Tata Memorial Centre (TMC), seeing over 50,000 new cancer patients a year, and Navya, a clinical informatics and patient services organization focused on complex decision making, announced Experience Engine (XE), a machine learning solution to structure experiential knowledge relevant for treatment decision making. The learning solution derives a similarity metric for patients who have received similar treatments and predicts treatment decisions that experts are likely to recommend. XE goes beyond evidence, solving patient cases that clinical trial data does not.
Promising results from the first trial of organized learning from past tumor board decisions at TMC and UCLA-OVMC to predict treatment decisions that oncologists would make for a new set of patients are being presented at SABCS 2016. The abstract titled, “Building an Experience Engine to Make Cancer Treatment Decisions Using Machine Learning,” is being presented on Wednesday, December 7, 2016.
Dr. Rajendra A. Badwe, Director, TMC said, “The Experience Engine captures the way experts think and outputs treatment options for each patient in line with what they would recommend. This is how we can scale access to expertise. The resulting database of opinions is also an excellent companion for online training.”
Being a tertiary care referral center, TMC’s experts treat highly complex, nuanced and rare cases from across the Indian subcontinent, Asia and Africa. Multidisciplinary tumor board decisions at TMC represent an unparalleled wealth of intelligence and experience, currently trapped in the minds of experts and electronic medical records.
Dr. Naresh Ramarajan, graduate of Harvard College and Stanford School of Medicine, and Founder and Chief Medical Officer at Navya said, “Tata Memorial Centre and Navya create a new source of knowledge. The Experience Engine has implications for training oncologists, standardizing cancer care across the world and driving accurate decisions for complex patients not addressed by the evidence.”
Navya’s Engines synthesize evidence specific to a patient and learn from relevant tumor board decisions to make treatment decisions. The Navya Evidence Engine (EE) was validated in three clinical trials at Tata Memorial Centre and UCLA-OVMC. Results showing 98 percent concordance between the EE decisions and TMC and UCLA-OVMC tumor board decisions were published at the SABCS in 2014 and American Society of Clinical Oncology (ASCO) in 2016.
Tumor boards at tertiary centers like TMC and UCLA-OVMC provide solutions to complex cases not addressed by high quality evidence. Experts intuitively retrieve patterns from years of experience to make treatment decisions. Short of personal consultations, there is no way to access this vast “experience database.”
Richness of Navya’s ontology represents each patient with 690 individual features. XE uses relevance learning to identify the core set of highly informative features for decision making.
Multiple similarity distance metrics were systematically evaluated for each decision point. When a new patient was presented to XE, the learned similarity metric was used to identify similar patients. XE then predicts a decision based on the treatment received by these similar patients.
XE’s predicted decision was compared with the expert’s actual decision. The primary endpoint of comparison was accuracy (defined as AUC – Area Under Curve). In addition, state of the art multiclass classification algorithms were also evaluated. Winning XE algorithms were chosen specific to each decision point. The algorithms were used on a completely new prospective group of patients who were seeking an online opinion from tumor board experts of Tata Memorial Centre.
Accuracy of prediction for each decision point was significantly (~40 percent) more accurate than baseline of weighted random guessing. When XE predicted whether a patient needed standard evidence based therapy or a non-standard experience based therapy, it was highly accurate (70 percent to 99 percent based on the decision point).
The XE is a truly novel source of knowledge, containing learning from patients with significant comorbidities, multiple lines of prior treatments and poor performance status for whom standard evidence-based treatments from randomized control trials are not applicable. The analysis of hundreds of similar patients to these complex patients uncovers new insights into possible treatments.
Further, XE enables oncologists to evaluate why similar patients may receive different treatments. Variations in practice patterns, treatment centers, expert preferences, affordability of patients and patient preferences, are features that influence decision making. These are considered by XE, but not possible to consider by medical evidence guided by randomized clinical trials.
About Tata Memorial Centre
Tata Memorial Centre, founded in 1941, leads the Indian subcontinent in cancer care by evidence based practice of oncology, and research and services which are affordable, innovative, and relevant to the needs of the country. Every year nearly 50,000 new patients visit TMC from all over India and developing countries in Asia, Africa. Approximately, 70 percent of these patients are treated almost free of charge. Visit: https://tmc.gov.in.
About Navya
In 2009, Navya was founded in Cambridge, MA by graduates of Harvard, MIT Sloan, and Stanford. Navya’s patented system uses clinical informatics, predictive analytics and machine learning technologies. It combines several clinical information sources as inputs – and outputs a treatment decision most applicable to a unique patient. For the first time, quick and affordable access to evidence and experience based expert treatment decisions is available to every cancer patient. Navya’s Online Expert Opinion Service has been used by 8000 patients in 42 countries. Visit: www.navyanetwork.com.
Barbara Jacoby is an award winning blogger that has contributed her writings to multiple online publications that have touched readers worldwide.