Measuring multimorbidity beyond counting diseases: systematic review of community and population studies and guide to index choice

Abstract Objectives To identify and summarise existing indices for measuring multimorbidity beyond disease counts, to establish which indices include mental health comorbidities or outcomes, and to develop recommendations based on applicability, performance, and usage. Design Systematic review. Data sources Seven medical research databases (Medline, Web of Science Core Collection, Cochrane Library, Embase, PsycINFO, Scopus, and CINAHL Plus) from inception to October 2018 and bibliographies and citations of relevant papers. Searches were limited to English language publications. Eligibility criteria for study selection Original articles describing a new multimorbidity index including more information than disease counts and not focusing on comorbidity associated with one specific disease. Studies were of adults based in the community or at population level. Results Among 7128 search results, 5560 unique titles were identified. After screening against eligibility criteria the review finally included 35 papers. As index components, 25 indices used conditions (weighted or in combination with other parameters), five used diagnostic categories, four used drug use, and one used physiological measures. Predicted outcomes included mortality (18 indices), healthcare use or costs (13), hospital admission (13), and health related quality of life (7). 29 indices considered some aspect of mental health, with most including it as a comorbidity. 12 indices are recommended for use. Conclusions 35 multimorbidity indices are available, with differing components and outcomes. Researchers and clinicians should examine existing indices for suitability before creating new ones. Systematic review registration PROSPERO CRD42017074211.

4. If the index uses a list of diseases, does it describe the selection process for this list?
Statistical methods (maximum ++) 5. Are the statistical methods used clearly described?
6. Is a sample size calculation included?
Validity (maximum +++) 7. When outcomes were included, were outcome raters blinded to the variables used in the index? 8. Was there a test for inter-rater or test-retest reliability of the index? 9. Was the index validated, either in this paper or elsewhere?
Funding source (maximum ++) 10.  Addition of various variables (e.g. count of physician visits and unique prescriptions). No weighting Not applicable Summary statistics for specific expenditure-related outcomes given

Byles 2005 [22]
Weights generated using hazard ratios for mortality and odds ratios for admission. Weights with and without health self-rating Lists of odds and hazard ratios given but no coefficients or intercepts 51% admitted to hospital (485 patients), 7% died (n=59 in derivation sample, n=29 validation)

Bayliss 2005 [23] Disease Burden Morbidity Assessment
Conditions weighted by selfreported severity Comorbidities include "senile and presenile organic psychotic conditions." Diagnoses from claims records Does not specify subtypes of hospital admissions (e.g. psychiatric) Not separately examined

Stanley 2017 [49]
Comorbidity variables: alcohol abuse, anxiety and behavioural disorders, dementia, drug abuse, major psychiatric disorder, mental and behavioural disorders due to brain damage, mental retardation. Diagnoses from routinely collected healthcare data Comorbidities include dementia (self-report), depression (PHQ-9) and mental aspects of HRQL (SF-12) None Alzheimer's and other dementias included as comorbidity (identified by prescription of drugs for dementia) Does not specify whether outcome includes psychiatric hospitalisation MDBI reported as 100% sensitive and 100% specific for Alzheimer's and other dementias when measured against medical records Lee 2006 [22] Two functional questions (difficulty managing finances and personal hygiene) refer to "health or memory problems" None Difficulty bathing and managing finances each assigned two points in overall model

Byles 2005 [37]
Depression and forgetfulness included as comorbidities. Self-reported with severity rating Mental Component Score (MCS) of SF-36 Increasing scores on all versions of the multimorbidity index were associated with worse scores on the SF-36 MCS

Bayliss 2005 [24]
None Depression screen from Behavioural Risk Factor Surveillance System Being less depressed was significantly negatively correlated with self-reported disease burden (P<0.001) and number of conditions (P=0.002) but not with Charlson index or RxRisk score

Selim 2004 [25]
Comorbidity variables include self-reported schizophrenia, depression, bipolar disorder, anxiety disorder, post-traumatic stress disorder, alcohol abuse Mental health outpatient visits from administrative data, SF-36 Mental disorders on comorbidity index correlated better with the mental than physical scale of the SF-36. Comorbidity index including mental disorders was not significantly associated with mortality

Pope 2004 [26]
Comorbidity variables: Drug or alcohol psychosis, drug or alcohol dependence, schizophrenia, major depressive, bipolar, and paranoid disorders. Diagnoses taken from claims data None Not separately examined

Crabtree 2000 [41]
Anxiety/depression (self-report) included as one comorbidity . In external validation on 212 acute geriatric inpatients, mean (SD) age 81 (7.3) years, 62% Self-rated health. Reported as a statistically significant association between MDBI scores and decreasing self-rated health (P<0.001) [47] Publication and name of index

Citations since publication3
Citations per year Validation and/or comparison Predictive accuracy measurement in original paper

Performance (original outcomes) Additional outcomes tested in external validation
Later validation papers by different authors compared to CIRS-G, Charlson and condition count [46] and tested association with mortality and selfrated health [47] female, prediction of three-month mortality or readmission: MDBI=2.99 (0.99 to 9.03), CIRS-G=1. 2 (1.1 to 1.3) Regular updates [75] Predicting healthcare expenditure (same as original) Updated annually with amended ICD-10 mappings and software Regular reports; 2018 report lists detailed predictive expenditure accuracy for combinations of conditions of CMS-HCC. [69] Ratio of one-year predicted to actual expenditure=1.00 (but gives caveat that this is an average in a very large group)  Romano 1993 [88] (Dartmouth-Manitoba)

Publication and name of index
Predicting mortality (same as original) Adapted for use with administrative data using ICD-9-CM codes; broader definitions than Deyo Compared to other scales in original Farley paper (see eTable 9) [21] D'Hoore 1993 [89] Predicting mortality (same as original) Uses first three digits of ICD-9 In this paper, C-statistic for in-hospital mortality=0. 83 Ghali 1996 [90] Predicting mortality (same as original) Assigns new weights to Deyo's system, according to study-specific mortality. Includes only five conditions In this paper, C-statistic for in-hospital mortality=0.74, original Charlson index=0.70 Quan 2005 [91] Predicting mortality (same as original) Adapted for ICD-10, includes 12 conditions. Later revision assigns new weights to conditions [92] In this paper, C-statistics discriminating inhospital mortality in one cohort: Quan=0. 83

What is a systematic review?
A systematic review is a common type of research. It's a way of finding lots of published research articles and summarising them together.

What do we know about this topic already?
It's common for people to have two or more chronic conditions at once. This is often called multimorbidity. Researchers and clinicians measure multimorbidity in many different ways.
What questions does this review ask?
1. What methods exist for measuring multimorbidity?
2. How good are they?
3. Do they include mental health?

How was the search carried out?
We decided in advance which topics to include. In October 2018, we searched seven online medical research databases. Two researchers separately checked 5,560 article titles. We discarded irrelevant articles.

What did the review find?
We ended up with 35 papers, each describing a way to measure multimorbidity. Most of them combined the number of chronic conditions with other things like age. Some counted people's prescribed drugs and others included medical test results.
Most of the tools aimed to predict health in some way. For example, 18 of them looked at death rates, 13 at hospitalisations and six at quality of life.
Nearly all the papers considered mental health, with 18 counting it as part of multimorbidity.
Eleven measures aimed to predict some aspect of mental health.
Only one paper mentioned including patients in their research design.
How good were the papers?
We graded each paper according to set standards. Six were high quality, 22 were satisfactory and seven were low quality. Three of the papers didn't mention who funded their research.
Four were funded by drug companiesthis might make them biased. The other 28 papers had no funding bias.

What does this mean for patients?
These tools might help make other research more relevant to people with multiple conditions.
For example, drug trials use very healthy people who are not like most patients. Researchers could use these tools to account for multimorbidity in a wider range of people.
Healthcare officials can also use the tools to predict how services will be used and plan how to fund them.