Negotiating full-risk specialty care arrangements is one of the highest-stakes maneuvers in value-based care. Because the margin for error is razor-thin, minor variations in look-back windows, clinical exclusion criteria, or qualifying encounters can easily compound into millions of dollars of unmanaged financial exposure across multi-year contracts. Operators cannot afford to wait weeks for static actuarial models to run every time a single variable shifts during a negotiation.

In this HTN Expert Learning Event sponsored by Ursa Health, Dr. David Johnson (Co-Founder & Chief Physician Executive of Atlas Oncology Partners) and Dr. Robin Clarke (CEO of Ursa Health) pull back the curtain on what it takes to bring a de novo, total cost of care specialty model to market. Alongside Katie Sylvester (AVP of Analytics Engineering at Ursa Health), the discussion dives into how Atlas moved past slow, static actuarial models to embrace a dynamic, parameterized data architecture that allows them to simulate, negotiate, and operate complex risk portfolios at speed and scale. 

This is a summary of key takeaways from our conversation. You can check out the full recording below.

This summary was developed with support from AI.

13:00 In oncology and other complex specialties, traditional episodic value-based care models — such as the Oncology Care Model (OCM) and Enhancing Oncology Model (EOM ) — fail to capture true clinical value because their boundaries are fundamentally mismatched with the longitudinal disease trajectory. Under an episodic framework, the triggering event for an episode is typically the initiation of systemic therapy itself. But because the vast majority of the episode's cost is the predetermined, sunk cost of the drugs, the actual addressable spend left for providers to optimize is heavily compressed.

14:00 Furthermore, clinical heterogeneity within identical tumor sites makes static episodic pricing actuarially unstable. For instance, grouping all breast cancer episodes together ignores massive variances in disease complexity, progression, and indicated medical modalities. As models get increasingly granular to account for these clinical variations, they run directly into the "small N problem," where patient volume per cell drops too low to safely assume financial risk.

To build a viable business model around reducing addressable spend, Atlas Oncology shifted from rigid episodes to longitudinal, diagnosis-based population health risk. This model tracks attributed cancer patients from initial diagnosis all the way through end-of-life, capturing the natural waxing, waning, remissions, and recurrences of the disease. Rather than just taking accountability during a discrete window of chemotherapy, for example, the specialty entity manages populations longitudinally across lines of business, focusing on addressable cost levers through ongoing high-touch engagement, proactive symptom management, and directly addressing medical and psychosocial comorbidities by a care team embedded in the oncology practice.

15:00 When designing specialty risk models, health tech enablers often try to copy existing longitudinal frameworks from nephrology or primary care ACOs, but these comparisons fall apart under clinical scrutiny. Dr. David Johnson highlighted the stark divide in predictability and velocity of innovation:

16:00 Primary care capitation and TCOC models present a different structural flaw: accountability sits with the wrong clinical agent. Primary care models excel at chronic disease prevention and broad population health management, but when a patient develops a high-acuity condition like cancer, the primary care physician (PCP) ceases to be the primary driver of utilization.

"When a patient has cancer or heart failure, the specialist is really the one that's influencing care and, most importantly, driving the engagement," Dr. David Johnson explained. Once diagnosed, the specialist becomes the patient's central point of contact and coordinates the care that dictates hospitalizations, imaging, and downstream costs. If the specialist is excluded from the underlying financial arrangement, the model cannot effectively influence clinical behavior or spend. Specialty VBC requires a specialty-native data model that accounts for these distinct acuity profiles, secular trends, and care-delivery nuances rather than forcing an administrative primary care overlay onto specialty clinics.

26:00 The historical bottleneck in validating value-based contracts is the slow, manual loop of actuarial modeling. Traditionally, changing an attribution rule or adjusting a look-back period required data engineers to manually rewrite extensive ETL (Extract, Transform, Load) code, rerun queries, and hand off static outputs to business teams. This "black box waiting time" severely slows down negotiations with payers.

27:00 To overcome this, a model was built that maps varied payer claims packages into a unified, core data model, abstracting the underlying contract terms into dynamic parameters within a user interface. Instead of hardcoding logic into siloed data steps, engineers and business analysts utilize numeric, Boolean, and structural toggles directly within the platform UI.

Katie Sylvester described these parameters as "dials we can turn … fine-tuning a contract. You have a set of dials in front of you. Turn the dial, rerun the model, and then every step of the funnel instantly recalibrates.” Because these parameters are maintained as metadata directly within the analytics platform UI rather than isolated spreadsheets, the exact configuration settings travel seamlessly with the data assets. Analysts reviewing performance metrics like Per Member Per Month (PMPM) spend or risk scores can immediately see the exact parameters that generated those figures without needing to dig into back-end code.

32:00 While parameterization makes it easy to modify a risk model, it introduces a downstream engineering challenge: running a new configuration typically overwrites the previous results, erasing the baseline data needed for comparison. During multi-payer contract negotiations, business users need to simultaneously evaluate multiple scenarios side-by-side to understand the exact financial implications of shifting risk terms.

33:00 To solve this, a database architecture was engineered to allow a single claims data package to run through the same attribution model under multiple distinct parameter configurations simultaneously. Each unique combination of parameters is saved and tracked as a distinct "scenario.”

This simultaneous execution causes a "grain size explosion" across all upstream and downstream assets in the data model, scaling the data rows linearly from 1X to 3X or more depending on the number of scenarios active. Managing this volume requires an infrastructure that runs attribution components such as membership criteria, disease identification, and engagement rules independently and in parallel within a unified, longitudinal timeline object.

34:00 By evaluating attribution step-by-step for each member-month in parallel, the platform avoids sequential processing bottlenecks. The real business value comes at the presentation layer: analysts can build a single dashboard to slice and dice financial metrics like PMPM trends, and then simply split the visualization by the scenario parameter. This eliminates the need to export, manually align, and stitch together separate Excel workbooks for every minor contract iteration.

42:00 A primary friction point in specialty value-based contracting is the mismatch in operational speed between agile health tech startups and slow-moving payer actuarial teams. Startups can iterate on risk models in real-time, but payers often require multi-week review cycles to validate assumptions internally. To bridge this gap and establish trust, the risk-bearing entity must take on the heavy lifting of the initial opportunity analysis, delivering completely transparent and reproducible data logic rather than a proprietary "black box.”

43:00 The technical solution lies in changing how ETL pipelines are built. Instead of deploying a single, monolithic script that ingests raw data and directly outputs a final cohort table, the platform's architecture forces the creation of discrete intermediate tables for every single step of the attribution funnel.

For example, one intermediate table maintains the broad clinical population; the next table isolates and drops members already attributed to a full-risk primary care ACO; a subsequent table applies specific clinical exclusion parameters. Providing these explicit intermediate checkpoints allows payer actuaries to easily audit the logic, reconcile data differences, and run matching queries on their internal files. This radical transparency changes the actuarial conversation from a defensive debate over data validity to a collaborative, win-win refinement of the contract.

21:00 A common and costly error for health tech enablers is treating pre-contract modeling and post-contract operations as two completely separate software challenges. Companies often build custom actuarial models in standalone toolkits to close a deal, only to face massive technical debt and months of delay when trying to rebuild that logic into an ongoing population health management and contract reconciliation platform. Atlas Oncology avoided this trap by deploying an infrastructure designed to pivot directly from pre-deal simulation into daily market operations.

The exact data funnel used to back-test historical claims and simulate risk parameters remains fully active after the contract is signed. When live, concurrent claims data streams into the platform, the system continues running patients through the parameterized funnel. In a post-go-live environment, this serves several critical operational functions:

37:00 While this technical framework was battle-tested through an oncology case study, the architecture itself is specialty-agnostic. This shift from blunt primary care overrides to a dynamic, parameterized data model can serve as a template for any high-acuity, longitudinal specialty, whether the goal is to better manage congestive heart failure, behavioral health, severe inflammatory bowel disease, dementia, or any other complex disease.

The core infrastructure requirement remains identical across clinical domains: operators must possess a data model capable of ingesting varied payer packages, mapping them into standard clinical timelines, and exposing contract variables as real-time configurable parameters. By moving past hardcoded logic and siloed spreadsheets, risk-bearing specialty partnerships can de-risk their financial exposure, align cleanly with slow-moving payer actuarial processes, and immediately bridge the gap between pre-deal simulation and daily EMR workflow execution.

7:00 If your organization is navigating the complexities of specialty value-based care contracting and requires the structural flexibility to simulate, negotiate, and operate advanced risk frameworks seamlessly, reach out to the team at Ursa Health.

CEO, Dr. Robin Clarke: [email protected]
AVP of Analytics Engineering, Katie Sylvester: [email protected]