The Prevalence of Disease Clusters in Older Adults with Multiple Chronic Diseases – A Systematic Literature Review

1The growing interest in the concept of multimorbidity, which refers to the co-occurrence of multiple (often chronic) diseases or medical conditions within one person, is motivated by the rising prevalence of multimorbidity, its negative health consequences, and the challenge to manage multimorbid patients in health care settings, often family medicine practice.

2Managing patients with multimorbidity is much more complicated than managing patients with a single condition. Clinical evidence-based guidelines have been developed to provide recommendations for patient management, to define standards of care, and focus efforts to improve quality. However, most clinical guidelines address single diseases, and do not always provide guidance for patients with multimorbidity. Simply combining the current disease oriented guidelines might result in a complex, inconvenient or even conflicting treatment regime, in terms of interactions between drugs and diseases, conflicting management strategies, and polypharmacy. To support health care providers in daily practice, guidelines for combinations of diseases are thus warranted, especially for the most prevalent combinations with complex or incompatible regimes.

3Despite the increasing body of research that has been conducted in the field of multimorbidity, there is still no clear, uniform operational definition for multimorbidity, and thus no clear picture of common multimorbidity combinations. Over the years, various methods have been developed and employed to measure multimorbidity. There are indices available that estimate a multimorbidity-score by weighting a range of diseases (e.g. Charlson Comorbidity Index or Cumulative Illness Rating Scale). Other applied multimorbidity measures are the Chronic Disease Score, RxRisk Model, or the Duke Severity of Illness Checklist. Furthermore, multimorbidity can be assessed by simply counting the number of co-existing diseases within a person, using a predefined list of medical conditions. As disease counts are easy to use, it is presumably the most common approach to define multimorbidity.

4Two recent systematic reviews described the available measures of multimorbidity in more detail and pointed out that the choice of a measure depends on the outcome of interest and the type of data available. Overall, these methods are employed to predict health outcomes, for instance, disability, quality of life, health care utilization or mortality. Additionally, these methods are often applied to assess prevalence rates. Prevalence estimates vary widely depending on the study population, setting, data sources, the type of the diseases considered and the number of conditions included in the analysis.

5Although evidence for the overall prevalence of multimorbidity is accumulating, insight into the prevalence of specific disease combinations (i.e. disease clusters) is limited. A few studies explored disease clusters of multimorbidity by conducting statistical cluster or factor analysis. These studies identified several broad clusters of diseases, but it remained unclear which specific combinations of diseases were most frequently occurring, taken into account the variation in prevalence rates. To the best of our knowledge, there are no systematic reviews that have investigated multimorbidity clusters, and therefore, a complete overview is still lacking.

6With this current systematic review we aim to describe the prevalence of disease clusters in older patients with multimorbidity, as found in published studies. In addition, we intend to acquire information that can be supportive in the process of developing multimorbidity guidelines that could assist patient management and improve quality of health care.

7To find eligible studies we consulted the electronic databases MEDLINE/PubMed, Embase and Cochrane Library. A search strategy was developed for each database, using a combination of key words and Medical Subject Headings (MEDLINE) or Emtree terms (EMBASE and Cochrane Library). Since the term multimorbidity does not have an equivalent in the database’s thesaurus, it was only searched as a key word. Until recently, the term comorbidity was used interchangeably with multimorbidity, as it also refers to the co-existence of multiple conditions. Hence, both terms and their spelling variations were included in our search algorithm. We combined search terms relating to multimorbidity (e.g. “multimorbid*”, “multiple chronic diseas*”, “multiple illness*”), comorbidity, chronic disease, and the definition or measurement (e.g. “index”, “definition”, “measurement”, “list”, “instrument”). The search strategy was developed iteratively to identify a combination of terms with an acceptable level of sensitivity and specificity. We restricted the search to articles with an available abstract, published in English or Dutch, and those published between January 2000 and September 2012. Before the year 2000, only a few articles had been published on the concept of multimorbidity. We did not restrict the search to a specific study type. To be complete, we also screened reference lists of all included articles. The final search strategy for MEDLINE is given in Appendix S1.

8The selection of studies followed several steps. First, different inclusion and exclusion criteria were specified for the selection of studies by title, abstract and full-text (Table 1 ). Second, a random sample of fifteen titles was screened by two authors (JS and JK) to control for unclear formulated inclusion and exclusion criteria, before screening all titles of the yielded articles; there was no disagreement or vagueness. Subsequently, one author (JS) screened all titles for relevancy, based on the defined inclusion and exclusion criteria (Table 1). Third, two authors (JK and JS) independently appraised a sample of twenty abstracts. There was no disagreement between the two authors, after which all remaining abstracts were screened for eligibility by one author (JS) and, when necessary, by a second author (JK or JB). Last, full-text articles were independently screened for eligibility by at least two authors (JS screened all the full texts, and JK and JB both screened half of the full texts). To evaluate the full text articles on the inclusion and exclusion criteria, both authors appointed to screen the full text article filled out a self-constructed checklist. Discrepancies and ambiguities were solved by discussion between the two authors and, when necessary, by a third author.

9After titles and abstracts had been screened, all remaining articles had an observational design. Therefore, quality assessment of the articles was based on several items of the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) checklist, which we included in our checklist. The items that were required to be described in the articles were (1) the study design; (2) the setting; (3) the study size; (4) eligibility criteria of participants; (5) the type of diseases included to measure comorbidity or multimorbidity; (6) the data collection method; and (7) outcome data related to the prevalence of combinations of diseases. These items, with specific conditions, were also considered as inclusion and exclusion criteria (see also Table 1). In addition, to be retained in our review, only those articles that met our inclusion and exclusion criteria, and thus our specified quality standard, were selected.

10For each included study, the following data were extracted:

111 Study characteristics: First author, year of publication, country, study size, setting, population age;

122 Information relating to the number and types of diseases examined;

133 Information relating to (the prevalence of) the presented disease clusters.

14The checklist was employed to gather data about the study characteristics. These data were tabulated and ordered according to the population setting and the presence or absence of a specific index-disease. A mean age was given or calculated, but when impossible the age range was given. Subsequently, all possible diseases, and disease combinations as described in the included studies, were gathered, counted, and tabulated. In addition, the accompanying prevalence rates for each combination were collected and presented. When necessary, odds ratios were converted into prevalence rates. All given prevalence rates concerned the total study sample, and if not, prevalence rates were converted to relate to the total sample.

15In total, 3070 potentially eligible articles were identified, of which 2410 remained after exclusion of duplicates, see Figure 1. After screening of titles and abstracts, 279 articles remained to be read completely. Of these articles, 212 were excluded because they did not meet our inclusion criteria, as shown in Figure 1. Additionally, 45 articles were found to be an abstract or supplement for a congress and were excluded, 1 article was excluded because of double publication of part of the results of the same research project, and of 2 articles we had no access to the full-text. As a result, 19 articles remained. One of these articles focused on multimorbidity in different settings and described the data of these populations separately. These different settings were regarded as 5 individual studies and therefore, our final sample for analysis represented 23 studies. All 23 studies fulfilled our inclusion criteria and met our quality criteria.

16All 23 studies had an observational design and were conducted in either the general population (n =13), primary care (n =7) or ambulatory care setting (n =1). Two studies were based on data of the Veterans Health Administration system (VHA) (Table 2 ). The population size of the studies varied from 599 to over one million individuals. Except for two, all studies reported clusters of two diseases. In five studies patients were only included when diagnosed with a specific disease (i.e. index-disease). In 8 studies prevalence rates were converted to provide comparable prevalence rates of the disease clusters. In one study, odds ratios were converted into prevalence rates.

17Sixty-three different diseases were found, of which some were defined rather broadly (e.g. heart disease, gastrointestinal disease), while others were described in more detail (e.g. cataract, atrial fibrillation). Diabetes mellitus was the most frequently measured disease (described in 19 out of 23 studies). Other commonly assessed diseases were hypertension, cancer, stroke, and depression (Figure 2 ). Besides the 63 diseases, 165 combinations of two diseases (i.e. disease pairs) and 50 combinations of three diseases (i.e. disease triplets) were reported in the studies. Of the disease pairs, 20 were described rather frequently (≥ 3 studies), see Table 3. The disease triplets could not be replicated in any of the other studies and were therefore not further analyzed.

18The rank in frequency of diseases examined in the included studies depended on the definition of the diseases. As displayed in Figure 2, various diseases of the circulatory tract were examined frequently (6 diseases in the top 20). However, the definition of these diseases differed in level of detail. If heart failure, coronary artery disease and heart attack/angina were defined as heart disease (this broad definition could comprise the separate diseases), heart disease was examined in 17 studies instead of in 6 (in some studies coronary artery disease and heart failure were both examined), making it the third most commonly assessed disease. This also applied the category COPD/asthma and the separate diseases asthma and COPD. If the specific diseases were grouped into the broad combined category, then COPD/asthma was investigated in 14 studies, instead of in 9 studies.

19The most frequently assessed combinations concerned 12 different diseases (Table 3). Regarding these diseases, several clusters were identified. Of the assessed diseases, depression was most frequently clustered, and was paired with 8 other diseases. Additionally, hypertension and diabetes mellitus were also found to be commonly clustered in the available studies (with 6 different diseases). Although depression was the disease most frequently assessed in pairs, the highest prevalence rates were found for disease pairs including hypertension, highest for its combination with osteoarthritis (20%). The top ten disease combinations with the highest prevalence rates all included the diseases hypertension, coronary artery disease, and diabetes mellitus. In the studies that focused on a specific index-disease, mainly studies concerning depression, even higher prevalence rates were identified; 57% of the patients with a major depression were also diagnosed with hypertension (see Table 4 ).

20Per study, varying prevalence rates for each disease combination were found. Especially for depression with hypertension (from 1.2% to 12.9%), and for cancer with hypertension (from 1.0% to 10.6%). Further, the highest prevalence values were often found in studies in which the morbidity data were collected via interviews or surveys. These studies almost always concerned the general population. Nearly all studies that applied electronic medical records (EMRs) to collect morbidity data were executed in a primary care setting.

21While multimorbidity in older people seems to be the rule rather than the exception, evidence on the prevalence of specific disease clusters in patients with multimorbidity is limited. In this systematic review 19 articles were included, representing 23 studies, that described 63 diseases and 165 disease pairs. Twenty disease pairs, comprising 12 different diseases, were examined rather frequently. Of the assessed diseases, depression was the disease most frequently clustered, and was paired with 8 different diseases. Hypertension and diabetes mellitus were found to be the second most commonly clustered diseases, and were combined with 6 different diseases. The combinations with the highest prevalence rates included hypertension, coronary artery disease and diabetes mellitus.

22The prevalence estimates of disease clusters differed widely among studies, a result that is in line with findings reported in other reviews. We will discuss two main possible explanations. First, differences in the population under study may affect the prevalence of multimorbidity and related disease clusters, like age, income, or ethnicity. Multimorbidity is strongly associated with age. Although we focused on older adults, the population’s mean age still varied considerably (from 56 years to 85 years). Further, multimorbidity seems more common among people living in socioeconomically deprived areas or people with a low income. Second, variation in prevalence rates might be due to the applied definition of the diseases, the applied data collection method and the study setting. In our review, some diseases were defined very broadly (e.g. cancer, heart disease) while other diseases were defined in more detail (e.g. osteoarthritis, atrial fibrillation). Studies executed in a primary care setting often applied medical records with information on a detailed level, yet they applied different classification codes with different definitions or based on different diagnostic methods (e.g. depression). In contrast, studies applied in the general population often used surveys or interviews, all inquiring about diseases differently. Other diseases, like obesity, are not always considered as a disease and therefore not included. As a consequence, few disease combinations and accompanying prevalence rates were identical.

23With our current results we have identified combinations of diseases that are likely to co-occur and thus, a suitable treatment plan needs to be developed. Existing clinical practice guidelines, however, do not often address multimorbidity, and following all guidelines for all individual diseases may lead to a considerable treatment burden and to contradictory drug and self-care regimes. Indeed, Boyd et al. reported that several potential medication interactions were found for a pattern that consisted of the diseases hypertension, diabetes mellitus, osteoarthritis, osteoporosis, and COPD. Contradicting life-style recommendations were found for osteoporosis and diabetes mellitus.. As it is reasonable that our identified disease pairs are highly common in (elderly) adults, it would be useful if guidelines address potential drug interactions and contradicting treatment recommendations (drug-disease interactions, and disease-disease interactions) for these disease pairs.

24This systematic review has some limitations. We used the term multimorbidity in our search process. This term is not well indexed in literature databases, and we might have missed some studies. To compensate for this constraint, we combined an extended list of text words referring to the term multimorbidity and we included the term comorbidity (with its possible spelling variations) to our search strategy. Next, we developed a scoring method based on several items of the STROBE checklist, and added these items to our strict inclusion and exclusion criteria, in order to obtain a minimal quality standard of all included studies. As a result, we could not differentiate further between levels of quality. Last, with this type of study we were restricted to merely describe the most frequently explored disease pairs in patients with multimorbidity, and not necessarily the most occurring disease pairs. Yet, the 12 identified diseases do represent highly prevalent diseases internationally, and the accompanying combinations of these diseases are also likely to be highly prevalent.

25Reflecting on our findings and limitations, more effort should be made to establish a multimorbidity disease list with uniformly defined diseases. Only by doing so, heterogeneity between study results can be diminished, and information about the prevalence and burden of multimorbidity will be more genuine and comparable. It seems also important to have a better understanding of specific treatment conflicts concerning certain disease clusters, and not merely by scrutinizing the existing guidelines, but by actually assessing daily practice according to guideline recommendations. In this regard, it seems practical to start with the most frequently occurring diseases. Furthermore, it is still valuable to gain more insight into (the prevalence of) specific co-occurring disease clusters, especially of clusters of three, and four diseases, as a large proportion of the elderly population is diagnosed with more than two chronic conditions. For the development of a multimorbidity guideline, however, it might be easier to take into account rather small disease clusters instead of broad, comprehensive disease clusters.

26Management of care for (older) patients with multimorbidity can be challenging, or even burdensome. To be more concrete, health professionals need to strike a balance between the various disease-specific guidelines before one can develop an appropriate treatment plan with feasible recommendations and advices, taking the patient’s personal abilities into account. The disease clusters that we have distinguished, could serve as a first priority setting towards the development of multimorbidity guidelines. A likely option is to start with the most frequently occurring disease combinations, as regards the evaluation of potential treatment conflicts, the adjustment of existing clinical guidelines, or even the development of new guidelines.