Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Health Care Utilization Responses to Insurance Coverage

Abstract. Health insurance alters the price of care at the point of service, the financial feasibility of treatment, continuity of access, provider revenues, and the allocation of medical resources. Consequently, utilization responses to coverage changes arise through several channels whose empirical magnitudes and welfare implications differ across services, populations, institutions, and horizons. This monograph offers an original source-critical synthesis of randomized experiments, quasi-experimental coverage changes, administrative utilization records, international health financing evidence, and methodological work. It separates insurance acquisition from generosity of coverage; eligibility from enrollment; the probability of any use from the intensity, composition, and quality of use; and patient demand from provider capacity and treatment decisions. The RAND Health Insurance Experiment establishes clear cost-sharing sensitivity, while the Oregon Medicaid lottery provides unusually credible estimates of insurance access, medical utilization, emergency-department use, diagnosis, and financial protection. Medicare eligibility discontinuities, U.S. Medicaid expansions, Massachusetts reforms, employer deductibles, and programs in Mexico, Thailand, and Ghana reveal outcome-specific and institutionally conditioned effects. Experimental and observational findings are evaluated according to their identified estimands, counterfactuals, follow-up periods, measurement instruments, and external validity. Across the evidence, improved coverage frequently raises contacts and service uptake and lowers financial exposure; measured health outcomes depend on the clinical intervention, baseline deprivation, supply response, and observation period. Reductions in utilization after greater cost sharing cannot be equated automatically with efficiency, and increases following enrollment cannot be assumed automatically to represent waste. The central analytical conclusion is that the effect of insurance is a vector of changes in access, episodes, service mix, provider behavior, health, and financial protection. Policy evaluation must estimate and interpret these dimensions together while respecting the distinct causal evidence supporting each.

Keywords: Health insurance; health care utilization; medical demand; extensive margin; intensive margin; insurance coverage; Medicaid; Medicare; RAND Health Insurance Experiment; Oregon Health Insurance Experiment; cost sharing; copayment; deductible; emergency department; primary care; inpatient admissions; pharmaceuticals; preventive services; provider behavior; supply constraints; universal health coverage; financial protection; randomized experiment; difference-in-differences; regression discontinuity; causal inference; treatment heterogeneity; international health systems.

Introduction

A health insurance contract or statutory entitlement changes the circumstances in which illness becomes a medical encounter. It can make a doctor visit financially affordable, allow a physician to order a diagnostic procedure, permit treatment through a contracted hospital, or provide a drug without a payment that exceeds the household's available cash. It also changes the incentives faced by patients, providers, insurers, and public financing bodies. These simultaneous changes are the central scientific problem of health care utilization responses to insurance coverage.

The fundamental unit is a person experiencing a changing coverage environment in a specific health system over a defined interval. Utilization can then be counted as any contact, distinct visits, prescriptions filled, days of hospitalization, procedures, preventive screenings, or recurring episodes of care. Each count means something different. A rise in an outpatient contact rate is distinguishable from longer treatment within the same contact; an observed increase in hospital payments is distinguishable from an increase in admissions because prices and case mix may change. Insurance generosity, network access, workforce capacity, and administrative eligibility must be recorded before a comparison has a stable interpretation.

The public-health economics of insurance has long considered uncertainty, clinical knowledge, provider agency, and financial risk. Arrow's analysis of uncertain medical care (1963) identified why health services differ from transactions in standardized commodities. The subsequent literature shows that making care cheaper to the patient changes utilization while offering protection against unpredictable costs. Nyman's access-motive account of insurance (1999) emphasizes access to treatment that a household could not buy outright. Both considerations belong in a complete evaluation of utilization responses, especially for high-cost or specialized treatment.

This dossier evaluates three linked empirical questions. First, how does obtaining or retaining insurance change the probability of using a service, and how does generosity change the number or intensity of services among users? Second, how do physicians, hospitals, pharmacies, and insurers mediate a patient's response through prices, organization, diagnosis, network design, and availability? Third, what can observed service use tell us about access, clinical benefit, financial protection, and equity when these outcomes operate on different time scales?

The evidence comprises primary peer-reviewed research, original institutional data documentation, and comparative reporting by health authorities. Randomized assignment receives particular attention for causal identification, followed by clearly specified quasi-experimental research. Observational associations are included only with their identification limitations. Figures retain their study years and denominators. No new formal theory, proprietary doctrine, original patient-level estimate, or claim of established scientific novelty is introduced. The original contribution consists of comparative synthesis, reconciliation of apparently discordant results, and an operational specification of what must be measured to evaluate insurance-induced changes responsibly.

Foundations and Source-Critical History

The object of a utilization response

An insurance-induced response is the difference between utilization under one coverage regime and utilization under a relevant counterfactual regime for the same target population. Coverage may be binary for an initial enrollment decision, but its institutional content is multidimensional. A formerly uninsured person may gain a public program with a specific provider network and no premium, or a private policy with deductibles and narrow specialist access. A continuously insured person may experience a higher copayment, an expanded benefit, a provider contract change, or an altered reimbursement schedule. These treatments cannot be collapsed into a single undifferentiated insurance indicator when causal interpretation matters.

One must distinguish insurance eligibility from actual coverage, coverage from knowledge of benefits, eligibility from the ability to locate an accepting practitioner, and service receipt from an improvement in health. An entitlement can exist with incomplete enrollment and high claim denial rates; a visit can take place without clinically appropriate follow-up; appropriate care may be unavailable even when it is formally covered. The scientific object therefore extends from the insured person's demand response to the provider's production of medical care, while maintaining utilization as the primary outcome rather than allowing the dossier to become a general survey of health insurance financing.

A concise conceptual inventory separates four potentially measurable layers. The exposure layer records plan membership, duration, copayments, annual deductibles, benefit exclusions and networks. The access layer records a usual source of care, appointment availability, travel time, accepted insurance, needed services delayed, and clinical indications. The use layer records the probability, frequency, intensity, mix, and continuity of encounters. The consequence layer records health, financial protection, out-of-pocket expenditure, patient experience, public spending, and equity. Each layer is observationally and conceptually distinct; drawing an inference across layers requires evidence of the connecting mechanism.

Historical development of the evidence

The RAND Health Insurance Experiment was a landmark because families were assigned to insurance plans with differing patient cost-sharing obligations. In their controlled-trial interim findings, Newhouse and colleagues (1981) observed greater utilization with free care through both more people using services and more services among users. The final randomized demand study by Manning and colleagues (1987) quantified the expenditure and elasticity response. The design isolated individual cost-sharing changes in a historical U.S. fee-for-service setting; contemporary plans differ in provider concentration, pharmaceutical innovation, managed care, and out-of-network restrictions.

The introduction of U.S. Medicare provided a different type of evidence: a large historical expansion of public financing for older adults. Finkelstein and McKnight's initial Medicare impact study (2008) identifies substantial financial risk protection from the program's adoption but no discernible aggregate elderly mortality impact in its first ten years. Card, Dobkin, and Maestas (2008), examining the modern age-65 eligibility threshold, document increases in service use in their Medicare utilization study. These investigations differ in time, treatment contrast, population, and outcome. They furnish complementary historical observations rather than competing estimates of one immutable Medicare effect.

The 2008 Oregon Medicaid lottery introduced randomized encouragement to apply for publicly financed coverage among previously uninsured low-income adults. The first-year Oregon study by Finkelstein and colleagues (2011) and the subsequent clinical outcome experiment by Baicker and colleagues (2013) distinguish access, use, financial burden, and health measures under different follow-up intervals. Oregon's specific lottery mechanism matters: selected individuals could apply, and only eligible applicants were enrolled. Assignment to the lottery and actual insurance receipt represent distinct treatment variables.

Further reform evidence uses geographic and age thresholds, staggered Medicaid expansion, employer plan changes, and cluster-level program introduction. These designs bring wider institutional coverage but also more identifying assumptions. A central historical lesson is that causal knowledge has progressed through diversification of research designs, not through the discovery of one constant utilization elasticity transferable across diseases and systems.

Research design and the meaning of a causal effect

Random assignment identifies an intention-to-treat effect of an offer or assignment when treatment allocation and outcome measurement are preserved. If a coverage lottery increases enrollment for some people, dividing the reduced-form effect on utilization by the enrollment first stage may identify an instrumental-variable local average treatment effect under exclusion, monotonicity, and other design-specific assumptions. The first is the effect of receiving the opportunity to obtain insurance; the second concerns the effect of coverage on those whose enrollment changed because of the offer. Their values and policy populations are different.

A regression discontinuity at a statutory eligibility threshold compares people close to that threshold and depends on continuity of other determinants and a well-characterized policy bundle. Difference-in-differences compares outcome changes across exposed and unexposed groups; it depends on defensible untreated counterfactual trends and is vulnerable to simultaneous reforms. Event studies, falsification tests and pre-policy trends can test implications of a design, though they cannot guarantee the identifying assumption.

Survey comparisons among insured and uninsured individuals require special caution. Illness may cause both insurance enrollment and additional service use. Earnings may determine eligibility, ability to pay, and regional access. Motivation or health literacy may affect both enrolling and seeking preventive care. Even sophisticated matching cannot necessarily remove unmeasured selection. In the Ghanaian study of national insurance and care seeking, Blanchet, Fink, and Osei-Akoto (2012) apply propensity-score methods and find higher formal care and prescription use among enrollees; these results provide relevant context while remaining exposed to unobserved confounding.

Extensive and Intensive Utilization

Coverage, price, access, and formal decomposition

Let Y be the nonnegative number of a defined medical service event per eligible person during a fixed follow-up period; I(Y > 0) indicates any use. The population mean admits the identity:

E[Y] = Pr(Y > 0) × E[Y | Y > 0].

This is an accounting identity, not an independently proposed theoretical model. It distinguishes an extensive margin, the probability of at least one use, and an intensive margin, the mean quantity among observed users. A coverage change may alter either or both terms. Because the set of users can itself change, the second conditional mean is defined over different populations before and after treatment. A naive comparison of intensity among users may therefore contain selection even when assignment is randomized. Principal-stratum inference or an estimand explicitly conditional on post-treatment use is required to interpret the intensive margin causally.

For expenditure on service type k, an additional accounting decomposition is helpful: total paid spending equals the sum over services of quantity multiplied by the corresponding transacted prices and payment components, subject to consistent unit definitions. A rise in spending is not itself proof that more encounters occurred. Rates for treated patients, drug prices, clinical severity, visit coding, and insurer payments can shift even when visits remain unchanged. The clinical intensity of an encounter—diagnostic tests, medicines, procedures, referrals—should be recorded separately from the number of encounters.

The patient's relevant price is a schedule, not ordinarily a single constant. Under coinsurance, the patient pays a share of an allowed price; under a fixed copayment, a nominal amount per service; under a deductible, the marginal price may fall sharply after annual expenditure reaches a threshold. A maximum annual out-of-pocket obligation may alter subsequent marginal prices. Expectations about future care, household liquidity, informal payments, benefit comprehension, and provider billing also influence the perceived price. Estimating an elasticity without identifying the price concept and population leaves the estimand ambiguous.

A local price elasticity can be defined conventionally as the proportional change in service demand divided by the proportional change in the relevant patient price for a specified range. It should not automatically be used to predict behavior between free care and a high deductible, between insured and uninsured states, or for one-time catastrophic surgery. The marginal price under zero cost sharing and the access value of an otherwise unaffordable treatment complicate simplistic constant-elasticity extrapolation.

RAND and the response to cost sharing

The RAND trial provides direct causal evidence that consumers respond to the financial terms of insurance. Newhouse and colleagues' interim study reported roughly 50 percent more spending under full coverage than under income-related catastrophic coverage in the interim comparison, while its other results showed that once hospitalized, expenditure per admission did not differ significantly across experimental plans. The final RAND demand estimates (Manning et al., 1987) found that a catastrophic policy reduced medical spending by approximately 31 percent relative to free care, with a commonly reported price elasticity near −0.2. The different numerical contrasts refer to different samples, designs, periods and plan comparisons within the research program; they must be quoted with their original definition rather than averaged mechanically.

The RAND evidence also challenges two misleading shortcuts. First, the fact that prices affect utilization does not by itself determine whether the forgone encounters were beneficial or low value. Second, inpatient and outpatient care are linked clinically, but a reduction in outpatient use does not imply an automatic rise in hospital expenditure of equal or greater magnitude. Manning and colleagues rejected an aggregate hospitalization-offset hypothesis for their experimental conditions. The claim is specific to that study's population, time, benefit structures, and statistical power.

Clinical subgroup results sharpen the distinction. Brook and colleagues' randomized adult health assessment (1983) found that an undifferentiated claim of improvement in all measured health dimensions was unsupported. Keeler and colleagues' hypertension analysis (1985), however, identified lower blood pressure among hypertensive patients receiving free care, particularly those with lower incomes. The relevant comparison is therefore clinically indicated services and outcomes among populations for whom affordability constrains care, not simply the overall count of encounters.

For interpretation today, the RAND experiment remains foundational for patient cost-sharing responsiveness. A modern analysis must separately examine organized provider payments, drug formularies, digital consultations, chronic disease regimens and contemporary network restrictions. Extrapolating a single parameter across these systems would discard the most valuable methodological lesson from the experiment: price effects are defined by a treatment contrast and a medical service environment.

The Oregon lottery and new coverage

Oregon offers evidence about the gain of publicly financed insurance among low-income previously uninsured adults. In the first-year research, being selected by lottery raised Medicaid coverage probability by approximately 25 percentage points; the selected group had higher use of primary and preventive care and hospital services and lower financial strain. The randomized offer is therefore a clean source of variation, but the estimate attributable to actual coverage requires accounting for incomplete enrollment. An improvement in care among lottery winners is an intention-to-treat effect unless an instrumental-variable conversion with explicit assumptions is used.

Baicker and colleagues' two-year Oregon clinical evaluation (2013) measured outcomes in 6,387 adults selected through the lottery and 5,842 not selected. Medicaid increased several preventive services, diabetes detection and use of diabetes medication, and substantially reduced catastrophic out-of-pocket expenditure. Average blood pressure, cholesterol, and glycemic control did not show statistically significant improvement over the measured interval. The proportion screening positive for depression declined by approximately 9.2 percentage points, or 30 percent relative to the control group. The absence of a statistically significant average change for the specified physical-health outcomes is a limited empirical result; it does not demonstrate that every potentially important physical-health effect is zero or that use produced no value.

The emergency department provides especially instructive evidence. Using hospital records for approximately 25,000 lottery participants over roughly 18 months, Taubman and colleagues' randomized emergency-use study (2014) estimated an increase of 0.41 emergency-department visits per person, approximately 40 percent relative to the 1.02-visit control mean as reported in the original study. This result is an estimate for a particular Medicaid expansion, setting, care market, and period. The later Oregon emergency follow-up (Finkelstein et al., 2016) reported continued elevated emergency use through a longer follow-up period. The available evidence rejects the assumption that gaining Medicaid immediately substitutes physician office care for all emergency encounters.

Oregon reconciles logically consistent outcomes: insurance can increase access and service use, improve the detection and treatment of some conditions, reduce risk of catastrophic spending, and still show limited detectable effects on selected average biomarkers after two years. These outcomes are measured on different scales with different precision. A patient may receive valuable diagnosis and financial protection before a chronic biomarker registers a measurable improvement; a randomized study may also have limited power for rare or slow-moving health outcomes. Scientific inference must respect both possibilities.

Medicare and clinical service composition

At age 65 in the United States, Medicare eligibility creates a sharp insurance transition for many people. Card, Dobkin, and Maestas (2008) found increases in routine doctor visits and hospital admissions, with different patterns by previous insurance status and procedure. Their study shows that a new statutory entitlement interacts with prior coverage and supplementary generosity. For previously uninsured individuals, routine access is particularly salient; for some previously insured individuals, supplementary coverage may make expensive elective procedures more accessible. The observation concerns the Medicare policy bundle and should not be interpreted as an isolated effect of an insurance card independent of benefit design.

The same researchers investigated seriously ill patients admitted through emergency departments near age 65. Their hospital-treatment mortality analysis (2009) identified small but significant changes in hospital procedures and charges and estimated an approximately one-percentage-point decrease in seven-day mortality at the Medicare threshold for a carefully defined severely ill population. This localized result differs from Finkelstein and McKnight's historical finding of no discernible aggregate mortality change following Medicare's introduction. The first examines acute treatment for selected emergency admissions near an age threshold; the second examines population-level mortality over an earlier decade. Reconciling the two does not require discarding either study; it requires preserving their distinct estimands.

A defensible evaluation records whether measured utilization consists of a preventive appointment, repeated chronic-care visit, emergency episode, elective operation, hospital stay, or high-cost intervention. Their clinical needs, price sensitivity, and feasible substitutes differ. The evidence supports service-specific rather than uniform predictions about the sign, speed, and health consequences of coverage effects.

Provider and Patient Responses

Clinical decisions, demand, and welfare

Medical care utilization is jointly determined. The patient recognizes a symptom or preventive need, chooses whether and where to seek care, and supplies time, transportation, consent, and any required cost sharing. Providers determine diagnostic intensity, treatment plans, referrals, follow-up intervals, coding and, in some financing arrangements, the services billed to a third-party payer. Insurers and public purchasers influence the transaction through covered benefits, fee schedules, authorization, networks, claims procedures, quality requirements, and provider payment contracts. A change in coverage may affect several decisions at once.

The resulting encounter is a negotiated clinical and institutional outcome rather than a direct expression of a patient's unconstrained demand. When a plan increases reimbursement or permits previously unaffordable treatment, a physician may recommend an intervention that was previously deferred. When the plan imposes a restrictive network, a patient may appear to have low claims utilization despite persistent unmet need. Referral rules or authorization delay can alter the timing of the same treatment, shifting its observed count into another calendar year. Medical necessity, treatment indication, provider availability, and billing practices therefore belong in the data specification.

Health insurance has at least two established channels of economic value. The risk-protection channel transfers uncertain medical expenditure liability from the individual to an organized pool. The access channel finances treatment beyond immediate household resources. Nyman's (1999) account makes the latter explicit, while the literature initiated by Arrow (1963) emphasizes uncertainty, asymmetric information, and professional relationships. An increase in medically indicated, previously inaccessible services can improve welfare even if it raises measured expenditure, and a reduction in unnecessary procedures may improve value without decreasing medical need.

Evaluation requires both the cost of real resources and the outcome obtained. Insurer reimbursement is a transfer to providers for accounting purposes but also finances the provider capacity that produces services. Patient cost sharing is a transfer component with a distributional and behavioral effect. Hospital staff time, drug ingredients, imaging capacity, and opportunity costs are resources. A short-run change in claims spending aggregates changes in quantity, price, intensity, and risk composition; it cannot substitute for a welfare calculation. A cost-effectiveness evaluation additionally needs credible information on clinical outcomes and intervention-specific value.

Prescription use, prevention, and chronic disease

Prescription utilization illustrates the difference between access, adherence, and treatment outcomes. Filling a prescription is an observable event; possessing the medicine does not prove ingestion; a dispensed dose does not establish that treatment was indicated or effective. Insurance may alter the price paid, formularies, prior authorization, pharmacy location, clinician prescribing, and continuity of coverage. A statistically significant rise in pharmacy claims following enrollment can reflect treatment initiation, better refill continuity, expanded coding, or substitution among drugs. Each mechanism must be assessed using appropriate medication histories, clinical indications, and follow-up data.

Oregon's increase in diabetes diagnosis and medication use after Medicaid acquisition is a concrete instance of a clinically meaningful diagnostic pathway without a clearly detectable contemporaneous change in mean glycated hemoglobin. This sequence is compatible with discovery and initiation of treatment before a measurable health response, but the specific causal claim of delayed biomarker benefit requires longer-term outcome evidence. Researchers should report diagnosis rates separately from disease incidence and should account for surveillance effects: more visits reveal previously undiagnosed disease, increasing recorded prevalence without implying a new wave of disease.

Evidence from U.S. reform settings reinforces the distinction. Wherry and Miller's early Medicaid expansion study (2016), using national survey comparisons around the 2014 expansion, reported changes in primary care visits, overnight stays, and newly diagnosed conditions among low-income adults. Sommers and colleagues' three-state coverage follow-up (2017) associated sustained expansion with improved access to medical care, affordability, and selected medication and preventive-care outcomes. These findings have different survey instruments, geographic treatment definitions, and statistical uncertainties from Oregon's randomized comparisons. They should be read as evidence extending the inquiry to state policy changes, not as exact replications of lottery estimates.

Preventive utilization is similarly heterogeneous. Vaccinations, blood-pressure screening, cancer screening, prenatal visits, and routine checkups differ in clinical evidence, appropriate intervals, availability, and downstream consequences. Increased screening may improve early detection; it may also lead to false positives, follow-up procedures, overdiagnosis, or incidental costs. Evaluation should use intervention-specific evidence of benefits and harms rather than treating all preventive visits as equivalent beneficial units. Underinsurance can suppress effective prevention, while a zero-price benefit can raise both appropriate and low-value testing depending on provider and patient incentives.

For people with established chronic disease, repeated contact may represent successful management. Continuity and clinical control can matter more than the count of visits. Where geography or workforce shortages create long delays, a nominally generous insurance benefit may have little effect on adherence because treatment capacity remains scarce. Researchers must therefore distinguish patient financial barriers from pharmacy stock availability, specialist schedules, and the care delivery infrastructure.

Hospital systems, employment, and payer incentives

Hospitals respond to insurance-induced demand through capacity allocation, staffing, scheduling, service specialization, negotiated payment rates, and collection practices. Their responses are often slower than the individual decision to schedule a visit. Expanded insurance can reduce uncompensated care, improve payment reliability, make service expansion financially feasible, and increase the demand placed upon emergency and outpatient departments. Whether total hospital activity increases depends on available beds, clinical workforce, capital constraints, and substitution between outpatient and inpatient settings.

Meille and Post's hospital response to Medicaid expansion (2023) used U.S. hospital data for 2010–2016 and reported increased emergency-department and outpatient visits in expansion states along with evidence of increased nursing staff employment. Their estimates indicate that supply-side adjustment can accompany changed coverage. An observed utilization difference following expansion may therefore include both demand changes and provider production responses, and these responses should not be mistaken for statistical confounding when they are genuine causal mediators of the policy.

Provider payment design influences the direction and intensity of those mediating responses. Fee-for-service remuneration can create incentives to expand separately billable contacts or procedures. Capitation and global budgets may encourage the management of care within a fixed financial envelope but require safeguards against insufficient services and risk selection. Episode-based payment can alter admissions, readmissions, length of stay, and coding. A coverage expansion under one payment regime is consequently a different institutional intervention from expansion under another. These incentive channels require local contract and claims evidence; their existence alone is not proof that a particular provider changed practice.

Kolstad and Kowalski's Massachusetts hospital-reform analysis (2012) found changes in hospital insurance composition, length of stay, and emergency-origin admissions after the state's 2006 reforms. This evidence should not be merged mechanically with Oregon's rise in total emergency-department contacts. A reduction in admissions originating in emergency departments measures a different pathway from the total number of emergency visits; system organization, timing, and insured population also differ. The apparent contrast is an instructive example of how care-setting transitions can coexist with different utilization patterns.

Insurance contracts affect provider revenue, but utilization analysis should avoid equating an increase in reimbursed encounters with an increase in total care. Some newly covered encounters replace previously uncompensated care, informal self-treatment, charity care, or privately financed visits. Hospital billing records may show higher insured claims even if the total number of patients changes much less. Linked all-payer encounters, household self-report, and provider files can help distinguish displacement, formalization, and genuine incremental utilization.

Behavioral mechanisms and treatment substitution

The conventional moral-hazard account considers how lower patient cost-sharing increases demand when a person does not bear the full social resource cost of a medical service. Price responsiveness can occur for necessary and unnecessary services alike. Its existence does not determine the clinical value of the marginal care. Baicker, Mullainathan, and Schwartzstein's behavioral-hazard analysis (2015) incorporates mistakes and limited decision quality into established health insurance welfare theory. It explains why even insured individuals may underuse high-value care and why a higher patient price can cause clinically harmful reductions. This is an existing research contribution, not a new model introduced by the present monograph.

An especially important empirical test comes from Brot-Goldberg and colleagues' deductible-switch study (2017). Employees of a large self-insured firm moved from a plan with free care to a high-deductible arrangement. The study reported an 11.8–13.8 percent reduction in aggregate firm-wide spending, attributed to reduced service quantities, with no evidence of price shopping after two years. Service reductions spanned categories with differing likely value. The study establishes strong behavioral responsiveness for the observed workforce and benefit change, but its employees, prior zero-price baseline, employer conditions, and nonlinear deductible design do not describe every privately insured or publicly insured population.

Treatment substitution cannot be assessed by a simple outpatient-versus-hospital binary. Patients may switch between branded and generic medicines, clinic and telehealth encounters, early and delayed diagnostic testing, urgent care and emergency departments, or formal care and home management. A reduction in one setting can be accompanied by improved care elsewhere, a delayed costly complication, or no material change in health. The direction is an empirical question requiring linked outcomes and suitable observation windows.

Chandra, Gruber, and McKnight's elderly cost-sharing offset study (2010) examined a policy change affecting retired California public employees' supplemental coverage. Physician visits and prescription use fell with higher cost-sharing, while hospital utilization rose, imposing costs largely on Medicare. This contrasts with RAND's lack of a general hospitalization offset under its own historical experimental conditions. A possible explanation is the difference in patient age, baseline chronic disease, benefit arrangements, and coordination between supplemental and primary payers. These factors are hypotheses about heterogeneity consistent with study contexts; their precise contributions are not identified by simply comparing coefficients across the two papers.

The proper analytical conclusion is that health care utilization is a system of clinically differentiated responses. Changes in utilization can move downstream costs between settings and financing entities. A policy that reports savings to one insurer but greater hospitalization to another requires an all-payer accounting to establish net system effects.

Effects of Access Constraints

Financial and administrative impediments

Insurance coverage can remove important financial barriers while leaving administrative obstacles intact. Premium costs affect joining and retaining a plan; deductible obligations affect the decision to seek care even after membership is established; prescription copayments affect continued treatment; network restrictions and medical necessity review alter the set of practically reimbursable providers. Coverage interruptions caused by eligibility reassessment or employment change can disrupt long-term care even if the individual returns to insured status later.

Financial protection and utilization have separate dimensions. A household may gain substantially through lower variance and lower tail exposure of medical expenditure without making additional visits. The 2008 Medicare history study and Oregon clinical study both support this separation. Conversely, a person may receive multiple new services but still incur high expenditures from uncovered medicines, transport, out-of-network bills or household income loss. Reporting only a coverage rate conceals such residual financial constraints.

Out-of-pocket exposure must be defined from actual financial responsibility rather than charges alone. The patient amount can comprise deductibles, coinsurance, copayments, noncovered services, and costs outside a formal maximum. Under administrative claims, amounts billed, allowed, paid by insurer, and paid by household may differ substantially. Subsidies, informal financial support, unpaid debt, and deferred bills can change economic burden without changing coded visits. A consistent analysis therefore connects coverage and utilization to a separate financial-protection outcome, while preserving their distinct denominators.

A study of eligibility-to-enrollment transitions must also measure take-up. If reform eligibility rises but eligible people fail to register, the observed intention-to-treat effect on service use may be small despite a sizable effect of actual coverage among compliers. If an enrollment process simultaneously provides navigation assistance, observed changes can reflect the combined intervention. Such effects are policy relevant, although attributing them solely to an insurance card would misstate the mechanism.

Waiting times, networks, and geographical supply

Medical care requires available professionals, facilities, equipment, and medicines. Coverage expansion can increase expressed demand faster than supply grows. The consequence may be longer waits, changed provider mix, triage, or geographical displacement. When the supply of an essential service is fixed over the evaluation window, insurance can reallocate available appointment slots between patients; an observed gain among newly insured patients may coincide with a decline in timely access for others. The net effect must be measured at an appropriate population and provider-market level, not presumed from individual insured-patient claims alone.

The OECD's Health at a Glance 2025 distinguishes cost, distance, and waiting-time constraints. Its detailed unmet-needs chapter reports that, in 2024, an average of 3.4 percent across 28 countries with comparable data reported unmet medical care needs because of these constraints; waiting time alone accounted for an average of 2.3 percent. The average share with unmet needs in the lowest income quintile was about 2.5 times that in the highest quintile. These are measured self-reported needs with a defined set of countries and years, not causal estimates of the effect of coverage expansion or all-world incidence.

An insured person may be unable to find a local clinician willing to accept the plan's payment rate. Network adequacy and geographical reach can thus mediate treatment. Researchers should measure provider participation and appointment lead times, including whether the source reports legal network membership or actual capacity for appointments. Travel time, language access, disability accommodation, telehealth availability, and medication distribution can alter effective access. The appropriate interpretation varies with the treatment being studied: expansion of financial coverage alone differs from a reform that also hires medical staff and funds hospitals.

A further challenge is that unmet need and utilization can move in the same direction. Previously untreated individuals may become aware of additional required care once a clinician evaluates them, raising reported need while utilization also grows. Alternatively, a fall in recorded unmet need could result from discouraged responses or changed survey expectations rather than improved clinical treatment. Combining self-report, administrative encounters, clinical indications, and waiting measures improves interpretation.

International comparative experiments and reforms

The international evidence shows how financing and delivery arrangements condition causal responses. Mexico's Seguro Popular expansion was evaluated through cluster-level randomized encouragement and accompanying facility improvements. King's Mexican randomized assessment (2009) documented reduced catastrophic expenditure but no detected increase in utilization, medication spending, or health outcomes over approximately ten months. That result cannot be generalized to permanent long-term absence of use effects. The intervention combined insurance encouragement and provider upgrading, had incomplete compliance, and was evaluated at a short horizon. Its principal significance for this dossier is that financial protection can appear sooner or more strongly than measured use.

Thailand's 2001 “30 Baht” reform changed user payments and directed additional funding to hospital providers. Gruber, Hendren, and Townsend's Thailand infant-mortality study (2014) found increased utilization, especially in poorer areas, and a reduction in the prior association between provincial poverty and infant mortality. Their design highlights the role of supply-side hospital funding; the results should not be attributed exclusively to a lower copayment. Thailand demonstrates that coverage policy and provider financing can reinforce each other when a government organizes both sides of access.

Ghana's National Health Insurance Scheme presents additional evidence from a financing transition toward formal coverage. Blanchet, Fink, and Osei-Akoto's (2012) observational study reported higher probabilities of clinic attendance, formal treatment seeking, and prescription use among insured women in Accra, after propensity-score adjustment. These associations are substantively consistent with access improvements but cannot fully eliminate differences in baseline health, socioeconomic position, proximity, or enrollment selection. Their evidential weight differs from randomized trials; retaining them broadens international institutional understanding without inventing causal precision.

U.S. reforms also display variation across settings. Massachusetts expanded coverage across a broad population under one policy package, and mortality after Massachusetts reform (Sommers, Long, and Baicker, 2014) was evaluated using a geographically comparative quasi-experimental design. This ecological treatment contrast is affected by assumptions about comparable trends, migration, and concurrent care-system changes. The study's reported association with reduced mortality is relevant to long-run health consequences but must be interpreted with its methods, rather than converted into an estimated number of encounters required to save one life.

The U.S. ACA Medicaid expansions provide yet another treatment contrast: low-income adults in states expanding eligibility compared with those in other states. Their quasi-experimental findings generally support improved coverage and use of selected services, but they are subject to different pretrends, state choices, administrative rules, managed care arrangements and post-policy measurement periods. Distinct national and subnational systems provide evidence of heterogeneity; they do not create one globally invariant effect size.

Distributional and longitudinal consequences

Utilization changes are often largest where prior barriers were highest. Previously uninsured patients, low-income households, people with untreated chronic illness, rural residents, and individuals requiring specialist procedures may respond along different margins. A positive average treatment effect may conceal a large gain among one subgroup and negligible or negative changes elsewhere. Yet subgroup estimates are frequently less precise and exposed to multiple-testing problems. Heterogeneity claims should specify whether they were prespecified, the comparison was statistically tested, and the sample supports estimation.

The distributional question has several levels. A household experiences out-of-pocket changes, care access, health effects and time costs; a provider experiences new revenue or increased pressure; a public insurer or employer experiences changes in budget obligations; a society experiences effects on health, productivity and fiscal resources. Transfers between financial entities can cause reported program savings to diverge from total costs. These incidence relationships matter especially when an expanded public program reduces charity care or when supplemental cost-sharing moves hospitalization expenditure onto a primary public payer.

Time modifies nearly every mechanism. In the first months after enrollment, previously deferred appointments and diagnoses may generate a temporary utilization surge. Thereafter, improved continuity can stabilize follow-up, chronic disease treatment, and preventive care. Some interventions avert later complications; others reveal new need, leading to sustained utilization. Hospital staffing and capital investment occur on slower schedules. A credible longitudinal design should specify immediate, medium-term and longer-term estimands instead of interpreting a six-month response as a steady-state equilibrium.

The 2025 WHO–World Bank UHC report places national utilization studies in a broader setting. Its revised indicators show a world service coverage index of 71 in 2023, with approximately 4.6 billion people lacking full coverage for essential health services, and financial hardship affecting about 2.1 billion people in 2022. These statistics describe separate global service and financial-protection dimensions. They do not tell us how many extra visits an insurance reform would cause or whether any individual patient's additional encounter was clinically valuable; they establish the scale and unevenness of unmet protection motivating evidence-based evaluations.

Data, Measurement, and Validation

Estimands, counterfactuals, and statistical inference

The empirical statement that insurance “raises utilization” is incomplete until the causal quantity has been specified. At least four estimands commonly appear in coverage research: the effect of eligibility or an offer, the effect of enrollment among those whose coverage changes because of the offer, the effect of a particular plan design among already enrolled people, and the population effect of a reform that changes provider financing or service capacity as well as insurance. These are scientifically distinct interventions. A randomized offer can identify an intention-to-treat difference between assigned groups even when enrollment is incomplete. Under the exclusion, monotonicity, and relevance assumptions, the offer can sometimes instrument actual enrollment to identify a local average treatment effect among compliers. The interpretation requires precisely defined populations and careful attention to whether the offer itself changes knowledge of benefits, appointment navigation, or care-seeking behavior.

For a binary outcome such as any physician contact during a year, the outcome is the fraction of people with one or more recorded events. For a count outcome, it is the number of contacts per person in that year, retaining zeros for those with no contacts. An intensive-margin outcome conditions on having a contact. Comparing intensity across two treatment groups after conditioning on the post-treatment event of any use can introduce selection: becoming a user is itself affected by coverage. Separate analyses of extensive and unconditional counts, supplemented by suitable principal-stratum or distributional assumptions where defensible, provide a clearer causal interpretation than treating the observed users in both groups as exchangeable.

In a stylized randomized design, let Z indicate an insurance offer, D actual insurance receipt, and Y a utilization outcome. The offer effect is E[Y|Z=1] − E[Y|Z=0]. If the design supports an instrumental-variable interpretation, the coverage effect for compliers is represented by the ratio of the offer effect to E[D|Z=1] − E[D|Z=0]. This ratio requires both numerator and first-stage denominator to be estimated on compatible populations and observation periods. Oregon's first-year experimental analysis (Finkelstein et al., 2011) illustrates why reporting the first stage is essential: the lottery's effect on obtaining coverage was approximately 25 percentage points, so an effect among lottery winners and an estimated effect of actual coverage have different magnitudes and meanings. The ratio is a formal identification expression, not a new analytical model.

In discontinuity designs, investigators compare people near a precisely defined policy threshold while assuming potential outcomes would otherwise evolve smoothly. Card, Dobkin, and Maestas's Medicare utilization study (2008) uses the age-65 threshold to examine changes in insurance and treatment. Such a discontinuity estimates a local effect for people close to age 65 under institutional conditions operating at that threshold. It does not identify treatment effects for younger children, working-age adults far from retirement, or people living in differently financed systems. Age-related health gradients, anticipatory scheduling, program changes and overlapping retirement decisions should be examined as threats to the specific comparison.

Difference-in-differences designs compare changes over time in reform and comparison populations. Their credibility depends on a defensible counterfactual trend, appropriate handling of changes in composition, and attention to anticipation and concurrent policies. State-level Medicaid expansion comparisons include potential migration, differing economic conditions and preexisting trends in access. Wherry and Miller's ACA coverage evaluation (2016) and Sommers and colleagues' three-state access assessment (2017) identify outcomes using different designs and respondent samples; they should be compared by outcome definition and assumptions before their estimated effects are juxtaposed. Confidence intervals, clustered or robust standard errors, prespecified primary outcomes, attrition patterns and the number of comparisons matter as much as the point estimates.

The inferential distinction between a precisely estimated zero, a confidence interval that includes zero, and an estimate too imprecise to discriminate material gains from losses is essential. Oregon's two-year clinical evaluation (Baicker et al., 2013) found increased use and improved protection but did not detect statistically significant average changes in selected biomarkers. These results support a conclusion about measured effects and study power under its design; they do not prove the permanent ineffectiveness of the health care newly received. Equally, a statistically detectable change in appointments does not demonstrate better health unless outcomes and causal timing are measured appropriately.

Household surveys, claims, and administrative records

Measurement begins with data generation. Household surveys collect insurance status, self-reported doctor visits, medication use, unmet need, expenditures, family income, access perceptions, and sometimes diagnoses. The U.S. Agency for Healthcare Research and Quality documents the sample design, interview structure and medical-event components of the Medical Expenditure Panel Survey. Such surveys can connect household financial conditions to utilization beyond the universe of one insurer's claims. Their limitations include recall errors, sample attrition, changing survey modes, and differences between respondents' perception of coverage and administrative eligibility.

Administrative insurance files record paid or adjudicated claims, enrollment periods, coded clinical encounters and insurer payment. They offer high-frequency observation and larger samples for relatively rare events, but they capture the payer's reporting universe rather than every episode of care. Out-of-network spending, unpaid services, informal care, services below a deductible, denied claims, bundled payments and privately paid medicines may be missing or coded differently. Medicaid T-MSIS Analytic Files document a major federal effort to standardize state submissions; their use still calls for validation of state-year completeness, enrollment categories, managed-care encounter reporting and diagnostic coding.

Hospital discharge records count admissions in participating institutions rather than all clinically indicated care. Pharmacy claims measure paid dispensings or fills, not ingestion, adherence, or therapeutic response. Electronic health records can supply laboratory values and clinician decisions while excluding care provided outside the health system. Mortality records permit linkage to a hard endpoint but may vary in completeness and causes-of-death coding. Population denominators are therefore integral to data interpretation. An encounter rate computed among active enrollees is sensitive to plan switching; a rate among all eligible residents measures a different object; an annualized rate among members enrolled for at least twelve months selects a more stable, potentially healthier population.

Linkage between administrative insurance records, surveys and clinical data permits questions that no single source resolves. It can connect eligibility to a specific appointment, treatment to blood pressure, or insurance expansion to mortality. Miller, Johnson, and Wherry's Medicaid mortality analysis (2021), using survey data linked to death records, reported an estimated annual mortality reduction of 0.132 percentage points among the defined near-elderly population, approximately 9.4 percent relative to the sample mean. Its causal interpretation rests on the comparison of states and identification of likely eligible individuals; it does not make the numerical estimate transferable to all Medicaid populations or reveal which specific treatments produced the reported reduction. The study adds an important longer-horizon dimension to assessments focused only on visits.

Cross-national comparisons require another layer of documentation. WHO's Global Health Expenditure Database organizes financing aggregates across countries using health-accounting conventions. Those aggregates describe expenditure and funding, rather than unique patient encounters. National health insurance benefit definitions, public spending, fee schedules, informal payments and survey reporting conventions may differ even when an identical term such as “covered” is used. Converting expenditure totals into visits through average prices is generally unreliable when intensity, case mix and negotiated prices vary. International synthesis should therefore compare the concepts and mechanisms that can be matched before comparing numerical effect sizes.

Measurement consistency and source limitations

Three kinds of denominators must be kept separate. Financial protection denominators typically refer to households and household resources. Utilization denominators refer to persons, visits, treatment episodes or covered member-months. Clinical outcome denominators refer to eligible patients with a defined diagnosis or baseline risk. A care-access study can legitimately use all three, but its reported percentages and effect sizes are not interchangeable. For example, an increase of 0.41 emergency department visits per person across an approximately 18-month observation period, as reported in the Oregon emergency study (Taubman et al., 2014), is a count difference during that study interval. It must not be represented as an annual increment without an explicit conversion, and even a conversion would assume a temporal pattern that may be inappropriate.

Historical research creates additional comparability issues. The RAND experiment tested particular indemnity plans and cost-sharing schedules in an earlier U.S. delivery environment. Its final randomized demand estimates (Manning et al., 1987) are strong evidence of responsiveness in the tested contracts, but modern managed care, high-cost specialty pharmaceuticals, and provider-consolidation arrangements produce different effective prices and constraints. The natural experiment of Medicare's introduction differs from the insurance expansions of the 2010s, and the two should be separated rather than combined mechanically into a single effect without accounting for policy and population differences.

Some outcome definitions are themselves subject to clinical and institutional change. An emergency department visit can be categorized as nonurgent or preventable using diagnostic algorithms, but severity is incompletely captured by discharge coding. An outpatient visit can include a traditional appointment, telehealth contact or brief integrated care episode. Counting all as identical encounters may conceal changes in resource intensity. Preventive care changes as guidelines evolve; a screening classified as recommended in one year may be restricted to a narrower risk group later. A credible comparison archives definitions and, where possible, restates older observations using stable groupings.

The WHO–World Bank UHC monitoring report illustrates why even well-governed indicators need explicit interpretation. Its 2023 service coverage index of 71 is a composite health-system score, not a statement that exactly 71 percent of the world's people obtained all essential services. Its separate estimate of 4.6 billion people lacking complete essential health service coverage applies a distinct coverage conversion. The estimate of 2.1 billion people experiencing financial hardship relates to 2022 household out-of-pocket spending and has a different data basis. Combining these numbers into a one-dimensional insurance effect would erase both the definitions and the evidence limitations disclosed in the original report.

Sensitivity, falsification, and evidence grading

The first validation task is internal consistency. Researchers should reproduce the published sample sizes, assigned and enrolled group definitions, baseline rates, percentage-point changes, ratios, follow-up dates, and denominators from the original tables and methods whenever source access permits. A large relative difference can be a small absolute change when the control prevalence is low; an aggregate decline in spending can reflect prices rather than quantities. The RAND, Oregon, Medicare and deductible studies should be reported with the exact experiment, coverage contrast and service outcome linked to each estimate.

The second task is design falsification. Randomized studies warrant checks for balance at assignment, missing outcome records, attrition correlated with treatment, contamination of controls and compliance with assignments. Threshold-based designs warrant density and covariate continuity checks, bandwidth sensitivity and plausible alternative thresholds. Difference-in-differences work warrants event-study pretrend inspection, sensitivity to comparison populations, reform timing, and region-level correlation. Observational matching requires explicit covariate overlap and assessment of the possibility that unmeasured care preferences or illness risk simultaneously affect insurance enrollment and service use. These tests do not guarantee correct causal identification; they document the degree of evidentiary support.

The third task concerns multiple outcomes. Coverage research often reports physician contacts, emergency visits, hospitalizations, prescriptions, screening, self-reported health, biomarkers, financial distress and mortality. Reporting only statistically favorable outcomes creates selection risk. A complete synthesis records positive, null, and adverse estimates; the units and confidence intervals; and which outcomes were the original investigators' primary endpoints. Subgroup claims require additional restraint when based on small samples. A measured use increase may have strong causal support and uncertain welfare consequences; those are compatible statements.

The fourth task is institutional external validity. A randomized estimate from low-income adults in Oregon during a particular Medicaid expansion has strong internal identification yet a specific treatment and market context. Mexico's experimental evaluation, Thailand's financing reform, and Ghana's observational insurance study differ in exposure, delivery system, sample selection, and measured endpoints. Evidence grading should reflect this combination of causal credibility and context match. In general, directly verified randomized contrasts carry more causal weight for the exact treatment evaluated than a contemporaneous correlation, while detailed official national documentation carries greater weight for a country's statutory coverage definitions than an external summary written for another jurisdiction.

Finally, a comprehensive audit asks whether every cited link identifies the actual work invoked and whether every externally supported claim has a corresponding full bibliography entry. Names, years, institutional authors, journal volume, page span, DOI, link stability and article titles are substantive traceability elements. This dossier records its sources and their verification basis in a companion evidence file. It presents methodological findings at the level supported by original research records, without pretending that literature synthesis constitutes a newly conducted randomized trial, a new administrative database analysis or independent replication of the authors' raw data.

International Synthesis and Scientific Conclusions

Convergent findings and meaningful disagreements

Across major randomized studies, moving from higher to lower direct patient prices generally increases the use of some medical services. This result is robust in its direction for many contacts but conditional in scale and service composition. The RAND randomized cost-sharing findings (Manning et al., 1987) quantify a price response in a set of historical plans. The Oregon randomized Medicaid evidence (Baicker et al., 2013) establishes the behavioral consequences of obtaining public insurance for a previously uninsured, low-income sample. Their treatment contrasts are related but different: the first evaluates prices within controlled benefit arrangements; the second evaluates the wider insurance bundle, including the move from uninsured care to Medicaid enrollment.

More generous coverage can expand emergency care alongside scheduled outpatient care. This is demonstrated by Oregon's emergency-use findings (Taubman et al., 2014), which documented a 0.41-visit increase over approximately 18 months and no simple offset between office and emergency care. It would be an error to interpret an additional emergency encounter as clinically unnecessary solely because the visit occurred outside primary care. Some encounters reveal previously deferred illness; others may occur because the service has become more affordable. Their value is assessed using clinical indications and outcomes, not using the institutional label of the encounter alone.

Stronger patient cost-sharing can reduce service quantities without demonstrably generating sophisticated price shopping. The employer deductible study (Brot-Goldberg et al., 2017) documented a substantial spending decline driven largely by lower use rather than lower selected prices. The elderly cost-sharing study (Chandra, Gruber, and McKnight, 2010) identifies an important possible offset: reduced outpatient and pharmaceutical care can be associated with higher hospitalization elsewhere in the financing system. These findings differ from a naive assumption that a utilization decrease improves resource allocation. The composition of the forgone care and the incidence of downstream costs matter.

Insurance can improve financial protection even before measurable increases in service volume. The Mexico randomized evaluation (King et al., 2009) found a reduction in catastrophic spending in the study horizon without detectable aggregate utilization increases. The historical Medicare assessment (Finkelstein and McKnight, 2008) likewise highlights protection against out-of-pocket expense risk. A public policy may therefore be economically significant because it reduces the likelihood of a financially disastrous medical bill even when visit counts change little. Such effects belong directly to the insurance function of transferring financial risk.

Longer-run health estimates also differ across designs. The Medicare severe-admission study (Card, Dobkin, and Maestas, 2009) examined selected emergency admissions close to age 65, while the ACA Medicaid mortality study (Miller, Johnson, and Wherry, 2021) evaluated adults approaching older ages in expansion versus nonexpansion states. Their reported mortality improvements do not supply a common clinical effect per added appointment. They instead show that study population, illness severity, treatment access and observation period fundamentally shape what can be estimated and which pathways are plausible.

International studies reinforce the interaction of demand with supply. The Thailand financing reform (Gruber, Hendren, and Townsend, 2014) affected both patient payments and provider funding, and found increased use particularly in poorer areas together with changes in infant mortality. Ghana's insurance-utilization analysis (Blanchet, Fink, and Osei-Akoto, 2012) described increased formal care contacts using observational adjustment. These studies support institutional comparison, while their distinct evidentiary designs prevent a claim of a single, globally generalizable causal magnitude.

The integrated assessment of consequences

A scientific assessment of insurance-induced care use begins with the definition of the coverage intervention and the probability that eligible persons actually become covered. It then measures unconditional treatment incidence, the probability of any use, conditional episodes among users, intensity per episode, treatment mix, clinical indications, quality, and discontinuities in care. It links these outcomes to patient resources, payments, provider supply and financial protection. Lastly, it evaluates morbidity, mortality and welfare over clinically meaningful periods. This sequence is an analytical specification of established measurement problems, not a new independent model or doctrine.

The observed service volume should be decomposed into its causal components. When claims spending rises by 10 percent, the increase may reflect higher treatment frequency, higher reimbursement prices, greater intensity per episode, changes in diagnostic mix, new treatment technologies, or a shift from uncompensated and off-claim care into recorded insurance payment. A corresponding decrease can reflect fewer contacts, lower contracted prices or selective plan enrollment. Claims data must therefore be matched to quantities and prices where possible. Expenditure changes alone do not identify the utilization response.

Social welfare requires additional information. For a clinically necessary treatment previously unaffordable to a low-income patient, insurance can create access whose value exceeds measured payments. For a low-value repeated test, greater use may impose real resource costs with limited benefit. For chronic medication, continuity can reduce later complications yet initially raise recorded pharmacy spending. A public insurer might incur a cost that a household or hospital previously bore; a private payer may realize savings while the patient loses access to effective therapy. These are distributive and causal questions, and they should not be answered using aggregate insurance claims in isolation.

Clinical guidance and disease burden influence the value of additional treatment. An increase in preventive screenings can be productive when the test is appropriate and follow-up treatment is available. Screening can also identify disease and thereby temporarily increase referrals, pharmacy use and admissions, so a rise in subsequent utilization may signal improved detection. Evaluation must distinguish treatment due to newly recognized need from avoidable repeat contact. Such interpretation requires disease-specific outcomes and appropriate clinical expertise rather than treating the expenditure sign as an outcome.

Provider behavior is part of the causal system. Plans pay clinicians through fee schedules, capitation, negotiated contracts or budgets; these arrangements affect willingness to accept patients and the services offered after a visit. A rise in enrollment may be accompanied by new hospital capacity or may encounter fixed supply. The effect of coverage on encounters consequently depends on provider payment and network implementation, not exclusively on the patient-facing premium or copayment. An insurance-based policy can be evaluated properly only after documenting the implementation that actually reached households and medical facilities.

Distribution matters both as equity and as research design. A reform reaching historically excluded patients can raise utilization because unmet need was greater at baseline. Aggregate means weighted by existing insured populations may obscure this response. Conversely, a limited increase in the mean can conceal serious delays in rural or low-income subgroups. The OECD unmet-care comparative evidence establishes that financial, waiting-time and geographical barriers coexist and disproportionately affect disadvantaged groups in its documented country sample. Cross-country policymaking should therefore examine which barriers the insurance reform addresses and which remain untouched.

Implications for insurance research and policy

For research design, the primary implication is measurement discipline. Studies should state eligibility rules, take-up, benefits, deductible and copayment schedules, provider arrangements, age and income composition, baseline utilization, patient observation time, geographic scope, and outcome-specific follow-up. They should distinguish intention-to-treat from effects among compliers and separate unconditional use from use conditional on entering care. These specifications make a study more useful for transfer to another insurance system than a headline percentage divorced from its denominator.

For policy assessment, the next implication is to evaluate affordability, availability and clinical effectiveness together. Coverage rules that eliminate a charge but leave insufficient primary-care clinicians may increase waiting times or shift patients toward emergency services. Cost sharing that reduces visits can reduce low-value care and valuable treatment simultaneously. The design of exemptions, benefit packages and provider payments should be assessed against observed clinical indications and financial burden rather than a presumption that more or fewer visits is invariably preferable.

For fiscal analysis, all-payer and household incidence should accompany insurer-specific accounting. New coverage can transfer spending from patients, hospitals or uncompensated care pools into a public insurer; that transfer changes recorded insurer expenditure even if real treatment is unchanged. Hospitalization offsets and drug adherence consequences require sufficient follow-up and linked data. A complete budget impact should specify which entities finance the services, which incur delayed savings or costs, and whether health improvements change long-term resource needs.

For international comparison, common conceptual definitions are more defensible than forcing numerical equivalence across unlike programs. The U.S. Medicaid eligibility expansion, a Medicare age threshold, a national universal coverage reform with new hospital funding, and voluntary private insurance enrollment can be meaningfully compared through identified mechanisms while retaining their separate treatment bundles. Research synthesis should publish differences in outcomes and uncertainty as substantive results. Studies that find financial protection without a detected utilization effect contribute to knowledge just as importantly as studies detecting increased visits.

The current global setting underlines the significance of this topic. The 2025 WHO–World Bank UHC report documents persistent incomplete access and financial hardship on a global scale. These gaps require interpretation of effective coverage, delivery capacity, and household financial protection, alongside conventional insurance enrollment counts. The scientific objective is to determine whether coverage changes enable people to obtain appropriate care, alter the mix of care responsibly, reduce uncompensated risk, and improve measurable outcomes under clearly specified institutional conditions.

Several research questions remain open across systems: the long-run causal effect of coverage on continuity of chronic care; the interaction of provider capacity with an enrollment shock; the value of newly obtained appointments relative to forgone care after cost-sharing increases; and the allocation of fiscal responsibility between households, insurers and public budgets. Digital records, linked mortality data and changing benefit designs improve the opportunity to answer them. Their resolution depends on further original data collection and context-specific designs, rather than inference beyond the evidence already reviewed.

Conclusion

Health care utilization responds to insurance through identifiable and interacting mechanisms: financial affordability at the point of service, entitlement to treatment, access to participating providers, information about medical need, treatment decisions, and the financing of medical organizations. The measurable consequence is a multidimensional change in the probability of entering care, contacts per user, quantity and intensity of treatment, setting and timing of services, financial exposure, and eventually health. Each component requires a stated intervention, denominator, period and counterfactual.

RAND demonstrates substantial responsiveness to patient cost sharing within tested historical plans. Oregon demonstrates that a randomized increase in Medicaid coverage can raise primary, preventive, hospital and emergency use while improving financial protection, even when selected biomarkers show no statistically significant mean improvement in a two-year window. Medicare threshold studies, expansion-era analyses and employer-deductible research show that different reforms and populations generate different service mixes and clinical consequences. Comparative evidence from Mexico, Thailand and Ghana confirms the importance of public financing institutions, provider capacity and baseline access conditions. These findings form a coherent body of evidence precisely when their variations are preserved.

The sign of a utilization response does not by itself identify economic efficiency or clinical benefit. An additional episode may treat long-neglected disease, improve diagnosis, or add care of limited value. A forgone episode may avoid unnecessary intervention or interrupt effective treatment. Financial protection can be valuable with little change in observed visits. Provider and system constraints determine whether stronger purchasing power leads to effective treatment or to queues and reallocations among patients. Accordingly, responsible evaluation integrates service-specific causal effects, quality and health outcomes, household financial exposure, and the incidence of public and private costs.

The original scholarly contribution of this monograph is a comprehensive, source-critical comparative interpretation and a clearly defined evidentiary specification of utilization responses to coverage. It establishes which inferences are warranted by randomized research, quasi-experiments, observational studies and institutional documentation, and identifies where further evidence is necessary. The established literature supports contextual, outcome-specific conclusions and rules for valid comparison. The manuscript introduces no new proprietary economic theory, clinical model, doctrine or unverified claim of scientific priority.

References

  • Agency for Healthcare Research and Quality (AHRQ). (n.d.). Medical Expenditure Panel Survey, Household Component: Survey overview. U.S. Department of Health and Human Services. Original source.

  • Arrow, K. J. (1963). Uncertainty and the welfare economics of medical care. American Economic Review, 53(5), 941–973. Original source.

  • Baicker, K., Mullainathan, S., & Schwartzstein, J. (2015). Behavioral hazard in health insurance. Quarterly Journal of Economics, 130(4), 1623–1667. Original source.

  • Baicker, K., Taubman, S. L., Allen, H. L., Bernstein, M., Gruber, J. H., Newhouse, J. P., Schneider, E. C., Wright, B. J., Zaslavsky, A. M., & Finkelstein, A. N. (2013). The Oregon experiment—Effects of Medicaid on clinical outcomes. New England Journal of Medicine, 368(18), 1713–1722. Original source.

  • Blanchet, N. J., Fink, G., & Osei-Akoto, I. (2012). The effect of Ghana’s National Health Insurance Scheme on health care utilisation. Ghana Medical Journal, 46(2), 76–84. Original source.

  • Brook, R. H., Ware, J. E., Rogers, W. H., Keeler, E. B., Davies, A. R., Donald, C. A., Goldberg, G. A., Lohr, K. N., Masthay, P. C., & Newhouse, J. P. (1983). Does free care improve adults’ health? Results from a randomized controlled trial. New England Journal of Medicine, 309(23), 1426–1434. Original source.

  • Brot-Goldberg, Z. C., Chandra, A., Handel, B. R., & Kolstad, J. T. (2017). What does a deductible do? The impact of cost-sharing on health care prices, quantities, and spending dynamics. Quarterly Journal of Economics, 132(3), 1261–1318. Original source.

  • Card, D., Dobkin, C., & Maestas, N. (2008). The impact of nearly universal insurance coverage on health care utilization: Evidence from Medicare. American Economic Review, 98(5), 2242–2258. Original source.

  • Card, D., Dobkin, C., & Maestas, N. (2009). Does Medicare save lives? Quarterly Journal of Economics, 124(2), 597–636. Original source.

  • Centers for Medicare & Medicaid Services (CMS). (n.d.). Transformed Medicaid Statistical Information System (T-MSIS) Analytic Files (TAF). U.S. Department of Health and Human Services. Original source.

  • Chandra, A., Gruber, J., & McKnight, R. (2010). Patient cost-sharing and hospitalization offsets in the elderly. American Economic Review, 100(1), 193–213. Original source.

  • Finkelstein, A. N., Taubman, S. L., Allen, H. L., Wright, B. J., & Baicker, K. (2016). Effect of Medicaid coverage on ED use—Further evidence from Oregon’s experiment. New England Journal of Medicine, 375(16), 1505–1507. Original source.

  • Finkelstein, A., & McKnight, R. (2008). What did Medicare do? The initial impact of Medicare on mortality and out of pocket medical spending. Journal of Public Economics, 92(7), 1644–1668. Original source.

  • Finkelstein, A., Taubman, S., Wright, B., Bernstein, M., Gruber, J., Newhouse, J. P., Allen, H., & Baicker, K. (2011). The Oregon Health Insurance Experiment: Evidence from the first year. NBER Working Paper 17190. Original source.

  • Gruber, J., Hendren, N., & Townsend, R. M. (2014). The great equalizer: Health care access and infant mortality in Thailand. American Economic Journal: Applied Economics, 6(1), 91–107. Original source.

  • Keeler, E. B., Brook, R. H., Goldberg, G. A., Kamberg, C. J., & Newhouse, J. P. (1985). How free care reduced hypertension in the health insurance experiment. JAMA, 254(14), 1926–1931. Original source.

  • King, G., Gakidou, E., Imai, K., Lakin, J., Moore, R. T., Nall, C., Ravishankar, N., Vargas, M., Téllez-Rojo, M. M., Hernández Ávila, J. E., Hernández Ávila, M., & Hernández Llamas, H. (2009). Public policy for the poor? A randomised assessment of the Mexican universal health insurance programme. Lancet, 373(9673), 1447–1454. Original source.

  • Kolstad, J. T., & Kowalski, A. E. (2012). The impact of health care reform on hospital and preventive care: Evidence from Massachusetts. Journal of Public Economics, 96(11–12), 909–929. Original source.

  • Manning, W. G., Newhouse, J. P., Duan, N., Keeler, E. B., Leibowitz, A., & Marquis, M. S. (1987). Health insurance and the demand for medical care: Evidence from a randomized experiment. American Economic Review, 77(3), 251–277. Original source.

  • Meille, G., & Post, B. (2023). The effects of the Medicaid expansion on hospital utilization, employment, and capital. Medical Care Research and Review, 80(2), 165–174. Original source.

  • Miller, S., Johnson, N., & Wherry, L. R. (2021). Medicaid and mortality: New evidence from linked survey and administrative data. Quarterly Journal of Economics, 136(3), 1783–1829. Original source.

  • Newhouse, J. P., Manning, W. G., Morris, C. N., Orr, L. L., Duan, N., Keeler, E. B., Leibowitz, A., Marquis, K. H., Marquis, M. S., Phelps, C. E., & Brook, R. H. (1981). Some interim results from a controlled trial of cost sharing in health insurance. New England Journal of Medicine, 305(25), 1501–1507. Original source.

  • Nyman, J. A. (1999). The value of health insurance: The access motive. Journal of Health Economics, 18(2), 141–152. Original source.

  • Organisation for Economic Co-operation and Development (OECD). (2025). Health at a Glance 2025: OECD Indicators. OECD Publishing. Original source.

  • Organisation for Economic Co-operation and Development (OECD). (2025). Unmet needs for healthcare. In Health at a Glance 2025: OECD Indicators. Original source.

  • Sommers, B. D., Long, S. K., & Baicker, K. (2014). Changes in mortality after Massachusetts health care reform: A quasi-experimental study. Annals of Internal Medicine, 160(9), 585–593. Original source.

  • Sommers, B. D., Maylone, B., Blendon, R. J., Orav, E. J., & Epstein, A. M. (2017). Three-year impacts of the Affordable Care Act: Improved medical care and health among low-income adults. Health Affairs, 36(6), 1119–1128. Original source.

  • Taubman, S. L., Allen, H. L., Wright, B. J., Baicker, K., & Finkelstein, A. N. (2014). Medicaid increases emergency-department use: Evidence from Oregon’s Health Insurance Experiment. Science, 343(6168), 263–268. Original source.

  • Wherry, L. R., & Miller, S. (2016). Early coverage, access, utilization, and health effects associated with the Affordable Care Act Medicaid expansions: A quasi-experimental study. Annals of Internal Medicine, 164(12), 795–803. Original source.

  • World Health Organization & World Bank. (2025). Tracking universal health coverage: 2025 global monitoring report. WHO. ISBN 978-92-4-011780-8. Original source.

  • World Health Organization (WHO). (2026). Global Health Expenditure Database: Documentation Centre and March 2026 release. Original source.