Medicine is undergoing a technological revolution. Current clinical treatment guidelines are the result of an average of trial population drug response, however, an individual’s response can differ significantly from that of the average, potentially leading to adverse drug reactions or ineffective treatments. The next era of medicine applies machine learning to clinical data and tailors the treatment to suit the needs of each individual patient. This is precision medicine.

Intuitively, you can think of drug benefit (i.e. net utility of drug efficacy and toxicity) to be dependent on two main factors – patient prognostic variables (e.g. age, sex, genetics, etc.) and treatment regimen (dose, frequency). It is at the intersection of these two factors where the maximum drug benefit is achieved, a treatment specific to each individual patient (Figure 1).

figure 1

Precision medicine uses data about a patient’s own biology and physiology to increase our understanding of the drug response surface to tailor the treatment in order to maximize drug benefit. By analyzing clinical data with mathematical models, precision medicine is beginning to show its immense potential in improving patient care.

The ultimate goal is to translate our fundamental and quantitative understanding of the drug response surface into the hands of healthcare practitioners. At InsightRX, our goal is to leverage mathematical models and machine learning to enable precision medicine at the point of care.

Addressing Clinical Needs with Models and Algorithms

Busulfan is a chemotherapy drug used to treat leukemia and to prepare patients for organ transplant. Correct dosing is both a patient care and economic issue. Busulfan overdosing can lead to veno-occlusive disease at an estimated cost of $300,000 per episode, while under-dosing may cause transplant failure necessitating a second transplant at a cost of up to one million dollars. By specifically tuning the dose of Busulfan to each individual’s needs, both outcomes can be avoided. Working closely with clinical experts at UCSF, our team developed a technology – InsightRX – to optimize and individualize Busulfan dosing using a mathematical model built on clinical data from Busulfan. The usefulness of this platform was immediately apparent across other therapeutic areas beyond Busulfan. Applying this technology to tailor drug doses and improve treatment precision has massive potential. Adverse drug reactions are one of the top ten causes of death in the developed world, representing a cost of more than $136 billion annually.

InsightRX is Computing at the Cutting Edge of Patient Care

InsightRX is a cloud-based software platform that brings data, mathematical models and machine learning to medical care (Figure 2).

figure 2

‍Figure 2: A screenshot of the InsightRX interface for optimal dosing

The platform leverages prior patient data through mathematical models of pharmacology built from clinical studies to individualize treatment for patients. Patient biomarker data is combined with Bayesian forecasting, a statistical algorithm, to improve the prediction accuracy of patient response over time. The tailored model allows for a deeper understanding of the patient’s response profile to a particular drug, enabling precise and accurate treatment selection and maintenance (Figure 3).

For example, a patient treated with the antibiotic Vancomycin will have their blood drawn at intervals to monitor the concentration of the drug in the blood. This can be used to assess a patient’s absorption, distribution, and elimination rates of Vancomycin. Combining Bayesian forecasting with the underlying model allows clinicians to make treatment adjustments based on the patient’s underlying biology, maximizing efficacy while minimizing toxicity.

figure 3 ‍ ‍‍Figure 3: Individual learning with biomarker leveldatapatient Continuous Learning at the Population Level

The power of Bayesian forecasting improves over time as data from more patients is collected (Figure 4). In effect, the system continuously learns. This is an improvement over current systems in three ways:

The predictions improve over time, leading to increased dosing accuracy and better outcomes We can investigate drug response patterns in patient populations often excluded from clinical trials (e.g. children, pregnant women, and the elderly) We can discover predictors of patient response (such as a genetic predictor).

figure 4

‍Figure 4 :Model-based precision dosing and continuous learning

InsightRX applied in the clinical setting

InsightRX is being used in hospitals across the United States and Europe to optimize treatments for cancer, infectious diseases, and bone marrow transplants. It is already being used to improve dosing of Vancomycin, for which as many as 60% of patients receive an incorrect dose. The use of InsightRX has led to increased dosing accuracy and improved therapeutic target attainment by more than 50%.

InsightRX also presents an opportunity for hospitals to investigate the new factors involved in drug response such as drug interactions, genetic variations, and disease sub types. The ultimate goal of the InsightRX initiative is to enable data-driven precision medicine to continuously improve patient care.

comments powered by Disqus