RESEARCH DOSSIER
Chronic Disease and Multimorbidity Risk in Health Insurance
Abstract. Chronic disease and multimorbidity reshape health insurance through persistent and correlated demands for services, non-linear expenditure distributions, overlapping treatment pathways, and long-term financial obligations that depend on coverage rules and care institutions. This monograph analyzes how the existence, combinations, severity, duration, and treatment of multiple chronic conditions affect prospective morbidity measurement, insurance spending, risk adjustment, benefit design, provider payment, financial protection, and public risk pooling. It distinguishes multimorbidity prevalence in representative populations from prevalence among care users and from diagnosis frequencies recorded in insurance claims. Primary studies in Scotland, England, and the United States; OECD patient-reported and administrative indicators; WHO financing assessments; and original actuarial and risk-adjustment literature establish the evidentiary basis. Formal representations separate patient-level disease states, observed contacts, reimbursed claims, and insurer cash flows. A transparent hypothetical cohort illustrates how disease combinations and coverage eligibility change expected claims, while an explicit longitudinal identity shows why coding intensity can change recorded morbidity without any corresponding change in underlying disease. Randomized evidence on integrated care is assessed alongside observational studies of utilization and expenditure, with attention to selection, confounding, competing events, incomplete follow-up, and treatment burden. The resulting analytical synthesis identifies enduring obligations of measurement, financing, and governance: condition lists and denominators must be fixed and reproducible; mortality and disenrollment must be handled separately from disease progression; risk scores require independent clinical and financial calibration; and coordination programs must be judged on health, patient experience, equity, and costs as distinct outcomes. The monograph introduces no new proprietary actuarial model or doctrine and makes no causal claim beyond the identification strength of the underlying studies.
Keywords: Chronic disease; multimorbidity; multiple chronic conditions; health insurance; medical claims; morbidity measurement; risk adjustment; longitudinal expenditure; care pathways; care management; integrated care; financial protection; insurance pooling; adverse selection; prospective prediction; treatment burden; actuarial uncertainty.
Introduction
A health insurance system converts heterogeneous medical needs into a financing arrangement that promises some specified share of eligible care expenditures. The consequences of chronic disease are particularly important because disease persists across benefit years, coexists with other conditions, changes the types and timing of services used, and interacts with the institutional rules governing which entity pays. Multimorbidity makes these relationships observable in combinations rather than in isolated diagnoses. A member can need coordinated pharmacotherapy, laboratory monitoring, specialist visits, rehabilitation, acute hospital services, behavioral care, and assistance with daily functioning within the same period. Several of those services may address more than one condition; several may remain outside the relevant insurance entitlement. A diagnosis count therefore provides an incomplete description of clinical complexity and an incomplete measure of the financial obligation arising from it.
The epidemiological importance is established by large, independently designed studies. Barnett and colleagues (2012) examined 40 morbidities in 1,751,841 patients across 314 Scottish general practices and showed that chronic conditions clustered with age, socioeconomic deprivation, and mental–physical health combinations. Their Scottish multimorbidity study also found that onset occurred substantially earlier in more deprived groups. A later English primary care cohort examined multimorbidity within general-practice records using a different population and condition taxonomy (Cassell et al., 2018). These investigations document association and distribution within specified sampling frames; they cannot, without additional identification, establish what insurance arrangements cause.
The insurance problem has four connected levels. The population level concerns the distribution of need among everyone eligible for coverage, including people not receiving care. The clinical level concerns interacting disease processes and the design of treatment. The financial level concerns prices, utilization, benefit rules, payment responsibilities, reserves, and uncertainty over future claims. The institutional level concerns the pooling of risk, consumer protection, selection incentives, and the allocation of responsibility among insurers, public budgets, providers, and households. These levels interact, but their measurements are not interchangeable. In particular, expenditure represents both disease burden and the opportunity to obtain paid care; its absence may reflect a service gap rather than good health.
The research question is how to measure, predict, finance, and govern chronic and multimorbidity exposure without mistaking an administrative proxy for the underlying condition or insurer expenditure for patient welfare. This monograph employs documented descriptive studies, formal identities, peer-reviewed methods, and carefully bounded comparative inference. The evidence hierarchy is explicit: randomized studies can estimate effects within the populations and implementation settings actually assigned; observational studies describe patterns and may estimate conditional associations; official statistical descriptions define indicators within stated reference years; professional standards govern specified professional tasks in their jurisdictions; formal calculations express transparent analytical relationships under their assumptions. Bibliographic records and claim-specific source mapping appear in the companion evidence ledger. Dates and figures refer to the original observation periods, not to the November 2026 research date.
Foundations and Source-Critical History
From comorbidity to a population-level insurance object
The terminology carries substantive consequences. The historical review by van den Akker, Buntinx, and Knottnerus (1996), Comorbidity or multimorbidity, helped distinguish the coexistence of multiple conditions from the description of additional disease in relation to one designated index disease. Valderas and colleagues (2009) clarified that comorbidity, multimorbidity, disease burden, and patient complexity answer different questions across epidemiology, clinical care, and financing. Their analysis of the definitions provides a durable warning against using the same count to perform fundamentally different tasks. For population insurance, multimorbidity normally treats the individual as the center of measurement; for an index intervention, comorbidity often asks how coexisting disease modifies the result of treatment of a specified condition.
The threshold of two or more long-term conditions is a common entry definition, but its numerical simplicity is deceptive. The NICE multimorbidity guideline (NG56, 2016) encompasses specified physical and mental health conditions and can include continuing conditions, symptom complexes, sensory impairment, and substance misuse. A claims analyst using only a restricted list of reimbursable diagnostic codes may count a different phenomenon. A medical insurer that excludes some types of care from its benefit package may financially observe still fewer conditions. Prevalence should always be accompanied by the denominator, eligible condition list, observation window, condition-confirmation algorithm, population, and sampling mechanism.
The historical shift from disease-specific risk factors to multimorbidity is also a shift in the structure of explanation. A single-disease clinical trial can estimate the effects of a particular therapy while excluding people with complex coexisting disease. Its treatment guidance may subsequently be applied to insured populations where multiple therapies compete for adherence, monitoring time, financial resources, and patient attention. The gap between single-condition evidentiary design and whole-person practice was a central observation of Barnett et al. (2012), while NICE (2016) explicitly advises consideration of the total treatment burden and the appropriateness of single-condition recommendations. For insurance researchers, the consequential question is whether payment and care incentives amplify fragmentation or permit decisions adapted to combined needs.
Prognostic classifications and their limits
Administrative insurance science inherited major methodological tools from studies originally designed for different endpoints. Charlson and colleagues (1987) introduced a weighted index for mortality-related comorbidity in longitudinal studies; the original Charlson validation used defined clinical cohorts and particular mortality horizons. Elixhauser and colleagues (1998) proposed separate administrative comorbidity measures associated with hospital outcomes; their hospital-data classification study argued against automatically collapsing all categories into a single index. Neither instrument is a universal measure of annual insurer liability. A score designed to predict death, in-hospital charges, or length of stay does not necessarily preserve ranking when applied to outpatient pharmaceutical expenditures, disability payments, or public-sector purchasing.
The appropriate target controls the measurement choice. A mortality-oriented index may be valuable as a severity covariate in an analysis of hospital outcomes. A diagnosis vector may better represent how particular conditions affect utilization of distinct services. A model calibrated to prospective plan payments must be evaluated against the precise payment and covered-benefit definition. The historical development of comorbidity measurement therefore supports plural, task-specific validation. It does not warrant accepting every published score as an actuary's proxy for clinical need or consumer entitlement.
The empirical scale and selection of observed populations
Comparative prevalence numbers illustrate the importance of denominators. The OECD PaRIS chronic-condition analysis (2025) reports that 52% of surveyed primary-care users aged 45 and older across its participating settings had at least two chronic conditions. The OECD reports around 107,000 survey patients in 19 countries in its first PaRIS survey announcement. This is a primary-care user sample, not a prevalence estimate for all residents of those countries. Its composition is directly relevant to care coordination but necessarily conditional on contact with primary care.
A 2026 report analyzing the U.S. 2024 National Health Interview Survey found 26.4% of civilian, noninstitutionalized adults had at least two of ten selected diagnosed chronic conditions, using 32,629 survey respondents (Black and Ward, 2026). The 2024 NHIS prevalence analysis reported 62.0% for those aged 65 or over, with design-weighted estimation. By contrast, U.S. Behavioral Risk Factor Surveillance System data from 2023 yielded 51.4% for two or more of twelve selected self-reported conditions and a larger, differently sampled population; the CDC's 2013–2023 analysis used a different instrument and condition set (2025). The difference between 26.4%, 51.4%, and 52% is a demonstration of measurement architecture, not evidence of contradictory disease prevalence. Country, year, population age, survey mode, diagnosis confirmation, condition list, and use of care all vary.
The OECD 2025 chronic-condition indicators further show how diabetes and related chronic disease patterns vary across settings, with lifestyle risks and demographic composition contributing to exposure. The WHO's noncommunicable disease overview (2025) places long-term conditions within the international burden of mortality and early death, particularly in low- and middle-income countries. National statistics about death, diagnosis prevalence, treated disease, and insurer cost must retain distinct denominators and dates. Mortality figures for noncommunicable diseases cannot be interpreted as counts of people with multimorbidity, because causes of death and concurrent chronic conditions are different statistical objects.
Insurance financing as a population-level obligation
The modern insurance relevance of multimorbidity lies in risk pooling and long-run payment commitments. WHO explains in its analysis of health-financing pools that pooling spreads financial risk across a population and that fragmentation limits redistribution. This institutional result is important for chronic disease because individual expected expenditures are persistent: the same individuals often require services across many consecutive years. In voluntary or weakly integrated arrangements, stable differences in expected cost can create strong incentives for selective enrollment, coverage segmentation, and benefit design that deters costly members. A compulsory, broad, sufficiently financed risk pool can spread these obligations among low- and high-need members, though it also requires workable purchasing and public accountability.
The WHO and World Bank's 2025 universal coverage report documents major gaps in both effective service coverage and financial protection. It estimates that 4.6 billion people lacked full essential-service coverage in 2023 and 2.1 billion people experienced financial hardship from out-of-pocket health spending in 2022. These are different reference years and global indicators; neither is a count of uninsured people with chronic illness. Their relevance is institutional: a formal insurance card or tax-financed entitlement alone does not establish that indicated longitudinal treatment is available, timely, or affordable.
Multimorbidity Measurement
Observed condition sets, severity, and disease interactions
A person-level multimorbidity indicator can be written as M_i(t)=1{sum_k D_ik(t) >= 2}, where D_ik(t) denotes whether disease k meets the declared ascertainment criteria for person i within window t. The expression defines a measurement rule, not a biological state. Replacing K=10 conditions with K=30, increasing the look-back interval, or broadening the qualifying codes will change M even when all clinical histories remain constant. The numerator is the set of eligible people classified as having at least two recognized conditions; the denominator is the clearly defined eligible or observed population. Every prevalence estimate requires time alignment of both sets.
A condition count B_i=sum_k D_ik can be useful for stratification, but it discards severity, functional impairment, mental–physical combinations, interactions, and treatment workload. Two persons with the same count may differ in expected payments by orders of magnitude because of unstable heart failure, renal replacement therapy, cancer treatment, severe mental illness, or the availability of low-cost maintenance regimens. Conversely, many stabilized chronic conditions can be managed at relatively low cost when services are accessible. Condition co-occurrence must be examined for positive and negative associations, because shared risk factors and sequelae imply dependence rather than independently sampled diagnoses.
A richer description uses a vector D_i(t), observed severity S_i(t), functioning F_i(t), mental health status H_i(t), and care access A_i(t). The analyst should distinguish clinical measures from coding intensity and distinguish a recognized condition from the number of treatment claims caused by that condition. Patient complexity includes frailty, social support, cognition, health literacy, and treatment burden. The purpose of modeling these dimensions is to avoid equating more recorded diagnoses with a mechanically higher disease burden. Documentation itself can improve when coverage expands, electronic records are linked, or clinicians adopt new billing practices.
Claims algorithms and reproducibility
The CMS Chronic Conditions Data Warehouse provides a concrete example of administrative ascertainment. Its condition-category methodology identifies cohorts using specified diagnosis, procedure, and related claims codes. The CCW chronic-condition definitions now identify 30 chronic-condition categories for file years 2017 onward and 27 legacy categories for 1999–2021, with different look-back periods and algorithm versions. The CMS source expressly notes changes around the 2015 transition from ICD-9 to ICD-10 and later algorithm refinement. A retrospective trend that stitches together these category sets without a bridge specification may detect a change of classification rather than a change of disease.
Claims diagnoses also depend on the opportunity to generate claims. A person with little access to covered care may be incorrectly assigned to the lowest morbidity class. An insurer or provider may improve diagnosis completeness through outreach, care coordination, or documentation review. That administrative improvement can increase measured multimorbidity even if treatment outcomes improve. An independent validation sample, linked clinical records, or a sensitivity analysis for diagnosis capture is therefore valuable. Detection bias is particularly important when evaluating policies that change the probability of clinical encounters.
The U.S. Medical Expenditure Panel Survey offers a contrasting measurement perspective. The AHRQ household survey design includes nationally representative information on health status, coverage, services and payments, while the MEPS expenditure definitions distinguish household-reported treated conditions from clinically confirmed disease and count payments differently from provider charges. These data can support population expenditure comparisons with appropriate survey weights and design-based uncertainty. They cannot be substituted unchanged for insurer claim transactions, because person selection, missing diagnoses, out-of-pocket payments, and the unit of observation differ.
State construction and exposure units
The correct exposure measure depends on the question. A prevalence study might use all eligible adults observed at a survey date. A prospective insurance cost study uses contract-covered person-time, often member months, because enrollment and benefit liability change within a year. If member i is covered for m_i months in an observation year, exposure E=sum_i m_i/12 in member-years. A member-year expected claim estimate is total covered incurred claims divided by E; per-member-per-month (PMPM) cost divides by sum_i m_i. A calendar-year total divided by the count of unique claimants overstates a cost rate relative to a full enrolled population because zero-claim members are excluded.
A complete person-level analytic record should distinguish first diagnosis, onset inferred from observations, period prevalence, treatment initiation, severity progression, remission or stability where clinically meaningful, acute exacerbation, recovery, enrollment change, disenrollment, and death. Disease states need not be absorbing: some diagnoses remain lifelong while clinical status and service intensity change. A six-month absence of claims is never by itself proof of recovery. Death is a competing terminal event for service trajectories; disenrollment is a separate coverage event. Conflating them can bias both disease-duration estimates and insurer liabilities.
Quantifying dependence and combinations
Condition-specific risks should be evaluated both marginally and jointly. Let X be the indicator for one diagnosis and Y for another. The difference P(Y=1|X=1)-P(Y=1|X=0) quantifies observed association in the defined data, subject to confounding and ascertainment. It does not establish causal disease interaction. For expenditures, a conventional illustrative regression E[C|X,Y,Z]=beta0+beta1X+beta2Y+beta3XY+g(Z) encodes an interaction term beta3, with Z including relevant demographics, function and plan conditions. The interaction is scale dependent: additivity of costs does not imply additivity of log costs or disease hazards. Any interaction analysis must state the link function and error distribution.
To illustrate why additive counts are insufficient, consider four hypothetical annualized insurer-claim groups under a common benefit contract: neither of two conditions, $1,000; condition A alone, $3,000; condition B alone, $4,000; and both, $9,000. The incremental joint interaction on the dollar scale is $9,000-$3,000-$4,000+$1,000=$3,000. This is an arithmetic example chosen to make the interpretation explicit. It is not an estimate derived from any real insurer. It may represent acute interactions, extra specialist visits, medication complications, or service pricing, but those mechanisms must be separately demonstrated in empirical research.
Prediction scores and clinical meaning
A diagnosis-based risk score is constructed for a specified prediction target and observed information window. It may combine conditions, interactions, demographics, prior utilization, and contract effects, with model parameters fitted to prior observations. The prediction target might be next-year allowed medical spending, insurer-paid claims, preventable admissions, disability incidence, or mortality. Different targets imply different estimands and different governance requirements. Reusing a score trained to forecast public payment transfers as a clinical triage score is an unsupported step unless it is separately evaluated for clinical discrimination, calibration, and harm.
The HHS-HCC development study (Kautter et al., 2014) documents how health-condition categories and a risk-transfer structure were developed for individual and small-group insurance markets. The health-status information affects resource transfers among plans, rather than determining the whole value of a person's entitlement. Geruso and Layton (2020) demonstrated in their Medicare diagnosis-coding study that coding incentives can increase plan-reported risk scores relative to traditional fee-for-service comparators. These findings prohibit the simplistic inference that a higher risk-adjustment transfer necessarily indicates a sicker population. Observed differences combine underlying need, diagnostic capture, contract selection, and administrative incentives.
Longitudinal Disease Pathways
Disease histories as time-dependent processes
The relevant insurance horizon extends from a single care episode to many coverage years. For an enrollee with several chronic conditions, medical expenditure may depend on prior health status, current severity, medication adherence, future deterioration, treatment availability, coverage changes, and prices. A longitudinal representation therefore begins with a dated sequence of clinical states rather than a fixed binary indication that the person has ever carried a diagnosis. One possible descriptive state vector is S_i(t)=(D_i(t), severity_i(t), functioning_i(t), treatment_i(t), coverage_i(t)). The observed transition from S_i(t) to S_i(t+1) is driven partly by biological and care processes and partly by record availability. The transition probabilities must be estimated for the specific population and care setting; they should never be inferred merely from the annual prevalence table.
For formal analysis, a multi-state process may represent clinically stable disease, worsening or complication, acute admission, post-acute recovery, and death. If p_a(t) denotes the probability of state a and q_ab(t) a transition intensity from state a to state b, then a continuous-time formulation has dp_a/dt=sum_{b != a} p_b q_ba - p_a sum_{b != a} q_ab, under the stated Markov assumptions. Many chronic disease histories violate simple Markov assumptions because previous exacerbations, disease duration, and treatment histories predict later risk. Researchers can extend the state or use semi-Markov, recurrent-event, or landmark approaches. The formal equation is a methodological description, not a fitted clinical model for this dossier.
Competing endpoints matter. An individual who dies can no longer accumulate ordinary future health service claims, although end-of-life episodes may be costly and contract liabilities can persist for incurred services. A person who switches insurers leaves one payer's prospective risk pool, but the chronic disease has not disappeared. A person who obtains public eligibility may shift expenditure from private insurance to a statutory scheme. Each event affects different mathematical objects: health outcomes, coverage exposure, payer claims, and broader social expenditure. A coherent study specifies the payer perspective before selecting a censoring or competing-risk method.
Annual cost versus lifetime need
An annual premium is typically attached to an eligibility period; the disease pathway may last decades. Define C_i,t as insurer-paid claims incurred for member i during a specified covered period t and E_i,t as eligible member-time. The aggregate annual expectation is E[C_t]=sum_i E[C_i,t | coverage rules, observed history]. For a longitudinal perspective, expected discounted covered outgo over H periods can be expressed as V_i=E[sum_{t=1}^{H} v_t I_i,t C_i,t], where I_i,t indicates eligibility and v_t is a declared discount factor. The expression is illustrative: cash flows, insurer liabilities, statutory obligations and accounting conventions must be defined consistently before interpreting it as an actuarial reserve or premium. It is not a statement that an insurer has already incurred the entire future lifetime cost of a chronic condition.
The chronic nature of disease contributes to persistence in risk classification, but it does not guarantee persistence in insurer exposure. Group coverage may change with employment; individual contracts may terminate; public eligibility can begin at retirement or disability; managed care contracts can be reprocured. A prospective cost model should therefore assess two pathways: conditional medical expenditure if covered and the transition of coverage responsibility. Pool-level forecasts additionally require incoming members and evolving case mix, rather than an extrapolation of costs among existing high-utilization members alone.
Annual cost concentration must be differentiated from persistent high cost. A large claim in year one may be associated with temporary surgery, a cancer treatment phase, or acute deterioration. A person with a moderate but stable chronic burden can generate cumulative cost over many years. An analysis that ranks people by last year's claims and assumes the same rank next year will be affected by regression to the mean, changes in treatment, mortality, and discontinuous catastrophic events. Financial risk management should study transition matrices among annual expenditure quantiles, condition states, and insurance eligibility, with uncertainty around the matrices. Such work is particularly important when choosing care management cohorts or designing stop-loss arrangements.
Causal pathways, confounding, and treatment response
Multimorbidity is associated with health spending partly because multiple diseases cause additional treatment needs, and partly because common determinants—age, deprivation, obesity, access to care, provider patterns, and medication adherence—affect both diagnosis and utilization. Clinical encounters themselves increase the probability that further diseases are recorded. Observational regressions adjusted for a limited set of demographic variables therefore cannot automatically isolate an independent causal effect of the second or third diagnosis. The correct causal question must specify the intervention: preventing an illness, reducing its severity, improving access, coordinating care, modifying payment, or changing the coding process. These interventions operate through different channels.
A transparent conceptual pathway runs from underlying burden to symptoms and functional limitations; from those states through access and professional decisions to services; from services through contractual prices and benefit rules to expenses; and from expenses through payment and reporting procedures to recorded insurer claims. Social deprivation, geographic supply, household resources, and administrative eligibility influence several links. Conditioning on the presence of a submitted claim can create selection bias because being observed as a claimant depends on both health needs and access. Conditioning on hospitalization can similarly distort associations between diagnoses and later spending. Claims-based studies should therefore identify variables that are confounders, mediators, outcomes, or potential colliders before designing adjustment strategies.
The retrospective observational association between multiple conditions and service use is well documented. Lehnert and colleagues (2011) examined patterns of utilization and costs among older people in their multimorbidity cost review. McPhail (2016) synthesized how coexisting disease is associated with greater resources and expenditure in the chronic disease cost analysis. The literature supports material cost concentration and complexity, but estimates are heterogeneous by country, sample selection, condition classification, and resource valuation. Descriptions of near-exponential spending increases in some reviewed settings should not be universalized into an invariant law of disease-count cost escalation.
Patient pathways and payer pathways
Consider a patient with diabetes, hypertension, depression, and chronic kidney disease. A single clinical encounter may address glucose control, renal monitoring, blood pressure, medication interactions, and depression. For the patient, one encounter can reduce treatment burden. For the insurer, it may generate one office claim, laboratory claims, pharmaceutical expenditure, and a management payment. The payment identity is affected by how services are bundled or carved out. Under capitation, the insurer may incur a stable negotiated amount despite volatile underlying service intensity; under fee-for-service, every billable service changes gross allowed spending. A patient-level intervention can reduce total utilization yet raise some insurer-covered spending if previously foregone beneficial care becomes available.
A longitudinal case study must distinguish three cost perspectives. Total resource cost includes all real inputs used to deliver care, irrespective of payer. Health system expenditure reports valued payments or costs within the health accounts boundary. Insurer-paid claims comprise amounts contractually payable under a defined program and may exclude cost sharing, certain out-of-network services, social assistance, and non-covered long-term support. Summing multiple payer datasets without reconciliation can duplicate expenditures, while looking only at the insurer can overlook shifted burdens. That is central to interpreting apparent savings from disease-management programs.
Treatment burden, adherence, and indirect protection
Multimorbidity exposes patients to multiple appointments, monitoring regimes, drug regimens, travel obligations, administrative reviews, and choices among competing clinical priorities. The NICE NG56 recommendations give priority to patient goals, polypharmacy review, coordination responsibility, treatment burden, and unplanned care. They are clinical guidance within their jurisdiction; they are not a universal health insurance benefit mandate. For insurance analysis, however, they identify dimensions neglected by models that count only reimbursement. Reducing appointment burden, inappropriate prescriptions, or adverse drug interactions may improve welfare even if no material reduction in covered claims is visible within one budget year.
The WHO ICOPE second-edition handbook (2025) presents person-centered assessment and care pathways for older people, including social needs and changes in intrinsic capacity. Its institutional logic illustrates the overlap between medical insurance and long-term care. Some systems assign home support to social budgets, others cover it partially through health schemes or long-term care insurance. A reduced hospital stay can be a real clinical benefit but may require new community expenditure, support from unpaid family caregivers, and cross-budget transfer. Financial evaluation should identify all relevant perspectives and avoid celebrating a payer-specific cost reduction achieved by shifting expense onto households.
Care Management and Spending
Distinguishing coordination from measured effectiveness
Care management describes a broad family of interventions: case identification, medication reconciliation, chronic disease registries, shared care plans, clinical review, home support, remote monitoring, specialist coordination, care navigation, and behavioral health integration. Their targets and intensity differ markedly. An insurance program that merely labels a subgroup as high risk has not necessarily delivered a clinically effective intervention. Evaluation requires a description of program components, enrollment procedure, implementation fidelity, usual-care comparator, period of follow-up, and outcomes that matter to the patient and financing institution.
The strongest direct counterweight to automatic cost-saving claims is randomized evidence. The 3D pragmatic cluster trial (Salisbury et al., 2018) enrolled 1,546 adults with three or more chronic conditions in 33 UK practices. After 15 months, it found no difference in the prespecified primary quality-of-life outcome (adjusted EQ-5D-5L mean difference 0.00; 95% CI −0.02 to 0.02). A companion economic evaluation of 3D reported wide uncertainty and no meaningful cost or quality-adjusted life-year difference; its mean cost difference was £126, with a 95% confidence interval from −£739 to £991. The intervention nevertheless improved some measures of patient-centered experience. This contrast is scientifically important: experience, health status, resource use, and insurer expenditure are independently measured outcomes, and an improvement in one does not logically establish another.
The 2021 Cochrane multimorbidity review (Smith et al.) found mixed results across primary-care and community interventions, including limited evidence on costs and little or uncertain change in service use. Some interventions improved mental health or patient-reported outcomes, particularly when targeted to specific combinations such as depression with other chronic conditions. The review's uncertainty is consistent with heterogeneity of interventions rather than proof that coordinated care can never work. It supports careful targeting and prespecified evaluation, not universal saving assumptions.
Decomposing health insurance expenditure
For an insured group, annual covered spending can be expressed as C=sum_j n_j p_j a_j, where n_j is the number of services of type j, p_j the allowed price per service, and a_j the covered fraction the insurer is liable to pay under the relevant contract, subject to deductibles, caps, member cost sharing, and payment rules. Because benefit functions are often nonlinear within a year, a_j may depend on cumulative spending and benefit thresholds. Under capitation, direct fee-for-service summation may not represent the insurer's contractual obligation. The expression is therefore a decomposition for appropriately homogeneous benefit classes, not a universal payment formula.
A change in spending can arise from a change in disease mix, service quantity, provider fees, treatment intensity, benefit generosity, coding, administrative eligibility, or the distribution of cases across payers. A disease-management program may reduce emergency admissions while increasing primary care consultations and long-term pharmaceuticals. If benefits move from out-of-pocket to insured coverage, payer cost rises even as financial protection improves. Analysts should report utilization changes in natural units, actual allowed prices, insurer-paid shares, patient-paid shares, and total resource costs before attributing an aggregate trend to biological worsening.
Under a capitation arrangement, providers may receive a risk-adjusted amount per eligible person instead of reimbursement for every clinical event. The insurer's cost may become more predictable, but providers can face incentives to avoid complex patients or reduce service intensity unless quality and access safeguards operate. Under retrospective fee-for-service, the same disease combination may generate more billed activity. The correct assessment includes who bears financial risk and whether clinically necessary care occurs. Removing reimbursement from the claim ledger does not remove the cost of delivering services.
Insurance design, household resources, and access
Chronic conditions make recurring copayments, drug coverage restrictions, specialist network gaps, and repeated authorization processes consequential. A low annual probability of a catastrophic accident differs from a high probability of recurring medically necessary services. Contracts with narrow formularies or weak outpatient coverage can impose persistent out-of-pocket obligations despite formal enrollment. Conversely, expanding coverage can increase recorded use and insurer-paid expenditure by financing services previously unavailable. These outcomes should be considered jointly rather than treating all increased claims as a decline in insurance performance.
The 2025 WHO–World Bank UHC assessment emphasizes two dimensions: essential service coverage and financial hardship. It reports 2023 coverage and 2022 financial-hardship evidence using revised UHC indicators. The spending and health dimensions need consistent but separate monitoring. An insurance product may offer a nominal reimbursement rate while excluding the services required by people with multiple conditions; a publicly financed program may offer broad entitlements while geographic supply constraints delay treatment. Effective protection is therefore partly contractual and partly dependent on institutions that deliver care.
Equity analysis must also account for persistent social disadvantage. Barnett et al. (2012) found a substantially earlier onset of multimorbidity in deprived Scottish communities. Such differences imply that age-only risk stratification can miss members with intense need at younger ages. When subsidies, care management, or access rules rely exclusively on older age, they can systematically underrepresent multimorbidity among working-age people in disadvantaged groups. Whether an insurer can lawfully use particular characteristics for pricing or eligibility depends on local law, and a fair risk-transfer system need not use all predictive variables in consumer premiums.
Risk adjustment, selection, and coding incentives
Competing insurers with guaranteed enrollment rights may still differ in expected spending because of their enrollment populations. Risk equalization seeks to reduce selection incentives by redistributing money according to approved risk characteristics. The HHS-HCC methodological paper explains one U.S. approach to estimating differences in expected cost for individual and small group markets. Its categories are a policy and payment design under particular legal arrangements, not an epidemiological census of all chronic disease. Transfer systems can reduce incentives for selection while creating incentives to increase diagnoses that contribute to risk scores.
Geruso and Layton's 2020 coding-incentive analysis estimates diagnosis-based risk-score differences in a Medicare context, which demonstrates that recorded morbidity is partially endogenous to payment. This matters for longitudinal insurance comparison: a payment reform can increase observed risk scores in one period even if true morbidity and clinical outcomes remain stable. Conversely, better access can bring previously undocumented conditions into view. An independent audit should evaluate coding intensity, risk-score distribution, residual selection, predictive error, and indicators of under-service and patient outcomes. Adjustment must reflect allowed uses of health data and local governance rules.
One design choice is whether to use concurrent diagnoses or prior-year diagnoses to predict current payments. Concurrent information can correlate closely with current use but may intensify coding incentives. Prospective information better separates the prediction period from the payment period but loses information about newly emerging disease. The appropriate decision balances payment accuracy, selection incentives, clinical timing, administrative cost, and sensitivity to missing care. There is no single universally optimal time window across national health insurance systems.
Risk allocation beyond the annual claim
Long-term claims arising from chronic illness affect pricing and planning through claim persistence, cost trend, high-cost episodes, disability transitions, and long-term services. A risk pool should project not only mean claims but distributional tails and joint exposures to inflation, pharmaceutical innovation, and labor supply in health care. Disability and long-term care benefit triggers can depend on functioning, activities of daily living, waiting periods, and contract-specific definitions; they should not be inferred from a diagnosis count. Reinsurance and stop-loss may protect the insurer against extreme individual claims but cannot substitute for an adequately financed base population or accessible chronic care.
Administrative provision for incurred claims deserves separate treatment. The Actuarial Standards Board's ASOP No. 5 on incurred claims is a 2017 adopted U.S. professional standard covering estimation or review of incurred health and disability claims. It requires attention to service and reporting dates, data quality, claim seasonality, provider arrangements, and documentation within its scope. An insurer's accrued unpaid claims for already incurred services differ from the expected cost of services a member may need in future benefit years. Confusion between these categories can produce a substantial actuarial reporting error.
Data, Measurement, and Validation
Choosing an estimand before choosing a model
The target quantity must be explicit. For descriptive research, the estimand might be the proportion of a defined resident population with two or more specified conditions on a certain date. For insurance, it might be expected covered claims PMPM over the next 12 months among members enrolled at a fixed baseline, or the difference in spending if an eligible member receives an intervention compared with usual care. These are fundamentally different quantities. A validated prevalence instrument does not automatically predict expenditures; a spending model may be predictively useful while providing no causal information about how to reduce treatment burden.
A model protocol should specify population membership; index date; diagnosis history and permissible look-back; whether outcomes are allowed, paid, or incurred; target horizon; observation of zero-claim members; provider-network and benefit constraints; inflation and currency treatment; relevant outcome transformation; and the statistical mechanism for death and disenrollment. Comparisons should preserve calendar time and comparable observation intensity. External validation in a different market must check both clinical coding and the definition of insurer liability. Recalibration can be legitimate, but it should be recorded rather than treated as evidence of direct transportability.
Time alignment and leakage
Temporal leakage occurs when the predictor uses information that would not have been available at the prediction date. If next year's hospital claims are included in an index-year diagnosis score, the model may seem exceptionally accurate while failing the task of prospective allocation. Similarly, a diagnosis first documented during the outcome window is not a valid baseline predictor unless the design deliberately estimates concurrent risk. Backdated coverage entries and long claim submission delays can make these errors subtle. Data engineers should separate clinical service date, claim submission date, payment date, diagnosis record date, and coverage effective date.
The CCW algorithm version descriptions and AHRQ MEPS condition ascertainment notes are practical examples of why temporal and methodological version control matters. A shift from 27 to 30 chronic categories, or a change in survey questions, may create artificial breaks in a trend. Reproducible inference requires preserving the condition catalog, code mapping, algorithm version, eligibility file, data vintage, and source of each payment. A published figure should be traceable to both the numerator and the exact denominator used at the time of calculation.
Predictive performance and distributional fit
A prospective spending model should be assessed using an independent test period or other held-out design appropriate to the task. Aggregate calibration compares predicted with observed expenditures across the portfolio; subgroup calibration examines errors by multimorbidity count, combinations, age, socioeconomic proxy, coverage type, and cost quantiles, where lawful and feasible. Rank performance can help prioritize outreach but can conceal systematic underprediction for small clinically complex groups. Tail errors deserve explicit reporting because expensive admissions and specialty treatments can dominate portfolio variance and reinsurance attachment behavior.
Prediction intervals, not only point estimates, are needed when financial decisions involve uncertain medical spending. Claim distributions are commonly right-skewed, may contain many zeros, and show variation between members and across time. A two-part approach can separately model the probability of use and conditional positive cost. A hierarchical model can capture provider- and region-level variation. A compound frequency–severity model can represent service events and their financial impact. Choice among these approaches depends on the target outcome, observed data, and calibration; their existence is not evidence that one automatically dominates. Under any approach, the actuary should investigate whether model residuals show temporal dependence, structural breaks, or unobserved benefit changes.
A model predicting insurer-paid claims must reconcile the allowed-charge and payment processes. A patient deductible can concentrate liability late in the benefit year; a stop-loss treaty can change the insurer's net exposure; a retroactive eligibility correction can change the apparent claims denominator. Poor coding of zero-claim members and omitted member-months can make predicted PMPM claims materially unreliable. Back-testing should report both gross and net liabilities as applicable and describe whether outcomes include hospital, pharmaceutical, dental, mental health, and long-term support benefits.
Causal assessment of management programs
To evaluate care coordination, the estimand might be the intention-to-treat effect of an offer of management services among eligible members. If assignment is randomized, differences in outcomes can be attributed more credibly to the offer, subject to contamination, attrition, clustering, and missing data. If treatment receipt differs from assignment, an effect among participants is not equal to the randomized effect of the offer. Trials should report their primary outcomes as prespecified and distinguish exploratory subgroup results from confirmatory findings.
Without randomization, a prospective quasi-experimental design can exploit staggered program implementation, eligibility thresholds, or plausible external variation, if the identification assumptions can be defended. A difference-in-differences estimate requires careful scrutiny of parallel pre-intervention trajectories and composition changes. Matching on prior spending alone does not fully adjust for disease severity or motivation to participate. Regression to the mean is particularly severe when the intervention selects members following an unusually expensive year. A decline in their subsequent mean expenditure may occur without any intervention effect. The 3D trial's null primary outcome and the Cochrane synthesis of interventions demonstrate why substantive claims require tested comparators.
Evaluation should record the intervention's administrative costs, care-team expenses, patient travel and time, medication changes, hospitalizations, preventable harms, treatment burden, experience, and financial exposure. Short-term cost savings and longer-term clinical benefit may move in opposite directions. Study duration must be sufficient to observe plausible consequences, while analysis of limited follow-up must openly distinguish interim outcomes from permanent prevention of disease complications. The payer perspective and society-wide perspective should be reported separately whenever resource shifts are likely.
Sensitivity, subgroup, and missing-data analysis
An auditable analysis tests definitions and assumptions most likely to move the result. Relevant sensitivities include requiring one or two confirming diagnostic encounters; varying the disease list; changing the look-back window; separating physical-only from mental–physical combinations; treating death as competing risk rather than ordinary censoring; varying costs with inflation and service prices; and adjusting for unobserved coverage gaps. An alternative sample restricted to continuously enrolled members may reduce missing outcomes but introduce selection if healthier people are more likely to maintain enrollment.
Subgroup analyses can expose important variation in older adults, working-age people, low-income groups, rural populations, and individuals with behavioral health conditions. They should be driven by prespecified scientific questions and the sample's precision, not by a search for statistically striking differences. Cross-national statistical comparisons require uncertainty intervals, sampling weights, age standardization where appropriate, and explicit caution around survey and benefit definitions. The OECD's PaRIS primary-care sampling frame is highly informative for patient experience but cannot simply be reweighted to describe all uninsured or untreated residents without a defensible population frame.
The evidence ledger and research boundaries
This dossier's source ledger assigns a distinct internal identifier to each verified institution, original paper, or professional standard; records the original URL, the fact checked, the part of the manuscript where it is used, and the practical access scope. These identifiers are kept out of the reader-facing work. The bibliography provides authors, titles, venue or institution, original year, identifiable issue or document number when established, and the same direct source link. This means source traceability can be audited independently of the typography of the scientific text.
The present dossier synthesizes published studies and official sources. No patient-level dataset was downloaded or fitted here, and no source was represented as having been independently replicated from microdata. Survey prevalence values remain attached to their original target populations; randomized effect estimates remain attached to their trial designs; methodological formulas are transparent identities or illustrative examples. This evidence boundary is a substantive scientific result: reliable insurance analysis requires a clear account of which claims concern measurement, which concern association, and which concern causal intervention effects.
International Synthesis and Scientific Conclusions
The institutional diversity of risk sharing
Health insurance systems vary in their sources of funding, eligibility rules, benefit entitlements, provider payment, and the position of private plans. The same person's clinical state can lead to different insurer or household expenditure under two systems because they allocate cost differently. In one jurisdiction, chronic drug treatment may be heavily subsidized through a national scheme; in another, repeated copayments, exclusions, or network restrictions may impose substantial private spending. Cross-country comparison must identify who pays, who remains uninsured or undercovered, and which services enter each recorded expenditure series.
WHO's health financing and pooling analysis supplies the relevant institutional principle: financial risk protection depends on prepayment and the structure of shared pools, not only on individual diagnosis prediction. The global UHC monitoring report shows that actual service coverage and household financial protection are simultaneous dimensions of performance. Multimorbidity makes the distinction especially important because the need for care recurs, often involving several medical and social-service systems. When medical benefits are narrow, ordinary household expenditure can remain severe even if the insurer's claims experience appears favorable.
The U.S. offers detailed claims-based classification and financial risk-adjustment mechanisms, illustrated by CMS's condition algorithms and the HHS-HCC model design. England provides a prominent clinical reference for whole-person management through NICE multimorbidity guidance. The OECD's international PaRIS evidence documents experiences among primary-care users in participating countries. These materials answer different questions: coding and financial transfers, clinical management, and patient-reported care experience. Their findings should be integrated analytically rather than collapsed into a cross-country ranking that implies common data generation.
Avoidable hospitalization and an access paradox
The OECD's 2025 avoidable-admissions indicators identify potentially avoidable hospitalizations for selected conditions such as asthma, COPD, congestive heart failure, and diabetes as informative markers of primary-care access and quality. These indicators have important interpretive boundaries. A low hospitalization rate can reflect excellent disease management, but it can also result from restricted access to hospitals or incomplete recording. Risk-standardized measures must consider disease prevalence, severity, and local service capacity.
The insurer's spending perspective creates a parallel paradox. A plan that discourages costly specialist visits may show reduced claims in the short run, while delayed treatment may worsen health and increase future costs borne by the same or another payer. A plan that enables previously unmet care may show higher paid expenditure while improving outcomes. Neither event should be classified solely from its claim trend. A defensible performance assessment combines patient-reported access, avoidable admissions, disease outcomes, treatment burden, utilization, and financial protection, with a declared time horizon.
Interpreting persistence, inequality, and selection
Disease burden accumulates through life, but the age trajectory is socially patterned. The Scottish population evidence found earlier multimorbidity in more deprived communities; the U.S. 2024 NHIS estimates show material age and coverage differences. These sources should be used to construct appropriate risk strata, not to turn observed associations into deterministic forecasts about individuals. Multimorbidity at working ages calls for policies sensitive to coverage continuity and employment-related transitions, while advanced age adds considerations of frailty, functional support, and the division between health and long-term care finance.
Risk adjustment can distribute resources to plans with higher expected burden, but care quality must remain observable outside the payment score. The coding-incentives evidence demonstrates that insurance payments can shape clinical documentation. A robust system separates audit of diagnostic completeness from evaluation of patients' actual health and coverage. It also tests residual selection, such as plans' ability to attract people with uncomplicated disease combinations while discouraging those with complex mental–physical needs. Predictive accuracy alone does not resolve such incentives.
Substantive findings across evidence types
First, multimorbidity measurement is a function of the chosen definitions, data source, and observation system. Different published prevalence values are informative only relative to their condition lists, sampling frames, and age distributions. Second, chronic conditions generate expenditure through service access, clinical decisions, provider payments, and entitlement design. A high diagnosis count is neither a sufficient cost forecast nor a measure of welfare. Third, disease processes and insurance liability evolve on different clocks. The patient's clinical history may continue after eligibility changes, while previously incurred claims can remain payable after treatment dates.
Fourth, care management outcomes are multi-dimensional. Clinical experience and patient-centered care may improve without detectable quality-of-life or payer-cost gains in a finite trial. Fifth, payment systems shape which clinical information is recorded and how expensive members are financed. Independent risk-score calibration, coding audit, and residual selection analysis are indispensable. Sixth, the financial protection purpose of insurance requires attention to recurring out-of-pocket obligations and treatment access, including services outside the claims data of one payer. These conclusions follow from the combined source record and transparent analytical distinctions; none constitutes a newly named theory or newly estimated causal effect.
Implications for research and editorial reproducibility
Further research should link appropriately governed longitudinal clinical and claims histories while retaining reproducible condition taxonomies and coverage records. Analyses should evaluate duration, severity, and interacting disease combinations rather than rely exclusively on count thresholds; use competing-risk handling for death and explicit exposure for coverage changes; and separate the performance of risk scores for payment allocation from their usefulness in clinical triage. Comparative work should examine which parts of care costs shift between households, insurers, long-term care programs, and public budgets. For intervention evidence, longer follow-up and predeclared multi-outcome trials are more informative than uncontrolled pre/post savings estimates.
Every reported statistic should identify the setting, age group, eligibility, date, denominator, condition definitions, and source version. Every quantified cost effect should identify currency, price year, spending perspective, contractual benefit, and uncertainty. Study design should explicitly govern language: association, predictive performance, and causal effect are separate claims. These requirements are feasible within existing scientific and actuarial methodology and provide an immediately auditable foundation for subsequent comparative health insurance research.
Conclusion
Chronic disease and multimorbidity are persistent clinical realities whose financial consequences are mediated by insurance architecture. Scientific measurement must begin with a defined population, explicit condition inventory, dated disease and coverage histories, and an expenditure outcome that matches the institution under study. The evidence from representative surveys, clinical cohorts, administrative algorithms, and expenditure records demonstrates that different measurement systems can produce substantially different prevalence and risk estimates without factual inconsistency. The evidence from randomized care coordination research also requires caution in claiming financial savings: improvements in patient experience, clinical health, and insurer cost are distinct empirical results.
A coherent health insurance response combines reproducible morbidity classification, longitudinal claims forecasting, careful benefit and payment design, independent evaluation of care management, protection against persistent household expense, and appropriately broad risk pooling. Professional actuarial claims standards and institution-specific risk adjustment mechanisms provide relevant tools but remain tied to their own objects, legal settings, and validation requirements. The central scientific conclusion is that health insurance risk associated with multimorbidity must be understood through the joint observation of disease, care delivery, contractual liability, and institutional distribution of expense over time. The manuscript establishes this conclusion through critically evaluated sources and explicit analytical relationships, without asserting a new proprietary theory or a causal result unsupported by the available research.
Technical Appendix: Auditable Illustrative Calculations
A. Hypothetical exposure and PMPM
Assume, solely for illustration, a covered cohort consisting of 600 persons insured for 12 months and 400 insured for six months. Eligible member-months are 600×12+400×6=9,600; member-years are 800. If total covered incurred claims equal $9,600,000, PMPM is $9,600,000/9,600=$1,000 and annualized claims per member-year are $12,000. Dividing by 1,000 unique individuals instead would report $9,600 per distinct person and would answer a different question. None of these values describes any actual health insurance portfolio.
B. Hypothetical joint-cost contrast
For illustrative mean annual covered claims of $1,000 (neither diagnosis), $3,000 (A alone), $4,000 (B alone), and $9,000 (both), the additive-scale joint departure is $9,000−$3,000−$4,000+$1,000=$3,000. It describes the assigned four group means and makes no causal statement about an individual disease interaction. Estimation in real data would need a defined population, adjustment and uncertainty around every group mean.
C. Hypothetical coding sensitivity
Suppose a fixed insured population contains 200 clinically confirmed people with two or more of a defined disease list, but the claims algorithm initially identifies 150. Its hypothetical sensitivity relative to that reference is 150/200=75%. After a documentation improvement it identifies 180 of the same 200, raising observed identification to 90%. Recorded counts increase by 20%, calculated as (180−150)/150, although true condition prevalence was held constant by construction. Such an effect demonstrates the possible direction of administrative ascertainment bias; it is not an estimate of coding intensity in any public program.
D. Hypothetical competing-risk survival
Assume constant annual cause-specific hazards for death of 0.04 and coverage disenrollment of 0.06, purely as an illustrative approximation. The probability of remaining alive and enrolled for one year under this simplified continuous-time model is exp(−0.10), approximately 0.904837. The probability of leaving the joint state within one year is therefore approximately 0.095163. The chance of death as the first exit is (0.04/0.10)×(1−exp(−0.10)), approximately 0.038065; the chance of disenrollment first is approximately 0.057098. Their sum equals the total exit probability. This demonstrates why an exit from an insurer's books is not necessarily clinical recovery or death; the hazards and memoryless assumption are hypothetical.
E. Evidence-to-decision traceability
The companion EVIDENCE.json identifies the external research records actually checked and the manuscript claims for which they are relevant. This document's PASSPORT.json records the internal review decision, scientific limits, source SHA-256, evidence SHA-256, and publication status. The Git commit is recorded outside this self-contained package after archival so that the file fingerprint remains stable. These materials form an editorial research package intended for batch release through an independently authorized publication process.
References
van den Akker, M., Buntinx, F., and Knottnerus, J. A. (1996). Comorbidity or multimorbidity: What’s in a name? A review of literature. European Journal of General Practice, 2(2), 65–70. DOI: 10.3109/13814789609162146. https://doi.org/10.3109/13814789609162146
Valderas, J. M., Starfield, B., Sibbald, B., Salisbury, C., and Roland, M. (2009). Defining comorbidity: Implications for understanding health and health services. Annals of Family Medicine, 7(4), 357–363. DOI: 10.1370/afm.983. https://pubmed.ncbi.nlm.nih.gov/19597174/
Charlson, M. E., Pompei, P., Ales, K. L., and MacKenzie, C. R. (1987). A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. Journal of Chronic Diseases, 40(5), 373–383. DOI: 10.1016/0021-9681(87)90171-8. https://pubmed.ncbi.nlm.nih.gov/3558716/
Elixhauser, A., Steiner, C., Harris, D. R., and Coffey, R. M. (1998). Comorbidity measures for use with administrative data. Medical Care, 36(1), 8–27. DOI: 10.1097/00005650-199801000-00004. https://pubmed.ncbi.nlm.nih.gov/9431328/
Barnett, K., Mercer, S. W., Norbury, M., Watt, G., Wyke, S., and Guthrie, B. (2012). Epidemiology of multimorbidity and implications for health care, research, and medical education: A cross-sectional study. The Lancet, 380(9836), 37–43. DOI: 10.1016/S0140-6736(12)60240-2. https://pubmed.ncbi.nlm.nih.gov/22579043/
Cassell, A., Edwards, D., Harshfield, A., Rhodes, K., Brimicombe, J., Payne, R., and Griffin, S. (2018). The epidemiology of multimorbidity in primary care: A retrospective cohort study. British Journal of General Practice, 68(669), e245–e251. DOI: 10.3399/bjgp18X695465. https://pubmed.ncbi.nlm.nih.gov/29530918/
Black, L. I., Ng, A. E., Zablotsky, B., and Ward, B. W. (2026). Prevalence of multiple chronic conditions among US adults, 2024. American Journal of Public Health, 116(6), 855–858. DOI: 10.2105/AJPH.2026.308427. https://pubmed.ncbi.nlm.nih.gov/41886700/
Watson, K. B., Wiltz, J. L., Nhim, K., Kaufmann, R. B., Thomas, C. W., and Greenlund, K. J. (2025). Trends in multiple chronic conditions among US adults, by life stage, Behavioral Risk Factor Surveillance System, 2013–2023. Preventing Chronic Disease, 22, 240539. DOI: 10.5888/pcd22.240539. https://www.cdc.gov/pcd/issues/2025/24_0539.htm
Organisation for Economic Co-operation and Development (OECD). (2025). Living with multiple chronic conditions. In Does Healthcare Deliver? Results from the First OECD Patient-Reported Indicator Surveys (PaRIS). OECD Publishing, Paris. https://www.oecd.org/en/publications/2025/02/does-healthcare-deliver_978507f1/full-report/living-with-multiple-chronic-conditions_0275c435.html
OECD. (2025, February 20). Healthcare systems need to better respond to patients’ needs, finds first international OECD patient survey. OECD press release, Paris. https://www.oecd.org/en/about/news/press-releases/2025/02/healthcare-systems-need-to-better-respond-to-patients-needs-finds-first-international-oecd-patient-survey.html
OECD. (2025). Chronic conditions. In Health at a Glance 2025: OECD Indicators. OECD Publishing, Paris. DOI for parent report: 10.1787/8f9e3f98-en. https://www.oecd.org/en/publications/health-at-a-glance-2025_8f9e3f98-en/full-report/chronic-conditions_e0110c98.html
World Health Organization (WHO). (2025, September 25). Noncommunicable diseases. WHO fact sheet, Geneva. https://www.who.int/en/news-room/fact-sheets/detail/noncommunicable-diseases
Lehnert, T., Heider, D., Leicht, H., Heinrich, S., Corrieri, S., Luppa, M., Riedel-Heller, S., and König, H.-H. (2011). Review: Health care utilization and costs of elderly persons with multiple chronic conditions. Medical Care Research and Review, 68(4), 387–420. DOI: 10.1177/1077558711399580. https://pubmed.ncbi.nlm.nih.gov/21813576/
McPhail, S. M. (2016). Multimorbidity in chronic disease: Impact on health care resources and costs. Risk Management and Healthcare Policy, 9, 143–156. DOI: 10.2147/RMHP.S97248. https://pubmed.ncbi.nlm.nih.gov/27462182/
National Institute for Health and Care Excellence (NICE). (2016). Multimorbidity: Clinical assessment and management, NG56. Recommendations. London: NICE. https://www.nice.org.uk/guidance/ng56/chapter/Recommendations
NICE. (2016, September 21). Multimorbidity: Clinical assessment and management. NICE guideline NG56, overview and scope. https://www.nice.org.uk/guidance/NG56
WHO. (2025). Integrated care for older people (ICOPE): Guidance for person-centred assessment and pathways in primary care. Second edition. Geneva: WHO. ISBN 978-92-4-010372-6. https://www.who.int/publications/i/item/9789240103726
Salisbury, C., Man, M.-S., Bower, P., Guthrie, B., Chaplin, K., Gaunt, D. M., et al. (2018). Management of multimorbidity using a patient-centred care model: A pragmatic cluster-randomised trial of the 3D approach. The Lancet, 392(10141), 41–50. DOI: 10.1016/S0140-6736(18)31308-4. https://pubmed.ncbi.nlm.nih.gov/29961638/
Salisbury, C., et al. (2019). A patient-centred intervention to improve the management of multimorbidity in general practice: The 3D RCT. Health Services and Delivery Research, 7(5). Southampton: NIHR Journals Library. https://pubmed.ncbi.nlm.nih.gov/30779534/
Smith, S. M., Wallace, E., O’Dowd, T., and Fortin, M. (2021). Interventions for improving outcomes in patients with multimorbidity in primary care and community settings. Cochrane Database of Systematic Reviews, 2021(1), CD006560. DOI: 10.1002/14651858.CD006560.pub4. https://pubmed.ncbi.nlm.nih.gov/33448337/
Centers for Medicare & Medicaid Services (CMS), Chronic Conditions Data Warehouse (CCW). (accessed 2026). Condition Categories: Administrative condition-identification methods and code-system change history. https://www2.ccwdata.org/web/guest/condition-categories
CMS, Chronic Conditions Data Warehouse. (accessed 2026). Chronic Conditions: 30-variable 2017-forward and 27-variable 1999–2021 algorithm documentation. https://www2.ccwdata.org/web/guest/condition-categories-chronic
Agency for Healthcare Research and Quality (AHRQ). (accessed 2026). Medical Expenditure Panel Survey: Household Component overview. Rockville, MD: AHRQ. https://meps.ahrq.gov/survey_comp/household.jsp
AHRQ. (accessed 2026). MEPS-HC Data Tools: Medical conditions and expenditure definitions. Rockville, MD: AHRQ. https://datatools.ahrq.gov/meps-hc/
Kautter, J., Pope, G. C., Ingber, M., Freeman, S., Patterson, L., Cohen, M., and Keenan, P. (2014). The HHS-HCC risk adjustment model for individual and small group markets under the Affordable Care Act. Medicare & Medicaid Research Review, 4(3). DOI: 10.5600/mmrr2014-004-03-a03. https://www.cms.gov/mmrr/articles/a2014/mmrr2014_004_03_a03.html
Kautter, J., Pope, G. C., Ingber, M., Freeman, S., Patterson, L., Cohen, M., and Keenan, P. (2014). The HHS-HCC risk adjustment model for individual and small group markets under the Affordable Care Act. Medicare & Medicaid Research Review, 4(3). PubMed record, PMID 25360387. https://pubmed.ncbi.nlm.nih.gov/25360387/
Geruso, M., and Layton, T. (2020). Upcoding: Evidence from Medicare on squishy risk adjustment. Journal of Political Economy, 128(3), 984–1026. DOI: 10.1086/704756. https://pubmed.ncbi.nlm.nih.gov/32719571/
Actuarial Standards Board. (2017). Actuarial Standard of Practice No. 5: Incurred Health and Disability Claims. Revised March 2017; effective September 1, 2017; document no. 186. Washington, DC: ASB. https://www.actuarialstandardsboard.org/asops/incurred-health-and-disability-claims_186/
WHO. (accessed 2026). Pooling revenues and reducing fragmentation. WHO Health Financing program, Geneva. https://www.who.int/activities/pooling
WHO and World Bank. (2025). Tracking Universal Health Coverage: 2025 Global Monitoring Report. Geneva: WHO. ISBN 978-92-4-011780-8. https://www.who.int/publications/i/item/9789240117808
WHO. (2025). UHC Global Monitoring Report 2025: Results and indicators. WHO institutional monitoring portal, Geneva. https://www.who.int/teams/health-financing-and-economics/global-monitoring-report/2025
OECD. (2025). Avoidable hospital admissions. In Health at a Glance 2025: OECD Indicators. OECD Publishing, Paris. DOI for parent report: 10.1787/8f9e3f98-en. https://www.oecd.org/en/publications/health-at-a-glance-2025_8f9e3f98-en/full-report/avoidable-hospital-admissions_2a8e2482.html