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 [2] were well below those of the current study calculated with an open list of diagnoses. The Netherlands study [2] analyzed data from a database of 60,857 patients from a registration network of family practices in the Netherlands [2] 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 [18] 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[19], 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 [13]. Differences in the severity of disease may lead to a different pattern of consultation or to differences in the timing of seeking medical attention [20]. 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 [7] 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 [27]. However, with the exception of psychiatric diseases, questions about the presence of diseases were specific in the CCHS [14] and facilitated recall. Conversely, using medical records alone may result in an underestimation of some symptom-based conditions [26]. 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.