Sex-specific patterns and lifetime risk of multimorbidity in the general population: a 23-year prospective cohort study

This study is embedded within the Rotterdam Study, a prospective, population-based cohort study to assess the occurrence and determinants of age-related diseases in the general population [11]. In 1990, all inhabitants aged 55 years and older from a well-defined suburb in the city of Rotterdam, the Netherlands, were invited to participate. This initial cohort comprised 7983 participants (response rate 78%). In 2000, 3011 participants aged 55 years and older were added to the cohort (response rate 67%). In 2006, 3932 participants were included, aged 45 years and older (response rate 65%). In 2016, additional 3368 participants aged 40 years and older were included (response rate 46%). In total, the Rotterdam Study comprises 18,294 participants aged 40 years or more.

We followed study participants for the occurrence of any of the following diseases: cancer (excluding non-melanoma skin cancer), coronary heart disease, stroke, chronic obstructive pulmonary disease (COPD), depression (including clinically relevant depressive symptoms), type 2 diabetes, dementia, asthma, heart failure, and parkinsonism (including Parkinson’s disease). The selected diseases are common age-related diseases that have a high burden of disease in older populations and are all in the core list of 20 diseases recommended for multimorbidity studies [10, 12]. The participants were selected from the first three cohorts of the Rotterdam Study (n = 14,926, 55 and older for the first two cohorts, 45 and older for the third cohort). To ensure that the ascertainment of disease is consistent across all participants, we excluded participants with a history of one (n = 2742) or more (n = 758) of the selected diseases at baseline. We further excluded 5019 participants who were incompletely screened at baseline for at least one disease of interest, and 313 participants who withdrew informed consent to data collection for a particular disease. The final population included 6094 participants. Baseline population characteristics were collected during the interviews and examinations conducted upon entering the study. Information about ancestry was obtained from whole genome sequencing, conducted as part of the Rotterdam Study [11].

The disease ascertainment at baseline and follow-up comprised of structured interviews and physical examinations by research physicians and affiliated staff at the research centre of the Rotterdam Study. Further information was collected from medical records maintained by participants’ general practitioner, letters from medical specialists, discharge reports and pharmacies, and through digital linkage to the Dutch national cancer registry and the public pathology database. During follow-up, active surveillance for the occurrence of any of the ten diseases took place with research assistants regularly verifying participants’ medical records and other sources for diagnoses of interest. Participants were invited for repeated examinations and interviews every 4 to 6 years. Diagnoses of depression, dementia and parkinsonism involved two-phase screening in which participants suspected of having the disease after the initial screening were invited for additional examination. Cancer diagnoses were confirmed based on pathology records. Diagnoses of depression included those that fulfilled DSM-IV-TR criteria (text revision of the fourth edition of the Diagnostic and Statistical Manual of Mental Disorders) as well as depressive episodes that were deemed clinically relevant, but did not meet DSM-IV-TR criteria. Final diagnoses were adjudicated by a consensus panel composed of research physicians and medical specialists affiliated with the Rotterdam Study. Disease-specific definitions, procedures, and data collection methods have been described previously, a summary is available in the Additional file 1: Ascertainment of diseases [13‐24]. We defined multimorbidity as the co-occurrence of at least two diseases out of the ten diseases included in the study.

We conducted the analyses in three steps. First, we focused on disease trajectories from a single disease to multimorbidity and visualised those trajectories for each participant in a Sankey diagram. Each participant is represented by a stripe connecting adjacent columns that represent the first three diseases in chronological order. The height of the columns is thus proportional to the number of people with each disease and in each state.

Second, we considered different combinations of co-occurring diseases recorded during follow-up, regardless of chronological order. We plotted these combinations in an intersection diagram to visualise combinations of co-occurring diseases and the number of participants with individual diseases and disease combinations [25].

Third, we calculated the lifetime risk of multimorbidity for participants aged 45 years and older, adjusted for age (as time scale). As screening for each of the ten selected diseases did not start on the same date, the follow-up start for individual participants was set to the participant’s age corresponding to the date on which follow-up data for all selected diseases were available. The event of interest was the diagnosis of the second disease in chronological order. The follow-up ended either at the age of diagnosis of the second disease, age of death, the age of loss to follow-up, or the age of the earliest end of follow-up among the selected diseases, whichever came first. Participants were considered lost to follow-up if their medical records could not be accessed. In this analysis, we defined three classes of multimorbidity, based on individual combinations of the first two diseases that led to multimorbidity. All combinations of diseases that involved depression were classified as somatic-psychiatric multimorbidity as the most diverse pair of diseases [26]. We then classified all other combinations as somatic multimorbidity. We further distinguished between multimorbidity only involving diseases affecting one organ system (somatic concordant) and multimorbidity involving somatic diseases affecting different organ systems (somatic discordant) [27]. Somatic concordant multimorbidity involved the combinations of COPD and asthma, coronary heart disease and heart failure, and parkinsonism and dementia. All other pairs of individual diseases were classified as somatic discordant. We used multi-state models to calculate the specific absolute risk for each of the three multimorbidity classes separately for men and women, accounting for the competing risk of death [28]. Finally, we calculated the overall risk of multimorbidity as the composite risks of the three multimorbidity classes.

We conducted five sensitivity analyses to assess whether our definition of multimorbidity classes affects the results. First, we grouped together dementia and depression as mental health diseases, following the International Classification of Diseases. We defined two corresponding classes of multimorbidity: multimorbidity involving mental health disease (any combination with depression or dementia), and multimorbidity involving only physical diseases (any other combinations) [10]. Second, we grouped together diseases that would be typically managed by one specialist to assess the potential risk of fragmentation of care: we grouped stroke together with parkinsonism and dementia (diseases managed by a neurologist or geriatrician); multimorbidity involving any combination of these three diseases was classified as somatic concordant. Third, we considered the combination of the first three diseases (rather than two) in our definition of multimorbidity classes to assess how many participants change their multimorbidity class when the third disease is added. Fourth, we used a stricter definition of depression and considered only DSM-IV-TR defined cases of depressive disorder (single or recurrent), dysthymic disorder, mood disorder due to medical condition and mood disorder not otherwise specified. Participants with clinically relevant depressive symptoms were considered free of depression. Finally, in the fifth analysis, we included in the study population also participants with one prevalent disease and recalculated the lifetime risk of multimorbidity.

The study was carried out following the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines. All analyses were performed in R, version 4.0.4.

In Fig. 1 (interactive version is available online https://​www.​ergo-onderzoek.​nl/​multimorbidity-velek-etal), we plotted disease trajectories for each participant from the onset of a single disease to multimorbidity. Of the study population, 27.9% were diagnosed with two or more diseases during follow-up, of whom 37.0% (10.3% of the total population) had three or more diseases. The maximum number of co-occurring diseases was seven, which occurred in two participants. In contrast, 35.5% of participants remained free from any of the diseases of interest during the follow-up, and 33.8% had one disease. We observed a greater proportion of women with depression as their first disease than men (odds ratio (OR) 2.82, 95% confidence intervals (CI): 2.38–-3.36). Conversely, men were more frequently diagnosed with cancer than women (OR 1.62 (95% CI: 1.37–1.92) or COPD (OR 1.75 (95% CI: 1.48–2.07) as their first disease.

The most frequent combinations of co-occurring diseases are shown in Fig. 2 (interactive version is available online https://​www.​ergo-onderzoek.​nl/​multimorbidity-velek-etal). The most common pair of co-occurring diseases is COPD and depression (11.7% participants with multimorbidity). Of the top ten most frequent combinations of diseases, five include depression (in combination with COPD, dementia, cancer, diabetes and heart failure) and three include cancer (with COPD, depression and diabetes). Among men, the most common pair of diseases is cancer and COPD (12.5% participants with multimorbidity). Of the top ten most frequent combinations, five include COPD (with cancer, coronary heart disease, hear failure, depression and diabetes), four cancer, and three coronary heart disease and diabetes. Among women, the most common pair of diseases is depression and dementia (14.9% participants with multimorbidity). Seven out of ten most frequent combinations include depression (with dementia, COPD, cancer, diabetes, heart failure, stroke and coronary heart disease).

The overall lifetime risk of multimorbidity for men in our sample is 66.0 % (95 CI: 63.2–68.8%) and 65.1% (95% CI: 62.5–67.7%) for women in our sample (p-value for the sex difference 0.33). The risk of multimorbidity increases steeply with age: from 8.5% (95% CI: 6.5–10.9%) at the age of 65 years for men and 7.9% (95% CI: 6.0–10.5%) for women, up to 26.1% (95% CI: 23.5–28.9%) for men and 22.2% (95% CI: 19.9–24.9%) for women at the age of 75 years. The estimated lifetime risk of multimorbidity involving depression (somatic-psychiatric multimorbidity) is almost twice as high for women as for men (30.9% (95% CI: 28.4–33.5%) vs. 17.5% (95% CI: 15.2–20.1%), p-value < 0.001). Conversely, men have higher lifetime risk of multimorbidity with two somatic discordant diseases (44.2% (95% CI: 41.3–47.2%) vs. 29.7% (95% CI: 27.3–32.1%), p-value < 0.001). Compared with somatic discordant and somatic-psychiatric multimorbidity, the lifetime risk of somatic concordant multimorbidity (i.e. a combination of disease affecting the same organ system) remain relatively low (4.2% (95% CI: 3.2–5.5%) for men and 4.5% (95% CI: 3.5–5.7%) for women, p-value for sex difference 0.35) (Fig. 3).

Sensitivity analyses that involved alternative grouping of diseases and alternative definition of our multimorbidity classes resulted in increased risk of somatic concordant multimorbidity, but did not change substantially the sex-specific differences. The lifetime risk of multimorbidity dropped when we excluded cases of clinically relevant depressive symptoms (from 66.0 to 61.2% for men and from 65.1 to 55.7% for women). Yet, the relative difference between men and women in the risk of somatic-psychiatric multimorbidity was similar (12.0% (95% CI: 10.3–13.9) for women vs 6.4% (95% CI: 5.0–8.1) for men). Including participants with one prevalent disease at baseline, increased the overall risk of multimorbidity but the sex-specific differences pertaining to somatic-psychiatric multimorbidity did not change (17.9% (95% CI: 16.4–19.7%) for men and 33.9% (95% CI: 32.1–35.9%) for women). See Additional file 2: Sensitivity analysis for complete results.

In line with previous research, we found that both the risk of multimorbidity and the differences between men and women concerning multimorbidity with mental health diseases increased rapidly with age [1, 26]. Our estimates of the overall risk of multimorbidity are lower than existing estimates of prevalence for corresponding age groups; we also did not observe a higher risk of overall multimorbidity among women, as reported by some studies [1]. This may be due to the different number and type of diseases considered, as we did not include diseases with a low burden of treatment (such as hypertension) and locomotor diseases with high prevalence among elderly (such as osteoarthritis). Our finding that women have a higher absolute risk of multimorbidity with depression is in line with a US study on incident multimorbidity that found that women have a higher risk of depression than men across all ages and a higher risk of multimorbidity with depression [7]. Another US research observed 1.7 higher prevalence of multimorbidity involving a mental health disease among women [26]. A systematic literature review of disease clusters in older adults identified depression among the top twelve diseases that co-occur with other diseases most often [6]. A cross-sectional study from Canada found that non-psychotic mood and anxiety disorders co-occur more frequently with other diseases among middle-aged women than among men [5]. A cohort study from the USA showed that middle-aged and older women (but not men) with depression and/or anxiety have a higher risk of accumulating other chronic conditions as compared to women without depression or anxiety [34]. A Danish study on the effect of prior mental health disease on subsequent physical diseases found a higher risk for women than men for pulmonary and neurological diseases but not for circulatory diseases [37]. Our finding regarding the relatively low risk of multimorbidity with two concordant somatic diseases confirms previous results. The most frequent and most prevalent pairs of co-occurring diseases identified by other studies predominantly involve diseases affecting different organ systems [5, 6], although different selection, definition and ascertainment of diseases across studies limit direct comparison of results.

The major strength of the study includes continuous, long-term follow-up in a single population, which enabled us to overcome the limitations of studies that rely solely on electronic health records, such as inconsistent recording or detection bias. Consistent methods of disease ascertainment combining medical files with interviews and physical examination minimised the risk of misclassification and enabled us to identify participants’ trajectories from a single disease to multimorbidity and estimate the absolute risk of multimorbidity.

There are several limitations of this study. First, it is possible that non-responders to the invitation to participate in the Rotterdam study (28%), participants with incomplete data (34%) or participants with two or more prevalent diseases at baseline (5%) may have had a higher risk of multimorbidity which would lead to underestimation of our results (see Additional file 3: Table A1, A2 and A3 for population characteristics of excluded participants). Second, we did not have continuous follow-up available on chronic kidney disease, liver diseases, chronic pain, sensory impairments or osteoarthritis, which largely affect older adults; we also did not have data on other types of psychiatric diseases such as anxiety or personality disorder. Including these diseases would likely lead to a higher lifetime risk of multimorbidity. Depression is thought to be underreported in men [38]. Potentially, this could have artificially increased the sex differences in multimorbidity with depression. Within the Rotterdam Study, depression is assessed at each research visit with validated questionnaires and interviews in addition to participants’ medical records to minimise underreporting as much as possible. Third, we did not consider recurrent stroke or secondary cancer as separate diseases. Including these data would increase the risk of multimorbidity, though the increase would be relatively small: the risk of secondary cancer in our study population is 11.8%, and a recent estimate of 10-year risk of stroke recurrence is 13% [39]. Finally, as the study participants have predominantly European ancestry, findings have limited generalisability to other ethnicities.