Article Text

Original research
Complex multimorbidity and mortality in Japan: a prospective propensity-matched cohort study
  1. Daisuke Kato1,
  2. Ichiro Kawachi2,
  3. Junko Saito3,
  4. Naoki Kondo4
  1. 1Department of Family Medicine, Mie University Graduate School of Medicine Faculty of Medicine, Tsu, Mie, Japan
  2. 2Department of Social & Behavioral Sciences, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
  3. 3Behavioral Science Division, Center for Public Health Sciences, National Cancer Center Japan, Chuo-ku, Tokyo, Japan
  4. 4Department of Social Epidemiology, Graduate School of Medicine and School of Public Health, Kyoto University, Kyoto, Japan
  1. Correspondence to Dr Daisuke Kato; d.kato0417{at}


Objectives There are limitations to defining multimorbidity (MM) based on a simple count of diseases. To address these limitations, the concept of complex MM (CMM) focuses on how many body systems are affected in a single patient, rather than counting comorbid conditions. This study compared the prediction of mortality among older Japanese adults between CMM and conventional MM.

Design A population-based prospective cohort study.

Setting The Japan Gerontological Evaluation Study, a nationwide longitudinal cohort study, which ran from 2010 to 2016.

Participants Functionally independent individuals who were older than 65 and had complete illness data at the time of baseline survey were eligible.

Outcomes measure CMM was defined as the coexistence of 3 or more body system disorders at baseline. We calculated the propensity for each individual to develop CMM based on a wide array of characteristics, including socioeconomic status and health behaviours. Individuals with and without CMM were then matched on their propensity scores before we estimated overall survival using a log-rank test.

Results Our 6-year follow-up included 38 889 older adults: 20 233 (52.0%) and 7565 (19.5%) adults with MM and CMM, respectively. In the MM-matched cohort (n=15 666 pairs), the presence of MM was significantly associated with increased mortality (HR 1.07; 95% CI 1.01 to 1.14; p=0.02 by the log-rank test). A similar mortality association was found in the CMM-matched cohort (n=7524 pairs, HR, 1.07; 95% CI 0.99 to 1.16; p=0.08 by the log-rank test).

Conclusion This is the first study to report the association between CMM and mortality among older adults in Japan. MM and CMM predict mortality in older adults to a similar degree. This finding needs to be replicated with more precision in larger samples.

  • social medicine
  • epidemiology
  • geriatric medicine
  • multimorbidity

Data availability statement

Data are available on reasonable request. Data are from the JAGES study. All inquiries are to be addressed to the data management committee via email: All JAGES datasets have ethical and legal restrictions for public deposition due to the inclusion of sensitive information from human participants.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See:

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Strengths and limitations of this study

  • This is the first study to directly compare the associations between two alternative formulations of multimorbidity—conventional multimorbidity and complex multimorbidity—and survival in a nationwide cohort of older Japanese adults.

  • We used propensity score matching to minimise confounding bias when comparing the survival of individuals with and without multimorbidity.

  • One limitation is that we did not take into account the severity of disease at baseline, which may have underestimated the impact of comorbid conditions.


There are limitations in defining multimorbidity (MM; the co-occurrence of diseases in the same person) based on a simple count of diseases,1 and a new concept of ‘complex MM’ (CMM) has thus been proposed.2 CMM focuses on the impact across the different body systems rather than counting comorbid conditions.

In CMM, diseases are categorised by the body system they affect. Because impairments of the same body system often have similar interventions, their impacts on patient prognosis are expected to be similar. Therefore, it makes biological sense to combine closely related diseases (eg, osteoporosis and fractures) as affecting a single body system (ie, musculoskeletal and connective disorders) rather than counting them as two separate diseases when evaluating the impact of multiple comorbid conditions. In turn, disorders of different body systems should be counted separately because they need more complex and extensive treatment, and the treatment of one disease may adversely affect another. Furthermore, from a methodological perspective, focusing on body system disorders may be more reliable method for collecting patient self-report data as patients are apt to misclassify individual conditions (eg, osteoarthritis vs rheumatoid arthritis or asthma vs COPD (chronic obstructive pulmonary disease)) but they are unlikely to mistake the affected body system. CMM is also likely to be a more reliable method as it may avoid issues of whether a clinician sees two very similar diseases as distinct and thereby avoids the issue of some clinicians recording a single condition while others record two.

A growing number of studies have demonstrated the negative impact of MM on patient outcomes, showing that MM is associated with mortality, reduced quality of life, lower physical functioning and so on.1 3–6 In many reports claiming these associations, researchers have attempted to weight diseases according to severity. Although the MM approach is better than conventional medical care that tends to focus on a single disease at a time, the new concept of CMM that focuses on multiple body system disorders is expected to result in stronger predictions of patient outcomes.

There is little evidence on the impact of CMM on mortality.7 Although functional disability is associated with mortality, no previous studies have evaluated the impact of CMM by considering baseline activities of daily living status. Furthermore, previous reports that include both MM and CMM mainly performed descriptive statistics, not inferential statistics.8–11 Against this background, we used CMM and conventional MM to compare the predictions of mortality among older Japanese adults.


Data sources

We conducted this study using the longitudinal nationwide cohort data from the Japan Gerontological Evaluation Study (JAGES),12 which was established in 2010. This study focuses on adults in Japan over 65 years of age and aims to establish a society of healthy longevity.

Study population

Self-administered questionnaires for the baseline survey were mailed to 95 827 older adults in Japan between August 2010 and January 2012. Adults were sampled from 13 municipalities in 7 of the 47 prefectures in Japan. All adults were functionally independent, which was defined as not receiving public long-term care (LTC) insurance. The municipalities were from three of the four major islands of Japan (Hokkaido, Honshu, Kyushu).

Among the target population, 62 426 individuals responded to the survey (response rate, 65.1%). We included individuals who were functionally independent and not receiving any nursing care or home care assistance to avoid reverse causality between MM and functional disability, which is a key factor in mortality. We included individuals who had valid ID, sex, and age information, and who were linked to LTC insurance certification registers. We excluded individuals whose functional disability status at baseline was unknown, or who were already receiving nursing care or home care assistance, or whose data on the history of present illness was missing. Finally, we identified a cohort of 38 889 individuals. Further details of the cohort flow diagram are shown in figure 1.

Figure 1

Flow diagram of the sample group. ADL, activities of daily living.

MM and CMM

At baseline, 19 diseases were surveyed in the JAGES. Among them, as noted in table 1, we analysed the following 17 diseases to calculate MM and CMM: heart disease (including arrhythmia), stroke, high blood pressure, diabetes (including mild type), obesity, dyslipidaemia, impaired vision, gastrointestinal disease, liver disease, impaired hearing, mental disease, sleep problems, osteoporosis, joint disease/neuralgia, injury/fracture, cancer and respiratory disease. The remaining two symptoms, difficulty swallowing and difficulty with bowel movements, were excluded from the disease list in this study because they have aspects of dysfunction not disease. The JAGES did not survey diseases of the nervous system.

Table 1

Definition of body system categories in CMM and diseases surveyed in JAGES

MM was defined as having two or more of the aforementioned diseases concurrently. For CMM, the diseases surveyed were categorised according to the body system they affected.2 13 For example, heart disease and diabetes were individually categorised into disorders of the circulatory system and endocrine system. Next, CMM was defined as the coexistence of 3+ body system disorders at baseline (see table 1).


The outcome of this study was the 6-year incidence of mortality. We ascertained vital status from 2010 to 2016 by linking the cohort participants to the mortality records of the national LTC insurance database (follow-up rate=96.2%). The mean follow-up period was 5.6 years, and we observed 5183 (13.3%) deaths during the period.

Statistical analysis

Estimation of missing data

Given that the missing data was missing at random, we conducted multiple imputations using a bootstrapping Expectation-Maximisation algorithm.14 We analysed 20 multiply imputed datasets, taking the low missing rate of the cohort (approximately 5%) into consideration.15 Lastly, we combined all estimators by Rubin’s rule.16

Propensity score matching

We used propensity score matching to compare overall survival among individuals with and without MM/CMM. To address potential confounding bias, we conducted propensity score matching within a logistic regression framework. The participant information included in estimating the propensity score consisted of 44 variables: age, sex, smoking status, alcohol consumption, marital status, pension, dental health, employment status, consumption of meat of fish/fruits or vegetable, education, city code and so on (see online supplemental table S1).

We performed a 1:1 matching between individuals with and without MM/CMM using the nearest-neighbour method within a calliper (0.2 of the SD of the logit of the propensity score).17 18 We evaluated the covariate balance after matching using standardised differences. An absolute standardised difference of less than 0.1 was considered negligible in the groups (see tables 2 and 3).

Table 2

Demographic characteristics of the cohort study

Table 3

Standardised mean differences with or without MM/CMM, before and after propensity score matching

Survival data analysis

We estimated the overall survival using Kaplan-Meier curves.18 We also compared overall survival between matched with and without MM/CMM groups using a log-rank test.

Sensitivity analysis

While the definition of MM we adopted in this study is one of the most commonly used definitions in previous studies,2 we analysed this cohort data with a more sensitive approach. Specifically, we analysed the association between the number of diseases or body system disorders and the mortality by multivariate analysis with the covariates used in the propensity score calculation. The results of this analysis did not change the direction or significance of the MM/CMM effect (data not shown).

We used R software packages (V.4.0.1) for all statistical analyses, and the statistical significance level was 0.05 for all analyses.

Patient and public involvement

This was a nationwide cohort study focusing on community-dwelling individuals. No patients and the public were involved in this research.


Baseline population characteristics

Among the current cohort study, 20 233 (52.0%) participants out of 38 889 had MM and 7565 (19.5%) had CMM. Table 2 presents the demographic characteristics of the cohort study. Table 3 summarises the background characteristics of the participants between the two groups before and after matching. Populations with MM/CMM were more likely to be older, were more likely to have fewer teeth, and were more vulnerable to financial insecurity (worries about unexpected expenses) compared with those without MM/CMM. Furthermore, compared with populations with MM, populations with CMM were more likely to be female, to have lower education, to eat meals alone and to be unmarried.

MM outcome

After the 1:1 propensity score matching, 31 332 patients were recruited and evenly classified into propensity-matched MM and propensity-matched non-MM groups. The C-statistics before matching for evaluation of the discriminatory ability of the propensity score model was 0.64 (95% CI 0.63 to 0.64).19 The two matched cohorts were well balanced (see table 3). The populations with MM had a 7% higher mortality than those without MM as shown in figure 2 (HR 1.07; 95% CI 1.01 to 1.14; p=0.02 by the log-rank test).

Figure 2

Kaplan-Meier curve for overall survival comparing patients with and without MM. MM, multimorbidity.

CMM outcome

After the 1:1 propensity score matching, 15 048 patients were recruited and evenly classified into propensity-matched CMM and propensity-matched non-CMM groups. The C-statistics before matching for evaluation of the discriminatory ability of the propensity score model was 0.69 (95% CI 0.68 to 0.69).19 The two matched cohorts were well balanced (see table 3). The populations with CMM had slightly higher mortality than those without CMM as shown in figure 3 (HR, 1.07; 95% CI 0.99 to 1.16; p=0.08 by the log-rank test).

Figure 3

Kaplan-Meier curve for overall survival comparing patients with and without CMM. CMM, complex multimorbidity.


To the best of our knowledge, this is the first study to report the association between CMM and mortality among older adults in Japan. MM and CMM predict mortality in older adults to a similar degree.

MM is both an individual and a social issue. Low socioeconomic status (SES) individuals develop MM roughly 10–15 years earlier compared with high SES individuals.20 Therefore, to evaluate whether the presence of MM/CMM is causally related to mortality, SES should be considered as a confounding factor. There were larger intergroup differences in baseline variables for the CMM-matched cohort compared with the MM-matched cohort. Although CMM was already known to be associated with lower SES,21 the current findings indicate that CMM may be more closely related to social factors than MM.

We found that the impact of MM and CMM on mortality was similar. Furthermore, CMM was marginally statistically significantly associated with mortality. This may be partly because the current study did not consider disease severity or disease status except in the baseline survey. That is, it may not sufficiently represent body system disorders in terms of the number of disease groups affected. This finding needs to be replicated with more precision in larger samples.

There are several limitations to this study. First, the self-administered questionnaire was the basis for disease information, which may have led to recall bias. This reporting error may lead to bias in either direction because its extent depends on the type of disease and age.22 Second, although the results are based on a nationwide cohort study, the participants were not nationally representative, and hence external generalisability is not assured. The response rate (around 65%) was comparable to that of other cohort studies for community-dwelling individuals. Third, because this study was observational, our findings cannot be interpreted as indicating causality. Nonetheless, we attempted to minimise confounding bias through the use of propensity score matching.18


Both MM and CMM predicted future mortality among older adults in Japan. These findings indicate the importance of the interactive effects of multiple diseases.

Data availability statement

Data are available on reasonable request. Data are from the JAGES study. All inquiries are to be addressed to the data management committee via email: All JAGES datasets have ethical and legal restrictions for public deposition due to the inclusion of sensitive information from human participants.

Ethics statements

Ethics approval

JAGES participants were informed that participation was voluntary and that their consent to participate in the study was shown by returning the questionnaire via mail. The Nihon Fukushi University Ethics Committee (no. 10-5), National Center for Geriatrics and Gerontology (no. 992-2), and Chiba University Ethics Committee (no. 2493) approved the parent JAGES protocol.


The authors are grateful to all of the staff who conducted the baseline and follow-up surveys, all survey participants, and JAGES group members.


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Contributors All authors (DK, IK, JS and NK) made contributions to the conception and design of this study. DK analysed the data and authored the paper. IK checked the results and advised edits. All authors agreed on the final version of this manuscript.

  • Funding This study used data from JAGES (the Japan Gerontological Evaluation Study). This study was supported by the Japan Society for the Promotion of Science, KAKENHI Grant (JP15H01972), Health Labour Sciences Research Grant (H28-Choju-Ippan002), Japan Agency for Medical Research and Development (JP17dk0110017, JP18dk0110027, JP18ls0110002, JP18le0110009, JP20dk0110034, JP20dk0110037), Open Innovation Platform with Enterprises, Research Institute and Academia (OPERA, JPMJOP1831) from the Japan Science and Technology (JST), a grant from Innovative Research Program on Suicide Countermeasures(1-4), a grant from Sasakawa Sports Foundation, a grant from Japan Health Promotion & Fitness Foundation, a grant from Chiba Foundation for Health Promotion & Disease Prevention, the 8020 Research Grant for fiscal 2019 from the 8020 Promotion Foundation (adopted number: 19-2-06), a grant from Niimi Universit (1915010), grants from Meiji Yasuda Life Foundation of Health and Welfare and the Research Funding for Longevity Sciences from National Center for Geriatrics and Gerontology (29-42, 30-22).

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.