Development of a comprehensive mortality database in Japan through data linkage of population census and vital statistics

Abstract

Objective

No official framework exists for linking the Population Census with Vital Statistics in Japan, limiting the ability to monitor health-related inequalities. We aimed to develop a new methodology for data linkage to describe sociodemographic patterns of mortality in the Japanese population.

Methods

The 2020 Population Census (n = 120,721,239) and Vital Statistics (death records: n = 2,949,946) between October 2020 and September 2022 were linked using a multi-stage deterministic linkage algorithm. Linkage variables included sex, birth year/month, residential address code (basic unit block, census block, and municipality), marital status, and age of spouse. Residential address codes (basic unit blocks: the smallest geographic unit at the Census) of death records were identified through geocoding of the exact residential address. We compared age-standardized mortality rates (ASMRs: per 100,000 person-years) before and after linkage to assess the validity of mortality. We also estimated ASMRs across multiple sociodemographic indicators, including marital status, household size, educational level, occupation, and area deprivation index (ADI).

Results

Of the total death records, 2,253,228 (76.4%) were linked to the 2020 Census. The 2020 Japanese census-linked mortality database ultimately comprised 109,119,620 non-institutionalized Japanese individuals (90.4% of the total population) and 2,047,152 non-institutionalized death records (69.4% of total deaths). Crude all-cause mortality rates in most 5-year age categories from the database were 10–15% lower than those in the complete mortality registry without linkage; however, the difference was markedly greater in older age groups (aged ≥ 85 years), indicating less successful linkage in the age groups that had the highest mortality rates. ASMRs differed according to educational level, marital status, household size, occupation, and ADI (e.g., higher ASMRs among individuals with lower educational levels).

Conclusions

The new linkage framework, which incorporates detailed address information as a linkage variable, has greatly increased the included population and the completeness of linkage. This approach to developing the census-linked mortality data provides a platform for comparing population health across socioeconomic groups within Japan and Japan’s health-related inequalities with those in other countries.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12963-026-00479-0.

Introduction

Health statistics should be as comprehensive, comparable, and sustainable as possible to facilitate effective monitoring of population health and health-related inequalities. National-level analyses of medical services and health-related inequalities are widely conducted using an individual-level linked resident registry, with health-related outcome data serving as a powerful tool for such analyses. These individual-level linkage datasets typically integrate demographic information from a census (or population registries) with health-related data sources, such as death records. Census (or registry)-linked data have been developed in many countries—including England and Wales [1, 2], the United States [3, 4], Canada [5], Denmark [6], Italy [7], Belgium [8], Australia [9], New Zealand [10], and the European consortium [11–13]. In Japan, although several studies have reported socioeconomic inequalities in health using nationally representative health surveys without data linkage [14–18], few monitoring activities or studies of inequality have been conducted at the national level due to the lack of a linkage framework [19, 20]. Moreover, Japan’s national health initiative, “Health Japan 21” has included the general goal of “reducing health inequalities” since 2013; however, detailed numerical target values to guide plans and actions are lacking [21].

The Population Census and Vital Statistics (including death records) are complete statistics and, therefore, the most representative statistics in Japan [22, 23]. However, due to the lack of an official linkage variable, such as a personal identifier between these datasets, investigations of nationwide population health and health-related inequalities have been hindered. To address this, census-linked mortality data, which link death records to individual census questionnaire responses, based on the 2010 Census were developed using a “linkage key” generated from sex, birth year, birth month, address (municipality code only), marital status, and age of spouse [19, 20]. The 2010 census-linked mortality data comprised death records occurring during 2010–2015, linked to some 8 million individuals (approximately 10% of Japanese men and women aged 30–79 years) enumerated in the 2010 Census [19]. The linkage of death records to census questionnaire responses revealed educational inequalities in cause-specific mortality [19]. However, the generated “linkage key” raised concerns regarding selection bias, as address information was identified only at the municipality level [19]. We hypothesized that if data linkage using more detailed address information—such as at a census unit level within each municipality—were feasible in Japan, it would enable more accurate linkage.

To overcome these challenges and limitations, we developed the 2020 Japanese census-linked mortality database (hereinafter, “the census-linked mortality database”) using the latest and most comprehensive data (i.e., detailed address information). This article aims to describe the methodology used to develop and evaluate the novel census-linked mortality database. Additionally, we demonstrate its utility in investigating the sociodemographic patterns of premature mortality in Japanese men and women.

Method

Data sources

The Ministry of Internal Affairs and Communications (MIC) of Japan conducted a population census every five years as of October 1 and published the results, such as the 2020 Population Census (hereinafter, “the Census”) [22]. The original dataset included data on 126,146,099 census questionnaire responses (including foreign nationals). After excluding 2,402,460 foreign nationals, 2,202,484 unknown nationalities and 819,916 Japanese individuals with missing information on birth year and month, we included 117,827,253 non-institutionalized Japanese individuals (57,475,369 men and 60,351,884 women) and 2,893,986 institutionalized Japanese individuals (e.g., residents of geriatric health services facilities, school dormitories, Self-Defense Force bases, and hospitals) without missing information on birth year and month (Supplementary Fig. 1) [22]. We obtained microdata from the Census through MIC following due procedures for microdata use. The Ministry of Health, Labour and Welfare (MHLW) of Japan annually compiles and publishes the Vital Statistics [23]. We also obtained 2,949,946 death records between October 2020 and September 2022 (1,510,091 men and 1,439,855 women) from MHLW following due procedures for microdata use. Foreign nationals residing in Japan were excluded from the linkage.

The MIC and MHLW provided the latest data (the Census and Vital Statistics) after considering data management and protection compliance in accordance with Article 33 of the Statistics Act of Japan. This study was supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI (grant number: 23K16341).

Data collected

Table 1 shows the types of data collected in the Census and Vital Statistics (death records) that were available for data linkage and mortality analysis. The Census is the most comprehensive source of basic population-based data on various administrative measures (e.g., city planning and delineation of electoral districts) at the national and local government levels in Japan. It has been conducted approximately every five years since 1920 [22] The Vital Statistics of Japan are based on registry forms completed by municipal managers, including notifications of births, deaths (certified by a physician), marriages, and divorces, in accordance with each relevant act [23].

Table 1.

Type of data collected in the 2020 population census and vital statistics in Japan

	2020 Population census (as of October 1, 2020)	Vital Statistics (death record)
Linkage variable	✔ Sex ✔ Birth year, month* ✔ Marital status ✔ Age of spouse (living together)** ✔ Address code (basic unit block, i.e., a 14-digit identifier)	✔ Sex ✔ Birth year, month, (day) ✔ Marital status ✔ Age of spouse at death ✔ Exact residential address (text in Japanese)
Common variable	◆ Nationality	◆ Nationality
Specific variable	● Education attainment ● Number of individuals in the household ● Relationship to the householder ● Institutionalization status (e.g., geriatric health services facility, school dormitories, Self-Defense Force bases, and hospitals) ● Employment status ● Occupation (Japan Standard Occupational Classification) ● Industry (Japan Standard Industrial Classification) ● Kind of housing ● Period of living at the present domicile ● Address (municipality as of October 1, 2015) ● Place of work or school ● Means of transportation to the place of work or school	● Date of death ● Cause of death (ICD-10) ● Place of death

*Date of birth is not collected in the Census

**The age of spouse at death was calculated from the Census questionnaire responses based on the spouse’s birth year/month

ICD-10: the International Classification of Diseases 10th revision

2020 Population Census: href="https://www.stat.go.jp/english/data/kokusei/index.html"

Vital Statistics: href="https://www.mhlw.go.jp/english/database/db-hw/populate/index.html"

Data on dates of birth were not collected in the Census [22]. Marital status was classified as single, married, widowed, divorced, or unknown, and the definitions were completely consistent between the Census and death records [22, 23]. Whereas the age of spouse at death was recorded in the death records [23], it was calculated from the Census questionnaire responses based on the spouse’s birth year/month. If the spouse did not live in the same household, the age of spouse could not be determined from the Census. Exact date of death, cause of death, and place of death (not the residential address, but the location such as hospital, home, or geriatric health services facility) were available from the death records.

Data linkage

Figure 1 presents an overview of the data linkage. Owing to the lack of an official personal identification code in Japan, we used five variables as “linkage keys.” These comprised (1) sex, (2) birth year/month, (3) residential address code (basic unit block [kihontan-iku], census block [cho-aza], municipality, and prefecture), (4) marital status, and (5) age of spouse [19]. The residential address code used in the Census was defined as a 14-digit identifier, consisting of the following components: basic unit block (5 digits), census block (4 digits), municipality (3 digits), and prefecture (2 digits) [22]. Approximately 1.9 million census units (each comprising approximately 20–30 families: the smallest geographic unit) were designated as basic unit blocks. Japan has approximately 220,000 census blocks, organized under 1,718 municipalities and 47 prefectures [22].

Fig. 1.

Overview of the 2020 Japanese census-linked mortality database

For the current data linkage, we identified the residential address code for each death record from the exact residential address information using geocoding, a function of the geographical information system (GIS: ArcGIS Pro version 3.3). We successfully identified the residential address code (i.e., a 14-digit identifier), including basic unit blocks, for 98.3% of all death records, introducing a novel method for linking census and death record data in Japan. This procedure became feasible in this update because the digitization of address information in death records reached nearly 100% in the late 2010s.

Individuals with unique linkage keys were identified from the Census and included in the database (Supplementary Fig. 1). Of the total population, 9,158,818 individuals (7.6%) were excluded from the database due to duplicates of the linkage keys or missing marital status.

Death records were linked to the Census data using a multi-stage deterministic linkage method based on linkage levels (Table 2). The combinations of hierarchical address-based (I–III) and marital status-based linkage keys (a–c) were defined as linkage levels (I-a to III-b). We used hierarchical address-based linkage keys as follows: (I) basic unit block, (II) census block, and (III) municipality. We basically attempted linkage by the complete residential address code, including (I) basic unit block (i.e., a 14-digit address identifier), and most of linkage were successfully conducted (linkage level I-a, I-b, and I-c: Table 2). However, slight spatial inconsistencies—such as a spatial displacement of 50 m due to geocoding limitations—occasionally prevented an exact code matching of the basic unit block between the Census and death records. Performing additional linkage at broader geographic levels—such as (II) the census block (i.e., a 9-digit address identifier) and (III) municipality (i.e., a 5-digit address identifier)—helped reduce the influence of these spatial inconsistencies. Marital status and age of spouse were also incorporated as hierarchical linkage keys in the following order: (a) matching with age of spouse (for married only), (b) matching without age of spouse, (c) matching without marital status. For married individuals, the age of spouse at death may not be determined from the Census if they were living separately; therefore, hierarchical marital status-based linkage keys helped the linkage completeness in case of missing age of spouse.

Table 2.

Number of linked death records to the 2020 population census

	Linkage level	Matched linkage key variable	Death records (October 2020–September 2022)* Marital status					Total
		(common variable: "sex", "birth year/month", "prefecture")	Single	Married	Widow	Divorced	Unknown	n	%
(Total)			(268,752)	(1,212,720)	(1,223,916)	(242,813)	(1,745)	(2,949,946)**	-
Linked			192,842	1,082,015	808,766	168,812	793	2,253,228***	76.4
	I-a (the highest)	Common variable + "municipality" + "census block"+ "basic unit block"+ "marital status (married)" + "age of spouse"	-	822,871	-	-	-	822,871	(27.9)
	I-b	Common variable + "municipality"+ "census block" + "basic unit block"+ "marital status"	131,371	143,112	581,845	83,130	0	939,458	(31.8)
	I-c	Common variable + "municipality" + "census block" + "basic unit block"	16,623	10,109	63,777	40,496	672	131,677	(4.5)
	II-a	Common variable + "municipality" + "census block"+ "marital status (married)" + "age of spouse"	-	32,650	-	-	-	32,650	(1.1)
	II-b	Common variable + "municipality" + "census block"+ "marital status"	19,046	36,336	99,004	14,929	0	169,315	(5.7)
	II-c	Common variable + "municipality" + "census block"	16,721	11,724	41,527	20,045	121	90,138	(3.1)
	III-a	Common variable + "municipality" + "marital status (married)" + "age of spouse"	-	17,459	-	-	-	17,459	(0.6)
	III-b (the lowest)	Common variable + "municipality" + "marital status"	9,081	7,754	22,613	10,212	0	49,660	(1.7)
Not linked			75,910	130,705	415,150	74,001	952	696,718	23.6

*These records were linked to the individual census data of 109,119,620 non-institutionalized Japanese individuals (90.4% of the total non-institutionalized population: 52,897,148 men and 56,222,472 women) and 1,656,356 institutionalized Japanese individuals (57.2% of the total institutionalized population: 535,221 men and 1,121,135 women)

**2,949,946 death records between October 2020 and September 2022 (1,510,091 men and 1,439,855 women) from the Vital Statistics

***Among these, 2,047,152 death records (69.4% of the total deaths; 1,171,381 men and 875,771 women) were linked to non-institutionalized Japanese individuals, and the remaining 206,076 death records were linked to institutionalized Japanese individuals (184,574 death records) or non-institutionalized Japanese individuals who were excluded from the database (21,502 death records)

Basic unit block (kihontan-iku: coded as 5 digits): about 1.9 million census units (approximately 20-30 families per census unit)

Census block (cho-aza: 4 digits): about 220 thousand districts

Japan consists of 1,718 municipalities (3 digits) and 47 prefectures (2 digits)

We initially performed linkage at level I-a. For death records that could not be linked at this level, we proceeded to level I-b. In this manner, data linkage was performed sequentially from the highest (I-a) to the lowest linkage level (III-b). For example, the matchings of “linkage key” with basic unit block were defined as I-a (for married only), I-b (without age of spouse) and I-c (without marital status), see Table 2. Also, the matchings of “linkage key” with census block were defined as II-a (for married only), II-b (without age of spouse) and II-c (without marital status). The matchings of “linkage key” with municipality were defined as III-a (for married only) and III-b (without age of spouse). Linkage level III-c was excluded due to concerns about low accuracy of linkage. The highest degree of concordance indicates linkage with I-a for married individuals and I-b for single, widowed, and divorced individuals.

After data linkage, census-enumerated individuals living in school dormitories, Self-Defense Force bases, and hospitals (i.e., patients hospitalized for more than three months) were excluded from the database (n = 480,595). Individuals who were hospitalized were excluded because the hospital address is not recorded as a residential address on death records, resulting in linkage failures. Additionally, specific regions were excluded due to geocoding challenges, including (1) the evacuation zone in Fukushima Prefecture following the nuclear accident (n = 22,694), (2) nineteen small municipalities where geocoding was impractical due to unstructured address notation in rural area (n = 32,718), and (3) parts of Oita City and Beppu City (Oita Prefecture) due to unique local address notation (n = 250,438).

Sociodemographic characteristics

The census-linked mortality database contains general individual sociodemographic characteristics data collected in the Census [22], including marital status, the number of individuals in the household, their educational levels [19] occupations [17, 20] region (prefecture, municipality, and basic unit block), and region-level census-derived indicators (e.g., neighborhood socioeconomic conditions). Among the neighborhood socioeconomic conditions, which were derived as composite indicators of various census-based variables, the area deprivation index (ADI) is available to assess the area poverty level in this database. The ADI is a composite indicator of geographical socioeconomic status by municipality based on the weighted sums of several 2020 census-based variables [24–26]. Mortality in the Japanese population has been reported to be associated with ADI [24–26].

Mortality calculations

To confirm the validity of the census-linked mortality estimates, the all-cause crude mortality rates stratified by 5-year age categories were calculated and compared to the complete registry without linkage. When we considered mortality validations, we excluded institutionalized Japanese individuals from the census-linked mortality database because residential address information from the Census and the death records were generally difficult to match among those population (e.g., individuals who lived in social welfare facilities such as geriatric health service facilities).

Estimated crude mortality rates by educational level were described. We also compared the all-cause and major cause-specific age-standardized mortality rates (ASMRs: per 100,000 person-years) using both the census-linked mortality database and the complete registry without linkage. To assess regional variations, the correlations of ASMRs by 47 prefectures were evaluated between the census-linked mortality database and the complete registry without linkage. Finally, ASMRs for individuals aged 25–84 years were calculated by sociodemographic patterns of premature mortality. The underlying major causes of death were classified into four broad groups: all cancers (C00-D48), diseases of the circulatory system (I00-I99), external causes (including injuries and suicide; V01-Y98), and all other diseases. Person-years were computed as the total time accumulated by individuals at risk of all-cause death. Due to the absence of relevant data, no adjustments for censoring events—such as changes in residential address and marital status—were applied in the calculation. The direct standardized method was used, based on the 2015 Japan Standard Population (5-year age categories) [27].

As a sensitivity analysis, mortality rates of the non-institutionalized Japanese population without linkage were estimated using the number of deaths, excluding those occurring in geriatric health service facilities (from death records), and the population, excluding the residents of such facilities (from the Census). The estimated mortality rates of the non-institutionalized Japanese population without linkage were compared to the census-linked mortality data and the complete registry without linkage, respectively.

Results

Data linkage

Table 2 presents the number of death records from October 2020 to September 2022 linked to the 2020 Census, categorized by linkage level (I-a to III-b). Of the total death records, 2,253,228 (76.4%) were linked to the Census. Among these, 822,871 deaths (27.9%) were linked with the complete matching of sex, birth year/month, prefecture, municipality, census block, basic unit block, marital status (married) and age of spouse (i.e., linkage level I-a). An additional 939,458 deaths (31.8%) were linked with complete matching of sex, birth year/month, prefecture, municipality, census block, basic unit block, and marital status (linkage level I-b). Also, 131,677 deaths (4.5%) were linked with complete matching of sex, birth year/month, prefecture, municipality, census block, and basic unit block (linkage level I-c). The remaining deaths were linked at the following levels: 1.1% (II-a), 5.7% (II-b), 3.1% (II-c), 0.6% (III-a), and 1,7% (III-b).

The census-linked mortality database ultimately included 109,119,620 non-institutionalized Japanese individuals (90.4% of the total population: 52,897,148 men and 56,222,472 women) and 1,656,356 institutionalized Japanese individuals (57.2% of the total institutionalized population: 535,221 men and 1,121,135 women living in social welfare facilities such as geriatric health service facilities). The census‑linked mortality database for non‑institutionalized individuals exhibited sociodemographic distributions that were similar to those of all Japanese individuals (Supplementary Table 1). For mortality estimations of the non-institutionalized Japanese population, the census-linked mortality database included 2,047,152 death records (69.4% of the total deaths; 1,171,381 men and 875,771 women), as shown in Fig. 1. The remaining 206,076 death records were linked to institutionalized Japanese individuals (184,574 death records) or non-institutionalized Japanese individuals who were excluded from the database (21,502 death records). Therefore, we excluded these death records from subsequent mortality calculations (Supplementary Fig. 1). Hereafter, we present mortality rates pertaining to the non-institutionalized Japanese population.

Validations of mortality

Figure 2 illustrates the validations of mortality by comparing crude mortality before and after linkage, as well as the crude mortality rates stratified by educational level (three categories: high, middle, and low) using the census-linked mortality database. The crude all-cause mortality rates in most 5-year age categories from the census-linked mortality database were 10–15% lower than those in the complete mortality registry without linkage, which included institutionalized individuals; however, the difference was markedly greater in older age groups (aged ≥ 85 years), indicating less successful linkage in the age groups that also had the highest mortality rates. As shown in Supplementary Fig. 2, the sensitivity analysis restricted to the presumed non-institutionalized population (see method above) similarly showed greater mortality underestimation by the linked method compared with the unlinked method among both men and women aged ≥ 85 years. When the mortality analysis was further restricted to the presumed non-institutionalized population aged 25–84 years, the linked method underestimated all-cause mortality by 10.2% among men and 9.8% among women, relative to the unlinked method (see Supplementary Table 2). Although 69.4% of total deaths and 90.4% of the total Japanese population were included—corresponding to a 23.2% underestimation when all age groups are considered (based on 69.4/90.4 = 76.8%)—the crude mortality rates in the census‑linked dataset were only 10–15% lower than those in the complete mortality registry when analyses were restricted to individuals aged 25–84 years (see Supplementary Table 3). Crude mortality rates for the high-, middle-, and low-education groups increased with age in both men and women, and the relative differences decreased with increasing age; however, mortality was generally higher among those with lower educational levels (e.g., junior high school graduates).

Fig. 2.

Crude all-cause mortality rate calculated with the 2020 Japanese census-linked mortality database (non-institutionalized individuals) and complete registry without linkage (stratified by 5-year age category and educational level: aged 25–94 years)

All-cause and major cause-specific ASMRs (per 100,000 person-years) are shown in Supplementary Table 4 for all age categories and for individuals aged 25–84 years, respectively. The census-linked mortality database produced results that were more similar to the ASMRs calculated from the complete mortality registry when restricted to men and women aged 25–84 years (which covered 79.0% of all deaths [80.2% in men and 76.8% in women]). The correlations of ASMRs by 47 prefectures between the census-linked mortality database and the complete registry are shown in Supplementary Fig. 3 (Pearson’s correlation coefficients: r = 0.948; p < 0.01 for men and r = 0.881; p < 0.01 for women). Both the complete registry and estimates for the institutionalized Japanese population without linkage yielded similar ASMRs, except for “others” cause of death (Supplementary Table 2: sensitivity analysis).

Sociodemographic patterns of mortality

Figure 3 demonstrates ASMRs (per 100,000 person-years) according to sociodemographic characteristics in men and women aged 25–84 years (see also Supplementary Table 5). Marital status was strongly associated with ASMR; single, widowed, and divorced individuals exhibited substantially higher mortality compared with married individuals, especially for men. The all-cause ASMRs (per 100,000 person-years) for individuals who were single, married, widowed, and divorced were 1,475.5, 895.2, 1,195.8, and 1,499.0, respectively, for men, and 663.8, 429.6, 499.2, and 566.0 for women, respectively. Men living alone had higher ASMR compared with men living with family members. Educational level was clearly associated with ASMR, with a consistent tendency of lower ASMR among individuals with a higher level of education for both sexes. For men, all-cause ASMRs (per 100,000 person-years) were 802.1 (95% confidence interval [CI]: 797.7–806.5) for university graduates and 994.4 (95% CI: 990.9–997.8) for high school graduates. For women, all-cause ASMRs (per 100,000 person-years) were 378.1 (95% CI: 372.3–383.9) for university graduates and 450.1 (95% CI: 448.1–452.2) for high school graduates. ADI showed a clear gradient, with mortality increasing steadily from least to most deprived deciles for men, whereas the gradient was moderate for women.

Fig. 3.

Major cause-specific age-standardized mortality rate (95% confidence interval) and population distribution by sociodemographic characteristics (the 2020 Japanese census-linked mortality database: non-institutionalized individuals aged 25–84 years)

Discussion

Using individual data from the Census and Vital Statistics with detailed address information, we developed the 2020 Japanese census-linked mortality database, which consists of about 109 million Japanese individuals from the 2020 Census. Compared with the previous mortality data, which comprised about 8 million individuals from the 2010 Census [19], the included population increased by approximately 100 million. Improvements to the “linkage key” through digitalized address information have greatly increased the included population and linkage completeness. This new framework for data linkage addresses the lack of a national-level linkage framework in Japan between administrative data, social surveys, and resident registries, enabling investigations into nationwide population health and health-related inequalities.

The census-linked mortality database provides individual-based, comprehensive mortality data for the Japanese population, covering broad sociodemographic characteristics. This enables the investigation of health-related inequalities at both national and regional levels. In this study, clear mortality differences were identified across marital status, educational level, occupation, and the ADI for both sexes. For example, associations between the ADI and years of healthy life expectancy in Japan have been reported using an ecological study design [24]; however, the census-linked mortality database enables individual-based mortality analyses adjusted for neighborhood socioeconomic conditions (e.g., ADI), which is impossible with aggregated mortality data.

Because the date of death is available, the census-linked mortality database can be used for various mortality calculations and analyses, including survival analysis. Causes of death are classified according to the International Classification of Diseases, 10th revision (ICD-10), with no category restrictions. Using the census-linked mortality database, we can also calculate the slope index of inequality (SII), the relative index of inequality (RII), and the population attributable fraction (PAF) for sociodemographic indicators to assess socioeconomic inequalities (Fig. 1) [28].

The census-linked mortality database enables detailed international comparisons of education-based inequalities, which may verify hypotheses that Japan has smaller health inequalities in magnitude than other countries [7–13, 19]. Regarding educational inequalities in mortality, a similar finding was reported using census-linked mortality data from 2010 to 2015 [19]. Therefore, the new database serves as core mortality data for comparing population health within Japan and health-related inequalities with those in other countries. Moreover, the census-linked mortality database allows for the estimation of relative risks by performing multivariable adjusted analyses using age, marital status, household size, educational level, and prefecture as covariates.

The proposed framework facilitates the monitoring of sociodemographic differences in mortality at a lower cost and within shorter periods compared to cohort studies, which were previously used to assess health-related inequalities in Japan. For example, mortality risk differences by marital status were reported in the Japan Collaborative Cohort Study, which followed approximately 100,000 Japanese individuals for 10 years [29]. Our results showed comparable mortality differences by marital status—single men and women had about 1.5–1.6 times the all-cause mortality risk of married men and women [29]. The database successfully expanded these findings, covering over 90% of the general Japanese population and incorporating the latest data from the early 2020s.

We did not use names, exact dates of birth, or unique personal identifiers for data linkage, although such identifiers are commonly used for deterministic linkage in other countries [1–13]. The reason for their unavailability was that such information was not included in the Census database in the first place. Names were collected for administrative purposes related to census completeness, but they were excluded as part of the Census database. In addition, the birth is surveyed only with year and month; consequently, the original census microdata did not contain “day” information. Furthermore, Japan does not have a system for managing respondents using common unique personal identifiers across government statistics, such as Vital Statistics (including death records), or other health databases. Therefore, deterministic linkage based on unique personal identifiers is currently infeasible in Japan. Unless such identifiers are introduced in both the Census and Vital Statistics in the future, the approach adopted in this study will remain the most appropriate method for data linkage.

Strengths

The methodology used to develop the census-linked mortality database offers several advantages. First, GIS techniques were applied using exact residential address information linked to death records, enabling a substantial increase in the included population compared to the previous mortality data [19]. To our knowledge, this is the first instance in which census data for over 100 million individuals have been linked to death records in Japan. Consequently, the database represents the largest Japanese individual-level mortality dataset to date, covering approximately 90% of the Japanese population and including approximately 2 million death records over a two-year period. As a result, selection bias—previously a concern in the 2010 census-linked data—was reduced [19]. Additionally, improvements in the “linkage key” through the use of digitalized address information have greatly and significantly enhanced data linkage completeness, with over 60% death records linked at the basic unit block level, as reported in Table 2. Most death records of married individuals were successfully linked at linkage levels I-a and I-b, whereas those of single, widowed, and divorced individuals were complementarily linked at the subsequent linkage levels. A multi-stage deterministic linkage algorithm substantially improved linkage completeness across all marital status categories.

Second, the census-linked mortality database eliminates the effects of numerator–denominator bias (mismatch of death and population counts) by enabling prospective individual-based analysis starting from the Census [30]. Third, the database includes all combinations of cause-specific mortality (including deaths from COVID-19) and all variables collected in the Census. Fourth, the census-linked mortality database provides a cost-effective means of comparing population health and health-related inequalities over time within Japan and in relation to population in other countries.

Limitations

The most important limitation is that the census-linked mortality rate estimate was 10–15% lower than the rate in the complete mortality registry among the population aged 25 to 84 years, and much more than 10–15% lower among the population aged ≥ 85 years. The methodology will require further refinement to improve linkage among the oldest members of the population. This resulted in a proportional underestimation of crude mortality and the ASMR calculated using the census-linked mortality database. Although the previous mortality data in 2010 involved unstable mortality estimations, including both overestimation and underestimations [19] the current version exhibited unidirectional bias toward underestimation. As we explained above, we presented mortality rates pertaining to the non-institutionalized Japanese population aged 25–84 years as the main results referencing the complete death registry, including institutionalized individuals, to validate mortality estimations. The non-institutionalized Japanese population is generally healthier than the total population, which includes institutionalized individuals; therefore, lower mortality is attributed to restricted non-institutionalized individuals and missing census-death linkages derived from our methodology.

Mortality underestimation may have occurred due to missing census-death linkages. We speculate that the causes of missing census-death linkages (and subsequent mortality underestimation) may include: (1) changes in linkage key, (2) discrepancy between the residential address in the Census and the registered address (which is reported in death records) and (3) inaccuracies or missing data in the Census. If all individuals retained their residence and marital status, the death record would theoretically link perfectly with the Census record, although the risk of misreporting would remain. Changes in address and marital status after the Census can prevent exact census-death linkage and decrease the linkage rate, which may result in underestimation of mortality compared with data from the complete registry. This effect may be more pronounced among individuals with higher socioeconomic status, as those with a higher level of education are more likely to change their address [22]. This may cause underestimations of mortality rates for this population, resulting in overestimations of mortality inequalities [20]. Discrepancies between individuals’ exact residential address and their registered address can also cause underestimations of mortality, particularly among university students aged 20–24 years. We also assume that inaccurate answers or missing data in the Census are more likely in the older population. This could be related to specific causes of death and death after cognitive decline (e.g., senility), which may be associated with incomplete census responses and missing census-death linkages.

False data linkage may have occurred due to accidental matches of “linkage key,” arising from misreported census responses (e.g., misreported birth year/month) or changes in address and marital status after the Census. Additionally, it is plausible that death records, which should have been correctly linked to those excluded individuals, were erroneously linked to other individuals because the individuals with duplicates of the linkage keys or incomplete census responses were excluded from data linkage. However, the proportion of excluded individuals was relatively small (approximately 8% of the total population).

We conducted a sensitivity analysis in which the estimated mortality rates of the non-institutionalized Japanese population served as the reference for the validity of mortality (Supplementary Fig. 2 and Supplementary Table 2). However, calculating exact mortality for the non-institutionalized Japanese population without linkage was not feasible due to the presence of numerator–denominator bias. This is because not all institutionalized individuals necessarily die there, which results in a mismatch of institutionalization status between the numerator and the denominator. Moreover, institutionalization in geriatric health service facilities generally occurs among older people. Therefore, restricting the mortality analysis to individuals aged 25–84 years reduces the influence of institutionalized individuals, and the impact on the representativeness of the census-linked mortality database is likely to be minor. The exclusion of institutionalized or long‑term hospitalized individuals (2.4% of total population) may introduce bias into cause‑specific mortality estimates (e.g., those with sequelae after cerebrovascular disease); however, the magnitude of this bias in mortality inequalities is not substantial for individuals aged 25–84 years, with cause‑specific mortality underestimation ranging from − 5% to −15% (Supplementary Table 2).

Conclusions

The new linkage framework provides novel, individual-based, cause-specific census-linked mortality data for the Japanese population, encompassing a broad range of sociodemographic characteristics. Improvements in the “linkage key” through digitalized address information have greatly increased the included population and the completeness of linkage compared to the previous mortality data from the 2010 Census. The methodology could support a framework for nationwide data linkage in medicine and public health in Japan. The database also serves as a sustainable, low-cost method for comparing population health and health-related inequalities in the Japanese population over time, as well as with populations in other countries.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Author contributions

All authors contributed to the data analysis and the concept and design of the study. H. T. was responsible for the integrity of the data, the accuracy of the data analysis, and the drafting of the manuscript. T. N. managed the ADI dataset. All authors critically reviewed the manuscript. Y. K. and K. K. supervised the study and provided administrative, technical, and material support. All authors accepted responsibility for the decision to submit the manuscript for publication.

Funding

This work was supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI (grant number: 23K16341).

Data availability

The developed mortality data cannot be shared by the authors due to the data protection regulations. Nonetheless, similar mortality data can be reconstructed by linking death records to the Census data using the methodology described in this article. Microdata from the Census and Vital Statistics are available within Japan from the MIC and MHLW, respectively, in accordance with Article 33 of the Statistics Act. The MIC and MHLW may provide data after considering data management and protection policies, provided the researcher, including any international collaborators, meets certain required conditions. More information concerning the availability of microdata on national statistics in Japan is available (in Japanese) at https://www.e-stat.go.jp/microdata/data-use. A detailed database development protocol, which could generate a similar dataset, may be requested from the authors. For more information concerning the 2020 Japanese census-linked mortality database and its methodology, contact Hirokazu Tanaka (hitanak@ncc.go.jp). The authors do not claim copyright over the developed database.

Declarations

Ethics approval and consent to participate

Ethics approval was not applicable. This study is a secondary use of the Census and Vital Statistics data which were managed by the MIC and the MHLW. The MIC and the MHLW provided the data after considering data management and protection compliance in accordance with Article 33 of the Statistics Act, and we used the data in compliance with their instructions.

Competing interests

The authors declare no competing interests.

Disclaimer

Where authors are identified as personnel of the International Agency for Research on Cancer/World Health Organization, the authors alone are responsible for the views expressed in this article. These do not necessarily represent the decisions, policy, or views of the International Agency for Research on Cancer/World Health Organization.