The Impact of Generative AI on Drug Development, Healthcare, and Patients
You can’t swing a bat today without hitting something with “AI” tacked onto it. However, I usually see articles written about only one aspect of Generative AI use. Here, we will dive into the many other ways it can create a positive impact on our industry.
Let’s start at the beginning with drug discovery.
In Silico Drug Discovery
Today, many companies are employing in silico drug discovery. This refers to the use of computer simulations and computational models to identify and develop new pharmaceutical compounds. This approach leverages various algorithms, databases, and simulation techniques to predict the behavior of drug candidates, significantly accelerating the drug development process while reducing costs and the need for extensive laboratory experiments. Using this form of drug development, generative AI analyzes vast datasets to identify patterns and predict the activity of new drug candidates. It accelerates the identification and optimization of drug candidates compared to traditional methods. By identifying biological targets for new drugs and finding potential compounds that interact effectively with the target, new compounds can be developed that enhance effectiveness and reduce side effects.
Overcoming the Lack of Diversity in Clinical Trials
Clinical trials have long been criticized for their lack of diversity, with a significant majority of participants being white males.1 This lack of representation can lead to skewed data and treatments that may not be as effective for other demographics. Generative AI is analyzing existing clinical trial data to identify patterns of bias. By recognizing these patterns, it can highlight the underrepresentation of specific groups, such as women, minorities, and socioeconomically disadvantaged individuals. This analysis can inform trial designers about the necessary adjustments to make trials more inclusive. Pharma isn’t just leveraging generative AI to improve the design of their trials; it’s running FDA-approved clinical trials without a traditional control arm. They are using AI to crunch data from previous clinical trials, EHRs, and other medical databases to generate synthetic control arms. The training models are validated to ensure they accurately represent what would be expected in a real control group. Trials for rare diseases, where finding enough participants for a control group can be challenging, can be fully enrolled in less time and at a lower cost. And with all participants receiving the active treatment, the ethical concerns about withholding treatment from a control group evaporate.
Putting Theory into Practice
Luckily we don’t have to wait for ongoing trials to be completed and a drug to be approved before integrating generative AI into clinical care. It is being used right now to analyze data, detect disease, and make treatment recommendations. For example, the gold standard of care for breast cancer before generative AI utilized a combination of mammography, ultrasound, and biopsy for diagnosis. Radiologists manually examined mammograms and ultrasound images to identify potential tumors. Biopsies were performed on suspicious areas to confirm the presence of cancer. The standard treatment pathway for breast cancer typically followed a sequence of surgery (lumpectomy or mastectomy), followed by chemotherapy, radiation therapy, and hormone therapy depending on the type and stage of cancer. Treatment decisions were largely based on clinical guidelines, patient characteristics, and the experience of the oncologist. This approach involved variability in interpretation among radiologists, potential delays in diagnosis, and a one-size-fits-all treatment strategy that did not account for the molecular characteristics of the tumor. With generative AI, the standard of care has changed. Now, AI algorithms analyze mammograms and ultrasound images more consistently than human radiologists, reducing false positives and negatives. AI tools can identify subtle patterns and anomalies that may be missed by the human eye, leading to earlier detection of breast cancer. Generative AI has enabled the development of personalized treatment plans based on the genetic and molecular profile of the tumor. The algorithms analyze data from genomic tests (such as Oncotype DX2 and Mammaprint3) and integrate it with clinical data to recommend the most effective treatment regimen for each patient. This approach tailors therapy to the individual characteristics of the cancer, improving treatment outcomes and reducing unnecessary side effects. AI-driven predictive models assess the risk of recurrence and guide decisions about adjuvant therapies. These models use patient data to predict how likely the cancer is to return, allowing oncologists to make more informed decisions about post-surgery treatment. AI also aids in monitoring patient progress during and after treatment, providing real-time insights into how well the treatment is working and enabling timely adjustments.
Optimizing Health Systems
Health systems are training generative AI platforms to iterate over all kinds of data in order to generate recommendations. Not only to optimize care, but to reduce costs. EMR data, CMS data, ICD-10 codes, lab data, image data (X-ray, MRI, etc.), signal data (EKG, ultrasound, etc.), and many more sources are being ingested and processed. We’re talking exabytes of data. For example, a medical center could use a deep learning-based model to discern if a patient has COVID-19 based on their chest x-rays. They could also create a logistic regression-based model that attempts to identify whether or not a patient will need intubation. The data sets used could include chest x-ray and EHR data from other centers within a 150-mile radius. The value of the generative outputs could tell the center how many patients they can expect to treat and release from the ER, how many they can expect to admit, how many beds and ventilators they need, specific treatment protocols by demographic characteristics, and also projected rates of death based on other underlying patient comorbidities.
Managing Risk and Reimbursement
Various reimbursement models are used to pay providers for their services by insurers. These models influence how care is delivered and how providers manage patients. Below are just three models we use in the US: In a Fee-for-Service (FFS) model, providers are paid for each service they perform, such as office visits, tests, and procedures. Payments are made based on the quantity of care provided, not the quality. In this model, the providers have the least risk of cost overruns. However, it can lead to overutilization of services and may not incentivize quality or outcomes. In a Capitation model, providers are paid a set amount per patient per period, regardless of the number of services provided. This model aims to control costs by giving providers a fixed budget to manage the care of their patients. It encourages efficient use of resources, promotes preventive care, and holistic management of patients. While it’s more of a shared-risk model, providers bear the financial risk of high-cost patients. In a Global Budget model, providers receive a fixed total budget to cover all the healthcare services they provide to a specific population over a set period. This budget is intended to cover a wide range of services, including preventive care, acute care, and chronic disease management. Providers are motivated to eliminate waste, provide care efficiently, and focus on preventive care and early intervention. Providers may face financial challenges if the budget does not adequately cover the healthcare needs of the population. There is also a risk that providers might reduce necessary services to stay within budget, impacting the quality of care. So where does generative AI come in? AI can ingest enormous amounts of data to generate the ideal blend of models for any given health system and its payers. It can train on claims data, demographic data, provider performance metrics, utilization data, clinical data, outcomes and RWE data, regulatory and policy data, and more to generate optimal contracting recommendations down to the individual payer and provider level.
Empowering Patients
More than half of the US population is digitally savvy. The use of digital technology among people aged 50 to 64 is virtually the same4 as those 18 to 29. Digital natives, particularly teens and young adults, are increasingly utilizing generative AI and digital platforms to become more informed about their health therapies and treatment options. Conversational AI allows patients to simulate human-like interactions, providing instant answers to health-related questions and facilitating communication with HCPs. Screeners, for example, can take the form of a conversation with a nurse, versus a PDF with questions and check boxes, or a canned interactive online tool. When we humanize care and show empathy, we create an experience where patients interact with synthetic HCPs that have the bedside manner patients deserve.
Closing the Loop
When we leverage generative AI the right way, from the very beginning of product development all the way through to treating patients, everybody wins. Just as AI is constantly iterating, learning, transforming, and improving, so too can patient outcomes—our true goal.
