Systematic review on the instruments used for measuring the association of the level of multimorbidity and clinically important outcomes.

OBJECTIVES: There are multiple instruments for measuring multimorbidity. The main objective of this systematic review was to provide a list of instruments that are suitable for use in studies aiming to measure the association of a specific outcome with different levels of multimorbidity as the main independent variable in community-dwelling individuals. The secondary objective was to provide details of the requirements, strengths and limitations of these instruments, and the chosen outcomes.
METHODS: We conducted the review according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (PROSPERO registration number: CRD42018105297). We searched MEDLINE, Embase and CINAHL electronic databases published in English and manually searched the Journal of Comorbidity between 1 January 2010 and 23 October 2020 inclusive. Studies also had to select adult patients from primary care or general population and had at least one specified outcome variable. Two authors screened the titles, abstracts and full texts independently. Disagreements were resolved with a third author. The modified Newcastle-Ottawa Scale was used for quality assessment.
RESULTS: Ninety-six studies were identified, with 69 of them rated to have a low risk of bias. In total, 33 unique instruments were described. Disease Count and weighted indices like Charlson Comorbidity Index were commonly used. Other approaches included pharmaceutical-based instruments. Disease Count was the common instrument used for measuring all three essential core outcomes of multimorbidity research: mortality, mental health and quality of life. There was a rise in the development of novel weighted indices by using prognostic models. The data obtained for measuring multimorbidity were from sources including medical records, patient self-reports and large administrative databases.
CONCLUSIONS: We listed the details of 33 instruments for measuring the level of multimorbidity as a resource for investigators interested in the measurement of multimorbidity for its association with or prediction of a specific outcome. © Author(s) (or their employer(s)) 2021. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

This review builds on Huntley et al’s 2012 review article and provides an updated, comprehensive list of instruments that measure levels of multimorbidity in community-dwelling individuals.

A thorough literature search of three major electronic databases was conducted with the involvement of a health science librarian.

The review is reported based on Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.

This review excluded non-English language articles and grey literature.

Background

Multimorbidity is defined as the co-occurrence of two or more chronic medical conditions in an individual.1 It is a growing public health challenge and accounts for most of the expenditures in the healthcare system.2 The complex interactions of several coexisting diseases have profound implications on individuals3 4 and their healthcare providers.5 6

There are multiple instruments for measuring multimorbidity and many of them do not usually specify the severity of individual conditions.7 No gold standard multimorbidity measurement instrument exists and there is also no agreed categorisation of the available instruments. Sarfati8 9 classified the various measurement instruments into four broad approaches. They are as follows: (1) by simple counts of individual conditions (ie, Disease Count), (2) by organ or system-based approaches, (3) by weighting conditions and combining them into indices and (4) by other miscellaneous approaches. Most of these measurements are used to measure the prevalence or patterns of multimorbidity. However, they can also be used to predict an outcome or to evaluate an intervention for a desired outcome. A set of core outcomes of multimorbidity (COSmm) was proposed after consulting a panel of international experts in multimorbidity intervention studies using a Delphi process.10 Core outcome sets represent the minimum that should be measured and reported in all clinical trials of multimorbidity.11

Huntley et al12 published a systematic review in 2012 describing the instruments used to measure the morbidity burden in primary care and the general population. They found 17 different instruments from 194 articles. The most widely used instruments and those with the most significant evidence of validity were the Charlson Comorbidity Index (CCI), Disease Count and the Adjusted Clinical Groups (ACG) system.12 However, this review was conducted in 2009 and multimorbidity research has increased exponentially since then.

The present review was to build on the review article by Huntley et al12 in order to provide a current and comprehensive list of instruments that measure levels of multimorbidity for community-dwelling individuals. We used the term ‘level of multimorbidity’ to refer to the combined effects of multiple conditions on an individual. The main objective of this review was to list instruments for measuring the levels of multimorbidity. We specifically look for studies that measure the association of a clinically important outcome with different levels of multimorbidity as the main independent variable in community-dwelling individuals. Our second objective was to provide details of the requirements, strengths and limitations of these instruments, and the chosen outcomes in the studies so that clinicians and researchers can select or develop instruments that match their needs for predicting a specific outcome.

Methods

A protocol for this systematic review (CRD42018105297) was published online on PROSPERO.13 We searched MEDLINE, EMBASE, CINAHL and also manually searched the Journal of Comorbidity for potential studies. The medical subject headings and keywords used for the search are shown in online supplemental appendix 1.

We selected studies that included (1) adult patients from primary care or the general population as the majority of patients with multimorbidity are managed by primary care physicians14; (2) at least one specified outcome variable; and (3) published full-text articles from 1 January 2010 to 23 October 2020. Studies were excluded if they (1) selected patients from the hospital or nursing home only or patient data were drawn solely from the hospital or the nursing home; or (2) selected patients with an index condition; or (3) used level of multimorbidity as a covariate and not the main independent variable; or (4) were not written in English. We did not include a specific definition of multimorbidity instrument because, given a lack of consensus in the literature on the use of this term, we wanted to include a diverse range of studies on the above topic.

One reviewer (ESL) conducted a preliminary screen of titles and abstracts to exclude articles that were irrelevant. Abstracts of the remaining articles were screened independently by two reviewers (ESL and EQ-YH) according to the eligibility criteria. Disagreements were resolved through discussion until a consensus was reached. The full-text articles were then retrieved for the agreed list and independently assessed according to the eligibility criteria by the same reviewers. Disagreements were resolved through discussion with a third reviewer (TSH) until a consensus was reached. After agreement on the list of articles, the reference lists of included articles were hand-searched for additional eligible articles. We reported multimorbidity instruments that were described in all selected articles.

The risk of bias of the study design of selected articles was next appraised independently by three reviewers (ESL, EH and TSH) using the modified Newcastle-Ottawa Scale (NOS).15 16 Each article was assessed under the three broad categories: (1) selection, (2) comparability and (3) outcome (online supplemental appendices 2 and 3).

We contacted the authors, as needed, for additional information or clarification up to three times spaced 1 week apart. We contacted 25 authors and 19 of them replied. Any disagreements on the risk of bias were resolved among the three reviewers through regular meetings. HLK and FYW were responsible for tracking and updating the final outcome of the risk of bias assessment.

Patient and public involvement

This research was done without patient involvement. Patients were not invited to comment on the study design and were not consulted to develop patient relevant outcomes or to interpret the results. Patients were not invited to contribute to the writing or editing of this document for readability or accuracy.

Results

The number of included studies was 96, of which 69 were assessed to have low risk of bias. A summary of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow chart is depicted in figure 1. Forty-eight studies selected participants from the general population and the other 48 studies selected participants from primary care. Most of the studies in this systematic review were from Europe and North America with very few Asian studies. There were 44 cohort studies, of which 36 were assessed to have low risk of bias, and 52 cross-sectional studies, of which 33 were assessed to have low risk of bias. We found 33 unique instruments from the 96 studies. The instruments were categorised according to Sarfati8 9 into (1) simple counts of individual conditions; (2) organ or system-based approaches; (3) conditions that have been weighted and combined into indices; and (4) other approaches. A total of 150 outcomes were reported from all the studies. No studies were excluded for an outcome that was not deemed to be clinically important. Online supplemental appendices 4 and 5 summarise the risk of bias assessment of each study. Table 1 provides a summary of the study design, population source, age group, multimorbidity measurements, outcome measures and risk of bias assessment of all the studies.

Figure 1

PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram.

Table 1

Summary of included studies

Author (Year)	Study design	Population source	Age	Multimorbidity measurement	Outcomes measured	Risk of bias
Agborsangaya et al (2013)41	CS	GP	≥18	DC	HRQoL	Good
Bähler et al (2015)42	CS	GP	≥65	DC-ATC classification system	Total number of consultations	Good
Barile et al (2013)43	Cohort	GP	≥65	DC	ADL limitations, physically unhealthy days, mentally unhealthy days	Good
Barile et al (2012)44	CS	GP	≥65	DC	Physical HRQoL, mental HRQoL	Good
Barnett et al (2012)45	CS	PC	≥0	DC	Presence of mental health disorder	Good
Biehl et al (2016)46	Cohort	PC	≥65	ERA, CCI	Presence of critical illness	Good
Boeckxstaens et al (2015a)17	CS	PC	≥80	DC, CCI, CIRS	Disability (measured by ADL), frailty (five components)	Poor
Boeckxstaens et al (2015b)18	Cohort	PC	≥80	DC, mCCI, CIRS	Mortality at 3 years, hospitalisation at 3 years, functional decline at 19 months (ADL, physical, mental decline)	Fair
Brilleman et al (2014)47	Cohort	PC	≥18	QOF count, CCI, EDC count, ACG, RUB	Primary healthcare costThe EDC performed best followed by the QOF and ACG	Good
Brilleman and Salisbury (2013)48	Cohort	PC	≥18	QOF count, CCI, EDC count, ACG, RUB, prescribed drugs count	Mortality: The CCI was the best performing measure followed by the number of prescribed drugs.Number of primary care consultations (3-year period): The number of prescribed drugs had the greatest predictive validity followed by the ACG-based measures (ACG, EDC count and RUB).	Good
Caballer-Tarazona et al (2019)49	CS	GP	≥0	CRG	Expenditure of integrated healthcare (hospital, primary healthcare (PHC) and pharmaceutical prescription)	Poor
Carey et al (2013)50	Cohort	PC	≥60	Standard QOF, extended QOF, CCI (Khan)	Mortality (1-year period)The standard QOF score outperformed the CCI (Khan). The extended QOF score produced only a modest improvement in overall model performance.	Good
Chapman et al (2015)51	Cohort	GP	≥18	CCI, CCI-PSR	Mortality (5, 10, 15, 20, 25-year period)The CCI-PSR showed substantially better discrimination than the CCI.	Good
Charlson et al (2014)20	Cohort	GP	≥0	CCI	Healthcare cost, utilisation of services	Good
Chen et al (2011)52	CS	GP	≥18	DC	General health, mental distress, physical distress, activity limitations	Good
Chen et al (2018)53	CS	GP	≥45	DC	Health service utilisation	Poor
Chu et al (2018)54	CS	PC	≥40	DC, CIRS	Healthcare utilisation	Good
Clynes et al (2020)55	CS	GP	(Born in 1931–1939)	DC	Physical functioning	Poor
Crane et al (2010)56	Cohort	PC	≥60	ERA	Number of hospital visits, ED visits, hospital admissions, days hospitalised (1-year period)	Good
Crooks et al (2016)57	Cohort	PC	20–100	Comorbidity linked score, CCI, EI	Mortality (1-year period)The linked score had significantly improved discrimination and fit compared with the CCI and the Elixhauser Index	Good
Crooks et al (2015)58	Cohort	PC	≥20	CCI (Read), CCI (ICD-10), CCI (Read and ICD-10)	All-cause mortality (1–5 years)There was no large difference in the discrimination of the model for whichever codes that were used to derive the CCI.	Good
DiNapoli et al (2017)19	CS	PC	≥50	Organ systems with chronic disease	Presence of depressive or anxiety disorder	Good
Formiga et al (2013)59	Cohort	PC	85	CCI	Mortality (3-year period)	Good
Formiga et al (2011a)60	Cohort	GP	90 to 99	CCI	Mortality (5-year period)	Good
Formiga et al (2011b)61	CS	PC	85	CCI	Successful ageing	Good
Formiga et al (2016)62	Cohort	PC	85	CCI	Mortality (5-year period)	Good
Fraccaro et al (2016)63	Cohort	PC	≥18	CCI (Khan)	Mortality (1, 5, 10-year period), mortality (3, 6, 12-month period)	Good
Galenkamp et al (2011)64	CS	GP	57–98	DC	SRH	Good
Garin et al (2014)65	CS	GP	≥50	DC	QOL, disability	Good
Glynn et al (2011)66	CS	PC	>50	DC	Primary care consultations, hospital outpatient visits, hospital admissions, healthcare cost (all 1-year period)	Good
Gunn et al (2012)67	CS	PC	18–76	DC	Depressive symptoms (CES-D score)	Fair
Haas et al (2013)21	Cohort	PC	≥18	ACG, Minnesota Healthcare Home Tiering, HCC, ERA, CCC, CCI, hybrid model	Hospitalisation, ED visits, readmission within 30 days, healthcare expenditure (all 1-year period)The ACG model outperformed the other five models in all outcomes.	Good
Hanmer et al (2010)68	CS	GP	22 to 106	Additive model, minimum model, multiplicative model	Health utility (SF-6D)	Fair
Hu et al (2017)69	CS	PC	≥65	Age-adjusted CCI	Frequency of family physician visits	Fair
Hwang et al (2015)23	Cohort	GP	≥0	ACE-27, ACE-27 count	Healthcare expenditureThe model, using year 1 data to determine if an individual would be classified into the persistent high-user group for the following 3 years, indicates a very high level of accuracy in predicting membership in a high-user group.	Good
Isaacs et al (2014)70	CS	PC	18–101	DC	Prescription costs	Poor
Jennings et al (2015)71	Cohort	PC	≥75	DC	Count of fall-related injuries in the 24 months after the date of screening	Fair
Jia et al (2018)72	Cohort	GP	≥65	DC	Quality-adjusted life years (QALY)	Poor
Jia and Lebetkin (2017)73	Cohort	GP	≥65	DC	Quality-adjusted life years (QALY)	Poor
Jindai et al (2016)74	CS	GP	≥65	DC	Functional limitations (ADL, IADL, leisure and social activities, lower-extremity mobility, general physical activities)	Good
Kim et al (2012)75	CS	GP	≥65	DC	Quality of life (EQ5D)	Poor
Kojima et al (2011)76	CS	PC	≥65	DC	Fall tendency	Poor
Kristensen et al (2014)77	CS	PC	>0	RUB	Fee-for-services expenditures	Good
Lapi et al (2015)78	CS	PC	≥15	HSMI	Total mean healthcare cost per yearThe HSMI explained 50.17% of the variation in costs	Good
Lawson et al (2013)79	CS	GP	≥20	DC	Preference-weighted HRQoL	Good
Lemke et al (2012)80	Cohort	GP	≥0	CCI, ACG	Inpatient hospitalisationsACG-based predictive model was superior to CCI model.	Good
Li et al (2016)81	CS	GP	16–68	DC	Health-related quality of life	Poor
Loprinzi et al (2016)82	CS	GP	60–85	DC	Cognitive function	Good
Macinko et al (2019)83	CS	GP	≥18	DC (categorical 2 and 3 or more) (self-reported)	Primary care experience (self-reported)	Good
Marengoni et al (2011)84	CS	GP	≥75 (baseline)≥77 (follow-up)	DC	Disability	Good
McDaid et al (2013)85	CS	GP	≥50	DC	Disability, QoL, SRH	Good
Md Yusof et al (2010)86	Cohort	GP	64–85	CCI,	Mortality over 7 years	Fair
Milla-Perseguer et al (2019)87	CS	PC	≥18	CRG	Health-related quality of life (HRQL)—EQ-5D-3L	Good
Monterde et al (2020)88	Cohort	GP	≥18	Adjusted morbidity group (GMA), CCI, DC, CRG	Use of healthcare resources	Good
Muggah et al (2012)89	CS	GP	≥20	DC	Primary healthcare use	Poor
Mujica-Mota et al (2015)90	CS	PC	≥18	DC	Health-related quality of life (EQ5D)	Fair
Naessens et al (2011)91	CS	GP	18–64	DC	Healthcare cost	Poor
Østergaard and Foldager (2011)92	CS	PC	≥18	DC	Major depressive episode (measured by DSQ)	Poor
Palladino et al (2019)93	CS	GP	≥50	DC	Primary care use, reduced functional capacity, self-perceived health, hospital admissions, quality of life	Good
Pati et al (2019)94	CS	PC	≥18	Severity burden score (21 conditions)	Health-related quality of life (SF-12)	Good
Payne et al (2013)95	Cohort	PC	≥20	DC	Unplanned hospital admission, potentially preventable admission (all 1-year period)	Good
Payne et al (2014)96	Cohort	PC	≥20	DC	Unplanned hospital admissions (1-year period)	Good
Payne et al (2020)97	Cohort	PC	≥20	CCI, DC (37 read codes), Cambridge Multimorbidity Score	Mortality, unplanned inpatient hospital admission, primary care consultations	Good
Peters et al (2018)98	CS	PC	18–101	DC, DBIS	Quality of life	Fair
Quail et al (2011)99	Cohort	GP	≥20	DC, CCI (Quan), Elixhauser (Quan), number of different dispended drugs, CDS	Mortality (1-year period): Elixhauser (Quan) performed best followed by CCI.One or more hospitalisations; two or more hospitalisations: DC was the best performing measure	Good
Ranstad et al (2014)100	CS	GP	≥0	RUB	Registered active listing in primary care and all healthcare	Good
Reinke et al (2019)101	CS	PC	30–94	DC	Symptom burden (MSAS-SF), quality of life (Veterans RAND 12)	Good
Renne and Gobbens (2018)102	CS	PC	≥70	DC	Quality of life	Poor
Reyes et al (2014)103	Cohort	PC (men)	≥65	CCI	Hip fractures	Good
Ryu et al (2015)104	CS	PC	≥18	DC	Deficits of perceived general health, depressive symptoms	Good
Salisbury et al (2011)105	Cohort	PC	≥18	QOF count, EDC count	Primary care consultation rates, continuity of care (all 3-year period)	Good
Saver et al (2014)106	Cohort	GP	≥65	CCI (Romano)+Hypertension	Acute ACSH, chronic ACSH	Good
Shadmi et al (2011)107	CS	GP	≥18	ADG, CCI	Number of primary care physician visits, specialist visits, hospitalisationADG explained the largest percent of variance or in healthcare resource use	Good
Sibley et al (2014)108	CS	GP	≥65	DC	Self-reported falls in the last 12 months	Poor
Stanley and Sarfati (2017)109	Cohort	PC	≥18	M3 Index, CCI, Elixhauser (van Walraven)	Mortality, overnight hospitalisation (all 1-year period)M3 Index outperformed both CCI and Elixhauser (van Walraven)	Good
St John et al (2014)110	Cohort	GP	≥65	DC (0–36 conditions)	Mortality in 5 years	Good
St John et al (2019)111	Cohort	GP	≥65	DC	Functional impairment in 5 years	Good
Streit et al (2014)27	Cohort	PC	50–80	CCI, DC	Quality of cardiovascular preventive care, quality of preventive care	Good
Sullivan et al (2012)112	CS	GP	≥18	DC	Preference-based HRQoL	Good
Takahashi et al (2011)113	Cohort	PC	>60	ERA	Mortality, nursing home placement (all 2-year period)	Good
Takahashi et al (2016)114	Cohort	PC	≥18	Minnesota Tiering (ACG), enhanced model	Hospitalisation/ED visitsThe enhanced model is better	Good
Tyack et al (2016)115	Cohort	PC	≥18	DC	Health-related quality of life	Fair
Ubalde-Lopez et al (2016)24	CS	GP	F (mean): 35.9,M (mean): 37.9	MDMS	Sickness absence episodes taken in last 2 years	Good
van den Bussche et al (2011)116	CS	PC	≥65	DC	Frequency of contacts with physicians, number of different ambulatory physicians contacted (all 1-year period)	Good
van Oostrom et al (2014)117	CS	PC	≥55	DC	Number of contacts with general practice, medications prescribed, referrals	Good
Vos et al (2013)118	CS	PC	70–74	DC	Self-rated health (SF-36)	Poor
Wallace et al (2016a)119	Cohort	PC	≥70	Pra tool, modified Pra tool	Emergency hospital admission (1-year period)Both models demonstrated poor model discrimination	Good
Wallace et al (2016b)120	Cohort	PC	≥70	DC, Barnett conditions DC, CCI, prescribed drugs count, RxRisk-V	Emergency admission, functional decline (all 2-year period)All measures demonstrated poor discrimination	Good
Wei et al (2018)121	CS	GP	≥51	MWI	Subjective physical functioning, grip strength, gait speed, cognitive performance, ADL limitations, IADL limitations	Good
Wei et al (2019a)122	Cohort	GP	≥51	MWI	Physical functioning—SF-36, mortality	Good
Wei and Mukamal (2019b)123	Cohort	GP	≥51	MWI	Suicide mortality, health-related quality of life	Fair
Wei et al (2020a)124	Cohort	GP	≥51	MWI	Cognitive functioning	Good
Wei et al (2020b)125	Cohort	GP	≥51	MWI-ICD, DC, CCI, Elixhauser, health-related quality of life comorbidity index	Mortality, future physical functioning	Poor
Wei and Mukamal (2018)28	Cohort	GP	≥36	MWI, DC, CCI	Mortality (10-year period), future physical functioningMWI performed best in predicting mortality as compared with DC and CCI	Good
Wikman et al (2011)126	CS	GP	≥50	DC	QoL, affective well-being	Good
Wister et al (2015) 22	CS	GP	≥65	MM additive scale, MM weighted by HUI3, MM weighted by ADL scale, MM weighted by HUI3 betas	Life satisfaction, perceived health status	Good

ACE, Adult Comorbidity Evaluation; ACG, Adjusted Clinical Groups; ACSH, Ambulatory Care Sensitive Hospitalisation; ADG, Aggregated Diagnosis Groups; ADL, Activities of Daily Living; CCC, Chronic Condition Count; CCI, Charlson Comorbidity Index; CCI-PSR, Charlson Comorbidity Index-Psychosocial Risk; CDS, Chronic Disease Score; CIRS, Cumulative Illness Rating Scale; CRG, Clinical Risk Groups; CS, Cross-Sectional; DBIS, Disease Burden Impact Scale; DC, Disease Count (Unweighted); ED, Emergency Department; EDC, Expanded Diagnosis Clusters; EI, Elixhauser Index; ERA, Elder Risk Assessment; GP, General Population; HCC, Hierarchical Condition Categories; HRQoL, Health-Related Quality of Life; HSMI, Health Search Morbidity Index; HUI3, Health Utility Index; IADL, Instrumental Activities of Daily Living; ICD-10, International Classification of Diseases, Tenth Revision; mCCI, modified Charlson Comorbidity Index; MDMS, Multidimensional Multimorbidity Score; M3 Index, Multimorbidity Measure Index; MM, Multimorbidity; MWI, Multimorbidity-Weighted Index; PC, Primary Care; Pra tool, Probability of repeated admission risk prediction tool; QOF, Quality and Outcomes Framework; QoL, Quality of Life; RUB, Resource Utilisation Band; RxRisk-V, A Veterans Association adapted pharmacy-based case-mix instrument; SRH, Self-Rated Health.

Table 2 summarises the 33 instruments that were identified from all the studies. Table 3 provides a summary of multimorbidity instruments and their associations with the outcomes measured from all the included studies.

Table 2

Description of instruments used for measurement of multimorbidity and the data sources and resources required

Category	Instrument	System/Condition based	Weightage; Scoring method	Data sources and resources required
A: Count of individual conditions
A-1	DC	Condition (7–147)	Unweighted; condition count	ATC list of conditions, Elixhauser list of conditions, EMR, GP records, health service database, hospital discharge abstract, insurance claims or questionnaires— telephone, face-to-face, mailed surveys. Participant involvement required.
A-2	CCC91	Condition (6)	Unweighted; based on AHRQ’s clinical classification software and number of conditions for each category	EMR
B: Organ or system-based approaches
B-3	Organ systems with CDC	Organ system (17)	Unweighted sum of organ systems	EMR
B-4	CIRS127 128	Body systems (13)	1–5 (based on severity of the condition); different weightage for diseases	EMR
C: Weighted indices
C-5	ACE129	Condition (27)	1–3 (based on severity of most severe condition); highest score of single item	Insurance claims’ database
C-6	Cambridge MM Score97	Condition (20)	Weighted based on three different outcomes—primary care consultation, unplanned admission and mortality	EMR linked to mortality, hospital admission and socioeconomic deprivation
C-7	CCI20	Condition (19)	1–6 (based on impact on 1-year mortality (RR)—original); sum of weighted conditions	Administrative database, EMR, medical chart review, or interviews or postal questionnaire where participant involvement is required
C-8	CLS57	Condition (98)	Based on impact for mortality (HR); sum of beta coefficients of each category	Linked patients' records of all primary care events, hospital admissions and causes of death.
C-9	DBIS130 131	Conditions (25–28)	Weighted according to the degree in which each condition interferes with daily activities	Patient involvement in the questionnaire is required
C-10	EI (original and modified)132	Condition (21–31)	Based on impact on in-hospital mortality; summing of beta coefficients	Insurance claims' or medical services database
C-11	ERA56	Condition (6–9)	Weighted (based on impact on future hospitalisation); sum of weighted regression coefficients	EMR and administrative database
C-12	HCC133	Condition (70)	Based on Medicare capitation payments for health expenditure; more severe manifestations of a condition dominating (and zeroing out the effect of) less serious ones. Other diseases are summed additively.	EMR and HCC software licensing and fees
C-13	M3 Index109	Condition (55)	Weighted based on 1-year mortality; summing of beta coefficients	Linked patients' records
C-14	MDMS24	Condition (7 chronic conditions, 2 health behaviours for first dimension and 5 symptoms for second dimension)	Weighted but not based on any specific outcome; sum of the value for the weighted absolute contributions of each of the dimensions.	Standardised medical evaluation (interviewer-administered); participant involvement is required
C-15	MM weighted by ADL Scale134	Condition (19)	Weighted based on OARS functional status scale measuring ADL; sum of weighted conditions	Face-to-face or telephone interviews where participant involvement is required
C-16	MM weighted by HUI135	Condition (19)	Weighted based on correlation with health utility index; sum of weighted conditions	Face-to-face or telephone interviews where participant involvement is required
C-17	MM weighted by HUI betas135	Condition (19)	Weighted based on correlation with health utility index and adjusted for age and sex; summing of beta coefficients	Face-to-face or telephone interviews where participant involvement is required
C-18	MWI136	Condition (81)	Weighted based on impact on SF-36 physical functioning scale; sum of weights	Interviewer-administered or mail questionnaire where participant involvement is required
C-19	QOF standard (weighted)50	Condition (14)	0–6, based on impact on 1-year mortality (RR); sum of weighted conditions	EMR
C-20	QOF extended (weighted)50	Condition (9)	1–3, based on impact on 1-year mortality (RR); sum of weighted conditions	EMR
C-21	Severity Burden Score130	Condition (21)	Sum of weights of diseases by the level of interference for each condition	Interviewer-administered structured questionnaire by nurses where participant involvement is required
D: Other approaches (D1=Case mix, D2=Pharmaceutical-based)
D1-22	ACG137 138	Condition (93 mutually exclusive ACGs. Some are modified to 68 ACGs)	Incorporated into ACGs based on impact on resource use (proprietary); variable	EMR and ACG software licensing and fees
D1-23	ADG137 138	Condition (32 groups)	Based on duration, severity, diagnostic certainty, aetiology and need for specialty care; variable	EMR and ACG software licensing and fees
D1-24	CRG26	NA; diagnostic categories derived from organ systems or clinical category (37)	Pre-formulated based on the 3M clinical risk groups and consists of 9 core health ranks	EMR—inpatient and outpatient and 3M Clinical Risk Grouping software V.1.6 and service fees
D1-25	Adjusted Morbidity Groups (GMA)139	NA; mutually exclusive categories (31)	Based on multimorbidity and levels of patient complexity	Registry data
D1-26	HM21	Condition (NS)	Only MN tier 4+MN tier 3 with ERA>10; variable	EMR, HCC software licensing, fees and administrative data
D1-27	HSMI78	Condition (73 chronic and acute conditions)	Based on yearly healthcare costs directly derived from primary care setting; sum of regression coefficients (range from −0.06 to 1.04)	EMR
D1-28	Minnesota Tiering137 138	Condition (NS)	Grouping patients into 'complexity tiers' based on the number of major condition categories; condition count	EMR or administrative data and MN Tiering software licensing and fees
D1-29	Resource Utilisation Band137 138	Condition (six mutually exclusive bands)	Based on ACG algorithm on impact on resource use (proprietary); variable	EMR and ACG software licensing and fees
D2-30	CDS33	Condition (17)	Weighted 1–5; sum of weights based on pharmacological database	Prescription drug database
D2-31	Drug Count	NA; variable. Some may be based on pharmacologic-therapeutic classification system	Weighted; medication count	Self-reported questionnaire where participant involvement is required
D2-32	Modified Pra tool using RxRisk-V34 119 140	NA; Pra tool+RxRisk V	Weighted due to RxRisk-V; 4 categories	GP medical record+linked pharmacy claims database
D2-33	RxRisk-V34	NA; WHO-ATC classification system	Weighted according to the diagnostic group of drugs to predict future healthcare costs; sum of weights	GP medical record+linked pharmacy claims database

ACE, Adult Comorbidity Evaluation; ACG, Adjusted Clinical Groups; ADG, Aggregated Diagnosis Groups; ADL, Activities of Daily Living; ATC, Anatomical Therapeutic Chemical; CCC, Chronic Condition Count; CCI, Charlson Comorbidity Index; CDC, Chronic Disease Count; CDS, Chronic Disease Score; CGI-S, Clinical Global Impression-Severity Scale; CIRS, Cumulative Illness Rating Scale; CLS, Comorbidity Linked Score; CRG, Clinical Risk Groups; DBIS, Disease Burden Impact Scale; DC, Disease Count; EDC, Expanded Diagnosis Clusters; EI, Elixhauser Index; EMR, Electronic Medical Records; ERA, Elder Risk Assessment; GP, General Practitioner; HCC, Hierarchical Condition Categories; HM, hybrid model (MN Tier+ERA); HSMI, Health Search Morbidity Index; HUI, Health Utility Index; mCCI, modified Charlson Comorbidity Index; MDMS, Multidimensional Multimorbidity Score; M3 Index, Multimorbidity Measure Index; MM, Multimorbidity; MWI, Multimorbidity-Weighted Index; OARS, Older Americans Resources and Services; Pra tool, Probability of repeated admission risk prediction tool; QOF, Quality and Outcomes Framework; RxRisk-V, A Veterans Association adapted pharmacy-based case-mix instrument; SF-36, 36-item Short Form Survey.

Table 3

Summary of multimorbidity instruments and their associations with outcomes measured from all the included studies

Multimorbidity measures	Association between outcomes and multimorbidity
	Evidence of an association	No evidence of an association
A=Count of individual conditions
DC (many different groupings ranging from 764 76 113 to 14766 conditions and some are further categorised21)	ADL limitations,43 activity limitations,52 affective well-being,126 cognitive function,82 continuity of care (3 years),105 deficits of perceived general health,104 depression,92 depressive symptoms,67 104 disability,17 65 84 85 emergency hospital admission (2 years),120 fall-related injuries,71 fall risk,76 frequency of contacts with physicians (1 year),116 functional capacity,93 functional decline (2 years),120 functional Impairment,111 functional limitations,74 future physical functioning,28 general health,52 healthcare costs,23 91 health-related quality of life,68 75 81 90 115 hospitalisation (3 years),18 hospital admissions (1 year),95 96 99 hospital outpatient visits (1 year),114 hospitalisation/emergency department visits,114 life satisfaction,22 mental distress,52 mortality (1 year),99 (3 years),18 48 (5 years),110 (10 years),28 number of contacts with general practice (1 year),117 number of medications prescribed (1 year),117 number of mentally unhealthy days,43 44 number of physically unhealthy days,43 44 number of different ambulatory physicians contacted (1 year),116 number of primary care consultations (1 year),48 (3 years),48 number of referrals (1 year),117 outpatient/Inpatient service use,53 physical distress,52 physical function,55 prescription costs,70 perceived health status,22 presence of mental health disorder,66 primary care consultations (1 year period),105 (3 years),105 primary care experience—self-reported,83 primary healthcare cost,47 primary healthcare use,89 potentially preventable unplanned admission (1-year period),95 quality-adjusted life years,72 73 quality of life,93 98 101 102 self-rated health,118 self-reported falls (12 months),108 symptom burden,101 self-rated Health,64 85 self-perceived health,93 total number of consultation,42 total health care costs42	Functional decline,18 quality of cardiovascular preventive care,27 quality of preventive care27
CCC	Healthcare costs,21 hospital admissions (1 year),21 number of emergency department visits (1 year),21 readmission within 30 days (1 year)21
B=Organ or system-based approaches
Organ systems with CDC	Presence of depressive or anxiety disorder19
CIRS	Disability,17 frailty,17 healthcare utilisation,54 hospitalisation (3 years),18 mortality18	Functional decline18
C=Weighted indices
ACE	Healthcare expenditure23
Cambridge MM Score	Mortality,97 primary care consultation,97 unplanned admission97
CCI	Ambulatory care-sensitive hospitalisations (acute and chronic),106 disability,17 emergency department visits (1 year),21 emergency hospital admission (2 years),119 frailty,17 functional decline (2 years),119 future physical functioning,28 healthcare expenditure,21 hip fractures,103 hospitalisation (1 year),21 64 80 99 109 hospitalisation (3 years),18 mortality (1 year),50 63 99 109 mortality (3 years),18 (5 years),51 63 (10 years),51 63 (15, 20, 25 years),51 number of primary care consultations (3 years),48 number of primary care physician visits (1 year),107 number of specialist visits (1 year),107 potentially preventable unplanned admission (1 year),96 presence of critical illness,46 primary healthcare cost,47 mortality (1 year),57 58 (3 years),48 59 (5 years),58 60 62 (7 years),86 (10 years),28 readmission within 30 days (1 year),21 successful ageing61	Functional decline,18 primary care visits,69 quality of cardiovascular preventive care,27 quality of preventive care27
CLS	Mortality (1 year)57
DBIS	Quality of life98
EI (original and modified)	Hospitalisation (1 year),99 109 mortality (1 year)57 99 109
ERA	Healthcare expenditure,21 mortality (2 years),113 number of days hospitalised (1 year),56 number of emergency department visits (1 year),21 56 number of hospital admissions (1 year),21 56 number of hospital visits (1 year),21 56 nursing home placement (2 years),113 presence of critical illness,46 readmission within 30 days (1 year)21
HCC	Hospitalisation (1 year),21 ED visits (1 year),21 readmission within 30 days (1 year),21 healthcare expenditure (1 year)21
M3 Index	Hospitalisation (1 year),109 mortality (1 year)109
MDMS	Sickness absence episodes taken in 2 years (male)24	Sickness absence episodes taken in 2 years (female)24
MM weighted by ADL scale	Life satisfaction,22 perceived health status22
MM weighted by HUI	Life satisfaction,22 perceived health status22
MM weighted by HUI betas	Life satisfaction,22 perceived health status22
MWI	ADL limitations,121 IADL limitations,121 mortality (10 years),28 cognitive performance,121 future physical functioning,28 125 grip strength,121 health-related quality of life,123 mortality,125 subjective physical functioning,121 suicide mortality123	Gait speed28
QOF (standard)	Mortality (1 year)50
QOF (extended)	Mortality (1 year)50
Severity Burden Score	Mental component score (SF-12)94
D=Other approaches (D-1=Case Mix, D2=Pharmaceutical-based)
ACG	Hospitalisation (1 year),80 mortality (3 years),48 number of primary care consultations (3 years),48 primary healthcare cost,47 readmission within 30 days (1 year)21
ADG	Hospitalisation (1 year),107 number of primary care physician visits (1 year),107 number of specialist visits (1 year)107
CRG	Healthcare expenditure,49 HRQoL using EQ-5D-3L87
Adjusted Morbidity Groups (GMA)	Use of healthcare resources88
HM	Emergency department visits (1 year),21 healthcare expenditure,21 hospitalisation (1 year),21 readmission within 30 days (1 year)21
HSMI	Healthcare cost (primary care)78
Minnesota Tiering	Emergency department visits (1 year),21 114 healthcare expenditure,21 hospitalisation (1 year),21 114 readmission within 30 days (1 year)21
Resource Utilisation Band	Fee-for-service expenditures,77 primary healthcare cost,47 mortality (3 years),48 number of primary care consultations (3 years),48 registered active listing in primary care,100 registered active listing in all healthcare100
CDS	Hospitalisation (1 year),99 mortality (1 year)99
Drug Count	Emergency hospital admission (2 years),120 functional decline (2 years),120 hospitalisation (1 year),99 mortality (1 year),99 (3 years),48 number of primary care consultations (3 years)48
Pra tool Modified using RxRisk-V	Emergency hospital admission (1 year)119
RxRisk-V	Emergency hospital admission (2 years),120 functional decline (2 years)120

ACE-27, Adult Comorbidity Evaluation; ACG, Adjusted Clinical Groups; ADG, Aggregated Diagnosis Groups; ADL, Activities of Daily Living; CCC, Chronic Condition Count; CCI, Charlson Comorbidity Index; CDC, Chronic Disease Count; CDS, Chronic Disease Score; CIRS, Cumulative Illness Rating Scale; CLS, Comorbidity Linked Score; CRG, Clinical Risk Groups; DBIS, Disease Burden Impact Scale; DC, Disease Count; EI, Elixhauser Index; ERA, Elder Risk Assessment; HCC, Hierarchical Condition Categories; HM, Hybrid Model (MN Tier+ERA); HRQoL, health-related quality of life; HSMI, Health Search Morbidity Index; HUI, Health Utility Index; MDMS, Multidimensional Multimorbidity Score; M3 Index, Multimorbidity Measure (M3) Index; MM, Multimorbidity; MWI, Multimorbidity-Weighted Index; Pra tool, Probability of repeated admission risk prediction tool; QOF, Quality and Outcomes Framework; RxRisk-V, A Veterans Association adapted pharmacy-based case-mix instrument; SF-12, Short Form-12.

Simple counts of individual conditions

Disease Count was based on the total number of all the conditions an individual had, usually from a prespecified list of chronic conditions. It was used in 59 out of the 96 studies (61.5%). Disease Count was reported to be associated with activity limitations, continuity of care, disability, healthcare cost, healthcare utilisation, medications, mental disorders, mortality, general health, physical function, quality of life and self-rated health (table 3).

Organ or system-based approaches

There were two instruments in this category. They were Cumulative Illness Rating Scale (CIRS)17 18 and Organ Systems with Chronic Disease Count (Organ-CDC).19

Weighted indices

There were 17 unique weighted instruments found in the included studies. The original CCI with its different modifications was the most frequently used instrument and was used in 29 studies. The CCI was based on Disease Count, but the 17 conditions were weighted originally based on their impact on 1-year mortality.20 The final score was derived by the summation of all the weighted conditions. There were many variations and modifications of the score including the addition of psychosocial factors. The CCI instrument was found to be associated with multiple outcomes other than 1-year mortality.

Most of the other weighted index instruments were novel, like the Multimorbidity-Weighted Index (MWI), in which the investigators built multivariable prognostic models from a set of potential predictor conditions and weighted the conditions based on an outcome of clinical interest. The most common outcomes chosen were mortality and physical function. Other outcomes included health expenditure,21 health utility index22 and severity of the most severe condition.23 The Multidimensional Multimorbidity Score (MDMS)24 was unique as it was weighted based on health behaviours and patient symptoms and not based on any specific outcome.

Other approaches to measuring multimorbidity

Other approaches included case-mix and pharmaceutical-based instruments. For case-mix approach, the ACG and Resource Utilisation Band were the most commonly used instruments.25 Most of the case-mix instruments required proprietary software licenses from the USA and obtained data from electronic medical records or administrative data. The Clinical Risk Groups instrument was similar but took into account the severity of individual conditions.26

The second group of instruments in this category was related to pharmaceutical data. The most frequent type was the unweighted Drug Count. The other three (Chronic Disease Score, A Veterans Association adapted pharmacy-based case-mix instrument like RxRisk-V and modified Probability of repeated admission risk prediction tool using RxRisk-V) were all weighted indices. Except for the Drug Count that was based on a self-report questionnaire, the rest required a prescription drug database to obtain the data.

Outcomes

We classified the 150 outcomes into 17 categories as reported in the core outcomes set of multimorbidity research (COSmm).10 The most commonly reported outcomes were healthcare use (n=45), mortality (n=18), health-related quality of life (n=18) and physical function (n=13). The different studies unanimously showed that higher levels of multimorbidity was associated with higher healthcare use and mortality, lower health-related quality of life and poorer physical function. Seven outcomes in the COSmm were not found in all the 96 studies. These were treatment burden, self-management behaviour, self-efficacy, adherence, communications, shared decision-making and prioritisation. There were 19 outcomes that were not described in the COSmm. These included cognitive function, risk of suicide, frailty and falls. The outcomes not found to have any association with the instruments for measuring the level of multimorbidity were preventive care,27 sickness absence episodes (female)24 and gait speed.28

Discussion

Summary of findings

Thirty-three unique instruments for measuring the level of multimorbidity were identified and categorised according to the classification by Sarfati.8 The most commonly used instrument was ‘Disease Count’. It was also the only instrument that was associated with the three essential outcomes from the core outcomes set of multimorbidity research (COSmm),10 that is, quality of life, mental health and mortality.

Comparison with previous research

Although the most common instrument identified in this systematic review was similar to that of Huntley et al,12 several instruments including Duke Severity of Illness Checklist (DUSOI) and Functional Comorbidity Index identified in their article were not found in this systematic review. The possible reasons for not finding these instruments in this review could be due to the lack of interest in the instrument by the research community in recent years (to our knowledge, the last publication using DUSOI was in 2004),29 or the exclusion of studies specifying an index condition.

Advantages and disadvantages of selected instruments

Disease Count

The advantage of using ‘Disease Count’ is its simplicity and the ease of data ascertainment with minimal resources required. However, using ‘Disease Count’ does not consider the severity of each condition where the complexity of multimorbidity may not be properly addressed.30 The other disadvantage noted was the lack of transparency in the operational definition of multimorbidity, especially regarding the list of conditions considered for multimorbidity and the cut-points used. Despite its simplicity, the level of multimorbidity measured using ‘Disease Count’ was the only instrument that was found to be associated with the three essential core outcomes (quality of life, mental health and mortality).

Weighted indices

The common weighted indices identified in this systematic review were CCI, Elders Risk Assessment (ERA), Elixhauser Index (EI) and MWI. These weighted indices were often used in prognostic models to build complex multivariable regression models in which the weights were calculated from hazard ratios, odds ratios or regression coefficients.31

The advantage of these weighted indices is that the weights allow the adaptation of an index to a specific outcome. An investigator could recalibrate the correct weight by creating a prognostic model to produce a contextualised instrument for a different setting. Prognostic models can provide clinically relevant risk stratification and help to allocate resources.32 The disadvantage of such indices is that calculated weights are greatly influenced by the population, outcomes used, and the instrument’s original conception and purpose, hampering the ability to compare across studies.

Case-mix

The ACG system has a good track record in the USA and several other countries, especially for measuring the outcomes of healthcare utilisation. However, the instrument is proprietary, and the exact algorithm of the instrument is not open to the public and may not be suitable in certain settings. The Clinical Risk Group (CRG) system has a good track record in Spain. It measures the severity of each condition and its algorithm is fully transparent. The common disadvantage of both systems is the financial costs involved in obtaining the license.

Pharmaceutical-based instruments

Medication-based indices include versions of the Chronic Disease Score,33 which later became known as the RxRisk,34 and its adaptation for use in the veteran population, the RxRisk-V.35 Like the Disease Count, its main advantage is the ease of use with minimal resources required. However, many studies were not transparent regarding which type of drugs were included.

Data sources

Data sources used by these instruments included medical record information, patient self-report, clinical judgement and large administrative databases. Each data source has its inherent advantages and disadvantages. For patient self-report, patients with cognitive impairment may under-report symptoms and may be seen less frequently by their physicians, resulting in an under-recognition or undertreatment of conditions.36 It has also been shown that health administrative data based on billing system underestimated the prevalence of many chronic conditions.37

The available data in a particular setting may strongly influence the ultimate instrument chosen for multimorbidity research. As there is currently no consensus on the gold standard for sources of data, it is difficult to assess which data source was superior from this review.

There were 17 multimorbidity outcomes identified by a Delphi process involving a panel of international experts in multimorbidity intervention studies.10 However, only 10 out of the 17 outcomes were reported in the 96 studies identified in this systematic review. The most common outcome that was investigated was healthcare use. The seven missing outcomes belong to ‘patient-reported impact and behaviours’ and ‘consultation-related’ outcome groups, most likely indicating that there is a dearth of multimorbidity studies looking at these two groups of outcomes measures.

Clinical implications

Ideally, a single instrument measuring the level of multimorbidity should be able to predict a variety of relevant outcomes. However, Byles et al38 reported that a single instrument could not be used to predict different outcomes, in different patient groups and settings, unless different weights were assigned to these factors in calculating a score. Such multiple-scoring instruments may be the way forward for validation of prognostic models for different outcomes and different populations with established multimorbidity instruments. For example, depending on the outcome, study population and setting, the choice of conditions included in the multiple-scoring instrument should include those with a high prevalence in that study population and the weights should be determined by their significant impact (ie, outcome) on the affected population.

For pragmatic reasons, the final selection of the conditions to be included in such a multiple-scoring instrument may still have to take into account the availability of relevant and reliable data. A certain degree of reductionism will also have to be accepted because a single instrument will not be able to encompass all the nuances of the different interactions of chronic conditions on an individual living in his/her unique milieu. We recommend that researchers perform validation studies using the instruments listed in this systematic review to adjust the weights according to the specific outcome of interest for the study population relevant to their setting.

Strengths and limitations of the study

The main strengths of this systematic review were the involvement of a health science librarian in our search strategy, a published protocol, adherence to the protocol without major changes during the systematic review process,39 and the critical appraisal of all the primary studies with a risk of bias assessment tool.

The systematic review had several limitations. We excluded grey literature and included only studies that were published in the English language. We also did not contact authors directly for a suggestion of studies, nor identified a list of instruments from the preliminary search and then performed an additional search using the same databases.40 Additionally, this systematic review did not review the validity and reliability of all the instruments as it was beyond the scope of the intended work. We have, however, included the references of the original articles or validation studies in table 2 for each of the instrument where available. Finally, this review specifically aimed to look at the association of the level of multimorbidity as the main independent variable and excluded the level of multimorbidity as a mediating, confounding or effect-modifying variable. This strict criterion excluded 17 studies (figure 1) as a result. Excluding these 17 studies did not alter the findings as the instruments used in all the 17 studies were Disease Count (n=9), CIRS (n=3), CCI (n=3), EI (n=1) and Aggregated Diagnosis Groups (n=1) where no new instruments were identified.

Conclusions

In this systematic review, we found 33 instruments for measuring the level of multimorbidity in community-dwelling individuals that predict or explore the association of multimorbidity with at least one specified outcome. Disease Count and weighted indices like the CCI, the ERA and EI were commonly used for measuring the level of multimorbidity. Other approaches to measuring the level of multimorbidity included case-mix or pharmaceutical-based instruments.

We found continuing interest in measuring the level of multimorbidity with Disease Count and Drug Count. There has also been a rise in the development of novel weighted indices using prognostic models or validation of existing well-established instruments like the CCI over the last few years. There is currently an absence of a gold standard for where to obtain chronic disease information. The most suitable instrument will depend on the specified outcome of interest, the study population and the type of data and resources available.

Finally, there is still much work to improve on the body of knowledge of multimorbidity when most investigators in the last decade measured multimorbidity without including some of the important outcome measures of multimorbidity. We also suggest that a clear description of the instruments is required in the publication of multimorbidity studies to counter the frequent lack of information currently seen so as to contribute to robust multimorbidity research in future.