RESEARCH DOSSIER
High-Cost Claimant Concentration and Health Insurance Stop-Loss
Abstract. Concentration of medical expenditure among a relatively small number of people is a recurrent feature of health financing, but the exposure retained by a particular risk-bearing institution depends on contract eligibility, pricing, service mix, risk-pool size, payment timing, and the structure of excess protection. This monograph examines the relationship between high-cost claimant concentration and health insurance stop-loss, with particular attention to the economic distinction between medical expenditure distributions and insured claims distributions. It synthesizes U.S. Medical Expenditure Panel Survey estimates, longitudinal studies of high-cost patients, contemporary employer health-benefit evidence, international health-financing classifications, the insurance supervisory principles governing risk transfer, and relevant U.S. and European legal and prudential frameworks. Its analytical methods are exposure-normalized concentration measures, conditional persistence analysis, frequency-severity reasoning, formal specific and aggregate indemnity equations, and transparent hypothetical stress tests. A four-state numerical illustration separates an extreme single claimant from a broader frequency shock and shows how specific recovery, aggregate attachment, and cover limits produce different retained-loss distributions. The analysis identifies contract definitions, claims maturation, laser provisions, disclosure, reimbursements, liquidity, counterparty performance, and retained capital as decisive variables. In the illustrative model, gross expected claims of $6.409 million become $5.214 million of expected retained claims after specified layers, before stop-loss premiums, expenses, recoverability adjustments, and the cost of capital. Those results are mathematical consequences of the stated hypothetical scenario, not empirical estimates or proposed optimal retentions. The central conclusion is that a measured concentration statistic cannot itself determine stop-loss adequacy: valid evaluation requires the distribution of contractually eligible member costs, inter-member dependence, the layered payment function, and the practical collectibility and timing of recoveries. The dossier offers a source-grounded scientific synthesis without introducing a new proprietary theory, doctrine, or statistical model.
Keywords: High-cost claimants; health insurance; medical expenditure concentration; extreme claims; catastrophic claims; specific stop-loss; aggregate stop-loss; excess-of-loss reinsurance; self-funded employer health plans; level funding; attachment point; risk pooling; persistence; claim severity; frequency-severity dependence; retained loss; reinsurance recoverables; solvency; capital adequacy; tail value at risk; loss development; actuarial model risk.
Introduction
Medical expenditure is distributed unequally across people and across periods. Some individuals incur extensive expense because of acute hospitalization, severe injury, a complex birth, specialty drugs, prolonged treatment, or interacting chronic conditions. Other individuals incur little or no expenditure during the observation period. Financing institutions convert these uncertain payment streams into premiums, contributions, appropriations, reserves, claims budgets, and contingent risk transfers. The relationship between the patient-level distribution and the financing institution's net risk is the scientific object of this dossier.
The AHRQ concentration analysis reports that in 2022 the highest-spending 1% of the U.S. civilian noninstitutionalized population accounted for 21.7% of recorded personal medical expenditure; the highest 5% accounted for 49.7%, and the lowest 50% for 2.8% (Hernandez-Viver and Mitchell, 2025). Those population statistics include payments by multiple financing sources. They illustrate concentration but do not identify the amount that a particular commercial insurer, public program, or employer plan would submit for stop-loss recovery. The contract-specific problem requires further transformations.
The U.S. KFF employer survey for 2026 reports that 71% of workers covered by responding employers' health plans were in self-funded arrangements, and that among covered workers in self-funded plans at firms with 200 or more workers, 75% were in plans with stop-loss insurance. These are survey estimates for specified employer groups, not population-wide measurements of insurance risk or evidence that every plan purchases identical protection (KFF, 2026). The distinction between health benefits funded by an employer and the stop-loss insurance purchased by that employer is central to interpretation.
The research has three questions. First, which definitions and observations adequately characterize high-cost claimants, concentration, and persistence? Second, how do specific and aggregate attachments convert gross eligible claims into insurer or employer retention and recoverables? Third, how should the resulting net exposures be assessed for financing, liquidity, solvency, and governance across institutional settings? The method combines primary institutional sources, peer-reviewed empirical literature and direct mathematical reasoning. Every quantitative scenario created in this monograph is identified as illustrative. No confidential claims or pricing data were available, and no causal treatment effect or market-wide optimal attachment is asserted.
Analytical scope and terminology
A claimant is an individual whose services give rise to covered charges or claims under the arrangement being studied. A high-cost claimant is identified using a stated threshold, rank, or quantile and a stated time interval. A medical expenditure can aggregate provider payments from patients, insurers, and public programs; a paid claim may be a narrower, contract-adjusted transaction. Stop-loss protection is an indemnity contract under which a defined payer receives an amount contingent on covered cost exceeding a specified attachment or aggregate threshold. Retention denotes the risk remaining with that payer after contractually eligible recoveries, before accounting for premiums and expenses. These distinctions are maintained throughout.
Foundations and Source-Critical History
Research object and units of analysis
An observed medical event has several potentially different monetary values: billed charge, allowed contracted amount, payer-paid amount, patient cost sharing, and the final amount qualifying for specific or aggregate reimbursement. Even when all these amounts arise from the same hospitalization, their ranking and threshold exceedance can differ. A study that uses comprehensive expenditure data supplies evidence about the cost distribution of the healthcare system; the stop-loss actuary needs a contract-specific distribution of eligible reimbursements. This distinction also defines an important institutional boundary. A self-funded employer faces benefit obligations it finances directly, while a fully insured employer purchases a health policy under which the insurer carries defined claims risk. An insurer may in turn buy reinsurance. Public purchasers may pool costs across budgets or territories without buying a conventional commercial stop-loss policy.
OECD health financing classifications distinguish financing schemes from financing agents and from revenue sources, and describe government, compulsory insurance, voluntary insurance, and out-of-pocket mechanisms (OECD, 2025). The classification matters because the entity recording care expenditure may differ from the entity bearing the ultimate financial risk. For example, the presence of a national health insurer does not imply that individual local purchasers face the same terminal loss distribution as the national pool. Likewise, an employer's stop-loss recoverable and a member's legal entitlement to benefits are distinct contractual rights.
The WHO health-financing framework emphasizes revenue mobilization, pooling, purchasing, and benefit policy (WHO, accessed 2026). The size and composition of the pool determine how predictable its aggregate costs are, subject to dependence and changes in expected utilization. Stop-loss protection changes the allocation of an already-defined layer of financial risk, while the underlying system determines which services are financed and how much providers are paid. These decisions interact economically, but they are not substitutes.
Historical and institutional development
The long development of insurance contains multiple ways to share exceptional losses. At the member level, pooled health financing absorbs ordinary variation. At the institutional level, insurers and employers can transfer higher layers to specialized counterparties. Modern employer stop-loss is particularly visible in the United States because a significant share of workplace health benefits is self-funded. The 2026 KFF survey's funding categories distinguish self-funded, fully insured, and level-funded arrangements. A level-funded arrangement combines fixed monthly payments, claims funding, administration, and stop-loss, and its measured economic risk retention depends on the actual terms rather than on the marketing label.
In the United States, ERISA's savings and deemer provisions under 29 U.S.C. § 1144 help explain why the regulation of a self-funded employee benefit plan differs from the insurance regulation of the insurer issuing stop-loss coverage. The statutory language preserves state regulation of insurance while generally restricting the classification of a covered employee benefit plan itself as an insurer for relevant state laws. Application depends on the kind of plan and statutory exceptions, including multiple employer welfare arrangements. The legal description of a particular arrangement therefore requires contract and jurisdiction-level analysis.
The National Association of Insurance Commissioners lists a Stop-Loss Insurance Model Act as Model 92, establishing a model-law reference point rather than a uniform nationwide enactment (NAIC, accessed 2026). The model-law context explains why attachment standards, product supervision, and treatment of stop-loss contracts can vary by state. The study of threshold adequacy must record the law actually in force in the relevant state and year, not import an indicative model figure as universal law.
At the insurer level, IAIS ICP 13 addresses the role of reinsurance and other risk transfers in strategy, risk appetite, solvency, concentrations, liquidity, counterparty exposures, and stressed recoveries (IAIS, 2024 online framework). Employer stop-loss and insurer-to-insurer reinsurance share certain mathematical characteristics, but the identity and regulatory duties of the buyer differ. The scientific analysis consequently uses common loss-layer algebra while preserving institutional classification.
Classification and evidence hierarchy
Four forms of evidence are used. Official survey statistics measure an identified population using specified weights and designs. Longitudinal studies evaluate transitions across expenditure levels in defined cohorts. Law and prudential standards specify duties or recognition criteria in their jurisdictions. Mathematical identities describe what follows from a particular contract payment function. These categories answer different questions. Empirical correlations cannot establish a contract's indemnity. A legal model cannot establish adoption in every state. A mathematical formula cannot establish the actual distribution or price of clinical claims.
One especially instructive comparison comes from Holle, Wolff, and Herant's 2001–2018 concentration study. It found that the share of the U.S. population generating half of total health expenditure was remarkably stable near 4%–5% across the examined years, although the concentration of prescription-drug spending shifted significantly (Holle et al., 2021). That is evidence about repeated cross-sectional distributional structure, not evidence that the same persons recur in the top spending group. The latter question requires tracking individuals through time.
High-Cost Exposure and Tail Metrics
Measuring concentration
Let X(i,t) denote qualifying expenditure for individual i during observation interval t. For n persons, total expenditure T(t) equals the sum of their X(i,t), and the observed per-person mean is T(t)/n. Order expenditures from largest to smallest, X(1,t) ≥ X(2,t) ≥ ... ≥ X(n,t). The top-q share, C(q,t), is the sum of the highest ceil(qn) values divided by T(t), with explicit weighting rules where survey weights apply. A top-1% share of 20% means that 1% of the observed or represented population generated one-fifth of the stated cost measure. It does not mean that a claimant's risk of exceeding a contractual threshold is 20%, nor that a stop-loss contract reimburses 20% of total costs.
For population surveys, MEPS Statistical Brief 560 identifies the 2022 entry point to its top 1% spending tier as $80,990 in annual expenditures, with mean spending of $147,071 for people in that tier (Hernandez-Viver and Mitchell, 2025). The threshold and mean are related but distinct statistics; the threshold is a distributional cutoff, and the mean describes the more variable conditional tail. The brief's figures represent all financing sources and the civilian noninstitutionalized population. They are unsuitable as a direct nationwide specific stop-loss pricing table for privately insured employees, because enrollment, covered service eligibility, age distribution, contractual discounts, and population sampling differ.
A concentration curve plots the cumulative share of total cost against the cumulative share of persons ranked by cost. The complementary Lorenz curve, sorted from least to most costly, supports Gini calculations. The Gini coefficient can be written as one minus twice the area beneath the Lorenz curve. For extremely sparse pools, however, a single coefficient obscures whether concentration arises from one extraordinary claim or many moderately high claims. The pair C(1%) and C(5%) provides more localized tail information, while exceedance counts above stated currency thresholds offer a direct bridge to stop-loss exposure.
Measures used for reinsurance decisions should therefore include absolute and inflation-adjusted maximum eligible member cost; quantiles of eligible member cost; numbers of members crossing candidate attachments; excess amounts above attachments; first and second moments where finite and stable; maximum and aggregate retention; and cash flow duration. Analysis may additionally report the ratio of the largest claim to annual total, the number of claimants generating 50% of expenditure, or the Herfindahl-type sum of squared claimant shares. These measures are descriptive and should be computed with denominators that match the specified insured group and accounting period.
Persistence and incidence of high costs
The temporal dimension changes risk interpretation. A patient can have a major acute event in one year and relatively modest costs the next; another may experience an expensive chronic treatment sequence for many years. Persistent high expenditure affects renewal underwriting, pharmacy contracts, ongoing claim reserves, and the benefit of multiyear coverage. Episodic catastrophic expenditure creates a different balance between severe incident shocks and recurring member costs.
The Canadian longitudinal three-year high-cost-user study reports that approximately 45% of people in Ontario’s highest-spending 5% in 2009 remained in that 5% group the following year, while about one-third stayed in the group for all three study years (Wodchis, Austin, and Henry, 2016). That study's publicly financed population, measurement of covered services, and cohort years constrain transferability. It nevertheless demonstrates why cross-sectional concentration must not be treated as a transition probability.
A different VA persistence analysis followed high-cost Veterans Health Administration patients using VA and Medicare records from fiscal 2008–2012. Among baseline top-decile patients, 32% were high-cost after one year and 7% remained high-cost in all four follow-up years; the authors also documented a substantial mortality component (published 2018). The different percentile cutoff, age and health composition, data scope, and follow-up period prevent a direct numerical comparison with Ontario or MEPS. These studies jointly support an important methodological conclusion: the appropriate longitudinal risk state must be defined within the same healthcare financing population as the cover being analyzed.
For a given threshold d, define H(i,t;d) = 1 when X(i,t) exceeds d. The observed transition P(H(i,t+1;d)=1 | H(i,t;d)=1) is a useful descriptive persistence statistic. It requires explicit handling of membership exit, death, benefit-plan changes, and missing information. If individuals who exit the pool are silently dropped, the persistence estimate can become a selection measure of survivors rather than an estimate of all originally covered persons. Separate analyses of incidence among previously low-cost members and persistence among already high-cost members are especially useful for attachment selection.
Severity, frequency, and dependence
For pool size n, a simplified annual claim total is S = sum X(i). This identity does not imply independent X(i). Utilization, infectious-disease waves, provider pricing, catastrophic treatment innovation, shared socioeconomic conditions, and plan design can generate dependence. Conditional on a common inflation or morbidity shock Z, costs may be positively associated even if independent conditional on Z. If variances exist, Var(S) equals the sum of individual variances plus twice the sum of all pairwise covariances. Ignoring these covariances can understate aggregate tail risk.
A frequency-severity representation of claims is also useful. Let N denote the number of eligible claim events and Y(j) their eligible amounts, so S = sum over j=1,...,N of Y(j). For an individual-specific stop-loss, however, events must be aggregated by the relevant insured person's contractual claim period before applying the attachment. A $100,000 hospital episode and a $180,000 specialty-drug episode for the same member may jointly cross a $250,000 annual specific attachment, even when neither event separately crosses it. Event-level modeling without member linkage would miss that reimbursement.
An annual tail estimate also depends on trend. A claim of $200,000 in a historical year does not automatically represent the same economic risk in a later year. Unit prices, utilization intensity, population mix, care site, exchange rate, provider contracting, and benefit expansion can move differently. An inflation adjustment should distinguish price, quantity, mix, and new technology. Overall medical inflation alone cannot reproduce the upper tail when the composition of care changes. The Holle et al. study specifically illustrates divergent trends in drug-spending concentration, reinforcing the case for service-line and condition-level segmentation.
Health-system sources of tail losses
Several clinical processes can produce concentrated expenditure: an unusually expensive single inpatient admission, organ transplantation and subsequent immunosuppression, neonatal intensive care, oncology treatment, gene or cell therapy, severe trauma, dialysis, or prolonged complex care. Claim size is shaped by the negotiated unit price and benefit design as well as clinical need. Consequently, a high-cost member count by itself does not diagnose inefficiency or predict avoidable spending. Quality-of-care assessment must distinguish necessary high-value treatment from inappropriate utilization, provider-price variation, and preventable complications.
The relevant ethical constraint is equally scientific. Reinsurance and risk adjustment are financing mechanisms. They do not establish that a high-cost person is responsible for the financing institution's risk or that access to effective treatment should be curtailed. Institutions can study variations in eligibility or cost management while separately measuring clinical outcomes, appeals, delayed care, and treatment discontinuity. The object is the distribution and transfer of payer-level liabilities, not a normative ranking of patients.
Specific and Aggregate Attachments
Specific stop-loss as a payment function
Specific stop-loss responds to an individually identified member's covered costs after the accumulated eligible amount exceeds the contractual attachment. Let x(i) be eligible member cost in the contract period, d the attachment, l(i) a limit on reimbursable excess for that member, and a(i) an applicable reimbursable participation share. In a simplified indemnity arrangement, specific reimbursement is R(i) = a(i) min(max[x(i) − d, 0], l(i)). Total specific recovery is the sum of R(i) over members. The payer's pre-premium net eligible claims are S − sum R(i), provided the contract does not alter other elements of the underlying benefit obligation.
This formula is a useful baseline, and actual policies specify materially different details. Some attachments are established by a uniform per-person deductible. Others apply individually named, increased attachment points, often described in employer stop-loss practice as lasers. A contract may also have a maximum reimbursable amount, exclude specified categories of expense, or define reimbursement according to paid claims only. Policies may require disclosure of known large claims, impose special pharmacy treatment, or determine the eligibility of post-termination claims under agreed run-out clauses. Each modification changes the payment function; therefore, the correct analysis starts with the signed contract, not merely the stated attachment amount.
A uniform specific deductible creates a kink in the payer's marginal retention: an additional eligible dollar below d remains with the primary payer, while an eligible dollar above d may shift to the stop-loss carrier until the coverage limit is exhausted. If there is a participation percentage a below 100%, the primary payer continues retaining 1 − a of additional loss in the reimbursable layer. An upper limit returns additional cost to the primary payer unless another contract responds. For a pool with extraordinary single-member severity, the upper limit can be more consequential than the attachment itself.
At the individual level, stop-loss is also sensitive to time conventions. Suppose a person's treatment extends from December to February. Under a claims-paid basis, payment date and check processing may determine the reimbursement year. Under a claims-incurred basis, date of service determines allocation, subject to reporting and payment conditions. An incurred-and-paid contract introduces two clocks. Coverage with a stated incurred period and limited paid run-out may leave incurred eligible costs unreimbursed when they are processed too late. A numerical attachment cannot be interpreted without its temporal basis.
Aggregate protection and interaction of covers
Aggregate stop-loss protects the annual or treaty-period sum of eligible retained costs above a portfolio-level attachment. Define B as the aggregate basis on which the attachment is applied, A as the aggregate attachment, L as the aggregate maximum reimbursement, and b as the reimbursement share. A simplified aggregate recovery is G = b min(max[B − A, 0], L). This form is consistent with the stop-loss definitions in KFF's survey, which distinguishes per-person and whole-plan thresholds (KFF, 2026). It is an idealized mathematical representation rather than a claim about every commercial policy.
The interaction with specific stop-loss is critical. If the aggregate basis B is eligible total claims net of specific reimbursements, then B = S − R_specific, subject to contractual adjustments. Aggregate recovery becomes G = b min(max[S − R_specific − A, 0], L), and ultimate retained eligible claims equal S − R_specific − G. If instead the aggregate contract is calculated on a different basis, such as a prescribed aggregate of eligible claims before specific recoveries, the ordering changes. The analyst must document the contract's allocation and anti-duplication clauses to avoid double counting or treating distinct coverage layers as additive when they are not.
Aggregate attachment may be defined in dollars or as a multiple of a premium or expected-claims base. For example, 125% of a contractually defined expected-claims fund is not a universal regulatory threshold and does not guarantee 25% protection against the true actuarial mean. It is a particular accounting and pricing definition. When enrollment changes, the attachment may be adjusted by monthly covered lives or expected claims. An apparent decline in aggregate loss ratio may result from the denominator adjustment rather than any decline in claim severity. Reporting should state both the currency attachment and the formula that generated it.
A contract with both specific and aggregate protection usually separates two types of uncertainty. Specific protection truncates the amount retained from very expensive individuals, up to limits. Aggregate protection can respond when many claims accumulate within individually retained layers. Thus a large outbreak, high utilization in many moderate-cost members, or unexpectedly broad pharmaceutical spending can stress aggregate retention even when no single individual crosses the specific deductible. This complementarity also means the two premiums cannot be evaluated independently of the joint claims distribution.
Contract periods, eligibility, limits, and claims administration
A complete stop-loss interpretation requires the following data for every covered claim: member identity and eligibility window, dates of service, claim reporting and payment dates, contracted allowed amount, benefit category, discounts, coordination of benefits, exclusions, deductible applied, maximum payable amount, recoveries already received, and pending disputed amounts. If the carrier reimburses only claims actually paid by the underlying plan, the purchaser may have to finance the full gross claim before receiving a stop-loss payment. Settlement and reporting lags create liquidity risk independently of ultimate indemnity.
A reimbursement estimate should distinguish notified and recognized recoverables, expected but unreported eligible losses, settled claims, denied or disputed reimbursements, and collectibility adjustments. Where specific stop-loss carriers provide advance-funding or accommodation arrangements, the analyst must identify whether these are binding obligations, discretionary advances, or loans repayable under conditions. A contract marketed as protecting a high layer can still leave a plan exposed to substantial short-term cash demands when reimbursement is delayed.
Contract interpretation also requires an explanation of claimant aggregation across family members, dependents, affiliated groups, reinstatement provisions, midyear employee turnover, acquisition or divestiture, and renewal. The scientific unit for a specific attachment is whatever legal unit the contract actually specifies, which may be a covered person rather than an employment contract or hospital episode. The same claim may be treated differently under separate stop-loss contracts written for successive periods. Explicit contract-version control is therefore part of data governance, not merely a legal formality.
Pricing, selection, and incentive effects
A stylized pure premium for specific stop-loss equals the expected reimbursable amount, E[sum R(i)], subject to the pricing measure used. The charged premium must also account for administrative costs, claim-handling expenses, commission, uncertainty loads, cost of capital, profit, and potential asymmetry of information. A corresponding aggregate pure premium is E[G], where G depends on any specific reimbursements and on enrollment or premium-base adjustment. The combined expected recovery cannot generally be inferred by adding two stand-alone cover estimates calculated against unrelated gross bases.
The expected excess above attachment d has an informative representation: for nonnegative random member cost X with a finite mean, E[(X − d)+] equals the integral from d to infinity of P(X > x) dx. This mathematical identity shows that stop-loss pricing depends on the entire tail above d, rather than only on the probability of crossing d. Two populations can have identical exceedance frequency but very different expected recovery if their conditional severities differ. Conversely, severe medical price growth can raise reimbursement even when the number of high-cost members stays unchanged.
Selection can occur at multiple levels. Employers select self-funded versus insured financing; stop-loss carriers price and underwrite particular employer groups; risk-bearing entities may choose specific versus aggregate protection; groups can change providers, benefits, and members over time. Observed differences between stop-loss purchasers and nonpurchasers therefore should not be interpreted as treatment effects of coverage without a credible identification design. In a suitable empirical comparison, one would seek policy variation external to the employer's underlying risk, test pre-trends, and control for contemporaneous changes in employee composition, benefit generosity, provider networks, and employer size.
There can also be incentives at contract boundaries. A payer retaining costs below a specific deductible may have a different marginal incentive to negotiate a late high-cost claim after the attachment has been exceeded. Reimbursement requirements, coordination clauses, audits, claims-management services, and experience-rated renewals can partly alter that incentive. An economic argument about incentives requires an actual causal channel and recognition that clinical decisions remain constrained by medical necessity, professional duties, and patient protections. It would be unwarranted to infer changes in care quality from indemnity algebra alone.
Net Retention and Capital
Accounting and economic decomposition
A payer's gross claims, reimbursements, and ultimate retained claims form an accounting identity, but the solvency consequence is more complex. Let S denote ultimate eligible gross loss, R the valid and collectible amount reimbursed by specific protection, G the valid and collectible aggregate reimbursement, and P the total stop-loss premium. Simplified net claim cost is N = S − R − G. The financing institution's total economic outlay before investment and administrative effects is N + P. If reimbursement is uncertain, the analyst should replace contractual R and G with cash-flow-valued recoverables that reflect payment probabilities, credit risk, settlement timing, and disputes. A contract with high nominal reimbursement can be inadequate in a liquidity crisis if collection is slow.
For insurers reporting under international standards, the IFRS 17 requirements govern recognition and measurement of contracts within scope and require separate information about insurance service results and insurance finance components (IFRS Foundation, accessed 2026). Applying that standard to any specific stop-loss transaction requires deciding whether the contract qualifies as an insurance or reinsurance contract held, whether the reporting entity is within IFRS accounting scope, and how separate rights and obligations are recognized. Gross underwriting liabilities and reinsurance assets should not be conflated merely because they are economically related.
In the European Union, Solvency II Article 81 explicitly requires reinsurance recoverables to account for timing differences between direct payments and recoveries and to be adjusted for expected losses from counterparty default (European Union, 2009, consolidated text). This regulatory feature is a direct answer to the financial issue: nominal risk transfer does not automatically make a cedent economically whole. Recognition is tied to the particular regulated insurer and applicable eligibility conditions.
Four-state numerical case study
Consider a hypothetical employer plan covering 1,000 people for one year. In the baseline composition, 990 members incur $3,000 each, nine incur $200,000 each, and one incurs $1.5 million. The totals are $2.970 million, $1.800 million, and $1.500 million respectively, yielding $6.270 million of gross eligible claims. This constructed portfolio has 1% of its members (ten people) responsible for $3.300 million, or approximately 52.63%, of gross spending. Its largest claimant alone accounts for approximately 23.92%. These are illustrative amounts designed to expose contractual mechanics, not observed market data or estimates for any identifiable employer.
Suppose the plan buys a uniform specific cover reimbursing 100% of eligible annual member costs above a $250,000 attachment, with no specific limit in the range of the four states. A single claimant with $1.5 million of eligible cost generates a specific reimbursement of $1.250 million. The nine $200,000 claimants do not exceed the attachment. Accordingly, gross baseline loss of $6.270 million becomes $5.020 million after specific reimbursement.
Now suppose the plan also buys aggregate protection applied to the member-level claims net of specific recoveries, attaching at $6.000 million and reimbursing no more than $2.000 million in the contract year. In the baseline state, net claims of $5.020 million do not cross the aggregate attachment. The aggregate layer pays zero. If an analyst mistakenly applied the same aggregate threshold to the gross $6.270 million without considering the contract's net-of-specific basis, the analyst would report a fictitious $270,000 recovery in this state. That difference represents a concrete analytical error, not merely a technical rounding issue.
Four states and their assumed probabilities are specified below. In the low-severity state, the expensive member incurs $300,000; in the baseline state $1.5 million; in the extreme-severity state $3.0 million. Other members are unchanged in those three states. In the frequency-shock state, 970 members incur $3,000, 29 incur $200,000, and one incurs $1.5 million. Probabilities are illustrative assumptions and are not inferred from any clinical or employer sample.
| State | Probability | Gross eligible claims (m) | Net after specific (m) | Final retained claims ($m) | | --- | ---: | ---: | ---: | ---: | ---: | ---: | | Low severity | 0.40 | 5.07 | 0.05 | 5.02 | 0.00 | 5.02 | | Baseline severity | 0.35 | 6.27 | 1.25 | 5.02 | 0.00 | 5.02 | | Extreme individual severity | 0.15 | 7.77 | 2.75 | 5.02 | 0.00 | 5.02 | | Frequency shock | 0.10 | 10.21 | 1.25 | 8.96 | 2.00 | 6.96 |
Weighted expected gross claims equal $6.409 million; expected specific reimbursement equals $0.995 million; expected aggregate reimbursement equals $0.200 million; expected final retained claims equal $5.214 million. These identities reconcile exactly: 6.409 − 0.995 − 0.200 = 5.214, in millions of dollars. Stop-loss premiums, administrative expenses, cost of capital, investment income, taxes, claims delay, and counterparty losses are excluded by construction. Thus $1.195 million is the expected contractual claims recovery in this illustration, not a net economic gain from purchasing the contracts.
Two substantive features emerge. First, specific protection almost entirely removes the severity variation of the single very expensive member in the first three states: each leaves $5.020 million of loss after the specific layer. Second, the frequency shock produces $8.960 million of claims after specific protection, because 29 moderately expensive members remain below $250,000. Aggregate protection pays its entire $2.000 million limit in that shock and leaves $6.960 million of retained claims. The layers respond to different forms of concentration even though all four states arise from one underlying population construction.
The example also shows that the presence of an attachment does not guarantee a maximum net retained loss. Aggregate limit exhaustion leaves further cost with the employer. If the frequency shock were more severe, the employer's final retention would continue to increase after the $2.000 million aggregate limit was reached. If a specific limit existed, sufficiently large individual claims could also return additional loss to the employer. Both limits should appear explicitly in any stress test or board reporting.
Counterparty, liquidity, and capital consequences
Pure risk reduction should be separated from counterparty substitution. A primary payer transfers a layer of medical expenditure uncertainty to a stop-loss carrier but may acquire credit and operational risk. The IAIS reinsurance framework states that effective reinsurance risk transfer also creates counterparty, operational, basis, and timing exposures; its requirements call for alignment with risk appetite, capital strategy, concentration limits, and liquidity planning (IAIS, 2024). Consequently, a lower theoretical net claim variance does not by itself establish a lower total risk exposure.
For regulated European insurers, the Solvency II health underwriting module separates non-life-technique health, life-technique health, and health catastrophe submodules (European Commission, 2015, consolidated text). The basic capital requirement framework also identifies counterparty default and other risk modules and uses a 99.5% value-at-risk calibration over one year for relevant modules. The formal regulatory calculation differs from simply inserting the illustrative four-state example into an arbitrary 99.5% quantile; a legitimate solvency calculation requires the actual supervisory formula, valuation base, risk classifications, and eligibility of mitigation.
Liquidity risk is often greatest before reimbursement. An employer plan may have to pay a hospital or provider according to its benefit arrangement on a short timeline while recovering a stop-loss amount after documentation and adjudication. The institution should stress gross claims due within 30, 60, and 90 days, available operating cash and committed facilities, expected reimbursement lags, and carrier default during stressed conditions. A plan can be solvent on an ultimate net basis yet unable to meet immediate payments. Adequate internal governance therefore measures both solvency and cash sufficiency.
Counterparty concentration can also accumulate if multiple high layers depend on one stop-loss insurer or closely related group. A practical exposure register should contain recoverables by insurer, security quality, collateral, disputes, aging, coverage year, and underlying claims. The Solvency II reinsurance-recoverables rule explicitly recognizes default losses and timing differences; it provides a jurisdiction-specific example of why reinsurance should be valued as a contingent cash-flow right rather than simply deducted from nominal gross claims.
Risk management and clinical implications
The appropriate retention depends on the sponsor's capacity to absorb volatility, risk appetite, financial resources, and the pricing of layers. A large diversified institution may retain more predictable high-frequency claims because it has stable cash flow and efficient scale. A smaller employer with thin liquidity may value contractual protection against even one major case more highly. Neither conclusion licenses a universal dollar attachment or a universally preferred type of coverage. The analysis should compare full distributional consequences against offered premiums and capital costs for the specific covered pool.
Claims-management programs can improve contracting, clinical coordination, and measurement, but the evidence must be evaluated on outcomes as well as gross paid expense. A payer might reduce its reimbursements by changing classification or delaying payments; such an accounting effect is not necessarily a reduction in health-system resource use. High-cost members deserve adequate clinical care and accurate adjudication regardless of which financing institution ultimately bears the loss. Sound governance therefore separates decisions about eligibility and contract reimbursement from medical decisions made in the patient's interests.
Data, Measurement, and Validation
Data architecture and denominators
The quality of a stop-loss analysis is primarily constrained by the quality of the link between covered persons, medical events, and contractual claims. A defensible research dataset should contain a stable encrypted person identifier, coverage start and end dates, plan and contract year, benefit category, service date, allowed amount, net payer-paid amount, provider reimbursement, medical diagnosis or procedure grouping where lawful, individual-specific attachment where applicable, stop-loss eligibility status, final payment date, and reconciliation status. Distinct identifiers are needed for claim line, service episode, person, family, employer group, primary insurer, and risk-transfer counterparty. These identifiers need not reveal patient names for aggregate scientific analysis.
The sampling denominator must be explicit. A count of people ever enrolled during a year is different from average covered lives or member-years. If a plan has 1,000 members for six months and 1,200 for the next six, annual average membership is 1,100, while the number of unique individuals may be greater. Annual claims per unique member, claims per member-year, and claims per active member at year-end can therefore produce different estimates. An attachment calculated on actual member-years cannot be checked against a denominator constructed from unique claimants without adjustment.
A high-cost rank should use persons, not just people with positive claims. Excluding zero-cost enrollees inflates top-percentile cutoff values by conditioning on utilization and changes the population being described. Similarly, mixing primary insured payments with patient out-of-pocket obligations can create a measure of overall system cost when the intended quantity is the employer's covered liability. The MEPS statistical definitions explicitly identify their sampled civilian noninstitutionalized population and payments from multiple sources; researchers should preserve those definitions when reproducing its concentration measures (Hernandez-Viver and Mitchell, 2025).
A contract-level dataset must also record recoveries actually booked and received from stop-loss carriers. A measured net-to-gross ratio based on projected reinsurance can differ from the realized ratio after exclusions, audits, claims run-out, or counterparty settlement. For each observation period, report gross eligible ultimate claims, contractual expected recovery, approved reimbursable amount, paid recovery, and remaining recoverable separately. They answer different operational and accounting questions.
Statistical estimation and uncertainty
A scientifically useful tail analysis reports both the point estimate and its uncertainty, with a method matched to the data design. MEPS survey results require survey weights and design-aware variance estimation. Claims data from a single large employer can be treated as a complete observed portfolio, but inference about a future underwriting period still depends on sampling variation, medical trend, and model assumptions. Repeated annual observations from the same members are correlated; bootstrap resampling at the person level, rather than the claim-line level, can be appropriate for estimating longitudinal uncertainty when the entire person's history is resampled together.
Threshold estimates should be shown over a predeclared grid of attachments, such as the observed 90th, 95th, 99th, and 99.5th percentiles or a range of nominated dollar deductibles. Tail estimates should not be selected after seeing which threshold produces the preferred answer without acknowledging the selection. A rare-event frequency estimate from a small pool can have a wide confidence interval. A model that reports an exact expected loss above a high deductible from a handful of observations creates false precision unless the uncertainty in severity extrapolation is made explicit.
The upper tail may be modeled empirically, parametrically, or by combining both approaches. Empirical distributions preserve observed experience but are unreliable beyond observed maxima. Parametric lognormal, gamma, generalized Pareto, or spliced models may be useful for specified regions of a distribution, but their fit does not authorize indefinite extrapolation. A credible analysis reports the fitted range, threshold sensitivity, goodness-of-fit diagnostics, parameter uncertainty, exposure changes, extreme-value leverage, and the effect of caps or provider contracts. Forecasting of biologic therapies or changes in specialty-drug cost structure may require separate scenarios rather than simple historical inflation factors.
Expected recoveries can be estimated by applying the full contractual payment function to each simulated annual loss record. Recomputing aggregate attachment for each simulation, where the contract requires enrollment adjustment, is essential. When an analyst instead computes a single expected gross loss and then applies the stop-loss function to that mean, Jensen-type nonlinearity produces a generally incorrect reimbursement estimate: E[(S − A)+] is not equal to (E[S] − A)+. Similarly, an expected specific recovery requires integrating the individual tail above d, not applying d to the expected member cost.
Model validation, claims development, and governance
Model validation must be both mathematical and operational. The program should test exact attachment equality, zero claims, policy-year boundaries, two claims for the same member, excess above limits, aggregate recovery after specific reimbursement, negative adjustments, late claims, coverage changes, and double-counting prevention. Reconciliation identities should be checked independently: every scenario satisfies S = net retention + all accepted indemnity recoveries before additional accounting items. Validation should also independently rebuild a subset of actual claim histories from underlying payment records and signed contract clauses.
Out-of-sample evaluation can use temporal holdout years, but changing treatment cost and benefit designs make an unchanged loss distribution an inadequate benchmark. Backtests should compare predicted and realized frequencies of attachment crossing, aggregate loss distribution, claimant transition rates, maximum losses, and recovery cash flows. Extreme losses may be too rare to produce stable backtest failure rates. Scenario and stress testing therefore supplement statistical calibration; they should be disclosed as assumptions, not described as observed frequencies.
Claims-development timing is a particularly important source of error. Medical services provided near contract expiration may be reported or paid many months later. The value of a recovery under a run-out contract depends on the agreed payment window. A reserving model should therefore distinguish incurred-but-not-reported, reported-but-not-paid, and paid-but-not-reimbursed amounts where these categories are relevant. Development triangles aggregated at payer level can hide the distinct reporting and settlement patterns of rare high-severity claims; claimant-specific large-loss reports and contract-year reconciliation may be necessary.
Controls around data completeness should examine missing member identifiers, negative adjustments, duplicate claim lines, bundled services, pharmacy rebate timing, cross-plan coordination, and provider refunds. The analyst also needs to document whether claim reversals are reflected in recoverables, how rebates and subrogation receipts affect the eligible loss, and which contractual rules apply to specialty-pharmacy or other exceptional benefits. A model with precise mathematics and mismatched claims eligibility can be materially less accurate than a simpler model based on well-controlled data.
Source limitations and comparative identification
Cross-country comparison requires careful treatment of institutional roles and total spending. The OECD financing-scheme definitions describe how compulsory, public, voluntary, and household sources fund care, but those national shares do not disclose individual insurer attachment points or private stop-loss indemnity (OECD, 2025). Aggregate spending distributions may vary with demographic composition, clinical entitlements, provider reimbursement, pharmaceutical policy, and accounting scope. It is therefore misleading to compare the U.S. MEPS top 1% share directly with the loss ratio of a specialized insurer or with a local public budget.
Longitudinal data provide an additional caution. Wodchis, Austin, and Henry's Ontario expenditure trajectories track government-financed care for residents, while Yoon and colleagues' VA persistence study combines Veterans Health Administration and Medicare experience (Wodchis et al., 2016; Yoon et al., 2018). Their selected populations and cost definitions differ materially. The observed persistence percentages illuminate mechanisms rather than offering portable numerical assumptions for every health insurance market.
A causal assessment of stop-loss purchasing would require variation that plausibly separates an employer's choice of coverage from its expected claims and liquidity position. One possible empirical design would follow comparable groups before and after a regulatory or procurement change, measure preexisting expenditure trends, and assess concurrent changes to benefits and provider contracts. Such a design would still require confirmation that the policy variation affects stop-loss availability independently of unobserved group-level health shocks. This dossier does not possess an administrative dataset suitable for that identification and accordingly presents no causal effect estimate.
International Synthesis and Scientific Conclusions
United States, European Union, and other health-financing systems
The United States employer market supplies a clear institutional example of specific and aggregate stop-loss because many employers directly finance employee health benefits. KFF's 2026 plan-funding evidence documents a large self-funded segment and a distinct level-funded segment among smaller employers (KFF, 2026). The scientific question for these entities is the distribution of claims they are obligated to pay and the counterparty reimbursement supplied by an issued stop-loss policy. For a small employer, a single expensive case may affect liquidity and benefit financing disproportionately; for a large employer, dependence among moderately expensive claims and aggregate trends can remain important even with substantial risk pooling.
The American legal analysis also requires attention to ERISA preemption and savings and to the relevant state's treatment of stop-loss insurance. The NAIC model-law catalog provides a framework for identifying policy form regulation and the Stop-Loss Insurance Model Act, but it does not by itself prove uniform adoption. Coverage terms, insurer licensing, regulatory minima, and group-size rules must be tied to the applicable law in force. This case-by-case inquiry is necessary for any empirical comparison of attachment constraints or insurance-market substitution.
European Union insurers face a different prudential framework, in which reinsurance recognition depends on financial and risk-transfer conditions. The Solvency II health-risk classification governs health underwriting risk modules for regulated undertakings, and Article 81 on recoverables requires allowance for timing and counterparty default (European Commission, 2015; European Union, 2009). These rules are useful international comparison points but should not be attributed wholesale to employers or public health budget holders outside their regulatory scope.
Across health systems financed mainly by taxation or compulsory social insurance, risk pooling and transfers can operate through budgets, equalization funds, provider contracts, and reserves rather than by commercial specific stop-loss. The WHO's health-financing principles place emphasis on pooling and strategic purchasing, while the OECD's international scheme comparison documents differing shares of government, compulsory insurance, voluntary insurance, and out-of-pocket finance (WHO, accessed 2026; OECD, 2025). The financial function of pooling exceptional costs may be common even when the contractual and legal structures are different.
Comparative interpretation of financial protection
There are at least three separate beneficiaries of a sound financing structure. The patient receives a defined right to appropriate covered services and protection against personally borne health costs; the sponsor or insurer receives manageable volatility in its own claims obligations; and the wider system gains a more stable mechanism to finance high-cost treatment. These outcomes have separate measures. A specific stop-loss recovery can improve the sponsor's balance sheet without altering the patient's co-payment. Reducing patient cost sharing can improve financial protection without changing the sponsor's commercial reinsurance. A national government can subsidize expensive treatments while leaving risk-bearing institutions with highly variable annual reimbursement.
Contract design therefore deserves an integrated evaluation of the expected payment, upper-tail retention, premium, liquidity, counterparty exposure, coverage continuity, benefit adequacy, and claim-service performance. A lower attachment usually increases expected contractual recovery when all else is held constant, but it also may change premium, renewal pricing, and administration. An aggregate limit can contain a particular stress layer while leaving the payer exposed to more extreme aggregate experience. A new treatment can alter the relevant tail faster than the observed data can reveal, making contract renewal terms and scenario updates especially important.
A recurrent temptation is to declare a universal rule from a compelling concentration figure. The AHRQ top-1% share reflects a whole population and multiple payers; the employer stop-loss attachment reflects the legal and actuarial terms of one plan; the prudential capital outcome reflects still another set of rules. The coherent answer is to retain these units throughout the analysis. The same medical event may be severe for the patient, expensive for the provider, eligible for only part of an insurer's benefit, and fully or partly recoverable under a separate financial contract. Each stage produces a distinct value whose relationship must be reconciled.
Principal scientific findings and research agenda
This research establishes five source-grounded findings within its defined scope. First, high-cost medical spending is empirically concentrated, with the observed magnitude depending on population, period, service coverage, and method. Second, expenditure concentration and persistence are different statistical characteristics: the same cross-sectional spending share can coexist with materially different member transitions. Third, specific and aggregate stop-loss transfer distinct portions of the loss distribution, so their joint evaluation depends on the contractual sequencing and the dependence of member costs. Fourth, ultimate claims retention and cash-flow resilience differ because recoveries can be delayed, disputed, limited, or uncollectible. Fifth, a valid international comparison requires legal and institutional classification before interpreting attachment figures, reinsurance treatment, or capital relief.
Future work is particularly valuable on longitudinal claim transitions in commercially insured pools, the impact of evolving specialty therapies on extreme-value tails, the joint distribution of specific and aggregate reimbursement, renewal-related laser decisions, and actual recovery payment lags. Research can also compare employer liquidity losses under similar clinical claim shocks but different coverage terms. A rigorous study would require legally accessible anonymized member-level data, original policy and endorsement language, reliable claim-year reconciliation, and a transparent data governance process. These are research priorities rather than unresolved factual conditions on the mathematical and documentary findings presented here.
Conclusions
High-cost claimant concentration is a fundamental property of observed medical expenditure, and stop-loss is a contract-dependent financial response to the resulting uncertainty. Population expenditure statistics identify the distributional phenomenon; longitudinal evidence clarifies its persistence; contract payment functions determine actual recoveries; and prudential analysis determines whether risk transfer translates into durable financial resilience. The evidence demonstrates substantial high-cost expenditure concentration, and established research shows that the identity of high-cost patients changes through time to degrees that depend on the population studied. The mathematical structure demonstrates why specific reimbursement responds to exceptionally severe individual costs while aggregate coverage may be more responsive to correlated or widespread moderately expensive claims.
The central analytical conclusion is operationally precise: an attachment point is meaningful only alongside the definition of eligible claims, observation period, benefit and payer population, individual or aggregate basis, exclusions, limits, settlement rules, and counterparty performance. Risk managers should compare gross and net loss distributions, expected recoveries, capital and liquidity stresses, and patient-relevant outcomes in their respective units. The four-state case study illustrates these relationships with independently checkable arithmetic and intentionally separates model assumptions from observed facts. No general optimal attachment or universal premium is inferred. Effective science in this field integrates expenditure measurement, actuarial tail analysis, contract law, accounting, prudential capital, and health-system institutions into one reproducible description of who bears a claim, when, and under which conditions.
Technical Appendix: Mathematical Definitions and Reproducible Calculations
Concentration and persistence
For nonnegative eligible costs x_i for n covered persons and total T = Σ_i x_i > 0, define the top-k cost share C_k = (Σ_{j=1}^k x_(j))/T where x_(1) ≥ ... ≥ x_(n). When n is survey weighted, use the underlying published weighting and cutoff conventions; do not substitute integer ranks for fractional population weights. The maximum individual share is C_1. For repeated person histories, H_i(t,d) = 1{x_i(t)>d}; persistence is the conditional fraction of baseline high-cost persons observed in the same state in a later period, with censoring defined explicitly.
For a generic nonnegative random variable X and d ≥ 0, the stop-loss transform is E[(X−d)+] = ∫_d^∞ P(X>x)dx, when this integral is finite. Its (one-sided, where defined) derivative with respect to d is −P(X>d). This is a formal identity; its numerical application requires the right cost distribution. For total S = Σ_i X_i, the aggregate transform is E[(S−A)+]; this need not equal Σ_i E[(X_i−d)+] and depends on the joint distribution of individual costs.
Contract payment functions
For each member i with eligible annual cost x_i, specific attachment d_i, proportional reimbursement a_i, and recovery limit l_i, define R_i = a_i min{max(x_i−d_i,0),l_i}. Let R_S = Σ_i R_i. If the aggregate contract's basis is explicitly net of specific recoveries, let B = Σ_i x_i − R_S. For aggregate attachment A, reimbursement rate b, and limit L, define R_A = b min{max(B−A,0),L}. Final retained eligible claim cost is N = B − R_A. Each formula assumes that terms are applied to the same eligible base and that contractual exclusions and run-out rules have already been reflected in x_i. A different signed contract must be modeled with its own operator.
Four-state arithmetic
Common component across three severity states: 990×$3,000 + 9×$200,000 = $4.770 million. Adding one expense of $0.300m, $1.500m, or $3.000m yields gross totals of $5.070m, $6.270m, and $7.770m. At d=$0.250m, specific reimbursement for that member equals $0.050m, $1.250m, and $2.750m, respectively. Each state then leaves $5.020m after specific reimbursement. All three are below the stipulated aggregate attachment A=$6.000m.
In the frequency-shock state, 970×$3,000=$2.910m; 29×$200,000=$5.800m; one×$1,500,000=$1.500m. Gross S=$10.210m. Specific recovery is $1.250m, so B=$8.960m. The eligible aggregate excess is $2.960m; the $2.000m limit yields aggregate recovery $2.000m and retained claims of $6.960m. The scenario probabilities sum to 0.40+0.35+0.15+0.10=1.00.
Expected gross S = 0.40×5.070 + 0.35×6.270 + 0.15×7.770 + 0.10×10.210 = $6.409m. Expected specific recovery = 0.40×0.050 + 0.35×1.250 + 0.15×2.750 + 0.10×1.250 = $0.995m. Expected aggregate recovery = 0.10×2.000 = $0.200m. Expected final retained eligible claims = 0.90×5.020 + 0.10×6.960 = $5.214m. Identity check: $6.409m − $0.995m − $0.200m = $5.214m. These are nominal costs in a hypothetical annual period without trend, premium, expenses, credit, or discounting.
References
European Commission. (2015; consolidated rulebook accessed November 15, 2026). Commission Delegated Regulation (EU) 2015/35, Article 144: Health Underwriting Risk Module. European Insurance and Occupational Pensions Authority, Solvency II Single Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-5764_en.
European Union. (2009; consolidated text accessed November 15, 2026). Directive 2009/138/EC (Solvency II), Article 81: Recoverables from Reinsurance Contracts and Special Purpose Vehicles. European Insurance and Occupational Pensions Authority, Solvency II Single Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2169_en.
European Union. (2009; consolidated text accessed November 15, 2026). Directive 2009/138/EC (Solvency II), Article 104: Design of the Basic Solvency Capital Requirement. European Insurance and Occupational Pensions Authority, Solvency II Single Rulebook. https://www.eiopa.europa.eu/rulebook/solvency-ii-single-rulebook/article-2191_en.
Hernandez-Viver, A., & Mitchell, E. M. (2025, March). Concentration of Healthcare Expenditures and Selected Characteristics of People with High Expenses, United States Civilian Noninstitutionalized Population, 2018–2022. Statistical Brief No. 560. Agency for Healthcare Research and Quality, Medical Expenditure Panel Survey. https://meps.ahrq.gov/data_files/publications/st560/stat560.shtml.
Holle, M., Wolff, T., & Herant, M. (2021). Trends in the concentration and distribution of health care expenditures in the US, 2001–2018. JAMA Network Open, 4(9), e2125179. https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2784058. DOI: 10.1001/jamanetworkopen.2021.25179.
IFRS Foundation. (2017; amended 2020; current standards page accessed November 15, 2026). IFRS 17 Insurance Contracts. https://www.ifrs.org/issued-standards/list-of-standards/ifrs-17-insurance-contracts/.
International Association of Insurance Supervisors. (2024 online framework, accessed November 15, 2026). Insurance Core Principles and ComFrame: ICP 13, Reinsurance and Other Forms of Risk Transfer. https://www.iais.org/icp-online-tool/13524-icp-13reinsurance-and-other-forms-of-risk-transfer/.
KFF. (2026, October 22). 2026 Employer Health Benefits Survey. https://www.kff.org/health-costs/2026-employer-health-benefits-survey/.
National Association of Insurance Commissioners. (Accessed November 15, 2026). Model Laws: Stop-Loss Insurance Model Act, Model No. 92; State Adoption and Model-Law Framework. https://content.naic.org/model-laws.
Organisation for Economic Co-operation and Development. (2025, November 13). Health at a Glance 2025: OECD Indicators—Health Expenditure by Financing Scheme. OECD Publishing, Paris. https://www.oecd.org/en/publications/health-at-a-glance-2025_8f9e3f98-en/full-report/health-expenditure-by-financing-scheme_074bead0.html. Report DOI: 10.1787/8f9e3f98-en.
United States Congress. (1974; current codification consulted November 15, 2026). Employee Retirement Income Security Act, 29 U.S.C. § 1144: Other Laws (Preemption, Savings, and Deemer Provisions). Legal Information Institute, Cornell Law School, public legal text. https://www.law.cornell.edu/uscode/text/29/1144.
Wodchis, W. P., Austin, P. C., & Henry, D. A. (2016). A 3-year study of high-cost users of health care. CMAJ, 188(3), 182–188. https://pubmed.ncbi.nlm.nih.gov/26755672/. DOI: 10.1503/cmaj.150064.
World Health Organization. (Accessed November 15, 2026). Health Financing. https://www.who.int/health-topics/health-financing/.
Yoon, J., Chee, C. P., Su, P., Almenoff, P., Zulman, D. M., & Wagner, T. H. (2018). Persistence of high health care costs among VA patients. Health Services Research, 53(5), 3898–3916. https://onlinelibrary.wiley.com/doi/10.1111/1475-6773.12989. DOI: 10.1111/1475-6773.12989.