Newsletter Winter 2019 - Michael Cobb - Applications of Machine Learning in Healthcare: Scalable Treatment Solutions for Cancer Patients

Applications of Machine Learning in Healthcare: Scalable Treatment Solutions for Cancer Patients

Michael Cobb

Software Engineer, Navya Network

This email address is being protected from spambots. You need JavaScript enabled to view it.  

A decade ago, Ms. Gitika Srivastava and Dr. Naresh Ramarajan learned that their close relatives had been diagnosed with cancer. In response, the two co-founded a company that uses machine learning to transform how patients make cancer treatment decisions. 

There are many barriers to democratize access to cancer care. Finding the right diagnosis, the best doctors, and the most applicable treatment is already a challenging task for patients with access to insurance and healthcare resources in developed countries. In developing countries, such as India, each of these tasks becomes exponentially more difficult. In 2014, it was estimated that there was only one oncologist per 16,000 cancer patients in India. In stark contrast, the American Society of Clinical Oncology (ASCO) predicts that the United States will have one oncologist per 100 patients by 2020 [1].

Navya Network is one company that offers a service to reduce this disparity. Navya applies an innovative, machine learning-based informatics system to provide scalable treatment solutions to patients already diagnosed with cancer. Distinct from telemedicine solutions, which connect patients to doctors directly, Navya’s technology structures patients’ medical reports and preferences into an ontology. An ontology, in essence, is a standardized way of organizing the definitions and relationships between concepts. With patients’ data in a consistent format, Navya matches it with clinical trials and previous patients’ experience to create a case summary with an ordered list of treatment recommendations. Because of how tightly structured this summary is, an oncologist can review it and offer feedback in a matter of minutes. Navya sends this summary to multiple oncologists and aggregates their opinions into a treatment plan, which is written in language accessible to patients.

Clinical validation trials confirmed that treatment recommendations from Navya’s clinical informatics system are 98% concordant with predictions of tumor boards and expert panels at Tata Memorial Centre in India and Olive View-UCLA Medical Center in California [2]. The company has assisted over 26,000 patients. Although these patients are predominantly from India, the company is broadening its scope and has reached patients in 68 countries.

This article will first describe the gap analysis that led to the creation of the Navya approach, while highlighting how the company’s solution is distinct from telemedicine. It will then give an overview of how Navya has used machine learning to create a scalable approach to democratizing cancer treatment decisions.

Filling a Treatment Decision Support Gap for Doctors and Patients

The motivations of Navya’s founders, Ms. Gitika Srivastava and Dr. Naresh Ramarajan, were both personal and professional. The idea arose from the pair’s experience with cancer care and was refined with a careful gap analysis.

In 2007, each learned their close relatives had been diagnosed with cancer. Ms. Srivastava remembers how her family found the process of finding the right treatment agonizing, not least because of difficulties in finding reliable, accurate information. Despite a strong academic and technical background, she “had no way to know where to find papers [in academic journals], how to read them, and how to interpret their statistics, analysis, applicability, and complex words.” Ms. Srivastava emphasizes that the first goal of Navya was to provide an empathetic, personal service that would provide the clarity needed to simplify treatment decisions and the complex medical literature surrounding them.

As a resident at a tertiary care center, Dr. Ramarajan realized that finding reliable information was not just challenging for patients, but for doctors as well. He saw that one of the biggest barriers to scaling cancer treatment was the amount of time doctors had to spend sorting through data: synthesizing a patient’s case, staying abreast of the latest medical literature, and determining which treatments from literature apply to one instance of cancer. After processing the necessary data, the actual decision-making time of doctors is typically only a few minutes or seconds per case. The founders realized that if they could remove cancer experts’ burden of drudging through data and use them only for their decision making ability, they could give experts the means to reach an order of magnitude more patients.

A critical decision for Navya was to focus exclusively on treatment decisions, not diagnosis.  First, no one else was doing it. Dr. Ramarajan emphasizes that once a patient is diagnosed, that is when they have a complex tradeoff to make. At the time, no service approached treatment decisions that could simultaneously keep patient preferences in mind, take multiple expert opinions into account, and exhaustively factor in clinical evidence.

Navya entered a landscape where existing solutions, either for diagnosis or treatment, tried to replace a visit to a doctor, rather than taking advantage of specialists’ unique expertise and ability to collaborate. Telemedicine solutions, where an expert reviews a patient’s case files and either responds with a writeup or a short phone call, is one popular alternative to an in-person visit. However, patients are still left with the task of consulting multiple experts and synthesizing their opinions. The company recognized that, short of having access to a tumor board in a tertiary care center, patients did not have adequate resources to connect them to experts and help them come to a consensus opinion.

The second reason for focusing on treatment decisions was that diagnostic decision making is a far harder machine learning problem to crack. Diagnostic decision making is backed by heuristics, rules, and clinical practice based on incomplete information. Because treatment decisions could use a more data-driven model, Navya would be able to more easily evaluate whether it was producing correct results.

A third advantage was that Navya could work directly with patients and remain an analytics software company. Because patients would have already undergone the necessary tests to receive a diagnosis, Navya could directly help patients with their next need: making treatment decisions. In contrast, working with doctors on diagnoses would have inserted the company into the practice of medicine. This would force Navya to provide decisions based on incomplete information, would necessitate that Navya frequently advise patients to get further tests, and could have exposed the company to further regulation as a diagnostic service.

Similarly, Navya’s decision to focus on India made sense from both personal and business perspectives. Ms. Srivastava was motivated by a personal desire to give back to her country, especially to Tata Memorial Centre, the tertiary care center that had treated her family member. Ms. Srivastava knew that the hospital’s oncologists were “extremely evidence-based, research-based, and collaborative.” As one of the largest tertiary care centers in Asia, it also had the benefit of experts from many different specialties under one umbrella. Tata Memorial Centre agreed to commit its entire staff of oncologists as members of Navya’s Expert Panel in exchange for co-branding of Navya’s Expert Opinion Service.

India also proved to be an ideal setting to develop and test Navya’s concepts, as it provided an environment that was by turns complex, chaotic, and collaborative. India’s system of cancer care also had some of the largest disparities in the world. In 2014, India had only 0.6 physicians and 0.9 beds per 1000 people, over three times less than in the United States [1]. India’s disparity in quality of care, especially between rural and urban areas, could benefit greatly from an online service. In fact, patients in a prospective study incurred 40% of their care expenses before setting foot in a hospital [3].

Navya had decided its goal was to provide a service that would use technology to empower patients with treatment decisions, while minimizing the time physicians spent synthesizing information. India offered Navya the chance to partner with one of the world’s largest tertiary care centers, an ideal test environment, and a chance to make a real impact on patients’ treatment decisions.

Using Machine Learning to Make Treatment Decisions By Structuring Evidence and Experience into an Ontology

Navya’s machine learning solutions are designed to answer the question, “What is the best treatment for a patient’s case?” The process works as follows: patients upload their medical reports to Navya’s portal. After a clinically-trained analyst verifies that all necessary reports are in place, the case is run through Navya’s clinical informatics system, which returns a list of treatments ranked by applicability to the patient. This list is sent to organ-specific specialists relevant to the patient’s case, who tweak the recommendations as needed. Once all of the recommendations are finalized, Navya uses more machine learning to weigh and combine the opinions of all the experts who reviewed the case before synthesizing them into a report. The analyst verifies the report, tweaks the language to be as accessible as possible, and returns it to the patient. This process takes a median time of 24 hours [4].

When processing a patient’s case, Navya uses machine learning in ways that can be grouped into “evidence” and “experience” based engines for making decisions.

The evidence engine serves two purposes. First, it synthesizes existing medical literature. For many cases, a number of different trials apply, but their results conflict. Navya uses machine learning metrics to rank the strength and the quality of these trials. Second, the evidence engine ranks the applicability of clinical trials to a single patient. Navya calls this metric the “applicability index.” For example, a clinical trial might take all comers over 18 years of age, but may not have enrolled anybody who is over 70 years of age, and so would be less applicable to elderly patients. From the results of this process, Navya can determine how well-represented one patient is in a trial or population of trials. Based on these results, the clinical informatics system can determine how likely it is that the results of a trial would be applicable to a specific patient.

While the evidence engine provides a solution for doctors who do not have time to sort through new clinical trials, the experience engine gives Navya the tools to support patients whose unique and complex cases are not described adequately in literature. In fact, a large number of patients’ diagnoses do not match any clinical trials. The evidence engine uses the data of past patients at Tata Memorial Centre, including past patients’ outcomes, treatment pathways, and decisions. Using this information, the engine can model a patient decision based on the current patient’s similarity to a cluster of patients that have already gone through similar treatments. One of the largest advantages of the experience engine is that it adapts its model using oncologists’ feedback. After a list of treatments is presented to an oncologist, the oncologist can tweak the system’s recommendations before sending it to the patient. This feedback is fed back into experience engine, which allows it to refine its model. This raises the confidence of future predictions and increases the breadth of cancers it can support over time.

Navya, then, can efficiently find all of the clinical trials that are relevant to a patient’s case and rank them by quality and applicability; it can also suggest treatments based off of previous patient outcomes. That said, what allows Navya’s clinical informatics system to achieve such a high degree of concordance with expert oncologists in real world settings? The answer is how Navya stores its data. Every piece of patient medical information, every treatment decision and outcome, and every clinical report is restructured into a curated ontology. The ontology is more than a list of terms and definitions; it describes terms’ synonyms, contexts, and relationships with other terms. The most powerful effect of this ontology is that it allows Navya to replicate the intuition that oncologists pick up after years of working on patient cases. Dr. Ramarajan describes that the ontology allows the informatics system to relate things that seem distinct, like a side effect from a treatment to a patient’s pre-existing medical condition. These relationships allow the system to make inferences, strengthening its predictions. As a result, Navya’s expert and experience engines can recommend treatments that replicate the conclusions of an experienced oncologist who is well versed in the patient’s medical history and the applicable literature.


Machine learning provides an innovative and readily scalable solution in the landscape of democratizing cancer care, one that can be uniquely tailored to patients facing treatment decisions for complex cancers. By eliminating the time it takes doctors to review the details of a patient's case, and allowing multiple doctors to collaborate asynchronously on a case, Navya escapes the one-doctor one-patient model that limits telemedicine solutions. Because of Navya’s use of a structured ontology, Navya’s machine learning solutions offer the ability to replicate a tumor board’s intuition using machine learning, providing support for patients whose cases are not covered adequately in literature, with 98% concordance to oncologist recommendations at Tata Memorial Centre and Olive View-UCLA Medical Center. Especially in areas where the geographic and monetary barriers of accessing a tumor board at a tertiary care center prevent patients from finding the best treatment option, Navya offers a solution that is truly scalable.


1. Goss PE, et al. Challenges to effective cancer control in China, India, and Russia. Lancet Oncol 2014; 15(5):489-538.

2. Ramarajan N, et al. International application of an online clinical informatics expert system for breast cancer. Journal of Clinical Oncology 2017; 34:15_suppl, e18037-e18037.

3. Mohanti BK, et al. Estimating the Economic Burden of Cancer at a Tertiary Public Hospital: A Study at the All India Institute of Medical Sciences. [discussion paper]. New Delhi: Indian Statistical Institute, Delhi Planning Unit; 2011.

4. Badwe RA, et al. Global impact of a clinical informatics system: Scalable delivery of on-time access to evidence-based multidisciplinary expert treatment decision systems for all cancers. Journal of Clinical Oncology 2017; 35:15_suppl, 6502-6502.