient settings.

Table 3. Supplied volume and number of patients.

	Supplied volume	Ratio	Patients number	Ratio
Cerebrovascular rehabilitation
Hospitalised	12 383 616	95.4%	224 705	68.4%
Outpatient	592 793	4.6%	103 586	31.6%
Musculoskeletal rehabilitation
Hospitalised	4 238 295	69.5%	187 965	31.1%
Outpatient	1 859 839	30.5%	416 978	68.9%
Respiratory rehabilitation
Hospitalised	619 166	96.3%	47 663	88.1%
Outpatient	23 573	3.7%	6417	11.9%

These figures highlighted the predominance of inpatient services for cerebrovascular and respiratory rehabilitation in terms of service units and patient numbers. Conversely, musculoskeletal rehabilitation showed a more balanced distribution, with a higher proportion of outpatient patients despite more inpatient service units.

Table 4 depicts the average number of monthly units per claim across the three study periods. For cerebrovascular rehabilitation, inpatient services showed an initial increase from 54.9 units before COVID-19 to 56.2 units during the COVID-19 period, followed by a decrease to 53.1 units post-COVID-19. Outpatient services demonstrated a consistent declining trend from 5.8 to 5.1 units across the periods. Musculoskeletal rehabilitation exhibited substantial increases in inpatient services, rising from 21.0 units before COVID-19 to 25.4 units during COVID-19, with a slight decrease to 24.6 units post-COVID-19 (overall increase of 17.1%). Outpatient services showed modest but consistent growth from 4.4 to 4.7 units (6.8% increase). Respiratory rehabilitation demonstrated the most pronounced changes, with inpatient services increasing from 12.1 units before COVID-19 to 14.3 units during COVID-19, then declining to 13.5 units post-COVID-19 (overall increase of 11.6%). Outpatient services showed initial decline from 3.8 to 3.2 units during COVID-19, followed by recovery to 4.1 units post-COVID-19.

Table 4. Changes in rehabilitation service provision.

	Mean (SD)						P value*
	Before COVID-19 (2014/4–2020/3)		After COVID-19 (2020/4–2023/4)		Post COVID-19 (2023/5–2024/5)
Cerebrovascular rehabilitation
Hospitalised	54.9	(70.3)	56.2	(72.3)	53.1	(71.0)	<0.01
Outpatient	5.8	(6.0)	5.6	(6.5)	5.1	(5.3)	<0.01
Musculoskeletal rehabilitation
Hospitalised	21.0	(29.2)	25.4	(33.8)	24.6	(33.2)	<0.01
Outpatient	4.4	(4.0)	4.5	(3.8)	4.7	(3.9)	<0.01
Respiratory rehabilitation
Hospitalised	12.1	(12.5)	14.3	(15.9)	13.5	(15.0)	<0.01
Outpatient	3.8	(4.0)	3.2	(3.0)	4.1	(4.0)	<0.01

^*ANOVA test.

These results indicate that the COVID-19 pandemic had lasting impacts on rehabilitation service delivery patterns, with inpatient services generally showing increased intensity while outpatient services demonstrated more variable responses across different rehabilitation types.

Observed-to-expected ratios

Observed-to-expected ratios during COVID-19 and recovery period

Figures2 3 present the temporal evolution of rehabilitation service utilisation through observed-to-expected (OE) ratios, revealing distinct patterns across rehabilitation types and settings. The analysis covered the complete period, encompassing the pre-pandemic baseline, the entire COVID-19 pandemic trajectory, the transition to Category 5 classification in May 2023, and the subsequent recovery phase.

Figure 2. Observed and predicted number of supplied services and patients.

Figure 3. Observed-to-expected ratio.

Cerebrovascular rehabilitation

Cerebrovascular rehabilitation services demonstrated moderate initial resilience but sustained negative impacts throughout most of the observation period. Inpatient unit provision exhibited relative stability with OE ratios fluctuating around ±5.0% until August 2021 (fifth wave), before experiencing a notable decline to −13.6% between the end of the fifth wave and the start of the sixth wave in early 2022. Following the Category 5 reclassification in May 2023, inpatient units showed gradual stabilisation but remained below expected levels through March 2024. Inpatient patient numbers demonstrated more pronounced sensitivity to pandemic disruptions, reaching the lowest OE ratio of −14.3%, immediately following the fifth wave and maintaining consistently negative ratios after the initial pandemic onset, with limited recovery evident by March 2024.

Outpatient cerebrovascular services experienced the most severe immediate impact, with OE ratios plummeting to −42.9% for units and −38.5% for patients during the first wave in May 2020. While unit provision gradually improved and temporarily shifted to positive territory in late 2021, it returned to negative values after January 2022 and showed only marginal improvement following the Category 5 transition. Patient numbers demonstrated a more sustained recovery trajectory, turning slightly positive after April 2021, maintaining improvement through 2022–2023, and showing stabilisation to neutral levels by early 2024.

Musculoskeletal rehabilitation

Musculoskeletal rehabilitation services showed cyclical patterns that were closely aligned with infection waves, with evidence of systematic recovery by the end of the observation period. Inpatient services recorded notable decreases during peak bed utilisation periods, with OE ratios of −9.8% during the first wave (May 2020), −7.9% during the fifth wave (August 2021) and −6.9% during the sixth wave (February 2022). Following the Category 5 reclassification, inpatient units demonstrated clear recovery trends, approaching neutral levels by March 2024. Patient numbers, which reached the lowest point of −25.7% during the first wave, showed robust recovery throughout 2023 and achieved positive OE ratios by early 2024.

Outpatient musculoskeletal services experienced the most dramatic initial disruption, with OE ratios declining to −40.9% for units and −36.3% for patients in May 2020. Recovery patterns showed unit provision transitioning to positive trends by March 2021, maintaining steady improvement through 2024, and achieving OE ratios of approximately +5% to +10% by March 2024. Patient numbers demonstrated similar recovery, achieving sustained positive ratios from September 2021 onwards and stabilising at positive levels throughout the post-Category 5 period.

Respiratory rehabilitation

Respiratory rehabilitation displayed the most distinctive utilisation patterns throughout the observation period. Inpatient services experienced substantial increases beginning in April 2020, likely reflecting COVID-19-related demand for respiratory support services. This surge peaked dramatically in September 2021, reaching OE ratios of +62.4% for units and +53.0% for patients—the highest positive deviations observed across all rehabilitation categories. Following a sharp decline in December 2021, inpatient respiratory services showed volatile patterns through 2022 before demonstrating normalisation trends post-Category 5 transition, approaching baseline levels by March 2024.

Outpatient respiratory rehabilitation services demonstrated the most persistent negative trends throughout the observation period. Following severe reductions during the first wave (−71.4% for units in June 2020 and −67.2% for patients in May 2020), these services showed limited recovery through 2022 and 2023. Even after the Category 5 reclassification, outpatient respiratory services remained substantially below expected levels, with OE ratios of approximately −30% to −40% persisting through March 2024, suggesting structural changes in respiratory care delivery patterns.

Recovery trajectories and post-Category 5 trends

The transition to Category 5 classification in May 2023 marked a turning point for several rehabilitation services. Musculoskeletal rehabilitation demonstrated the most complete recovery, with both inpatient and outpatient services achieving or exceeding expected levels by March 2024. Cerebrovascular rehabilitation showed partial recovery, with outpatient services approaching baseline levels while inpatient services remained moderately below expectations. Respiratory rehabilitation presented mixed patterns, with inpatient services normalising but outpatient services showing persistent deficits, suggesting lasting changes in care delivery models.

Cross-category analysis and healthcare system adaptation

The 5-year analysis revealed complex, type-specific responses in rehabilitation service provision during the COVID-19 pandemic and recovery period. The data demonstrate that while acute disruptions occurred across all rehabilitation types during 2020–2021, recovery patterns varied substantially. By March 2024, approximately 4 years after the initial pandemic impact, musculoskeletal rehabilitation had achieved full recovery, cerebrovascular rehabilitation showed partial recovery and respiratory rehabilitation exhibited mixed but improving trends. These differential recovery patterns suggest that pandemic impacts on rehabilitation services were not uniform and that some structural changes in care delivery may persist beyond the acute pandemic phase.

Discussion

Our results suggest a strong relationship between COVID-19 bed utilisation and decreased cerebrovascular and musculoskeletal rehabilitation services. The lowest OE ratio for inpatient rehabilitation coincided with peak COVID-19 bed utilisation after the third wave, except for cerebrovascular rehabilitation during the fifth wave, suggesting that hospital bed shortages were associated with constraints in rehabilitation service provision. In case of cerebrovascular rehabilitation during the fifth wave, for peak COVID-19 bed utilisation, OE ratio also showed a decreasing trend but the lowest timing was observed after the peak COVID-19 bed utilisation. Previous studies reported similar decreases in rehabilitation services during the early stages of the pandemic.^{35 36} Our study extends these findings to later periods of high global infection rates and acute hospital bed shortages. The decrease in services may be attributed to infection control measures within hospitals, as a German study found that 15.7% of the surveyed hospitals experienced partial or complete ward closures amid COVID-19 outbreaks.³⁷

Inpatient rehabilitation for cerebrovascular and musculoskeletal conditions showed high sensitivity to bed shortages and infection outbreaks. Conversely, outpatient rehabilitation experienced a sharp decline during the first wave but subsequently stabilised. This differential impact is notable, as inpatient rehabilitation accounted for 95.8% and 68.6% of the total volume of cerebrovascular and musculoskeletal services, respectively. These rehabilitation fields often require intensive inpatient care, followed by outpatient rehabilitation. Although rehabilitation may not directly prevent mortality, it substantially affects the patient’s quality of life and subsequent physical function. Ensuring stable provision of inpatient rehabilitation during health crises remains a critical challenge for healthcare systems. Conversely, respiratory rehabilitation showed a consistent upward trend until September 2021, likely reflecting the growing demand because of COVID-19. This trend aligns with those reported in previous studies, demonstrating the effectiveness of respiratory rehabilitation in improving outcomes for patients with COVID-19.^{38 39} The peak in the OE ratio occurred 1 month after the fifth wave, possibly attributed to the increased virulence of the delta variant. However, from October 2021 to February 2022, this trend shifted negatively, possibly owing to service constraints or a decrease in patients who are severely ill as a result of vaccination and reduced virulence of the Omicron variant.⁴⁰,⁴²

The study found increased rehabilitation provisions per inpatient month across all three areas post-COVID-19. For cerebrovascular rehabilitation, two possibilities were considered: (1) decreased patient functionality because of limited rehabilitation and daily activities during the pandemic⁴³ and (2) severe illness requiring intensive rehabilitation, possibly resulting from delays in emergency care and reduced availability of treatments such as mechanical thrombectomy.^{44 45} However, some studies have reported no differences in stroke severity during the pandemic,⁴⁶,⁴⁸ highlighting the need for further research to fully understand these trends.

Our study design has important limitations in establishing causality between bed utilisation and rehabilitation service changes. The SARIMA model, while effective for identifying temporal associations, cannot control for time-varying confounders that may have influenced service utilisation simultaneously.

Several demand-side factors likely contributed to the observed reductions. First, patients voluntarily avoided hospital visits due to infection concerns, particularly during emergency declarations. Second, strict social distancing policies limited patient mobility and healthcare access. Third, public health messaging about staying home may have discouraged non-urgent healthcare utilisation.

Additionally, supply-side factors beyond bed constraints may have played roles. First, staffing shortages occurred due to infection, quarantine or redeployment to COVID-19 units. Second, infection control protocols reduced operational capacity. Third, temporary closures of rehabilitation units followed COVID-19 outbreaks within facilities.

Our analysis cannot definitively separate these mechanisms. The observed temporal correlation between bed utilisation peaks and service reductions suggests healthcare system strain as a contributing factor, but patient behaviour changes likely also played significant roles. Future research using individual-level data with patient-reported reasons for care delays or cancellations would help clarify these pathways.

Another important limitation is the exclusion of patients aged 75 years and older. In Japan’s healthcare system, individuals aged 75+ years are covered under a separate insurance scheme (Late-Stage Medical Care System for the Elderly), preventing their inclusion in our National Health Insurance database. This age group constitutes the majority of rehabilitation service users, particularly for cerebrovascular and musculoskeletal rehabilitation. The exclusion means our results may underestimate the total impact of COVID-19 on rehabilitation services.

Despite these limitations, our findings have important implications for policymakers. Monitoring the bed occupancy rate for infectious diseases can help estimate the impact of novel infectious diseases on rehabilitation services. Therefore, policymakers should use these rates to quantitatively measure their impact on rehabilitation services. Second, a quantitative assessment of risks and benefits is essential. This study suggests a trade-off between bed utilisation for infectious diseases and rehabilitation services such as cerebrovascular and musculoskeletal services. During a pandemic, medical resources cannot be allocated to only one aspect of care. Policymakers can use quantitative indicators like quality-adjusted life-years to determine the appropriate balance. However, the calculation is complicated, requiring further research for rapid and timely assessment. Lastly, during the pandemic surges, bed shortages are inevitable; therefore, policymakers and healthcare providers should develop triage strategies to prioritise rehabilitation services.

Future research should develop strategies for maintaining access to rehabilitation services during health crises. Although the potential for online rehabilitation has been noted,^{1 49} caution is warranted because of challenges such as limited Internet access for older adults and the need for tailored services.² Long-term studies are required to assess the impact of such interventions on patient outcomes and healthcare system efficiency. Qualitative research is required to clarify the decision-making processes and healthcare delivery structures within individual medical institutions during bed shortages and patient surges. The healthcare system in crisis is a complex structure defined not only by the number of beds and healthcare workers but also by staff, resources, structures and systems.⁵⁰ Decision-making during an emerging infectious disease outbreak often relies on limited information and differs from typical decision-making processes.⁵¹ Understanding these processes can help develop more effective strategies for maintaining essential services, including rehabilitation, during future health crises.

Data availability statement

Data may be obtained from a third party and are not publicly available.