Executive Summary
Every employer, insurer, and government agency paying for healthcare is paying two bills at once. The first is for the price of each service. The second, larger bill is for how much care actually gets delivered. Almost nobody buying health insurance today can see those two bills separately.
This study separates them.
Using 8.2 million commercial insurance episodes across four Texas metros, grouped into 131 clinically defined episode types and attributed to 72,776 individual provider care groups, the analysis breaks total cost into a price component (what providers charge per unit of service) and an efficiency component (how many services are actually delivered). Patient status is normalized for age, sex, zip code, complications, and clinical classification. The results overturn a widely held assumption about what really drives healthcare costs.
Five Findings
- Variance is enormous. For common conditions, the risk-adjusted cost of a single episode can differ 20× to 40× between the lowest-cost and highest-cost provider in the same city.
- Efficiency drives the variance. Differences in how efficiently providers deliver care account for 88% of total cost variance when episodes are weighted equally. Unit prices account for 12%. When weighted by actual dollars spent, efficiency accounts for over 99%.
- Conditions dominate spending. Condition episodes (diabetes, depression, low back pain, asthma) represent roughly four of every five dollars spent. For conditions, essentially 100% of cost variance is driven by provider efficiency.
- Procedures tell the opposite story. For procedural episodes (knee replacement, cataract surgery, coronary bypass), unit price drives 89% of the variance. But procedures represent only about 17% of total spending.
- Physician and patient preference explain the efficiency variance. Physician practice style accounts for 30–35% of geographic spending variation. Patient preferences explain approximately 5%.
Across 8.2 million episodes in Texas, roughly 88% of the variance in total cost between providers is driven by how efficiently care is delivered, not by the prices providers are paid. For condition-based spending, that figure is effectively 100%. Physician practice patterns are the primary reason this efficiency variance exists.
Why This Study Matters
Healthcare costs have grown faster than wages, faster than corporate revenue, and faster than general inflation for most of the last three decades. For a typical mid-sized employer, healthcare is now the second-largest expense after payroll, and the one they have the least ability to manage.
Every major buyer has tried the same levers: negotiating lower prices, narrowing networks, shifting cost to employees, introducing transparency tools. These strategies have slowed growth in places. They have not solved the problem.
One reason: they have all targeted the wrong bill.
The Institute of Medicine estimated in 2009 that approximately 30% of U.S. health spending ($750 billion) was wasted on unnecessary services, excessive administration, fraud, and other problems. A 2019 JAMA systematic review updated those figures to $760–$935 billion annually, roughly 25% of total spending. Of that, $75.7–$101.2 billion was attributed specifically to overtreatment or low-value care.
Those waste estimates did not answer who was responsible for the overtreatment or why it persisted. This study, combined with the research literature summarized below, provides that answer.
What episodes reveal that other approaches cannot
Price transparency of the kind required by federal rule is useful but incomplete. Knowing that two MRI facilities charge different rates does not tell you which physician's diabetes patients end up with half as many emergency department visits, or which orthopedic practice's total joint replacement episodes produce the most complications and the longest stays.
Episodes of care capture the full clinical journey for a single clinical problem: the physician, the facility, the testing, the drugs, the follow-up. Measured at the episode level, the variance becomes visible and actionable.
The Data Set
The study is built on a large-scale commercial medical claims data set, analyzed at the episode level.
- Geography. Four Texas metropolitan areas: Dallas–Fort Worth, Houston, San Antonio, and Austin. Restricting to a single state removes differences in regulation and insurance mandates; restricting to metros removes rural-vs-urban distortions.
- Claims base. Commercial medical claims representing approximately 40% of the commercial insurance market in the four metros.
- Care groups. 72,776 provider care groups. A care group is anchored by a physician and linked to the specific facility where care was delivered.
- Volume. 8.2 million episodes across 131 episode types covering cardiology, orthopedics, gastroenterology, behavioral health, women's health, dermatology, ENT, urology, endocrinology, and other specialties.
- Risk adjustment. Each episode is risk-adjusted for age, sex, ZIP code, clinical classifications, and episode-relevant complications.
- Credibility screening. Care groups below a minimum episode volume were excluded.
How the decomposition works
For every care group treating a given episode type, the study calculates two indices relative to the market average. A price index captures whether that care group's unit prices are above or below market. An efficiency index captures whether that care group delivers more or fewer services than the market average for the same risk-adjusted patient. Because total episode cost is the product of the price paid and the quantity of services delivered, the two indices multiplied together describe the full episode cost relative to market.
Finding 1: The Variance Is Enormous
Within the same metro area, for the same clinical problem, and after risk adjustment, the difference between the cheapest and most expensive care group for a single episode is routinely in the tens of times.
| Episode | Metro | Low | High | Ratio | Care Groups |
|---|---|---|---|---|---|
| Cataract Surgery | Austin/SA | $1,311 | $13,314 | 9.1× | 169 |
| Colonoscopy | Houston | $640 | $8,006 | 12.5× | 168 |
| Knee Replacement | Austin/SA | $21,269 | $85,052 | 4.0× | 46 |
| Hip Replacement | Dallas–FW | $37,043 | $125,146 | 3.4× | 65 |
| CABG (cardiac bypass) | Dallas–FW | $106,570 | $266,346 | 2.5× | 7 |
| Depression & Anxiety | Austin/SA | $1,522 | $44,355 | 28.1× | 1,180 |
| Depression & Anxiety | Dallas–FW | $950 | $26,492 | 26.8× | 2,212 |
| Acute ENT Conditions | Dallas–FW | $547 | $22,882 | 31.7× | 2,863 |
| Contact Dermatitis | Houston | $646 | $26,064 | 39.3× | 812 |
A patient in Houston with the same insurance and the same clinical presentation for dermatitis can end up in an episode costing $646 or $26,064, depending on which provider they see. None of this shows up on an explanation of benefits, a provider directory, or a price transparency file. It is visible only after episodes are assembled and risk-adjusted.
Finding 2: Efficiency, Not Price, Drives Most of the Variance
The central analytical exercise of this study is separating the two drivers of total episode cost.
Across all 131 episodes (equal weight)
| Driver | Share of Variance | Meaning |
|---|---|---|
| Efficiency (volume of care delivered) | 88% | Practice pattern differences between providers |
| Unit price (fee schedule) | 12% | Contracted rate differences between providers |
This result reframes the cost problem. The levers most buyers have pulled for thirty years address, at most, the 12% bucket.
Weighted by actual dollars spent
| Driver | Dollar-Weighted Share | Translation |
|---|---|---|
| Efficiency | 99.2% | Virtually all economically material variance |
| Unit price | 0.8% | Negligible when measured in dollars |
Measured against the dollars that flow through the commercial healthcare system, the fight over unit prices is a fight over roughly 1% of total spending variance.
Finding 3: Conditions and Procedures Tell Opposite Stories
Conditions: efficiency drives everything
| Condition episodes | Share of Variance | Dollar Impact |
|---|---|---|
| Efficiency (utilization) | ~100% | $5.4 billion above market |
| Unit price | ~0% (mild offset) | ($112) million offset |
For conditions, the prices providers charge are roughly in line across care groups. What varies is how much care they deliver. Two primary care practices managing the same diabetic population can produce episodes differing in total cost by 10× or more, while billing at roughly market rates for each individual service.
Procedures: unit price dominates
| Procedure episodes | Share of Variance | Dollar Impact |
|---|---|---|
| Unit price | 89% | $176 million above market |
| Efficiency | 11% | ($103) million offset |
Most existing cost-management strategies were built with the procedural episode in mind: reference-based pricing, bundled payments, centers of excellence, travel-surgery programs. They work on the 17% of spending that comes from procedures. They leave largely untouched the 83% that comes from conditions, where variance is driven almost entirely by how care is delivered.
Finding 4: Why Efficiency Varies — The Role of Physician Preference
The Oxbridge study quantifies efficiency variance. The next question: why does it vary so much between providers treating the same condition in the same city?
The foundational evidence
The Dartmouth Atlas of Health Care (1993–2019) documented systematic geographic variation in medical resource use across more than 300 Hospital Referral Regions. Supply-sensitive care — services whose use varies with local resource availability rather than clinical evidence — accounted for well over half of Medicare spending and was marked by systematic overuse in high-capacity areas.
Physician practice style: the dominant driver
The most rigorous studies use natural experiments: physician and patient migration across regions, physician exits forcing patient reassignment, within-institution comparisons holding environment constant. Their findings are consistent:
- Badinski, Finkelstein, Gentzkow, Hull, and Williams (2023, NBER) — Physicians account for approximately 30–35% of geographic variation in utilization, conservatively three times as important as non-physician supply-side factors.
- Cutler, Skinner, Stern, and Wennberg (2019) — Physician beliefs unsupported by evidence drive 35% of end-of-life spending variation and 12% of all-enrollee spending variation.
- Finkelstein and colleagues (2020) — PCP practice style explains 41–89% of primary care spending variation.
- Van Parys (2016) — Physicians at the 75th percentile of spending spend 20% more than those at the 25th percentile for comparable patients.
- Molitor (2018) — 60–80% of physician behavior is shaped by the local environment; 20–40% is an intrinsic practice style that travels with the physician.
Components of physician preference
| Component | Definition | Estimated Magnitude |
|---|---|---|
| Clinical beliefs | Beliefs about treatment efficacy not supported by evidence | 35% of end-of-life spending variation |
| Training and practice origin | Medical school culture, residency training embedded in practice style | 20–40% survives migration |
| Organizational culture | Peer influence and group norms within practice groups or hospitals | Elasticity of ~0.27–0.30 |
| Supply availability | Physician behavior responsive to local resource availability | 60–80% of behavior explained by environment |
| Financial incentives | Fee-for-service inducement to increase volume | Varies by specialty |
| Patient-physician concordance | Trust dynamics modulating ordering intensity | 3% increase in SDM reduces expenditures ~10% |
Financial incentives
Under fee-for-service payment, physicians are paid more when they deliver more services. This creates a structural incentive for higher utilization that operates alongside, and sometimes reinforces, clinical beliefs about appropriate care intensity.
In maternity care, a $100 increase in the fee differential between C-section and vaginal delivery was associated with a 3.4% increase in the primary C-section rate. In spine surgery, the Lown Institute's 2025 analysis identified more than 200,000 unnecessary back surgeries over three years, costing approximately $2 billion. In orthopedics, implant costs for total knee replacement range from $1,797 to $12,093 across surgeons within the same institution.
Finding 5: Patient Preference — Smaller but Real
Physician preferences dominate patient preferences as a cost driver. But patient preference is not zero, and it operates through a distinct mechanism with real policy implications.
Baker, Bundorf, and Kessler (2014, Health Affairs) directly measured patient preferences using a survey-based instrument linked to spending at the Hospital Referral Region level. Patient preferences explain approximately 5% of total Medicare spending variation.
When patients receive full information through structured decision aids, a substantial share reverse their stated preference for surgery. For benign prostatic hyperplasia, only 14% of fully informed patients preferred surgery. This suggests that much of observed surgical variation attributed to "patient demand" is actually physician influence operating through information asymmetry.
Shared decision-making
Brown and Hurley (2023) found that a 3% increase in shared decision-making is associated with approximately a 10% decrease in expenditures. This impact doubled for Latinx patients seen by Latinx physicians and tripled for Black patients seen by Black physicians.
Physician preferences dominate patient preferences as a driver of cost variance by approximately 6 to 1 (30–35% vs. 5% of geographic variation). SDM programs can reduce expenditures by 10% for every 3% increase in SDM, making correction of the information asymmetry one of the highest-return interventions available.
Preference-Driven Variance in Practice: Episode-Level Evidence
Five episodes from the Oxbridge portfolio demonstrate the range of preference-related cost drivers.
Lumbar spine surgery
Lumbar spinal fusion is among the most expensive and fastest-growing surgical procedures in the U.S., with hospital costs reaching $12 billion in 2014. A 2023 JAMA Network Open study found that fusion rates rose from 67.4% to 90.4% despite two landmark NEJM randomized trials in 2016 finding no benefit of fusion over decompression alone. Academic surgeons kept costs 55% lower than private practice surgeons for the same conditions.
Knee and hip replacement
In a 29-hospital study, total knee arthroplasty costs varied 2:1 between the 90th and 10th percentile despite equal outcomes. Implant costs ranged from $1,797 to $12,093 across surgeons within the same institution. Programs implementing implant price capitation found cost decreases of 20–50% without adverse effects.
Coronary artery disease and PCI
For stable coronary artery disease, cost-effectiveness analyses show optimal medical therapy has the lowest lifetime costs at approximately $22,952 versus $25,081 for bare metal stent PCI and $25,536 for drug-eluting stent PCI. Per-episode cost impact: $2,000–$12,000 for PCI over OMT.
Autoimmune conditions and biologic prescribing
Biologic drugs exceed $25,000 per patient-year, and the most expensive therapies exceed $80,000. The timing of initiation, agent selection, dosing intensity, and switching behavior are almost entirely physician- and patient-driven.
Hysterectomy and uterine fibroids
The choice of surgical approach, concurrent procedures, and the decision to recommend surgery versus conservative management produce 50–68% of procedure rate variation unexplained by patient factors. Hospital length of stay adds $3,000–$6,000 per case for open versus minimally invasive approaches.
| Category | Primary Preference Channel | Preference-Driven Variance |
|---|---|---|
| MSK Surgical | Implant selection + surgical approach | 30–60% of episode cost variation |
| MSK Condition (Low Back) | Surgery vs. conservative + injection overuse | 40–65% of episode cost variation |
| Cardiac | PCI vs. OMT + device selection | 25–40% of episode cost variation |
| Autoimmune/Dermatologic | Biologic timing + agent selection | 30–70% of drug cost variation |
| Women's Health Surgical | Surgical approach | 50–68% of procedure rate variation |
| Maternity | C-section financial incentives | 10–30% of C-section rate variation |
| Chronic Conditions (HF, COPD) | Post-acute routing + readmission mgmt | 25–50% of 90-day episode variation |
A Quantitative Framework for Healthcare Cost Variance
| Variance Component | Share | Basis |
|---|---|---|
| Efficiency variance | 88% of episode-level variance | Oxbridge Texas study (8.2M episodes) |
| Physician practice style/beliefs | 30–35% of geographic utilization variance | Badinski et al. (2023); Cutler et al. (2019) |
| Other supply-side factors | ~20% of geographic variance | Badinski et al. (2023) |
| Patient preferences (isolated) | ~5% of spending variance | Baker et al. (2014) |
| Patient health/demographics | ~12% of spending variance | Baker et al. (2014) |
| Price variance | 12% of episode-level variance | Oxbridge Texas study (8.2M episodes) |
The 88/12 split comes from the Oxbridge episode-level analysis using a single national insurer's fee schedule. Under multi-payer assumptions, the price share may rise toward 20–25%, but efficiency accounts for the dominant share under any scenario, and for conditions it remains near 100%.
Strategic Options for Employers
The findings translate into concrete benefit design choices, arranged from least to most ambitious.
Option 1: Incentive-only program
The employer sets a benchmark (typically the mean or median care group price for each episode) and shares a portion of savings when employees select providers below the benchmark. Disruption is low; the overlay layers on top of the existing plan without a network change.
Option 2: Incentives plus guaranteed-price episodes
The employer and the employee know in advance what the financial outcome will be for each covered episode. The care group's price is locked in up front through a guaranteed-price arrangement, eliminating the surprise bill.
Option 3: Allowances plus guaranteed-price episodes
The most fully developed form. An episode allowance cap is built into the benefit plan design, typically set at the 70th, 80th, or 90th percentile of care group prices in a market. This design is actuarially scorable and reinsurable.
Deployment paths
Employers can offer Episode Benefit Plans as an option alongside existing plans, as a complete replacement (quoted at equivalent actuarial value), or as an incentive-only overlay bolted onto an existing program with the cooperation of the current plan administrator.
Strategic Options for Health Plans and TPAs
The variance within the networks a plan already owns or leases is substantially larger than the variance between networks. That variance becomes the basis for a new set of product and cost-containment strategies.
Provider market consolidation makes true price competition between organizations increasingly difficult, but the data show that significant cost variance exists between care groups within the same network, and even within the same hospital or system. Fee schedule negotiation targets the smaller bucket. The larger opportunity is to surface and steer around the efficiency variance inside existing contracts.
A white-labeled episode product can deploy in approximately 120 days, with the plan retaining brand, network, contracts, and financial control.
Episode Overlays Work With Any Plan Design
Episode-based incentives, guaranteed-price episodes, and episode allowances can be layered on top of any underlying benefit structure: PPO, HMO, POS, reference-based pricing, or traditional indemnity. The episode overlay addresses a dimension of cost the underlying plan does not reach.
A PPO negotiates unit prices. An HMO manages utilization through gatekeeping. A reference-based pricing plan caps what it pays per service. None of them measure or manage the total cost of a complete episode of care. The overlay does not replace the underlying plan's network, formulary, or cost-sharing structure. It adds transparency and accountability at the episode level.
A Note on Price Data and Sensitivity Analysis
This study used the contracted payment rates of a single national health insurer, creating more homogeneity in the price component than a multi-payer blend would produce. The 88/12 efficiency-to-price split may modestly understate price variance in a multi-payer environment.
Sensitivity analyses using alternative fee schedule assumptions confirm: the core finding holds. The price share may shift from 12% toward 20–25% under aggressive multi-payer assumptions, but efficiency consistently accounts for the dominant share. For condition-based spending, the efficiency share remains near 100% under any fee schedule assumption.
Conclusion
The conventional approach to healthcare cost management has focused on price. These strategies address, at best, the 12% of cost variance attributable to what providers charge per unit of service.
This study shows that the dominant driver is efficiency: how much care is delivered, by whom, in what setting, and through what combination of services. Efficiency accounts for 88% of total episode-level cost variance across 8.2 million commercial insurance episodes. For condition-based spending, it is effectively 100%.
Five decades of peer-reviewed research establish that physician practice style and beliefs are the primary reason efficiency variance exists. Physicians account for 30–35% of geographic utilization variance. Patient preferences contribute approximately 5% independently, but much of what appears as patient demand is physician influence operating through information asymmetry.
The episode of care is the unit of measurement that makes both price and efficiency variance visible, attributable, and actionable at the individual care-group level. Episode-based benefit designs address the larger portion of cost variance that conventional approaches leave untouched, and they work with any underlying plan design.
Methodology and Limitations
Study design
Commercial medical claims from four Texas metros: Dallas–Fort Worth, Houston, San Antonio, and Austin. 131 episode types spanning cardiology, orthopedics, gastroenterology, behavioral health, women's health, dermatology, ENT, urology, endocrinology, and other specialties. Each episode risk-adjusted for age, sex, ZIP code, clinical classifications, and episode-relevant complications.
Analytical method
The price and efficiency indices are multiplicative, not additive. The decomposition uses a logarithmic (log-ratio) approach, which allows the two components to sum exactly to total variance in log terms with no residual.
What this study does not claim
The study does not assert that low-cost care groups are, in every case, delivering better clinical care. What is well established is that higher cost does not produce better outcomes. Dollar attribution figures are specific to the four-metro Texas data set. Oncology episodes are excluded due to risk adjustment complexity.
About the research
The analysis was conducted by Oxbridge Health, which develops episode-of-care payment programs and has launched Episode Benefit Plans in initial markets, supported by reinsurers. The peer-reviewed literature cited draws on studies published in Science, JAMA, JAMA Network Open, JAMA Internal Medicine, the New England Journal of Medicine, the Quarterly Journal of Economics, Health Affairs, the Annals of Internal Medicine, the Journal of Health Economics, and other journals, as well as NBER working papers and the Dartmouth Atlas of Health Care.