Public health researchers in Japan have established a nationwide, individual-level mortality database by linking records from the 2020 Population Census to national Vital Statistics death records covering a two-year observation period. Japan previously lacked an official administrative framework or common personal identification code connecting comprehensive resident enumerations with mortality registries, a gap that historically limited the systematic tracking of health-related inequalities across the country. To address this methodological challenge, an investigative team led by Hirokazu Tanaka constructed a multi-stage deterministic linkage system that paired census microdata with death certificates recorded between October 2020 and September 2022. By integrating geocoded residential addresses down to small neighborhood units alongside demographic variables, the initiative successfully connected millions of records, yielding an empirical resource designed to measure how mortality risks differ across distinct socioeconomic environments.
Establishing reliable population-wide health statistics requires robust data integration, particularly in countries where medical services, demographic shifts, and health disparities must be monitored over extended intervals. In many nations, including Canada, Denmark, Italy, Belgium, Australia, New Zealand, the United Kingdom, and the United States, population registries or regular censuses have long been coupled with vital event registries through explicit personal identification numbers. In Japan, although national health surveys have periodically identified socioeconomic gradients in health outcomes, researchers faced enduring structural obstacles when attempting longitudinal or prospective evaluations across the entire populace. While Japan’s overarching health initiative, known as Health Japan 21, established the explicit national objective of reducing health inequalities starting in 2013, the absence of an overarching linkage mechanism left policymakers without detailed numerical targets or continuous baseline indicators.
An earlier attempt to bridge this statistical divide utilized data from the 2010 Population Census linked to death records occurring between 2010 and 2015. However, that pioneering effort depended on a linkage key constrained by geographic information available only at the broad municipality level. Consequently, the 2010 linked dataset captured approximately 8 million individuals, representing roughly 10 percent of Japanese residents aged 30 to 79 years. This restriction introduced considerable vulnerability to selection bias and produced erratic mortality estimates that fluctuated between overestimation and underestimation. Because address details were not resolved below the municipal tier, distinguishing individuals who shared identical birth months, birth years, and marital statuses within populous administrative jurisdictions proved exceedingly difficult, highlighting the necessity for a more granular spatial methodology.
To overcome these earlier limitations, the research team secured access to official microdata from the Ministry of Internal Affairs and Communications and the Ministry of Health, Labour and Welfare under the provisions of Article 33 of Japan’s Statistics Act. The baseline 2020 Population Census initially recorded 126,146,099 individual responses. From this starting pool, investigators excluded 2,402,460 foreign residents, 2,202,484 entries with unknown nationality, and 819,916 Japanese individuals missing recorded birth years or birth months. Furthermore, 9,158,818 individuals, representing 7.6 percent of the total population, were excluded due to duplicated linkage keys or unrecorded marital status. The final enumeration framework encompassed 117,827,253 non-institutionalized Japanese citizens and 2,893,986 institutionalized individuals, while the Vital Statistics component supplied 2,949,946 death records compiled between October 2020 and September 2022.
The foundational technical innovation of the 2020 linkage project involved processing the exact residential text addresses on death certificates through geographic information system software using ArcGIS Pro version 3.3. This geocoding procedure converted written addresses into a standardized 14-digit residential address code comprising four hierarchical tiers: the prefecture (two digits), the municipality (three digits), the census block or cho-aza (four digits), and the basic unit block or kihontan-iku (five digits). Basic unit blocks represent the smallest geographic delineation in the Japanese census, encompassing approximately 20 to 30 neighboring families across roughly 1.9 million designated units nationwide. Because the administrative digitization of death records reached nearly 100 percent completion in the late 2010s, the research team successfully assigned full 14-digit geographic codes to 98.3 percent of all submitted death certificates, providing a highly specific spatial anchor absent in previous iterations.
Because Japanese statistical collections do not survey the exact day of birth or retain personal names and unique identification codes within analytical files, data linkage was conducted through a hierarchical, multi-stage deterministic algorithm. The matching protocol evaluated combinations of five linkage variables: sex, birth year and month, residential address codes, marital status, and spouse age when applicable. Linkage proceeded sequentially through distinct hierarchical tiers, designated from I-a through III-b. The highest concordance tiers matched records at the basic unit block level, requiring spouse age for married individuals in tier I-a or matching without spouse age in tier I-b. Broader tiers matched at the census block level (tiers II-a through II-c) or the municipality level (tiers III-a and III-b) to accommodate spatial displacements of up to 50 meters that occasionally arose from geocoding limitations. Tier III-c was intentionally omitted due to concerns regarding unacceptable matching error rates.
Across the complete mortality registry, the deterministic matching framework successfully paired 2,253,228 death records, representing 76.4 percent of all deaths occurring during the two-year window. The highest degree of geographic specificity was achieved in the majority of matches, with 27.9 percent of all deaths linked under level I-a, 31.8 percent linked under level I-b, and 4.5 percent linked under level I-c. Secondary census block tiers accounted for 1.1 percent at level II-a, 5.7 percent at level II-b, and 3.1 percent at level II-c, while municipal tiers accounted for 0.6 percent at level III-a and 1.7 percent at level III-b. Following the exclusion of specific populations—such as individuals residing in long-term hospital care, military installations, school dormitories, and jurisdictions with irregular address notation—the finalized mortality analysis focused on 109,119,620 non-institutionalized Japanese individuals and 2,047,152 non-institutionalized death records, capturing 90.4 percent of the general population and 69.4 percent of all deaths.
To evaluate the validity of the resulting database, the researchers calculated all-cause crude mortality rates stratified into five-year age categories and compared them against unlinked figures derived from the complete national registry. Among non-institutionalized individuals aged 25 to 84 years, crude mortality rates in the linked dataset were approximately 10 to 15 percent lower than those observed in the complete registry. A sensitivity analysis restricting both registries to the presumed non-institutionalized population revealed an underestimation of 10.2 percent among men and 9.8 percent among women aged 25 to 84 years. However, this discrepancy widened substantially among individuals aged 85 years and older. In geographic evaluations across Japan’s 47 prefectures, age-standardized mortality rates derived from the linked database demonstrated strong positive correlations with unlinked registry benchmarks, showing Pearson correlation coefficients of 0.948 for men and 0.881 for women.
Analysis of age-standardized mortality rates among individuals aged 25 to 84 years revealed pronounced differences across multiple sociodemographic indicators. Marital status exhibited a strong association with mortality, particularly among men. Unmarried, widowed, and divorced men experienced all-cause mortality rates of 1,475.5, 1,195.8, and 1,499.0 per 100,000 person-years, respectively, compared with 895.2 among married men. Among women, corresponding rates were 663.8 for single, 499.2 for widowed, 566.0 for divorced, and 429.6 for married individuals. Higher levels of education were consistently associated with lower mortality: male university graduates exhibited an age-standardized rate of 802.1 per 100,000 person-years, compared with 994.4 among high school graduates, while female university graduates exhibited a rate of 378.1 compared with 450.1 among female high school graduates. Furthermore, area deprivation index deciles demonstrated an upward mortality gradient with increasing neighborhood socioeconomic deprivation, an effect that was particularly steady among men.
The authors highlighted several structural limitations inherent to the linkage methodology, chief among them being the systematic underestimation of mortality that disproportionately affected the oldest age cohorts. This pattern likely reflects residential transitions into specialized geriatric care facilities, cognitive decline leading to missing census details, or discrepancies between de facto living addresses and formally registered municipal residencies. Furthermore, post-census changes in marital status or geographic relocation can cause linkage failure, a factor that may differentially affect highly educated individuals who move more frequently. Despite these constraints, the integration of detailed geocoding eliminated classical numerator-denominator bias by establishing a defined prospective cohort of over 109 million individuals. The resulting census-linked mortality framework provides an empirical baseline for investigating domestic health disparities and facilitating standardized comparisons with international registries.
Subject of Research: Development and evaluation of an individual-level census-linked mortality database in Japan
Article Title: Development of a comprehensive mortality database in Japan through data linkage of population census and vital statistics
Article References: Tanaka, H., Katanoda, K., Nakaya, T., Togawa, K., & Kobayashi, Y. (2026). Development of a comprehensive mortality database in Japan through data linkage of population census and vital statistics. Population Health Metrics, 24(1), Article 34. https://doi.org/10.1186/s12963-026-00479-0
Image Credits: AI Generated
DOI: 10.1186/s12963-026-00479-0
Keywords: Japan, mortality database, population census, vital statistics, data linkage, geocoding, age-standardized mortality rates, health inequalities, sociodemographic factors, educational attainment, marital status, area deprivation index, epidemiology, public health, deterministic linkage, death records