we-solve (beta)

Econ-490 · economics · Economics

Vertical integration in health care

LLM status Open unverified

Statement

Original subject group: Health economics

Audit classification: Published research agenda. Checked source identity and appropriate agenda/background support; expanded intervention, units, population, definedness and causal restrictions. Exact targets and variants are editorial.

Definitions and precise research target

Editorial formalization. Consider Medicare beneficiaries receiving primary care in United States practices acquired by hospitals during 2010–2025. Define $a\in\{0,1\}$ as hospital ownership versus continued independent ownership of the same practice. For baseline beneficiary $i$, let $H_i(a)$ be five-year quality-adjusted survival, $C_i(a)$ total medical resource cost, and $T_i(a)$ travel and waiting burden. Fix a value $\lambda>0$ per quality-adjusted life-year and a monetary valuation $\rho>0$ per burden unit. The target is \[\tau_W=E[\lambda\{H_i(1)-H_i(0)\}-\{C_i(1)-C_i(0)\}-\rho\{T_i(1)-T_i(0)\}].\] Report contrasts by baseline rural residence and clinical risk without conditioning on post-acquisition referral choices. Identify acquisition effects under an explicit ownership counterfactual, retaining patients who move or change providers. Separate coordination changes from billing reclassification and referral steering where additional variation permits; ownership alone does not identify those mechanisms. Include effects on nonacquired practices if spillovers matter and show sensitivity to welfare valuations.

Define quality-adjusted survival as $H_i(a)=\sum_{t=1}^{5}\beta^t q_{it}(a)\mathbf1\{i\text{ alive at }t\}$, where $q_{it}\in[0,1]$ is an independently defined health index; deceased persons have zero subsequent survival contribution. $C_i(a)$ includes all real care costs, including providers changed after acquisition, and $T_i(a)$ records hours of travel/waiting on the same five-year discount convention. $\lambda$ has units dollars per discounted quality-adjusted life-year and $\rho$ dollars per discounted hour. Declare whose time and costs are included and require finite means. Site-of-service reimbursement changes alone are not real cost changes. Ownership assignment is at practice/system level; a treated-practice randomization or justified counterfactual must address endogenous acquisition and spillovers. The expression is a declared net-health-benefit objective, not an identified utilitarian welfare sum without further preferences. All expectations use a stated baseline population and probability law. Identification means every admissible data-generating model with the same observed-data law yields the same displayed target; otherwise report the identified set. Exact equations, horizons and assumptions are editorial, rather than source-given canonical conjectures.

Variants and assumption changes

  1. Editorial saturation variant. Include nonacquired practices and patients in market welfare $W(s)$, where $s$ is ownership prevalence. Compare with isolated-practice contrasts.
  2. Editorial payment-neutral variant. Cross acquisition with removal of site-of-service reimbursement advantages. The interaction identifies this channel only under credible variation in both regimes.

References and source audit

1. Consolidation quality and access evidence gaps verified; claimed exact geographic-proximity passage not independently retrieved. Locator: RAND RR-A1820-1, 30 September 2022; report-page Key Takeaways. Report metadata and primary RAND summary inspected; PDF follow-up retrieval timed out. The supplied p.13 proximity locator remains unverified. URL: https://www.rand.org/content/dam/rand/pubs/research_reports/RRA1800/RRA1820-1/RAND_RRA1820-1.pdf.
  1. Jodi L. Liu; Zachary M. Levinson; Annetta Zhou; Xiaoxi Zhao; PhuongGiang Nguyen; Nabeel Qureshi. 2022. Environmental Scan on Consolidation Trends and Impacts in Health Care Markets. RAND RR-A1820-1, 30 September 2022; report-page Key Takeaways. https://www.rand.org/content/dam/rand/pubs/research_reports/RRA1800/RRA1820-1/RAND_RRA1820-1.pdf. DOI: https://doi.org/10.7249/RRA1820-1.

Common conventions: Source mathematical conventions

These conventions apply to each entry unless explicitly replaced.

Models and identification. A primitive model $m$ specifies all probability laws, feasible choices, information, equilibrium and measurement rules used by its target. The admissible class $\mathcal M$ is fixed independently of outcomes. If $P_O(m)$ is its observable probability law and $\tau(m)$ the target, its identified set at observable law $P$ is \[ \mathcal I_\tau(P)=\{\tau(m):m\in\mathcal M,\ P_O(m)=P\}. \] Point identification means that this set is a singleton. An empty set signals incompatibility of data and assumptions. Coordinate bounds need not describe the joint set. Bounds are sharp only if every retained target is attainable or arbitrarily approachable by compatible models. Finite bounds require explicit restrictions on unobserved quantities; unrestricted confounding, hidden wealth, extrapolation or arbitrary equilibrium selection can yield uninformative sets.

Causal interventions. A potential outcome $Y_i(p)$ is the outcome under a complete policy regime $p$, including timing, financing, information and market spillovers. The target population and units are fixed before treatment. With interference, use the complete assignment vector or a justified exposure mapping; individual no-interference notation alone cannot identify an economy-wide intervention. A difference of mean potential outcomes does not require the cross-world joint distribution. Conditioning on post-treatment migration, survival, enrollment or employment changes the estimand and needs separate justification.

Statistical statements. An estimator, confidence set, forecast or test specifies a sampling law, model class and loss/coverage criterion. Uniform $(1-\alpha)$-coverage means $\inf_{m\in\mathcal M}\Pr_m\{\tau(m)\in C_n\}\geq1-\alpha$, or an explicitly asymptotic version. The whole parameter space trivially has coverage, so informativeness requires a stated width, risk or power objective. A forecasting improvement is relative to a named benchmark, information set and evaluation population.

Equilibrium and optimization. An equilibrium comprises feasible plans, optimality, beliefs where needed, budget constraints and all market/resource clearing. The primitives or a stated benchmark define each correspondence $\mathcal E(p;m)$. Nonemptiness is assumed or proved, never inferred just from compact policy sets. A selected-equilibrium target uses a declared selection rule; a robust target quantifies over all permitted equilibria. Use suprema or infima unless attainment is established. An $\varepsilon$-optimal policy is feasible and within $\varepsilon$ of the finite optimum value. Report an empty feasible set separately.

Welfare and accounting. Normative utility, interpersonal normalization, social weights, numeraire, discounting and the use of public receipts are primitives. Behavioral utility need not coincide with the welfare criterion. Household welfare includes ownership income and rents once; adding profits again to compensating variation generally double-counts them. A monetary equivalent is defined by an explicit transfer and price convention, with monotonicity and range conditions. Average effects, mechanism contributions, transfers and welfare losses are different targets. Infinite sums require convergence and dynamic feasible plans require terminal or no-Ponzi conditions.

Source versions. A working-paper date, revision date and journal publication year are distinguished. A DOI for a journal version is not automatically the DOI of an earlier manuscript. Locators refer to the stated version and printed pages where specified. A metadata-only check does not verify a claim about a full-text conclusion. Every entry's source subsection narrows the provenance claim accordingly.

Statement licence: CC-BY-4.0 · Source file

Contributions 1

Econ-490-A1 · Exploration · LLM-generated · gpt 6 astra ultra ·

GPT 6 Astra Ultra attempt on Econ-490

Hospital ownership and net health benefit: resource-cost bounds, attrition and mechanism limits

Problem ID: I11-486. Status: Unresolved; rigorous restricted results.

The target and what claims data omit

The target averages five-year health, resource cost, and travel/waiting changes over a baseline Medicare beneficiary cohort, with values $\lambda$ per discounted quality-adjusted life-year and $\rho$ per discounted hour. The RAND report [rand], whose primary summary was inspected, documents limitations of existing quality and access evidence. It does not provide identification of the target net-health-benefit measure.

We give a sharp valuation-and-cost sensitivity analysis after ownership effects on observable health and utilization have been identified. We also incorporate missing health indices without discarding movers or the deceased and exhibit a mechanism nonidentification result that survives factorial payment randomization. These are conditional results; no ownership instrument or parallel-trends assumption is claimed for the historical acquisition cohort.

Population, regime and experimental reduction

Before ownership assignment, fix a set of markets and a cohort of beneficiaries attached to the baseline practices in each market. Outcomes follow those beneficiaries across providers and moves. A regime $a\in\{0,1\}$ specifies hospital acquisition or continued independence of the designated practices, together with the within-market spillovers and continuation rules. Markets, rather than individual patients, are the independent assignment units. If one instead assigns practices inside markets, a justified full assignment-vector or saturation model must replace this scalar regime.

For each beneficiary define \[ H(a)=\sum_{t=1}^5\beta^t A_t(a)q_t(a),\quad T(a)=\sum_{t=1}^5\beta^t T_t(a),\qquad 0<\beta\leq1, \tag{1} \] where $A_t$ is survival, $q_t\in[0,1]$, and $T_t$ counts the stipulated patient and caregiver hours. Death sets subsequent health contribution to zero; it is not missingness. Let $Q_k(a)$ be discounted use of resource category $k=1,\ldots,J$, aggregated across all subsequent providers. Suppose \[ C(a)=\sum_{k=1}^Jc_kQ_k(a),\qquad c\in\mathcal C, \tag{2} \] where $\mathcal C$ is a known nonempty compact convex set of nonnegative unit resource costs. The same $c$ applies in both ownership regimes. This invariance is a substantive restriction; ownership-specific productivity changes would require separate cost vectors and additional data.

Assume regime assignment is independent of all potential outcomes, with positive known probabilities, and no cross-market spillovers. Sample markets from a fixed population and weight them using prespecified baseline cohort weights, never post-acquisition patient counts. Consistency then identifies the regime-specific distributions of the observed variables. The proof is immediate: for any measurable integrable observed outcome $Y(a)$, independence and consistency imply ${\mathbb{E}}[Y\mid A=a]={\mathbb{E}}[Y(a)]$, with the baseline weighting incorporated in $Y$. Analogous identification can be supplied by a defensible observational design, but it is not proved here.

Write $\Delta H={\mathbb{E}} H(1)-{\mathbb{E}} H(0)$, $\Delta T={\mathbb{E}} T(1)-{\mathbb{E}} T(0)$ and $\Delta Q_k={\mathbb{E}} Q_k(1)-{\mathbb{E}} Q_k(0)$. Billing payments need not equal (2). A site-of-service payment can change with no resource change at all.

Sharp welfare bounds with shared resource costs

The target conditional on $(\lambda,\rho,c)$ is \[ \tau(\lambda,\rho,c)=\lambda\Delta H-\rho\Delta T-c\cdot\Delta Q. \] Let $h_{\mathcal C}(v)=\max_{c\in\mathcal C}c\cdot v$ denote its support function.
Theorem (Shared-cost bounds and valuation frontier): Suppose $\Delta H,\Delta T,\Delta Q$ are identified, and the model imposes no restriction on $c$ beyond (2) and $c\in\mathcal C$. For fixed positive $(\lambda,\rho)$, the sharp welfare interval is \[ \left[\lambda\Delta H-\rho\Delta T-h_{\mathcal C}(\Delta Q),\quad \lambda\Delta H-\rho\Delta T+h_{\mathcal C}(-\Delta Q)\right]. \tag{3} \] Robustly nonnegative welfare is equivalent to \[ \lambda\Delta H-\rho\Delta T\geq h_{\mathcal C}(\Delta Q). \tag{4} \] For box restrictions $\underline c_k\leq c_k\leq\overline c_k$, \[ h_{\mathcal C}(\Delta Q)= \sum_{\Delta Q_k\geq0}\overline c_k\Delta Q_k+ \sum_{\Delta Q_k<0}\underline c_k\Delta Q_k. \tag{5} \]
Proof: Maximizing and minimizing the linear cost term gives (3), with attainment by compactness. To prove sharpness, retain the identified potential utilization and outcome laws and assign each admissible $c$ as the real unit-resource cost in both regimes. No observed variable depends on the unspecified $c$ under the maintained measurement model; every retained vector therefore generates the same observed law. Convexity of $\mathcal C$ gives every intermediate value. The interval's lower endpoint is nonnegative exactly under (4). In a box, each coordinate can be chosen separately: a nonnegative coefficient is maximized at its upper bound, and a negative one at its lower bound, proving (5). ∎

For $\Delta H>0$, (4) gives the explicit break-even health valuation $\lambda\geq\{\rho\Delta T+h_{\mathcal C}(\Delta Q)\}/\Delta H$, intersected with $\lambda>0$. If $\Delta H=0$, health valuation has no effect on the sign. If $\Delta H<0$, dividing reverses the inequality, so the robust-gain region may be empty for positive $\lambda$. Thus a contour in $(\lambda,\rho)$ is a normative sensitivity result, not an estimated preference distribution.

The shared cost vector matters. If hospital ownership reduces outpatient visits but increases inpatient days, the signs in (5) select different cost endpoints by category. Taking independent lower and upper bounds for total cost in each regime ignores the common $c$ restriction and generally gives wider, nonsharp bounds.

Missing health indices without survivor conditioning

Suppose survival $A_t$ is fully observed, but the health index is observed only when $R_t=1$. Let \[ L_a=\sum_{t=1}^5\beta^t{\mathbb{E}}[A_tR_tq_t\mid A=a],\qquad U_a=L_a+\sum_{t=1}^5\beta^t\Pr(A_t=1,R_t=0\mid A=a). \tag{6} \] In (6), $A$ without a time subscript denotes regime assignment; $A_t$ denotes survival. The distinction is purely notational. No missing-at-random assumption is made. Assume $T$ and $Q$ are fully recorded, and impose no relationship between missing $q_t$ and the other outcomes beyond $q_t\in[0,1]$.
Proposition (Sharp interval with health attrition): For fixed positive $(\lambda,\rho)$ and the preceding resource model, the sharp welfare interval is \[ \begin{aligned} [\ &\lambda(L_1-U_0)-\rho\Delta T-h_{\mathcal C}(\Delta Q),\\ &\lambda(U_1-L_0)-\rho\Delta T+h_{\mathcal C}(-\Delta Q)\ ]. \end{aligned}\tag{7} \] The statement allows arbitrary dependence of missingness on latent health and does not condition the target on surviving, staying with the practice, or retaining an observed health score.
Proof: Every observed health contribution is fixed by the observed law. An unobserved contribution among the living lies between zero and $\beta^t$, and among the deceased is zero. Summing yields ${\mathbb{E}} H(a)\in[L_a,U_a]$. Both endpoints are attained by assigning every missing live health score respectively zero or one, leaving observed outcomes unchanged. Because no cross-regime or cross-time restriction on these scores is imposed, the lower contrast $L_1-U_0$ and upper contrast $U_1-L_0$ are simultaneously feasible assignments of the two sets of potential health outcomes. They can be coupled arbitrarily with observed utilization and time; random assignment remains valid by construction. Choose resource-cost extremizers as in Theorem 1. Since the health completion and cost restrictions are independent in the declared class, this attains both endpoints of (7). Convex mixing of compatible completions and the convex cost set attains all intermediate welfare values. ∎

If scientific restrictions link longitudinal health, survival and utilization, they should tighten this set. Formula (7) is sharp only for the explicitly unrestricted completion class. A complete vital-status record can meaningfully improve it: treating every unobserved deceased person as potentially healthy would add impossible health mass.

Payment-neutral factorial variation is not automatic mediation identification

Suppose markets can additionally be randomized to a payment regime $b\in\{0,1\}$, where $b=1$ removes the specified site-of-service advantage. Randomization of $(a,b)$ identifies the four total-regime means of any observed outcome and their interaction \[ I_Y={\mathbb{E}} Y(1,1)-{\mathbb{E}} Y(0,1)-{\mathbb{E}} Y(1,0)+{\mathbb{E}} Y(0,0). \tag{8} \] This is a causal interaction of bundled regimes. It does not by itself isolate referral steering, coordination, or billing as mechanisms.
Proposition (Mechanism nonidentification despite a complete factorial): There exist two structural models agreeing on the entire joint law of ownership $A$, payment assignment $B$, observed referral $M$, and outcome $Y$ under every randomized $(A,B)$ regime, but disagreeing on the controlled effect of referral and on the natural indirect effect of ownership through referral.
Proof: Let $A,B$ be independently randomized binary variables. Both models set $M=A$. In model I define $Y=A$ for every intervention on the referral variable. In model II define $Y=M$. Payment $B$ has no effect in either model. Under every factorial assignment the observed variables satisfy $M=Y=A$, so their full distributions agree, not merely their means.

Under an intervention fixing $A=0,M=1$, model I yields $Y=0$ and model II yields $Y=1$, whereas both yield zero under $A=0,M=0$. Thus the controlled referral effect is zero in model I and one in model II. Since $M(1)=1$ and $M(0)=0$, the natural indirect contrast $Y(0,M(1))-Y(0,M(0))$ is likewise zero versus one. These models have valid bounded outcomes and no hidden random variable is required. They prove nonidentification even with observed referrals and full randomization of ownership and payment. ∎

The construction does not imply that payment-neutral randomization is unhelpful. It identifies (8) and can test whether the total ownership effect depends on payment incentives. To interpret that interaction as a particular channel requires exclusions ruling out other paths, or an additional intervention on the mediator with a defensible implementation. For actual financial accounting, if a billing change increases insurer payments and provider receipts equally at fixed utilization, their transfer cancels in a welfare account including both sides. Equation (2) appropriately leaves real cost unchanged. Distributional burdens and tax-financing costs require additional stated weights and primitives; they are not equal to medical resource cost.

The unresolved empirical work is acquisition identification over the fixed cohort, comprehensive resource measurement, market spillovers and validation of any mechanism exclusions. The bounds above remain available when point identification of net health benefit fails, and their width identifies which missing health or resource-cost information would be most valuable.

Completion Estimate

45%

This is a post hoc, heuristic estimate for the full catalogue target. The attempt derives sharp shared-cost and missing-health welfare bounds and shows mechanism nonidentification even under ownership-payment factorial randomization. Historical acquisition effects for the baseline Medicare cohort remain unidentified without a credible assignment design, comprehensive resource measurement, market spillovers, and defensible coordination or referral exclusions.

References

Go to contribution →