We used a cross-sectional, city-level, ecological design to test the hypothesis. We used 20 municipality cities designated by the government ordinance of Japan under the Local Government Act as a unit of analysis for two reasons. First, all these cities have populations above 500,000, and an independent local public health center on the frontline of infection control under the governance of the local mayor (19, 20). These cities also have urban economic and health infrastructures. Second, the numbers of COVID-19 cases and deaths per month in the designated cities are publicly available. These cities also manage standard citizen services and have greater autonomy in social welfare, public health, urban planning, management of national roads, and compulsory education (19).

Worldwide Governance Indicators are well-known comprehensive measures of country-level governance characteristics (21), although they may not be easily applicable to local governments, which have different powers and responsibilities than central governments. For example, “the kind of voice and accountability” refers to levels of democracy, such as freedom of political participation, expression, and elections, which may not be appropriate to apply at the local government level (22).

Because of the lack of standardized and applicable measures for our purpose, we chose a local policy of temporary benefit revocation for unpaid premiums under the national health insurance system to measure the extent to which local governments used administrative frameworks to manage policy consensus from various sectors and stakeholders. The rationale for this is as follows.

Japan’s public national health insurance system provides universal cover for community residents and is implemented by local government insurers under the National Health Insurance Act (23). The system mandates premium contributions according to household size and income level, and offers full medical, dental, and pharmaceutical coverage with a standardized copayment rate. However, Article 9 of the Act stipulates that individuals who default on their premium payments for a year or longer will be temporarily ineligible for coverage when utilizing healthcare services, and will be subject to full out-of-pocket medical visit expenses until the premium payments are no longer in arrears (24). During the period of ineligibility, the local municipality’s government insurers are mandated to issue a “certification of beneficiary status,” which guarantees that out-of-pocket payments will be reimbursed after confirmation of premium payments.

Although the issuance of status certification is generally mandated, an exemption may be applied if the local government makes a policy decision to suspend the temporary revocation of benefits. Such decisions are approved following unanimous consensus across the mayor’s office, city council, and other related parties, and are made for various reasons, including welfare (e.g., to ensure equal benefits for poorer households) or for management (e.g., avoiding the costs involved in temporarily revoking benefits).

Regardless of the reason for these decisions, implementing and maintaining the suspension policy necessitates complex policy processes that require coordination and unanimous consensus between the mayor, the insurance bureau, related sectors in the municipality office, the municipal council, and other internal and external actors whose interests and political values vary widely. Following the National Health Insurance Act, Local Councils for the Administration of National Health Insurance (statutory advisory councils), which comprise representative insured persons and medical providers, are required to submit an advisory opinion to the mayor on whether the suspension policy is relevant and feasible in ensuring equality of both benefits and contributions (24). The city councils have the power to determine the legitimacy of the suspension policy in relation to the city budget (25), and the prefecture functions as a supervisory agency for municipality insurer offices for jurisdiction and bureaucratic feasibility (24). These complex processes mean that local governments that intend to adopt the suspension policy must have the administrative capacity to align divergent perspectives across sectors and stakeholders. The required governance framework corresponds to the “collaborative governance” model, which involves collaboration across diverse stakeholders in multiple sectors to address public service issues through collective efforts, promoting flexible actions based on trust and legitimacy (26).

The suspension policy is implemented through processes involving multiple stakeholders and iterative activities across five interrelated functions (Figure 1): (i) groundwork laid by the National Health Insurance division; (ii) coordination between the legal, finance, and revenue-collection departments; (iii) deliberation by the statutory advisory council; (iv) decision-making by the mayor’s office; and (v) authorization and budgeting by the city council. These functions interact and repeat rather than unfolding as a fixed sequence of discrete tasks, and they require cooperation across both administrative and political entities.

In alignment with Emerson et al. (26) collaboration dynamics framework, this process unfolds through iterative interactions between three domains: principled engagement, which structures discovery, definition, deliberation, and determination; shared motivation, which involves building mutual trust, mutual understanding, internal legitimacy, and shared commitment; and capacity for joint action, characterized by concrete procedural and institutional arrangements, leadership, knowledge, and resources. We therefore consider the adoption of the suspension policy as a collaborative governance outcome that arises from iterative, mutually reinforcing interactions between principled engagement, shared motivation, and capacity for joint action throughout this governance process. Figure 1 illustrates how these functions relate to the three domains. Thus, we interpret this publicly visible decision as a stringent test of municipal collaborative governance capacity.

The implementation of such complex policymaking processes, particularly the suspension policy, is unlikely to simply reflect an ideological motivation to attract voting interests from liberal parties. This is because only 0.2% of households in Japan are subject to temporary benefit revocation (27, 28), and there is strong support for benefit revocation as a way of ensuring contribution fairness. Thus, contrary to the US discussion about the politicization of pandemic control, the suspension policy cannot be fully explained by the dominant ideology of local government.

Additionally, the suspension policy per se did not theoretically affect accessibility to medical treatment for COVID-19, as the Central Government Ministry announced conditional exceptions to the revocation policy for COVID-19-related consultations as of November 2020 (29), before the sampling period of the current study began. The aim of these exceptions was to facilitate equitable access to required medical services for people in poverty. Although access to medical care for non-COVID-19 conditions was discouraged owing to the revocation of benefit eligibility (30), this would not have affected the COVID-19-related mortality rate. Thus, any association between the policy of suspending benefit eligibility revocation and reductions in COVID-19-related mortality in a municipality should be related to municipal-level management of pandemic-related policy decisions (17, 31) rather than subsequent improvement of medical care access for people in poverty. We argue that the suspension policy may be an indicator of the governance style of local governments that used coordinated and collaborative policy management across health and non-health sectors to effectively respond to the ongoing challenges of the pandemic.

We treat policy outputs as observable manifestations of collaborative governance. Among potential outputs, suspension was selected as the indicator because it is formally decided and documented, requires non-sequential alignment across the domains described above, and is conceptually anchored in the three collaboration dynamic domains operating iteratively (Figure 1). This theory-guided operationalization provides content validity for municipal-level inference.

The monthly cumulative number of confirmed COVID-19-related deaths was the study outcome (32). Death data from December 2020 to November 2021 were compiled quarterly to reflect different COVID-19 variant outbreaks. Before December 2020, the original strain dominated the first period, and the Alpha variant dominated the second period in Japan. During the study period, the Delta variant became more prevalent; this was followed by the Omicron variant outbreak, during which the death count rapidly increased (16).

To measure city-level governance arrangements, we collected data on the suspension or continuation of temporary benefit revocation in each city by reviewing the minutes of meetings of its city council and the Local Council for the Administration of National Health Insurance. In July 2023, we confirmed that information by phone calls to each city. Cities with a policy of suspending temporary benefit revocation under the national health insurance system were assumed to have governments with administrative frameworks for flexible and collaborative policymaking (as argued above). Because significant changes in governance typically require more than 5 years (33), we classified cases as having a suspension policy from the beginning of the observation period, even if the policy was instituted during that time.

The following city-level covariates were measured: (1) proportion of the population aged 75 years or over (27), (2) number of acute care hospital beds per population (27, 34), (3) proportion of nursing home residents (27, 35), and (4) population density (in thousands of people per km²) (27, 36), as factors that may affect death rates, following a previous study (13). The specific covariate formulas are shown in Table 1.

We used negative binomial regression to address over-dispersion (37). This method was applied to assess the association between COVID-19-related deaths and the suspension of temporary benefit revocation, adjusting for covariates and periods across the 20 cities. We designated the first 3 months as the reference period. The logarithmic value of the number of deaths at the city level was set as the outcome variable. To account for differences in city population size, we included the logarithmic value of the city population as an offset term. Additionally, we clustered standard errors at the city level to account for within-city correlations across periods. All analyses were conducted using Stata/IC version 14.2 (Stata Corporation, College Station, Texas, USA).

We adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline (38). Institutional review board approval was waived for this study, which did not require informed consent, as it used deidentified, publicly available data.

This study had several limitations. First, the design was cross-sectional and ecological at the municipal level; therefore, causal relationships cannot be inferred and the findings are not interpretable at the individual level. Given the limited sample size and the likelihood of residual or unmeasured confounding, as well as potential policy endogeneity and reverse causation, the observed patterns should be viewed as associations and as indicators that help to generate future hypotheses. Second, categorizing governance arrangements according to a single policy may have led to problems with misclassification. Although other municipal-level indicators of collaborative governance may exist, in this setting we considered the suspension policy a practice-grounded proxy for municipal collaborative governance; broader indicator sets should be developed in future studies. Four cities implemented the policy before the observation period, and one city during this period. Additionally, 2 of the 20 designated cities changed mayors during the observation period (42). However, substantial changes in governance arrangements may require an extended period, such as ≥5 years, and we believe there were no good reasons for major governance style changes to occur during the observed 2-year window (33). Third, although socioeconomic status was not included in the primary adjustment set, we conducted a sensitivity analysis by adding a city-level socioeconomic status proxy—equivalized disposable household income per capita—and the results were essentially unchanged (Supplementary Table 2). A previous study identified associations between socioeconomic status and COVID-19-related mortality and incidence at the prefecture level in Japan; in particular, a higher unemployment rate was proportionally linked to higher mortality (13). However, the differences in unemployment rates across the 20 cities were less pronounced than the differences across the prefectures (27). Therefore, although the sensitivity analysis reduces concerns about confounding among socioeconomic status variables, the presence of some residual confounding cannot be ruled out. Fourth, this study included only 20 ordinance-designated cities, which typically have greater administrative and public health capacities than smaller municipalities. It may not be possible to generalize the findings to smaller Japanese municipalities, prefectural public health center jurisdictions, and international settings; the coordination mechanisms we propose may be transferable, but effect sizes are likely to depend on local capacity, provider market structure, and legal authority. Finally, owing to the lack of disaggregated data, we compared crude rates of COVID-19-related mortality, which may have ignored important variations in sex and age distributions among city populations.