Skip to main content
interwell health logo

CUSTOMER STORY

Giving kidney care teams their time back with real-time AI

Interwell Health

28 minutes to 18 minutes

Average after-call documentation time for care team members, with the AI assistant drafting notes and care plan updates

140 to 175

Patients supported per care team member, with quality performance remaining consistent across key internal metrics

~160,000

Patients with kidney disease under Interwell value-based care, supported by care teams and Interwell KidneyAI

Our care teams have an incredible impact on patients’ lives. Databricks enables our care teams to focus more time and expertise on our highest-risk patients.

- Sam Gopal, SVP, Product and Technology, Interwell Health

Interwell Health is on a mission to change the trajectory of kidney disease — slowing its progression, guiding patients to informed treatment choices, and taking on total cost of care risk for one of healthcare's highest-need populations. But the care teams at the heart of that model were spending valuable time documenting care that could otherwise be devoted to supporting patients. 

By building a real-time AI assistant on Databricks, Interwell moved post-call documentation from roughly 28 minutes to 18 minutes on average, freed care teams to increase the patients each team member can support, and enabled clinical conversations to be standardized into structured, analyzable data that supports care management activities — all while keeping a licensed clinician involved on every care plan. What began as a single efficiency win has become the foundation for how Interwell intends to scale personalized kidney care to approximately 160,000 patients and beyond.

The rising stakes of value-based kidney care

Interwell Health is a value-based kidney care company that takes total cost of care risk for patients with chronic kidney disease — stages 3b, 4, and 5 through to end-stage kidney disease and dialysis. Working with commercial payers, Medicare, and Medicare Advantage populations, Interwell's goal is to improve outcomes and quality of care while reducing the total cost of care for a population that has historically been complex to manage.

The stakes are national in scale. Patients with kidney disease carry multiple comorbidities and represent an outsized share of healthcare spending — roughly 6 to 7% of Medicare beneficiaries account for an estimated 20 to 25% of total Medicare spend. Because standard case-management approaches fall short for this group, the government created a dedicated model, the Kidney Care Choices (KCC) program, that enables nephrologists to partner with dialysis organizations and value-based care entities to address the root causes of patient need. As one of the largest participants in that program, Interwell is measured against its peers on hard outcomes — and the program's future depends on participants like Interwell demonstrating real cost and quality improvements.

"Last year's results drive the step-up in next year's targets," said Sam Gopal, SVP, Product and Technology, who leads Interwell's product and technology teams. The only way to keep pace, he explained, is continuous improvement in the organization's ability to deliver — a treadmill that never stops as the bar rises year after year.

Interwell’s two-pronged care model combines provider enablement and patient care, empowering its national network of more than 2,300 independent nephrologists while supporting patients through virtual and in-person care. Knowing who needs what care, when, and from whom — a nephrologist, a care coordinator, a dietitian, a social worker, or a nurse— is a large-scale data problem. Interwell needed more sophisticated risk stratification to ensure patients receive the right level of support, directing clinical resources to those with the greatest needs while supporting lower-risk members digitally. 

That was hard to do on the data foundation Interwell had inherited. Although the majority of its network uses Interwell Acumen™, the company’s Epic® Connect EHR, Interwell also needed to integrate clinical data from a range of other EMRs used by its distributed nephrology network. Dialysis treatment data came from partner systems. Interwell received rosters, claims, and labs from multiple sources in varying formats and on different cadences. Reference data arrived from CMS, and internal actuarial and population health teams each worked off their own views of the information.

"Prior to Databricks, our footprint was really a challenge," Gopal said. Teams spent much of their day on pipeline management, double-checking and validating data rather than working at the top of their license. Layered on top were the constraints unique to healthcare: protecting PHI, meeting HIPAA and regulatory requirements, and building a defensible architecture Interwell could point to contractually — one that keeps sensitive data inside its own environment rather than sending it to a black-box external model. Even hiring was harder, with the need for US-based data engineers narrowing the talent pool.

Nowhere was the cost of that friction more visible than in the daily work of Interwell's care teams. Focused on caring for patients, care teams were burdened with administrative documentation — often captured across notepads, spreadsheets, and other disparate places with no streamlined workflow, especially while on the phone. Every minute spent documenting after a call was a minute not spent providing clinical care to a high-risk patient.

Building real-time AI in the flow of care

Interwell's answer is a real-time assistant that works alongside care team members (nurses, registered dietitians, and social workers) during their calls with patients. Rather than a point solution, the AI assistant is part of the Interwell KidneyAI suite, a collection of capabilities that reflect and support the clinical work happening on every call, built on the Databricks Data + AI Platform.

As a nurse speaks with a patient, speech is transcribed using secure, HIPAA-compliant Databricks technology to assist with documentation. Layered LLM processes characterize and tag each call — patient call, multi-patient call, clinician-to-provider — and then turn the transcript into structured documentation drafts for clinical review: a care plan call note that standardizes the record and supports clinician-to-clinician handoff, suggested updates to the patient's care plan across problems, goals, and interventions, and a goal progress note for each goal, along with any barriers that surfaced during the conversation. The distractions of documentation are handled after the call, so the clinician can stay present with the patient during the call.

The real-time nature is not a nice-to-have — it's the point. To allow clinicians to move more quickly from one patient to the next without waiting, the AI assistant generates notes and drafts care plan updates in near real time, delivered seamlessly into the tools clinicians already use. Interwell set an internal SLA of 95% of updates in under two minutes and is comfortably exceeding it, using serverless compute, dynamic scaling, and warmed clusters to meet demand across staff working in different time zones. "We can't be sitting around for 10 minutes waiting for these updates," the team noted; the work has to land directly in the operational workflow, not in a separate application.

That workflow is the electronic health record itself. Interwell built an API-based integration to meet care teams where they are at the point of care.

The AI assistant never removes the clinician from the decision. Its outputs arrive as draft suggestions and starting points — editable, modifiable, and validated by a care team member before anything is attached to a patient's chart. Human in the loop is one of Interwell’s core guiding principles for deploying AI, and a clinician always has the final say on the care plan. Clinical judgment always remains with the clinician, and no AI-generated content is incorporated into the medical record without human review. In practice, the care team members describe the AI as capturing the essence of what they're trying to say, separating personal dialogue from clinical content, and leaving them just "a few tweaks" from done.

There's a strategic dividend, too. Because the AI drafts in a consistent problems-goals-interventions structure — where care team members once documented freeform — Interwell can now draw correlations between the care plans its teams put in place and the outcomes patients experience, feeding a virtuous loop that refines its standards of care.

Interwell delivers Interwell KidneyAI and its broader analytics on a deep set of Databricks capabilities:

  • Unity Catalog for schema-level governance and role-based access controls, bringing fragmented sources into one governed environment.

  • A medallion architecture that standardizes data from multiple payers into a single schema while preserving history and slowly changing dimensions.

  • Model Serving to securely deploy hosted and open-source models behind internal APIs — a key part of Interwell's defensible, data-stays-in-our-environment architecture.

  • Vector Search powering Interwell's RAG solutions, grounding answers in real internal documentation through embedding similarity, score filtering, and deduplication.

  • MLflow for experiment tracking, model registries, and the observability and explainability Interwell relies on to defend its models in a regulated setting.

  • Feature Store, with precomputed clinical and claims features reused across multiple predictive models rather than re-derived each time.

  • Databricks Asset Bundles for deployment and publishing models downstream.

Interwell KidneyAI predictive models — a suite of classical ML for hospitalization risk, readmission, and CKD progression — run on the same foundation, providing population health insights that assist Interwell’s care teams to prioritize outreach and allocate resources.

The platform also changed how Interwell's people work together. Using its time at Databricks' Data + AI Summit to train data engineering, population health analytics, and actuarial teams side by side, Interwell — a virtual company — built shared understanding and working relationships in one place. Follow-on lunch-and-learns and a deliberate choice to bring teams together rather than run siloed office hours have driven cross-pollination across the business, breaking down silos that previously slowed the organization down.

More time for patients — and a platform for what's next

The headline result is time returned to patient care. After-call documentation that once ran around 28 minutes now takes roughly 18, and the capacity that unlocks is substantial: patients supported per care team member moved from 140 to 175 with quality performance remaining consistent across key internal metrics. Across care team members supporting approximately 160,000 kidney patients, that shift lets Interwell reserve scarce clinical time for its highest-risk members while continuing to reach a large and growing population.

The teams closest to the work feel it. Care team members have anthropomorphized the AI assistant, referring to it by the internal name “AVA.” "I love her" is a common refrain — and their day-to-day feedback captures why: assessment documentation that was "so much less time consuming," notes that generate quickly during calls, and a process efficient enough that one nurse said she almost feels like she's forgetting a step. As one put it after her first full screening with AVA: "just a few tweaks and boom, done." 

Beyond efficiency, the standardization of documentation has given Interwell a cleaner, more holistic view of each patient and a data asset it didn't have before — the raw material for correlating care approaches with outcomes and sharpening its care model over time. And in the one arena where value-based kidney care participants are benchmarked head-to-head, the government's KCC program, Interwell is delivering measurable shared savings and quality results, including optimal dialysis starts. 

Building an AI assistant was the first big win, and Interwell sees it as a foundation rather than a finish line. The organization is building out its "gold layer" to answer a wider set of questions about high-risk populations and emerging trends, with the ambition to scale personalized care down to an individual journey for each of its 160,000 members. It is exploring  additional automation opportunities for administrative tasks — scheduling, education delivery, self-service engagement — to allow care coordinators to spend more time on meaningful patient interactions. And it is looking at extending Databricks Apps into its nephrology provider workflows, meeting providers where they are much as Interwell KidneyAI meets care teams today. Natural-language analytics through Databricks Genie, still early for this use case, is becoming an increasingly active part of how teams across Interwell explore their data.

"You're giving them the intelligence to spend more time with patients," is how the story lands — and for a company working to prove a new model of kidney care and change how care is delivered, that time is the whole point.

FAQ: Interwell Health and Databricks

Interwell Health is a value-based kidney care company that takes total cost of care risk for patients with chronic kidney disease, from stage 3b through end-stage kidney disease and dialysis.