Multimorbidity

Definition

Multimorbidity is the simultaneous co-occurrence of two or more chronic diseases in the same individual, with no single condition designated as the index disorder. It is distinguished from comorbidity (Feinstein 1970), which refers to additional diseases relative to a specified index condition and is organized around disease-specific prognosis. The multimorbidity framework shifts attention from single-disease prognosis to the whole person experiencing multiple overlapping conditions. Multimorbidity is the dominant health pattern of older adults in high-income countries, affecting $55$ – $98\%$ of persons 60 and over depending on definition and setting.

Key Ideas

Comorbidity vs. multimorbidity: Comorbidity asks how additional diseases affect the prognosis of an index condition. Multimorbidity removes the index condition entirely, treating all co-occurring diseases as jointly relevant and the person — not any single disease — as the unit of analysis.
Three operational definitions:
1. Simple count (2+ or 3+ concurrent chronic diseases): most used in epidemiological studies; parsimony allows additive modeling and preserves power for rare conditions; disadvantage: all diseases scored equally regardless of severity.
2. Cumulative weighted indices (Charlson Comorbidity Index, Index of Co-Existing Disease (ICED), Cumulative Illness Rating Scale (CIRS)): incorporate both count and severity; preferred in clinical studies identifying patients at risk for adverse outcomes.
3. Diseases plus functional/cognitive limitations: used when the aim is estimating complex care needs involving medical, cognitive, and social service providers.
Prevalence: $20$ – $30\%$ across all ages; $55$ – $98\%$ in persons $60+$ across studies (Marengoni et al. 2011). The wide range reflects variation in: number of conditions evaluated, age structure of study population, setting (community $\approx 55$ – $72\%$ , institutions $\approx 82\%$ ), ascertainment method (self-report vs. general practitioner (GP) records vs. clinical examination), and the threshold used ( $2$ , $3$ , or $4+$ diseases).
Risk factors (known): Increasing age, prior morbidity, lower education, lower socioeconomic status (SES). Protective: large social network. No studies have evaluated genetic background, biological causes, or lifestyle factors as determinants.
Coincidental vs. associative multimorbidity: Diseases may co-occur at the rate predicted by their independent prevalences (coincidental) or at higher-than-expected rates. Associative multimorbidity arises when diseases share risk factors or when the risk factor for disease X is correlated with the risk factor for disease Y. Common underlying mechanisms include chronic inflammation, oxidative stress, and metabolic syndrome — biological processes potentially linking seemingly unrelated diseases.

How It Works

Functional decline

More diseases $\to$ higher risk of disability and functional decline. The relationship is nearly universal across study designs and populations. Clusters of conditions may have synergistic (not merely additive) effects: specific disease combinations are associated with difficulty in particular functional tasks, suggesting specificity in the disease-to-disability pathway beyond simple accumulation.

Mortality — the count paradox

Disease count alone is an unreliable mortality predictor. Studies that find increasing multimorbidity raises mortality (Menotti 2001, Deeg 2002, Byles 2005) are not contradicted by studies that do not (Marengoni 2009, Landi 2010); the difference lies in whether disability status is controlled. Landi et al. (2010) found multimorbidity predicted 4-year mortality only when associated with disability. This implies that it is the functional impact of co-occurring diseases — not their count — that drives mortality risk. Severity, duration, and disease-disability interactions matter more than counts.

Quality of life

Consistently impaired. Physical function (Short Form 36 (SF-36)/Short Form 12 (SF-12) physical component) deteriorates with increasing disease count across studies. Mental health/depression also increases, though a simple count without severity weighting may miss this association. Severity-adjusted multimorbidity better captures psychological distress.

Health care utilization

Linear relationship: each additional chronic condition increases prescriptions, referrals, hospitalizations, and expenditures. In US Medicare ( $n = 1.2$ million, Wolff 2002): $65\%$ of $65+$ have multimorbidity; preventable complications rise with disease count. Schneider et al. (2009, $n = 2.2$ million): per-beneficiary Medicare payments scale with condition count.

The care guideline paradox

Clinical practice guidelines are designed for single-disease patients. Boyd et al. (2005) showed that simultaneously applying disease-specific guidelines for the most common conditions in an elderly multimorbid patient could produce adverse drug–disease interactions — not because any single guideline is wrong, but because guidelines are optimized for the single-disease case and their joint application has not been tested. Min et al. (2007) found an apparent paradox: quality-of-care indicators are satisfied more often in patients with more conditions ( $47\%$ with $0$ conditions $\to$ $59\%$ with $5$ – $6$ ) — likely because complex patients interact more with the health care system. This paradox reveals that standard quality metrics built around process compliance for individual diseases are measuring the wrong thing for multimorbid patients.

Why It Matters

Aging populations: Multimorbidity is the modal health state of elderly populations in high-income countries. As the Demographic Transition delivers large cohorts of very old adults, multimorbidity will dominate health care demand and cost.
Healthy Life Expectancy (HALE) gap: Multimorbidity is the primary mechanism by which aging populations experience years of life in less-than-full health. It drives the widening gap between life expectancy and Healthy Life Expectancy — living longer but not healthier. See Morbidity-Mortality Distinction.
Disability Insurance (DI)/disability programs: Musculoskeletal conditions and mental health disorders — the leading Global Burden of Disease (GBD) disabling conditions and leading DI qualifying diagnoses — are the most common multimorbidity-generating conditions at working ages. Multimorbidity among near-retirement-age workers increases DI applications independently of any single diagnosis. See DI Growth Decomposition.
Clinical paradigm failure: Medicine's single-disease model is structurally ill-suited to multimorbidity. Most randomized trials exclude multimorbid patients; most guidelines apply to single conditions; most quality metrics reward single-disease process compliance. This evidence vacuum is one of the largest structural gaps in clinical medicine.
Health care costs: Multimorbidity is the primary driver of high and rising per-beneficiary Medicare spending. The relationship between condition count and expenditure is nearly linear, with no sign of diminishing returns at high disease counts.

Multimorbidity in Serious Mental Illness Populations

Razzano et al. (2015) extend multimorbidity research to adults with serious mental disorders (SMI) — defined by Diagnostic and Statistical Manual of Mental Disorders, 4th Edition (DSM-IV) diagnosis with moderate-to-severe functional impairment, a population substantially overlapping with Supplemental Security Income (SSI) and Social Security Disability Insurance (SSDI) recipients. Using face-to-face National Health Interview Survey (NHIS)/National Health and Nutrition Examination Survey (NHANES) structured interviews of $457$ adults at publicly funded community mental health programs in $4$ U.S. states:

$87\%$ reported $\geq 1$ of 17 assessed physical conditions; $82\%$ reported $\geq 2$ ; $69\%$ $\geq 3$ ; mean $2.8$ conditions per person.
$14$ of $17$ conditions were significantly more prevalent than in the non-institutionalized U.S. adult population. Largest relative excesses: non-viral liver disease $6\times$ ; diabetes and chronic bronchitis $\approx 3\times$ ; viral hepatitis $2\times$ ; hypertension $1.7\times$ .
Top 5: hyperlipidemia ( $45\%$ ), hypertension ( $44\%$ ), asthma ( $28\%$ ), arthritis ( $22\%$ ), diabetes ( $21\%$ ).
Treatment gaps: treatment prevalence below $70\%$ for $8$ of $17$ conditions; particularly low for non-viral liver disease ( $30\%$ ), hepatitis A/B/C ( $33\%$ ), stroke ( $42\%$ ), and cancer ( $50\%$ ).

Within-SMI disparities: Controlling for age, study site, Medicaid status, diagnosis, and health self-efficacy, racial/ethnic minority group members were $\approx 2\times$ as likely as Caucasians to be diagnosed with hypertension and diabetes — a disparity-within-a-disparity result embedded in a population already characterized by dramatically elevated multimorbidity. Women were $\approx 2\times$ as likely as men to be diagnosed with diabetes.

Schizophrenia and under-diagnosis: Individuals with schizophrenia were approximately half as likely to report hypertension and arthritis as those with other psychiatric diagnoses. The authors attribute this to probable under-diagnosis via cognitive impairment effects on self-reporting, not biological protection — consistent with prior literature finding paradoxically low symptom self-report in schizophrenia.

The SMI multimorbidity burden is clinically distinct from the aging-population multimorbidity pattern documented by Marengoni et al. (2011): it is younger-onset, more concentrated in metabolic and respiratory conditions, and compounded by mental health system segregation from physical health care — creating structural barriers to integrated management. The SSI/SSDI program context for this population is addressed in the respective program articles.

Multimorbidity in the DI/SSI Beneficiary Population

Walker and Roessel (2019) extend multimorbidity analysis to the Social Security Administration (SSA) disability-program population using the 2015 National Beneficiary Survey (NBS; N=4,062; weighted: 12.9M DI and SSI beneficiaries), matched to SSA administrative records. This analysis is particularly relevant because disability-program recipients are a high-multimorbidity population by construction — they must demonstrate disabling impairments to qualify — yet SSA administrative records are structurally limited to two impairments per person (primary + secondary).

Prevalence: Two-thirds (67%) of beneficiaries report conditions in two or more impairment categories in the survey; only ~49% have two or more impairment categories in administrative records. The survey average is 2.1 impairment categories per person. The most common categories are musculoskeletal disorders (42%), mental disorders (35%), circulatory diseases (20.6%), and endocrine/nutritional disorders (17%).

Program-type heterogeneity: DI-only beneficiaries are heavier in musculoskeletal disorders (49% vs. 30% SSI-only), reflecting their older average age; SSI-only recipients are heavier in mental disorders (42% vs. 30%) and intellectual disability (8% vs. 3%), reflecting their younger, more mentally impaired population.

Admin-survey divergence and concurrence: The overall concurrence rate (admin→survey match) is 72%. Physical impairments are underrepresented in admin data relative to survey (musculoskeletal: 32% admin vs. 42% survey), because case development stops once one impairment justifies an allowance and secondary conditions may go unrecorded. Mental and intellectual impairments are overrepresented in admin relative to survey — stigma-based underreporting suppresses self-disclosure in survey settings. The forward concurrence ranges from 79% (musculoskeletal, highest) to 29% (intellectual disability, lowest).

Health and functional burden by count: Multimorbidity severity in this population shows a steep gradient: 85% of those with 3+ impairment categories rate their health "fair or worse"; 64% report 2+ activities of daily living (ADL)/instrumental ADL (IADL) difficulties. Work orientation declines monotonically with impairment count, from 65% (0 impairments) to 40% (3+ impairments).

Policy implications: Because admin data are capped at two impairments and systematically biased by impairment type, they are an incomplete predictor of functional outcomes and employment potential. Survey data — or richer case-analysis data from SSA's eCAT system, which allows unlimited impairment documentation — provide a more actionable picture of beneficiary disability burden. This has implications for targeting employment support services to beneficiaries who might benefit most.

Multimorbidity Measurement: The ACG System

Starfield and Kinder (2011) argue that counting chronic conditions — the dominant epidemiological approach — is an inadequate predictor of health resource use and propose the Johns Hopkins University (JHU) Adjusted Clinical Groups (ACG) System as a superior framework.

Architecture

The ACG System classifies all International Classification of Diseases (ICD) diagnoses into 32 Aggregated Diagnosis Groups (ADGs) based on clinical properties: acute vs. recurrent vs. chronic, likelihood of referral to specialist care, need for hospitalization, severity, and psychosocial dimensions. The 32 ADGs are collapsed into 12 Collapsed ADG (CADG) categories. The combination of CADGs a person holds maps them into one of approximately 100 mutually exclusive ACG cells, which function as a case-mix descriptor capturing the type of morbidity burden rather than mere count.

The Type-Over-Count Argument

The core claim is that the number of different types of illness — measured by ADG distribution — is a substantially better predictor of resource use than the raw count of chronic conditions. Evidence:

Resource use paradox (Israel data): The majority of individuals with any single chronic condition have below-average resource use: ~ $80\%$ of hypertensives, ~ $70\%$ with hyperlipidemia, ~ $40\%$ with diabetes, ~ $33\%$ with osteoporosis. This contradicts the intuition that having a chronic condition drives costs; costs are driven by co-occurring types of illness across ADG categories.
Morbidity burden > sum of disease costs (British Columbia): Using BC Canada administrative data, resource use (total costs, hospital costs, ambulatory costs, specialist visits, primary care visits) increases more than linearly with morbidity burden at every stratum of chronic condition count ( $0$ – $4+$ ). Total burden is not reducible to the sum of individual disease-specific costs.
Non-random clustering (Manitoba): ACG morbidity burden scores correlate with SES in expected directions, confirming that morbidity clusters non-randomly in disadvantaged populations — consistent with the associative multimorbidity mechanism documented in epidemiological literature.
Trajectory stability (Taiwan): In Taiwan's National Health Insurance program, approximately $80\%$ of the population remains in the same or adjacent ACG morbidity burden trajectory group across multiple years. This stability validates ACG as a tool for prospective risk stratification and resource allocation.

Policy Applications

As of 2011, the ACG System is used operationally in health systems across the US, Canada, Sweden, Spain, UK, Germany, Israel, South Africa, and Malaysia for: population morbidity profiling; equitable budget allocation; provider performance assessment (British Columbia fraud detection); high-risk patient identification; and monitoring of policy reform impacts (Sweden provider-choice legislation).

Limitations

Starfield and Kinder disclose that JHU holds copyright on ACG software and collects royalties from insurance plans; both authors are JHU faculty. The paper is methodological advocacy drawing on prior validation studies, not an original empirical analysis. Count-based approaches (simple $2+$ threshold, weighted indices like Charlson) remain more tractable for large epidemiological studies; ACG's advantage is primarily in administrative/clinical applications with electronic data.

Open Questions

What biological mechanisms explain disease clustering beyond shared observable risk factors? Inflammation, oxidative stress, and metabolic syndrome are hypothesized; evidence is limited.
Is a simple count ( $2+$ diseases) adequate, or must severity always be incorporated? Count-based prevalence is $55$ – $98\%$ in the elderly, which may be too broad to be clinically actionable.
What care models work for multimorbid patients? "Guided Care" (Boyd 2007) and community-of-practice models (Soubhi 2010) show feasibility but lack controlled-trial evidence of effectiveness.
Can multimorbidity be prevented, or only managed? A life-course perspective — when do the seeds of multimorbidity cluster — is largely absent from the evidence base.
Does multimorbidity's association with mortality reflect causal disease-disease interactions, or is disability the primary mediating path?