The Promise of Artificial Intelligence in Precision Medication Dosing

In the United States alone, drug-related problems in patients account for $177 billion in costs annually. Approximately $20 billion of this cost is ascribed to preventable adverse drug reactions, and 30–50% of these preventable side effects are due to dosing errors. It is one of many reasons why the movement from a one-size-fits-all approach to a personalized, precision drug dosing approach is such an important development in care delivery today.

These enormous costs reflect how complex the process of prescribing drugs has become for healthcare providers.  Medications often have potent effects on the body–both intended and unintended–and drug doses need to be precise to achieve optimal outcomes while avoiding side effects.

To inform dosing decisions, doctors have relied primarily on their clinical experience, knowledge of the medications they are prescribing, and paper-based recommendations for dosing from drug manufacturers and the FDA. However,  these recommendations are often imprecise, as they draw from clinical studies that may or may not accurately reflect an individual patient's response to the medication. Therefore, there is an upper limit to the precision with which medication can be dosed using traditional methods.

AI Transforms Dosing and Gives Patients a Personalized Fit

The most compelling approach to solving this important problem to date is with the application of artificial intelligence to enable precision dosing. Precision dosing is an umbrella term that refers to the process of transforming a "one-size-fits-all" therapeutic approach into a targeted one, based on an individual's demonstrated response to medication

Precision dosing has been identified as a crucial method to maximize therapeutic safety and efficacy with significant potential benefits for patients and healthcare providers, and AI-powered solutions have so far proven to be among the most powerful tools to actualize precision dosing.

In 2008, Dr. Donald M. Berwick, former Administrator of the Centers for Medicare and Medicaid Services, articulated what has been referred to as the triple aim of the U.S. healthcare system: to improve the experience of care, to improve the health of populations, and to reduce the per capita costs of healthcare. An October 2020 report by KLAS and the Center for Connected Medicine found artificial intelligence to be one of the most promising emerging healthcare technologies for clinical decision support, to help the US healthcare system achieve these aims.

Better Decision Support in Dosing Achieved

Despite significant promise, applications of precision dosing have tended to be difficult to scale (Source 1), due to factors that make precision dosing challenging to generalize, while maintaining efficacy (Source 2). In fact, even some of the highest-profile precision medicine efforts have had difficulty demonstrating efficacy at scale (Source 3).

Effectively dosing a drug is a multifactorial problem because it is difficult to create a series of rules that comprehensively accounts for all of the variables impacting a particular observed response. Past approaches have relied on clinicians' professional expertise to identify as many of those variables as possible, but not even the most skilled clinicians can incorporate all relevant factors. Without technology, decisions are made on the basis of associative reasoning driven by the clinician's training and interaction with the data–a mix of opinion, training, and experience. The quality of these factors varies enormously among individuals, and a prescriber may be tired, distracted, or pressed for time, creating increased variability in how decisions are made.

An AI-powered algorithm, on the other hand, is very consistent and primarily considers two factors in the decision-making process: the input and the outcome. Once an effective control algorithm is defined to achieve a certain outcome, it will consistently drive towards that outcome. By simulating the best of human intelligence with a mathematical algorithm, the results are consistent regardless of environmental factors.

5 Factors That Came Together to Make Now the Right Time for AI-Powered Dosing

Several factors have come together to create the necessary conditions to begin realizing the potential for AI-powered precision dosing:

Technological advancements in computing allow us to process large, complex datasets quickly, making AI solutions practical.

Public familiarity with artificial intelligence as an effective tool for solving complex problems makes physicians comfortable incorporating such tools in clinical settings.

Reliable data is now available in electronic medical records and is standardized in a manner that is much more ingestible by algorithms as compared to free-form paper medical records.

Big data analytics techniques have also made applying artificial intelligence and control algorithms to complex datasets much more practical and efficient. We can draw on data from millions of patients to design and test algorithms in silico to predict effectiveness and iterate quickly. This is a vast improvement on expert systems that are based on a clinician's smaller number of patients, possibly in the thousands or hundreds, that are generally only possible to test in much more costly and risky clinical trials.

Increasingly complex and powerful drugs have been developed that impact basic physiologic processes. Drugs that impact multiple physiologic processes and have a narrow therapeutic window (the "sweet spot" between toxicity and ineffective therapy) have become more prevalent. These are the types of drugs for which AI-powered drug dosing can provide the most benefit.

Chronic Anemia Offers an Especially Powerful Opportunity to Apply AI-Powered Dosing

There are more than 550,000 dialysis patients suffering from End-Stage Kidney Disease (ESKD) in the United States. Dialysis patients are at high risk for adverse outcomes, have complex medical problems, and are typically receiving multiple medications that can interact with one another. As a result, they need innovative approaches to manage the medications that they receive.

Almost 90% of dialysis patients experience chronic anemia and are treated with Erythropoiesis Stimulating Agents (ESAs). However, exposure to high doses of ESAs is associated with an increase in adverse cardiovascular events, so the primary clinical intent is to use the minimum amount of ESA necessary to prevent patients from requiring blood transfusions while avoiding potentially serious or fatal adverse cardiovascular events.

Before Dosis developed Strategic Anemia Advisor, dosing recommendations for ESAs were protocol-based directives implemented in dialysis units as a one-size-fits-many approach. The ability to make dosing highly personalized through the use of AI translates into significantly improved patient outcomes with far less harmful drug exposure for patients and reduced costs for dialysis units, insurance providers, and the U.S. healthcare system.

A Standard of Care for Chronic Disease Dosing in the Future

AI-powered precision dosing will likely be the standard of care for chronic disease management in the future. Artificial intelligence is a valuable tool that can enhance a physician's ability to practice and make the best judgements possible, improving the cost of care and the quality of care itself.

Dosing anemia drugs is only one, specific example of the impact that AI can have on medication prescribing. Dosis has already begun a trial of an AI-based intravenous iron dosing protocol, as an adjunct to Strategic Anemia Advisor. In addition, Dosis has developed a tool that informs the simultaneous dosing of three different types of medication that are used to manage mineral and bone disorder, a common comorbidity in kidney disease patients. This application will be the first of its kind, modelling three interdependent biological variables and three medications simultaneously that impact these values to return them to normal levels. Once AI for precision drug dosing is widely adopted, it will be extremely unlikely for the industry to revert back to previous dosing methods. The efficacy gap between AI-powered tools and legacy dosing methods will also only widen, as more data is incorporated into these tools.

In 10 years, AI-driven dosing models will likely be the standard of care across the healthcare spectrum, used for a wide variety of drugs like warfarin, insulin, and immunosuppressives. Indeed, any drug that is administered chronically and has a narrow therapeutic range is a good candidate for AI-driven dosing.  In addition, as more tools are developed and more opportunities to use those tools are identified, we will see exponential growth in the use of AI to drive therapies.

Author

Dr. George Aronoff is Chief Medical Officer of Dosis, Inc and co-inventor of Strategic Anemia Advisor, now commercialized and made available for use by Dosis Inc. He has over 30 years of experience in nephrology. He was previously Chief of Nephrology & Hypertension at the University of Louisville, where his research with Drs. Brier and Gaweda focused on using AI to dose ESAs in dialysis patients. He received his M.S. in Pharmacology and M.D. from Indiana University.