Prevalence estimates of multimorbidity: a comparative study of two sources
© Fortin et al. 2010
Received: 27 November 2009
Accepted: 6 May 2010
Published: 6 May 2010
Skip to main content
© Fortin et al. 2010
Received: 27 November 2009
Accepted: 6 May 2010
Published: 6 May 2010
Published prevalence studies on multimorbidity present diverse data collection methods, sources of data, targeted age groups, diagnoses considered and study populations, making the comparability of prevalence estimates questionable. The objective of this study was to compare prevalence estimates of multimorbidity derived from two sources and to examine the impact of the number of diagnoses considered in the measurement of multimorbidity.
Prevalence of multimorbidity was estimated in adults over 25 years of age from two separate Canadian studies: a 2005 survey of 26,000 respondents randomly selected from the general population and a 2003 study of 980 patients from 21 family practices. We estimated the prevalence of multimorbidity based on the co-occurrence of ≥ 2 and ≥ 3 diseases of the seven diseases listed in the general population survey. For primary care patients, we also estimated multimorbidity prevalence using an open list of chronic diseases.
Prevalence estimates were considerably higher for each age group in the primary care sample than in the general population. For primary care patients, the number of chronic diseases considered for estimates resulted in large differences, especially in younger age groups. The prevalence of multimorbidity increased with age in both study populations.
The prevalence of multimorbidity was substantially lower when estimated in a general population than in a family practice-based sample and was higher when the number of conditions considered increased.
As a result of different factors, including aging populations and advances in medical care and public health policy, a growing proportion of patients present multiple coexistent diseases, or multimorbidity . To estimate the magnitude of this problem, studies about the prevalence of multimorbidity have been conducted in different parts of the world: in Europe [1–5], the Middle East , Australia , the United States [8–10], and Canada [11–13]. These studies present diverse sources of data (e.g., questionnaires, chart reviews, administrative data), data collection methods, targeted age groups, diagnoses considered and study populations, making the comparability of prevalence estimates questionable. Some studies [5, 6, 8–11] focused on data from samples of the general population; others [1–3, 7, 13] from primary care practices. In a recent study  on the prevalence of multimorbidity in four different settings (population-based, general practice, hospital and nursing home), the authors concluded that the study population in research on the prevalence of multimorbidity required special attention. Their study, however, was limited to people ≥ 55 years of age, and the diseases considered for prevalence estimates of multimorbidity varied in the different study populations. Furthermore, information about all morbidities (complaints, diseases and disorders) was collected from medical charts for patients in general practice and from a limited list of chronic conditions for the general population . We believe that differences among the diagnoses considered could have influenced estimates of the prevalence of multimorbidity. Analysis of how the prevalence of multimorbidity in the general population is reflected in clinical practice has not yet been done. To make a valid comparison of prevalence estimates, data should be obtained simultaneously or within a short time frame from both general and primary care populations. In addition, the medical conditions taken into account and the operational definition of multimorbidity used should be consistent. Data from the general population, although useful for a comparison of different geographical areas, provide an incomplete picture of where chronic diseases are dealt with, for example, in the offices of family doctors.
A recent publication  in which prevalence estimates of multimorbidity were calculated using data obtained in 2005 from the general population of the province of Quebec, Canada, allowed us to make a comparison with prevalence estimates of multimorbidity using data from a 2003 family practice-based study  in the same province. The aim of the current study was to compare age- and sex-specific estimates of the prevalence of multimorbidity derived from these two populations, and to examine the impact of a variation in the number of different diagnoses considered on prevalence estimates within the practice-based population.
The prevalence of multimorbidity in the general population of Quebec comes from a Quebec publication  based on data from the Canadian Community Health Survey (CCHS). The CCHS is a series of general surveys that Statistics Canada http://www.statcan.gc.ca has carried out since 2000. Approximately 132,000 randomly-selected Canadians (26,000 from Quebec) 12 years and older participate, either in person or by telephone, in a 45-minute computer-assisted interview. For the current study, we used data collected in the 2005 CCHS  on the self-reported presence of seven diseases: arthritis, cancer, diabetes mellitus, hypertension, heart disease, obstructive lung disease and psychiatric problems. We limited calculations to results from participants aged 25 years and over to permit comparisons.
We obtained family practice-based data from a study conducted by our research group in the Saguenay region of Quebec in 2003 . In this study, trained research staff extracted diagnoses of chronic diseases from patients' medical charts. Details of the methods and sampling strategies used in this study are described elsewhere . The regional ethics committee approved the original Saguenay study. In brief, 980 (90%) of 1085 consecutive adult patients solicited during successive consultation periods from 21 family physicians' practices consented to participate.
For the current study, we limited chronic diseases to the same seven diseases included in the CCHS. We also calculated estimates of the prevalence of multimorbidity in the family-practice patients with an open list of chronic diseases, i.e., all the diseases that might have appeared in the patients' medical records.
We retained the two operational definitions of multimorbidity used in the CCHS report ; that is, we defined multimorbidity as (1) ≥ 2 diseases and (2) ≥ 3 diseases. Both definitions were previously used in several other studies [13–17]. Disease complications were considered as distinct diseases. For the comparison of prevalence data between the two studies [13, 14], we compared multimorbidity data for participants ≥ 25 years of age only because these data were available from both the Saguenay  and the CCHS  studies. Subjects were grouped by age as follows: 25-44 yr, 45-64 yr, 65-79 yr, and 80 yr and over. The analysis was done at the patient level only, as clustering of patients by physician in the Saguenay study was negligible.
We estimated age-specific and sex-specific prevalence, without adjustment for the cluster sample study design and subsequently calculated age-standardized prevalence using the general population as a reference. To determine the significance of different prevalence estimates we calculated Fisher's 95% confidence intervals looking for overlap and also compared them graphically.
Estimated population of Quebec
( N = 6,447,800)
Saguenay family practice-based sample
( N = 938)
Results of the current study suggest that estimates of the prevalence of multimorbidity based on a simple count of diseases for the general population are not equivalent to those for family practice-based populations. Sex- and age-specific prevalence estimates of multimorbidity as well as age-standardized prevalence estimates are substantially higher in the family practice-based than in the general population. The nature of the study population is, therefore, a major factor in the accurate interpretation of studies of the prevalence of multimorbidity. In the current study, for patients with ≥ 2 chronic diseases -- the classical definition of multimorbidity -- the difference in prevalence estimates of multimorbidity for the family practice-based and general care studies reached 40%. Based on age-standardized estimates, the prevalence was approximately three times higher for the primary care-based population. Differences were even more marked for patients with ≥ 3 chronic diseases.
Prevalence estimates of multimorbidity in the general population are important for reporting about the health status of the population. However, the results of the current study suggest that the clinical burden of multimorbidity is higher in family practice than would be expected from data collected for the general population, highlighting the importance of having prevalence estimates at the practice level, and the development and implementation of practice-based epidemiological research. Because of the large percentage of patients with multimorbidity, including geriatric patients, in the primary care population, primary care settings must be strengthened quantitatively and qualitatively to keep pace with this growing problem.
The operational definition considered in prevalence studies about multimorbidity is the second most important concern. In the current study, we found that the greater the number of diagnoses included, the higher the prevalence estimates of multimorbidity. Using the same classical definition of multimorbidity and the same age groups, but varying the number of diagnoses considered (a list of 7 conditions vs. an open list) to compare prevalence estimates for the same family practice-based population, we found large differences in these estimates across all age groups (Fig. 3).
Moreover, not only the number of diseases, but also the way they are documented is important. Prevalence estimates of multimorbidity calculated with a different definition in a Netherlands study  were well below those of the current study calculated with an open list of diagnoses. The Netherlands study  analyzed data from a database of 60,857 patients from a registration network of family practices in the Netherlands  to estimate the prevalence of the co-occurrence of ≥ 2 "active health problems". Health problems, based on ICPC codes related to rubrics, were defined as "active" if identified by the general practitioner or the patient, as reflected in current treatment, subsequent diagnostic investigations, disease monitoring, or the known progressive course of a disease. Prevalence of the co-occurrence of ≥ 2 active health problems for patients 20-39 years of age was 16.0% for men and 18.8% for women; for those 40-59 years of age, 33.6% for men and 35.9% for women; for those 60-79 years of age, 60.9% for men and 64.9% for women; for those ≥ 80 years of age 74.2% for men and 79.9% for women. Compared with the study in Saguenay, these different results may represent real differences in the prevalence of multimorbidity or may be the consequence, at least in part, of the way the diseases were documented in each study. Sampling also contributes to the difference. The Saguenay study recruited patients attending the practice. This could bring out a higher proportion of patients with complex needs as they consult more often and therefore have a higher chance of being selected. At the same time, this provides us with a good estimate of the burden at the practice level. On the other hand, the Netherlands study included all patients from the register (including those consulting less often). This could explain part of the difference. Furthermore, including complications of previous conditions could result in a higher count of diseases. For example, diabetes complicated by renal failure and neuropathy would have counted as three separate occurrences in the Saguenay study thus contributing to the higher numbers.
Many prevalence studies use a limited list of chronic conditions [4, 6, 7, 9, 11, 12]; however, not including frequent conditions could affect prevalence results. Similarly, the inclusion of medical conditions considered as risk factors is controversial. For example, hyperlipidemia and obesity are two common conditions frequently omitted from prevalence studies [4, 6, 12]. The real requirement for medical treatment of such conditions makes a strong argument for their inclusion in the count. Limiting the number of diagnoses considered when defining multimorbidity is of special concern because of the great heterogeneity in disease burden observed among patients with chronic conditions. Ideally, studies about the prevalence of multimorbidity should be based on a standard list of chronic conditions that include at a minimum the most frequent diagnoses. From their review of the literature, Bayliss and colleagues  compiled a list of 24 health conditions most frequently assessed for the measurement of co-morbidity to develop an instrument for the assessment of disease burden. Such a list could be a good reference point for estimating the prevalence of multimorbidity in different populations. If the International Classification of Primary Care-Version 2 (ICPC-2) is used, a good reference point for estimating the prevalence of multimorbidity could be the list of chronic conditions designed by O'Halloran and colleagues, based on the ICPC-2 (list available at http://www.fmrc.org.au/Download/DefiningChronicConditions.pdf). The list, although designed to identify chronic conditions managed in Australian general practice, could be used in other settings.
According to the present study, in the general population, there were more women than men with multimorbidity, but more men than women with multimorbidity are seen in primary care. In the general practice population, prevalence rates tend to be higher among men in the younger age groups . Differences in the severity of disease may lead to a different pattern of consultation or to differences in the timing of seeking medical attention . In general, other population-based studies [5, 6, 9] have shown a higher prevalence of multimorbidity in women. For general practice, some studies [2, 3] reported an age-specific prevalence that tended to be higher for younger men and higher for older women. Another study  reported no difference in the prevalence of multiple diseases between sexes in general practice. However, diseases were classified according to the Cumulative Illness Rating Scale (CIRS) morbidity domains and multimorbidity was defined as presence of morbidity in two or more domains rather than individual diseases. It could have contributed to the lack of differences among gender groups as diseases within the same CIRS domain would count only for one.
The strength of the current study lies in its comparison of two different sources of data that were collected within a relatively short time span, thus validating the comparison. When we look at the relatively slow rate of the incidence of chronic diseases and their long duration, the difference of two years between the two studies' data collections seems negligible for the comparison of two different populations.
This study has limitations. The comparison of prevalence estimates by sex was limited by the different age spans of the study populations (Table 1). Prevalence estimates in the current study may have been affected by the different methods of collecting the data in the two studies compared. Another limitation is that we estimated age-specific and sex-specific prevalence without adjustment for the cluster sample study design, however, it is unlikely that this has affected the findings in terms of the differences in results from the two methods because of the large differences found. In the 2005 study, trained research staff extracted diagnoses of medical conditions from patients' medical charts, making its data collection different from the self-report method used in the CCHS. Because several studies [21–25] have suggested that self-reported and record-based estimations of multimorbidity provide similar results, we considered that a comparison between these two studies was valid. However, other studies [26–28] have reported differences. The questionnaire used for the general population group is susceptible to a self-declaration bias; patients may underreport diagnoses of less importance to them or that they do not recall . However, with the exception of psychiatric diseases, questions about the presence of diseases were specific in the CCHS  and facilitated recall. Conversely, using medical records alone may result in an underestimation of some symptom-based conditions . The direction of the bias with different sources of data could go either way. However, any bias that might have been introduced in the current study is unlikely to affect the robustness of the conclusions, given the magnitude of the most important differences.
Prevalence estimates of multimorbidity for the family practice-based population were higher than those for the general population, in both men and women. Age-specific prevalence estimates were also higher for all age groups in the family practice population and had a higher rate of increase with age. Based on age-standardized estimates and a classical definition of multimorbidity, the prevalence in the practice-based population was three times higher. When the number of chronic conditions considered increased, the prevalence estimate of multimorbidity was higher. The use of a limited list of chronic diseases may introduce an important bias in the prevalence estimates of multimorbidity. Reporting prevalence estimates should always specify the reference population, the method of estimation, and the data source to allow comparison and accurate interpretation of prevalence studies about multimorbidity. The clinical burden on family practice was higher than expected from the data collected for the general population, highlighting the importance of prevalence estimates of multimorbidity at the practice level.
This study was presented in part as a poster at the Wonca Europe 2009 Conference, 16-19 September 2009. Basel, Switzerland.
We thank Linda Cazale and Nathalie Audet from the Institut de la statistique de Quebec for providing the data to complete our comparative analyses. This paper was prepared with the editorial assistance of Sharon Nancekivell, Guelph Ontario, Canada.
The original family practice-based prevalence study used for comparison in the current study was funded by the Fonds de la Recherche en Santé du Québec. The current study was funded by Pfizer Canada (independent research grant). Martin Fortin is funded by the Canadian Institutes of Health Research and partners (Chaire de recherche appliquée des IRSC sur les services et politiques de santé en maladies chroniques en soins de première ligne/Instituts de recherche en santé du Canada, Institut des services et des politiques de santé, Fondation canadienne de recherche sur les services de santé, Centre de santé et de services sociaux de Chicoutimi).
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.