Data sources

In Ontario, the costs of all medically necessary care are covered by public health insurance funded from general taxation. This includes all hospital and physician services as well as some home care and long-term care services. Drug coverage is provided to those aged 65 and over and all residents who receive government social assistance payments. All residents of the province are eligible for this medical coverage, and immigrants receive services after a three-month waiting period. Associated provincial health insurance claims databases offer a unique opportunity to identify all individuals who used the health care system and to retrieve information about their medical conditions for inclusion in our cohort. These data are housed and secured at the Institute for Clinical Evaluative Sciences (ICES) under data security and privacy policies and procedures that are approved by the Office of the Information and Privacy Commissioner of Ontario. We used these data to identify persons with chronic conditions who had contact with the healthcare system regardless of where they obtained care. All datasets were linked using unique, encoded identifiers and analyzed at the ICES.

Our cohorts were drawn from a series of linked health administrative databases, not a single insurance file. Ontario’s health care system typically provides services by sector (ie. inpatient, drugs, physician services), and the health administrative databases that capture this service provision are distinct. Therefore, in order to assemble a complete picture of the underlying diagnoses and conditions of Ontarians, various health administrative data were linked using unique encrypted identifiers at the individual level. The Discharge Abstract Database (DAD) consists of data from all hospital discharges in Ontario and the Ontario Health Insurance Plan (OHIP) claims database consists of billing claims for all physician encounters. Additionally, derived chronic condition cohorts developed at ICES using linked data algorithms were also considered [31-37]. These ICES cohorts have been validated and derived from the key databases (DAD, and OHIP claims). The Registered Persons Database (RPDB) was used to identify Ontarians eligible for health insurance coverage and to provide basic demographic information (namely age). Statistics Canada census data were also used to derive population estimates by age and sex in each year [38].

Identifying the chronic conditions

In order to be considered as cases, individuals were required to have a history of at least one of 16 common chronic medical conditions. We selected conditions based on their clinical relevance and burden, both in terms of cost and outcome (e.g. attributable deaths) as described in previous literature [1-4,24,39,40]. These 16 conditions included: arthritis (excluding rheumatoid arthritis), hypertension, asthma, depression, diabetes, cancer, chronic coronary syndrome (CCS), cardiac arrhythmia, osteoporosis, chronic obstructive pulmonary disease (COPD), congestive heart failure (CHF), renal failure, dementia, rheumatoid arthritis, stroke and acute myocardial infarction (AMI). While this represents a small number of possible conditions experienced by individuals, it captures many of the most substantial conditions from a population-based epidemiological perspective. Six of these conditions (AMI, asthma, CHF, COPD, hypertension, diabetes) were defined based on previously validated population-derived ICES cohorts [31-37]. For the conditions where a derived ICES cohort did not exist, we adopted a similar approach to the derivation algorithms (i.e. at least one diagnosis recorded in acute care, or two diagnoses recorded in physician records within a two-year period) to define the remaining chronic conditions: cancer, cardiac arrhythmia, chronic coronary syndrome, dementia, depression, arthritis (excluding rheumatoid arthritis), osteoporosis, renal failure, rheumatoid arthritis, and stroke. The full set of diagnostic codes used to define the conditions is included as Additional file 1. Multimorbidity was defined as the co-occurrence of two or more of these conditions.

Analyses

We reported the prevalence of multimorbidity in the Ontario population by age group, level of multimorbidity (2, 3, 4 and 5 or more conditions), and calendar year (2003 and 2009), by looking at concurrent frequencies. Thus, the prevalence was estimated by dividing the number of individuals with multimorbidity by the total population.

We assessed clustering of chronic conditions in two ways. First, for each selected condition, we measured the likelihood of co-occurring with 1, 2, 3 and 4 or more other conditions. The denominator was the total number of individuals with a specific condition and the numerator was the number of individuals with the same condition plus 1, 2, 3 and 4 or more co-occurring conditions. Secondly, we derived the five most common co-occurring clusters of conditions within each level of multimorbidity (i.e. pairs, triads, quartets, and quintets) and measured their prevalence (number of individuals presenting the most common cluster in the targeted level of multimorbidity divided by the number of individuals in this level).

All data analyses were performed with SAS package version 9.3 (SAS Institute, Cary, North Carolina).

The study received ethics approval by the Sunnybrook Health Sciences Centre Review Board. Participants’ consent was not necessary, as we used administrative data.

Multimorbidity crosses age groups

Overall, we found that in 2009, one in four Ontarians (24%) had at least two of the 16 conditions selected for the study. This proportion reached more than three in four people (81%) among those aged 75 years or more. The high prevalence of multimorbidity has also been reported in other settings and populations, primarily among older people. Like other studies [12,13,17,18,21,41], we found that multimorbidity significantly increases with age. While there is some (limited) evidence showing that young adults and children may also be affected by multimorbidity [19,21,41], most research has focused on older adults [5,9,18,19,42]. In our study, approximately 1.4% of those aged less than 18 years had multimorbidity, and the most common conditions in this age group were asthma, arthritis (excluding rheumatoid arthritis), depression, and cancer. Among young and middle-aged adults (18–64 years), multimorbidity varied from 7% to 35%, with most common conditions being, arthritis (excluding rheumatoid arthritis), hypertension, asthma, depression, diabetes and cancer. In older adults, besides arthritis, hypertension, and diabetes, chronic coronary syndrome, dementia and congestive heart failure are also highly common and contribute to the elevated prevalence of multimorbidity. While improved care and survival are likely the most important focus among older adults, healthy lifestyles and other behavioural factors are also important areas to focus on particularly among young and middle aged adults [3,7,43]. Physical inactivity, alcohol abuse, and an unhealthy diet have been reported as risk factors for chronic conditions such as cancers, diabetes, hypertension [44]. These factors, which are increasingly prevalent among young people [45,46], might contribute to the observed increases in the prevalence of chronic conditions and multimorbidity. There is a need to continue to advocate for prevention activities and healthy lifestyle and also find better ways to care for individuals with multiple conditions, regardless of their age.

Complexity of multimorbidity and the need of a patient-centered approach

One of the more interesting findings is that while single-disease prevalence has increased slightly, multimorbidity has increased significantly between 2003 and 2009 in Ontario. The increase in multimorbidity over the 6-year study period was 10 times higher than the increase of any single condition. Very few studies have analyzed trends over time in multimorbidity and/or illustrated trends in single conditions compared to trends in multimorbidity [18,47]. Uijen et al. [18] reported that the prevalence of single chronic disease(s) was stable over the 20-year period of their study, whereas there was a significant increase in multimorbidity, specifically in those with three conditions (67%) and four or more chronic diseases (288%). A similar trend was reported by Paez et al. [47] in their study based on self-reported data; there was a decrease in the prevalence of individuals with single conditions over ten years, compared to an excess of 1.1 and 5.9 percentage points, among those with two and three or more conditions respectively. We found a more remarkable increase in our study, as the prevalence almost doubled over the six-year period, for those with three or more conditions in Ontario. There are many potential reasons for this 40% increase in the prevalence of multimorbidity including unhealthy behaviors as well as improved understanding of conditions and advances in medical technology leading to longer survival and greater aptitude to live with many conditions [3,7,43]. Additionally, the enhancements in the management of administrative databases and improved tracking of diagnoses may continuously contribute to a greater number of individuals identified with any conditions overtime.

Our study also examined the clustering of conditions and found that there was no common typology among individuals with multimorbidity. Because most clinical programs or guidelines for chronic disease management focus on specific and single conditions, there is a growing concern that these programs may not be sufficient or effective for individuals with multimorbidity [5,10,48]. In order to inform the management of individuals with multimorbidity, it is crucial to identify the combinations of conditions among those with multimorbidity, as addressed in this study. Our findings highlighted the challenges in designing effective disease-oriented management programs, as individuals with multimorbidity do not exhibit dominant combinations of conditions. Developing disease management programs for individuals with multimorbidity will be challenging due to the unmanageable number of clusters of medical conditions. Among individuals with four conditions, the largest cluster of conditions represents only 5% of that population, while no distinct cluster has prevalence higher than 1.5% among individuals with 5 or more conditions. This is a daunting prospect for clinicians aiming to manage these individuals in clinical practice. The number and variety of observed clusters suggests that patient-centred care may be a more appropriate generalized clinical approach [10].

Strengths and limitations

This retrospective population-based cohort study has many strengths and some limitations in regard to the data sources, case definitions and analytical methods, including number of conditions selected or population coverage. The primary limitation could be the use of administrative data instead of definitive clinical information, but this also constitutes a strength for our population-based study. Our study relied on diagnoses recorded in hospitalization data and physicians’ claims to identify people with each of 16 selected conditions. Direct review of medical records has been suggested as the best source for measuring multimorbidity, since they are more comprehensive [11,13]. This is appropriate for individual clinical practices, but population-based analyses require a less resource-intensive approach. Administrative data have been demonstrated to be a relevant alternative [49-51] and a combination of multiple sources of data (hospital and physician) can contribute to produce more reliable estimates. In our study, multiple databases were used to ascertain the cases, including hospital stay (DAD), physician visits (OHIP), and validated disease cohorts.

With appropriate cases definitions, health administrative data can provide valid proxies of clinical status and allow for a suitable estimation of the prevalence or surveillance of chronic conditions [49,51]. An appropriate estimation of the prevalence of multimorbidity depends on the case definition (including selection of conditions), the source of information and the sampling or recruitment strategies [52]. The present study included 16 diagnostic groupings. For six conditions, cases were identified from ICES cohorts based on validated definitions. Validation studies of algorithms used to create ICES registries found sensitivity values from 72% (for hypertension) to 91% (for diabetes). Specificity values varied between 76.5% (for asthma) and 98% (for CHF) [31-37]. A similar approach to case definition was used for non-validated conditions.

In sum, the health administrative databases used in this study have been shown to adequately identify diseases with appropriate algorithms. Moreover the rules for tracking the medical information and the rigorous coding system ensure the accuracy and completeness of the databases. In fact standardized diseases classification and specific fee codes related to the provincial schedule of benefits are being used.

Health administrative databases also offer the advantage of providing information on individuals with multimorbidity at the population level and over time. This study included all of the Ontario population, including children and adults at all ages and found that multimorbidity was a significant concern for younger as well as older populations. The cohort enabled by the use of administrative data also provided sufficient numbers to rigorously assess the clustering of conditions. Thus, we were able to conclude that multimorbidity is a complex matter with few dominant patterns, even among those with two or three conditions. The availability of data in an ongoing manner also creates the possibility of assessing trends and evaluating the longitudinal transitions among individuals with multimorbidity. Future research with these data might detect particular trajectories of disease incidences and lead to identification of risk factors common to particular patterns of multimorbidity. The meaningful findings (increasing trends and age patterns), the multiplicity of data sources, the use of validated registries and the substantial coverage of the population provide some evidence that health administrative databases and the selection of 16 prevalent conditions are useful for measuring and monitoring multimorbidity in Ontario population.

The selection of a limited number of conditions could also be seen as a limitation. However, including all possible conditions is controversial and may be inaccurate since some conditions can be strongly related or constitute risk factors for others [15], leading to an overestimated count. These 16 conditions have a very high degree of overlap with those used in other studies on multimorbidity [24,53]. Most authors usually refer to a limited list of conditions of interest [8,25,29] while others consider an open list of all possible conditions [15,16,18,19,24]. Moreover, when using an open list, only a few and specific conditions contribute the most to the prevalence of multimorbidity [13,14,24,25]. van den Akker and colleagues [19] reported that only a small number of key conditions among a complete list of 335 different diagnoses contributed to the prevalence of multimorbidity, such that adding more conditions to the key conditions did not significantly alter the overall count of individuals with multimorbidity. Fortin and colleagues [11] concluded that it is preferable to use a list of at least 12 chronic diseases for measuring multimorbidity. Our selection of conditions was based on population impact, which is affected by population prevalence and the impact of the conditions on health. The population health impact of these conditions is well supported by previous studies [1-4,24,39,40]. In their review of multimorbidity indices, Diederichs et al. [54] recommend the inclusion of at least 11 of the most common diagnoses. All of these conditions were included in our study. Other researchers could decide to include additional conditions but that is unlikely to affect the general conclusions offered by the present study. In sum, the selection of the 16 conditions for our study constitutes a strength, considering the objective of studying the prevalence of multimorbidity at the population level. The inclusion of more conditions would likely not change the findings on the relative increase in multimorbidity over time or the lack of common clusters. However, extending the number of conditions (e.g. including more mental and physical impairments resulting from multiple chronic conditions) would likely lead to slightly higher estimates of the prevalence of multimorbidity and a greater complexity (ie. level of multimorbidity and clusters). As shown by McLean et al. [41] there is a higher likelihood that considering more mental health conditions would impact the prevalence and complexity of multimorbidity in younger ages.

Our work and that of others emphasizes the need for a more comprehensive approach to understanding multimorbidity, which needs to take into account a greater number of psychiatric conditions and social factors which influence coping and management [54].

An appropriate understanding of multimorbidity also needs to address the factors associated with this phenomenon. Though our study focused on the differences related to age, there are many other factors, including sex and socioeconomic status that contribute to the burden of multimorbidity [18,21,22,41,55]. In future analyses, the impact of sex and socioeconomic status on the prevalence and complexity of multimorbidity will be explored in greater detail.