How Generative AI Can Support Value-based Care

By Rahul Sharma, Chief Executive Officer, HSBlox
LinkedIn: Rahul Sharma
X: @RS_HSBlox

A 360-degree view of the patient is essential in providing clinicians, payers, and patients themselves with actionable insights that can lead to improved outcomes.

Artificial intelligence (AI) is increasingly playing major roles in data digitization, prediction analytics, and decision support systems. For example:

  • AI technologies like NLP (Natural Language Processing) and computer vision enable the data digitization process. NLP supports speech-to-text and vice versa, document and data conversions, patient notes, processing of unstructured data, and query support systems. Computer vision includes augmented reality (AR), virtual reality (VR), telehealth, and digital radiology.
  • Machine learning (ML) algorithms can help reduce claims denials by facilitating more accurate detection of errors in billing and coding and also are used to optimize the pharmaceutical supply chain.
  • Deep learning and cognitive computing tools can rapidly process huge data sets, helping to inform precise and comprehensive risk forecasting and providing recommended actions that improve patient outcomes.

Predictive AI versus Generative AI

Both Predictive AI and Generative AI rely on learning from data, but they do so in different ways and have different goals.

Predictive AI is used to forecast outcomes by identifying patterns in historical data and then using those patterns to predict trends. For example, a predictive AI model can be trained on a dataset of patients’ longitudinal health records (LHRs) and then used to predict which patients are most likely to run into a specific health scenario in the future.

Generative AI is able to create new content – such as text, audio, images, and code – by learning from existing datasets and then generating new content based off the training data. For example, Generative AI can analyze a patient’s genetic and molecular data to identify unique genetic markers. Based on this information, providers can recommend personalized treatment plans that target specific genetic mutations.

Using Gen AI to enhance VBC

Value-based care (VBC) is focused on outcomes rather than volume (as is the case in fee-for-service models). For a value-based program to work, however, many operational, financial, data complexity/accessibility, and technology-related challenges must be overcome.

Here are several applications for Gen AI within a VBC model:

Streamlining contract building. The process of developing, reviewing, and implementing VBC contracts today is manual and thus very time-consuming. Gen AI can be deployed to help streamline this approach.

By training large language models (LLMs) on past contracts, Gen AI can generate new contracts based on past patterns that can then be reviewed and finalized. LLMs can analyze individual components of complex and lengthy contracts, such as different variables and their values, pricing information, attributes of different clauses, and expiry dates. This information then can be extracted within seconds and presented with a simple-to-use workflow that allows users to finalize contracts within days.

Improving care management process(es). Successful patient care management relies heavily on the effective use of data, processes, and systems by a team usually comprised of physicians, nurses, CBOs (community-based organizations), care managers, and social workers. The strategy is to use timely interventions that reduce patients’ health risks and decrease the total cost of care.

There are four basic steps to a personalized care plan:

  1. Population stratification (using risk stratification techniques)
  2. Alignment of care management services to the needs of the patient (this can be accomplished through interacting with the patient in a personalized manner to ensure buy-in into the plan)
  3. Preparation of care plan and device monitoring for the patient for proactive care
  4. Assigning appropriate personnel to care plan team for execution, follow-ups, etc.

Gen AI isn’t necessary for patient risk stratification. The challenge starts with contacting and engaging the patient; communication channels (emails, phone calls, SMS/MMS messages, snail mail) must be used persistently to yield results. Healthcare organizations can use Gen AI to help personalize outgoing communication (conversational AI) based on past patient interactions, including any language translation preferences and level of education to ensure recipients understand the content of the communication.

Once a patient is engaged, the care team can prepare the personalized plan and put in place the device monitoring/data collection protocols. Non-clinical and administrative steps like medication reminders, scheduling appointments, scheduling check-ins for telehealth appointments, creation of alerts and notifications when things do not go as planned, prescription refills, prompting for daily exercise under the care plan – all can be personalized and automated using Gen AI.

Other healthcare use cases for Gen AI

The number of use cases being worked on using Gen AI is expanding. While some use cases are in the research/concept stages, a few have been moved into production. Use cases fall within the following categories:

  • Administrative tasks. Several companies (including Doximity, Abridge, and DeepScribe) are developing solutions that automate documentation, claims handling, preauthorization and appeals, patient onboarding, and scheduling. These solutions will help reduce the administrative burden on physicians, nurses, and the healthcare staff, which can reduce human mistakes.
  • Prevention of costly medical errors. Safety lapses kill thousands of patients every year. Gen AI, coupled with direct video stream surveillance, can monitor physicians, nurses, and hospital staff, compare their actions to evidence-based guidelines, and warn clinicians when it appears they are about to commit an error that could endanger a patient.
  • Medical education. Many patients today can easily find medical information online. Unfortunately, that information may be confusing and even incorrect. Gen AI-powered bots may evolve to be reliable sources of information for patients.
  • Clinical decision support. A clinical decision support system can simulate how an experienced physician would make the decision based on the data available.

Pharma use cases

In addition to clinical and administrative use cases, Gen AI is being used by pharmaceuticals for:

  • Drug discovery. NVIDIA now offers a set of Generative AI cloud services from which any company can do customization of AI foundation models to accelerate drug discovery and research work in the fields of genomics, molecular and cellular biology, and chemistry. This is already being utilized by startups, such as Evozyne and Insilico Medicine, as well as larger firms like Amgen.
  • Cancer research. Gen AI is being used to analyze patient tissue samples and employ functional precision oncology to improve patient outcomes.
  • Clinical trials and precision medicine. Gen AI can accelerate and improve clinical trials and precision medicine therapies.

The success of VBC models is heavily dependent on how easily healthcare organizations can share and leverage data to optimize patient and population outcomes. Gen AI can be deployed across numerous use cases in healthcare to meet VBC initiatives and goals, including contract-building, developing personalized patient care plans, reducing medical errors, clinical decision support, and streamlining administrative tasks. Pharmaceuticals can use Gen AI to accelerate drug discovery, improve cancer treatments, and support clinical trials.