Background
Multimorbidity is increasingly common in many countries worldwide, particularly those with higher life expectancy [
1‐
3]. Still, most healthcare systems and therapeutic guidelines rely on disease-centred approaches, losing sight of the complexity of multimorbid patients, and hampering patient-centred approaches in clinical decision-making and healthcare planning [
1]. The presence of multiple chronic conditions has been associated with lower quality of life and higher resource utilization and costs [
4‐
7]. Hence, there is growing interest in developing measures of multimorbidity that are useful for resource planning, patient selection and prioritization, and factor adjustment in research and benchmarking [
8‐
10].
The Charlson index, developed in the late ‘80s as a measurement of 1-year mortality risk [
11], was among the first tools proposed for quantifying multimorbidity, and it is still broadly used in healthcare and research settings. Since then, various tools for assessing multimorbidity and patient complexity have been proposed, including quantitative measurements based on the count of chronic diseases (e.g., the Quality and Outcome Framework of the NHS [QOF] [
8], the proposal of the Karolinska Institute for measuring chronic multimorbidity in older people [
12], and the healthcare cost and utilization project [HCUP] of the US Agency for Healthcare Research and Quality [
13]), and exhaustive pay tools for stratifying individuals into pre-established categories of multimorbidity (e.g., the Johns Hopkins Adjusted Clinical Groups [ACG®] [
14] and the 3 M™ clinical risk groups [CRG] classification system [
15]). Aside from marketed and/or nation-wide organizational tools, some authors and healthcare services nearby have explored alternative measures for summarizing the comorbidity burden and/or stratifying the population based on the health risk [
16,
17].
Irrespective of the approach used, various factors challenge the development of meaningful indicators of multimorbidity. First, the concept of multimorbidity typically gravitates around chronic diseases, whereas acute conditions (e.g., hip fracture, pancreatitis) may dramatically increase patient risk and complexity [
18]. Second, there is a lack of consensus regarding the criteria for identifying chronic diseases among all diagnostics [
7,
19]. Finally, some of the proposed indicators (e.g., the QOF, Karolinska measure, and HCUP) are based on unweighted counts of diseases, thus losing sight of the relative contribution of each comorbidity to patient complexity [
20]. While the Charlson index does provide a severity-driven weighted measure of chronic diagnostics, it is limited by the short list of diseases and severity categories considered [
19].
The implementation of centralized electronic records and administrative databases for billing control in many countries has paved the way for big data strategies that allow developing population-based tools for measuring multimorbidity. The deployment of a Catalan Health Surveillance System (CHSS) in our area in 2012 prompted us to develop a population-based tool for stratifying patients according to their morbidity burden. The tool, named morbidity adjusted groups (GMA,
Grupos de Morbilidad Ajustados), is based on the presence of chronic diseases, and it also considers recent acute diagnostic codes [
21]. Like other tools, such as the ACG® and CRG® systems, the GMA is a case-mix tool that allows grouping the population according to their comorbidity burden and taking it into account when assessing outcomes of care. However, the GMA provides additional outputs at the individual level, including the number of chronic diseases, the number of organ systems affected by a chronic disease, a clinical summary label, and the multimorbidity index (i.e., a weighted measure of all diagnostics, which allow quantitative health-risk stratification at a population level) [
22]. The GMA tool has shown good clinical performance—comparable with the CRGs [
23,
24]—, adequate capacity to predict resource utilization in our area [
25], and it has been validated in an external population using the ACG® and CRG® systems as a reference [
16].
In this analysis, we assessed the performance of the multimorbidity index provided by the GMA tool in explaining health outcomes typically associated with multimorbidity and compared it to that of other quantitative measures of multimorbidity such as the Charlson index and the number of chronic diseases according to the QOF, Karolinska, and HCUP systems.
Discussion
In this population-based, retrospective analysis of the general adult population and subpopulations of interest regarding chronic conditions, we compared the performance of various multimorbidity measures in explaining relevant healthcare outcomes associated with the management of patients with multiple chronic diseases. The baseline model of age, sex, and socioeconomic status, historically used for predicting healthcare resource utilization [
34], showed the lowest performance in all investigated outcomes. Of all composites of the baseline model and multimorbidity measures, the GMA multimorbidity index performed consistently better in all outcomes, across all subpopulations, and according to the various statistical estimates used. The GMA multimorbidity index has three main advantages that may explain these results. First, like the HCUP proposal, it exhaustively considers diagnostic codes potentially associated with chronic conditions; the CCS and CCI—morbidity indicators of the HCUP, also used in the GMA proposal for identifying and classifying chronic conditions—minimizes the likelihood of duplicities. Second, although relying on chronic conditions, the GMA tool also considers recent acute diagnoses (e.g., hip fracture, pancreatitis) that may increase patient complexity and even trigger an increase in resource utilization in the mid-time horizon [
18,
35]. Finally, in line with other multimorbidity measures like the Charlson index or ad hoc measures of weighted comorbidity [
36], the GMA multimorbidity index rates comorbidities according to the morbidity burden or severity.
Regarding the other measures of multimorbidity investigated, the Charlson index score performed particularly well in explaining mortality. This finding is consistent with the aim of this index, which was initially developed for predicting 1-year mortality in hospitalized patients. Conversely, the two measures of multimorbidity based on diagnostics count from a short pre-selected list (i.e., the Charlson index, and the QOF) were less accurate than measures that identify chronic conditions more exhaustively (e.g., Karolinska, HCUP, GMA) in explaining outcomes associated with healthcare resource utilization such as polypharmacy and ER or primary care visits. This result could be reasonably explained by the tendency of measures based on short lists of chronic conditions towards prioritizing disabling and life-threatening diseases. While these conditions are likely to influence hard endpoints, such as institutionalization or death, they may lose sight of less severe outcomes such as increased medication use or frequency of use of healthcare resources. The definition of a chronic condition has been identified among the most critical challenges of developing multimorbidity measures, and various authors have discussed the adequate trade-off between simplicity (e.g., use of short lists) and exhaustivity of the definition approach [
7,
10,
37]. In our experience, measures that consider all possible diagnostic codes (e.g., the HCUP and the GMA, both taking all diagnostic groups of the CCS) tended to perform better than those using predefined lists of diagnostics (e.g., QOF or the Karolinska proposal) in most outcomes.
Our analysis focused on multimorbidity measures that yield a numerical value (i.e., either a composite score or the number of chronic conditions) of comorbidity. While this approach excluded other complex tools such as the ACG [
14] or CRG [
15] systems, it allowed us quantitative comparisons of performance using statistics like the ROC-AUC. Of note, the multimorbidity index provided by the GMA tool has been previously compared with the ACG and CRG tools, showing better performance for all outcomes, except patients receiving polypharmacy [
16,
23,
38]. Taken together, the high performance of the GMA tool to explain differences in a variety of outcomes broadens its applicability, including the clinical sphere (e.g., identification of patients at higher risk through a population-based approach to proactively start a closer follow-up), benchmarking between healthcare centres or areas (e.g., adjusting for multimorbidity when comparing key indicators of healthcare delivery such as re-hospitalization rates), and healthcare resource allocation (e.g., anticipating resource needs or prioritizing in case of resource shortage [
39]).
Our analysis was limited by the use of administrative databases, which precluded us from investigating non-recorded outcomes such as quality of life or physical function. In fact, population-based multimorbidity tools, such as those included in our analysis, are typically designed for healthcare planning and resource allocation. Hence, although we have not tested the performance of the GMA for explaining differences in physical or quality of life decline at the individual level, other indexes are expected to perform better on these outcomes [
20,
36,
40]. Furthermore, the retrospective design provided an explanatory approach of healthcare outcomes; future studies shall assess the predictive capacity of these measures prospectively. On the other hand, the analysis was strengthened by the consistency of the main results across the various statistical estimates and subpopulations and the population-based approach, which allowed us to test the multimorbidity measures on a study population of over six million people. Of note, although the source datasets collect only resources afforded by the public insurance system, nearly all people with chronic diseases eventually visit the primary care resources to benefit from the pharmaceutical co-payment. Hence, the CHSS is unlikely to miss information on chronic diagnoses. The frequency of visits to the specialist, which is more likely to be biased in people with double healthcare coverage, was not included among study outcomes.
Conclusions
Our results show that the addition of a quantitative measure of multimorbidity to variables considered traditionally explanatory of healthcare outcomes—such as age, gender, and socioeconomic status—increases the performance of the model in explaining these outcomes. In our analysis, the GMA multimorbidity index performed better than other quantitative measures of multimorbidity in explaining relevant outcomes like all-cause death, total and non-scheduled hospitalization, primary care and ER visits, medication use, admission to a skilled nursing facility for intermediate care, and high expenditure. These findings provide policymakers and medical directors with strong evidence on the use of multimorbidity tools for clinical practice, resource planning, and public health researchers with useful insights for health risk stratification.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.