Scroll Top

Is Healthcare a Medical Science or a Data Science?
The merger of the two is the real answer

Data, from its collection to its interpretation, is dramatically transforming every industry—. Nowhere is that impact being felt more personally than in healthcare. Consumers are bombarded with ads for fitness trackers and smartwatches that promise to count steps and monitor heart health, and direct-to-consumer genetic tests that can tell you what your ancestry is and your likelihood of developing some cancersor chronic diseases. Insurance companies and health systems are leveraging claims data to predict which patients are will develop sepsis or who are at-risk for readmission after heart surgery in order to both cut costs and improve patient outcomes. Physicians are not immune to the challenges of working in a data-driven environment—in fact, even more than consumers and administrators, they are at the leading edge of the transformation that is taking place. From patient interactions to workflow processes, nearly every aspect of what a physician does is rapidly evolving as data collection and analytics, machine learning and artificial intelligence transform the practice of medicine.

Diagnosis, Prognosis, & Treatment

Historically, one of the first steps for an oncologist of a patient with a suspected cancer has been to biopsy the tumor. Pathologists would prepare slides of the excised mass, perform a histological examination using microscopy, and return a diagnosis back to the oncologist. Today, while this component of the general workflow remains the same, the steps involved have become more complex thanks to technological advances like digital pathology imaging systems. In 2017, FDA granted clearance to the Philips IntelliSite Pathology Solution (PIPS), a whole-slide imaging system for biopsied tissue. The PIPS system was the first of its kind, scanning and digitizing slides so that a pathologist could review and interpret the results. What makes the PIPS product so revolutionary is the digitization of images, which provides opportunities to minimize slide storage needs and maximize the efficiency and accuracy of slide review, including secondary evaluations and use in tumor boards.

Combined with digital pathology, advanced molecular tests, designed to identify genomic mutations and measure gene expression levels, are making the diagnostic and prognostic processes more detailed and complex. For example, it’s no longer sufficient to just provide a diagnosis of “breast cancer” without also noting if it is BRCA1+ or has normal HER2 expression. These added dimensions of data are more than just superfluous pieces of information—they are becoming necessary components for accurate prognosis and effective treatment plans. For many cancers, identifying the underlying mutation is now a routine part of care and added to clinical guidelines, as the number of targeted treatments grows, and new clinical trials are designed to focus on specific permutations, rather than simply the tissue of origin. New technologies like liquid biopsy are giving doctors a chance to monitor cancers as they evolve, in real-time, offering a glimpse of the future for oncology that is similar to HIV, where patients take a treatment regimen that preemptively heads off any mutations the cancer would likely take.

Artificial Intelligence

Even more remarkable is what is happening at the intersection of clinical data and artificial intelligence (AI), made possible only because of the digitization and aggregation of healthcare data. Many in healthcare believe AI to be years away from implementation in clinical care; however, AI-enabled imaging analysis is currently in place at a handful of top institutions and the FDA is meeting this growing trend head-on with recent approval and clearance for a variety of AI-based products. Arterys announced in February 2018 that it received FDA clearance for its Oncology AI suite for liver and lung imaging scans, while diabetic retinopathy, the leading cause of vision loss in patients with diabetes, can be identified with IDx’s IDx-DR, which can detect the disease without physician interpretation. And heart disease detection is being transformed by recent FDA 510(k) clearances of Bay Labs’ EchoMD AutoEF software and Zebra Medical’s coronary calcium scoring algorithm — both based on AI technology.

This trend to leverage AI and machine learning technology will only become more widespread in the upcoming years, as the technology continues to advance and having this capability puts health systems at a competitive advantage. AI-enabled image analytics is being used to help doctors sift through immense amounts of imaging data to identify almost imperceptible changes over time and to triage images that need the pathologists’ attention, such as with Arterys’ offering.  As a result, patients can be identified earlier, more accurately, and treatment can start sooner, particularly for difficult to diagnose cancers with poor mortality rates, like pancreatic or ovarian cancers.

To make AI and machine learning systems work requires massive quantities of data and sophisticated analytics. Before the Human Genome Project and the digitization of healthcare information, from imaging to lab results, this information was kept in data siloes and sharing or aggregating was virtually impossible—leaving physicians in an untenable position to keep up with the latest scientific advances. Consider IBM Watson for Oncology, just one of the AI solutions now on the market. Despite the ongoing, well-publicized issues about Watson’s abilities, IBM has integrated the data of more than 300 medical journals and nearly 15 million pages of text to support its function. This is part of what makes AI and machine learning both desirable and concerning. If an individual physician finds it impossible to keep abreast of all scientific advances, an AI system can aggregate all of that information, providing a distillation of what the physician needs as a valuable adjunct to their practice. However, the value of the information provided relies on the accuracy and relevance of the information that is fed into the system: garbage in will give you garbage out, no matter how sophisticated the algorithms behind it. So, if the AI system only trains its algorithms with patients who aren’t representative of the patients an individual doctor treats, the information may not be relevant to their clinical practice.

The Future of Oncology

At the heart of this growing reliance on data and its interpretation is the ability of doctors to better classify patients in order to treat them and obtain the best possible outcomes. Molecular diagnostics are just one tool to help in that process: classification first at the protein level, then at the genomic level. Distilled to its essence, it’s nothing more than a pattern recognition problem—and computers have been shown to outperform humans in this task. One recent example includes an AI calculator to predict death and risk of 18 post-operative complications from emergency general surgery with a better predictive value than ASA classification, NSQIP, or Emergency Surgery Score (ESS). This doesn’t mean that computers will replace doctors in the future. Medicine is both an art anda science and numerous barriers remain to a fully data-driven analytic process. For diagnosis and prognosis, the data revolution has made significant headway, but determining the right drug for a patient at the right time is still a work in progress, hampered by data hoarding and lack of interoperability. To develop accurate algorithms to identify which patient should have which drug requires a substantial amount of data: historical patient data including outcomes—information that has typically been maintained in proprietary databases by held by individual doctors, provider systems, or medical centers. Issues related to data privacy and bioethics have been addressed to allow for sharing of de-identified patient records, but more work remains.

Further, there are the economic costs associated with a data-driven practice and the incentives are not obvious. While some physicians are moving toward this new reality due to systemic changes at their institution, widespread change will only come when more patients recognize the benefits to this new practice of medicine and demand its use and when insurers appropriately reimburse for medical practices that improve patient outcomes. Some believe we are at the tipping point already: Erik Lefkofsky, co-founder and Chairman of Groupon and founder and CEO of Tempus has said “Asking doctors to treat cancer patients without the benefit of modern software is like asking someone to drive at night with no headlights.” In time, patients will be assigned to smaller and smaller categories that more completely describe their health condition and how it differs from someone else. AI-enabled software platforms will be able to provide doctors with the most up-to-date information regarding treatment modalities. Doctors will have vast amounts of information about the disease and historical patient data to help them better determine how the patient in front of them should be treated. The future of oncology looks a lot more like data science and physicians should be ready for this data-driven reality.