Oncology's Incessant Grip

AI, machine learning and cognitive computing may be changing healthcare, but it's not all plain sailing.



Over the past couple of years, much has been made of artificial intelligence (AI) and cognitive computing tools applied to healthcare diagnosis and treatment selection. These technologies have very different functions; AI goes beyond traditional hypothesis-driven approaches to find patterns in the data, while cognitive computing is trained to see the significance of patterns and relationships.

These tools can integrate disparate sources of data – from electronic medical records (EMR), family histories, imaging studies, genetic studies, molecular diagnostic testing, among many such sources – to support clinical decision making. The overarching concept is that the vast computation throughput of both approaches can augment experts to assure thoroughness and quality in diagnosis, treatment selection or placement in a clinical study. 

While much of the early work showed cognitive computing tools could be faster and more accurate in achieving a diagnosis than expert humans1, more recently doubts have arisen, most notably with news that collaboration between MD Anderson and IBM Watson had ended.2

There are a number of reasons why such events have initiated a combination of introspection, challenge and questions in oncology. First, oncology is both complicated and one of the most advanced clinical fields for evidence-based decisions. The National Comprehensive Cancer Network, American Society of Clinical Oncology/ASCO (CancerLinQ), Flatiron Health, US Oncology/McKesson (iKnowMed), Memorial Sloan Kettering (Drug Abacus), and other organizations have defined evidence-based approaches that guide clinical decisions. As more cancers are assessed as driven by genetic mutation, and as more therapies use immunoncological approaches that activate patients’ immune systems, there is very strong and specific evidence that can be brought to bear to make a diagnosis, select a treatment, and set expectations for response or treatment modification. 

Adding to this is, clinical trials are a critical part of the cancer treatment assembly. It is often cited that more than 30 percent of the entire clinical development pipeline of the pharma and biopharma industry is focused on cancer. So often, the innovations within that pipeline may represent a meaningful treatment option. The same technologies that apply to select of treatment were anticipated to identify clinical trials appropriate for consideration for the individual patient. 

Also, pharma, biopharma and medical device companies are also increasingly adopting these technologies as part of their own research, medical and next-generation commercial models. The assumption was that the move to ‘value and outcomes’ required use of some sort of analytics, AI or cognitive computing tools to be effective and attractive to healthcare providers. 

A look forward

Since some of these tools are only as good as the training they receive and the literature they access, there are limitations inherent in both what is known, the level of effort put into the initial training, and the frequency with which the tools are maintained. But some of this is even more nuanced. There are often very personal and individual choices that a clinical and patient may make together. Stated differently, there is rarely a single option to consider, rather a series of options within which many of the choices become persona to the patient and his or her family. 

Consider the 79-year-old man with non-metastatic prostate cancer. He and his clinician might jointly conclude that the rather slowly progressing nature of this cancer leads them to a decision to not treat, or treat minimally, versus suffering the quality of life compromises that more aggressive treatment may bring. In many cases, these different treatments may all be considered evidence-based and therefore as viable alternatives. 

Consistent with the need to make decisions that incorporate clinical evidence, patient-specific insights, and patient-specific options an array of other approaches are being deployed. For example, the Dana Farber Cancer Institute3 uses disease-specific groups of clinicians to curate the clinical literature and develop formal sets of treatment pathways, letting the pace meaningful new research determine the frequency of updating. Flatiron, a cloud Oncology EMR and analytics company, uses a combination of externally derived pathways, standardized data collection, with analytic support to provide their clinical practice customers treatment approaches and options for their patients. And, bioinformatics companies like QIAGEN’s IPA use curated ontologies in a cloud solution to aid molecular diagnostics labs interpret the results of next generation sequencing based diagnostic tests. While each of these examples is ‘technology-enabled,’ they are knowledge and expertise centric, versus being technology-driven. 

AI and other machine tools operate even more differently – in many cases the targeted value is seeing patterns and creating an insight that was not previously in the literature. But as any data scientist knows, there are patterns where correlations are spurious or artifacts of relationships unrelated to the patient’s disease or available treatment options. As such there is often work after the machine-enabled insight to understand what is behind it or its overall relevance. As such, this is highly valuable in an early-phase research program, but less apt in-line with clinical workflows. 

It’s still all about value and outcomes

This is still all about value and outcomes4. In this case, it is value for the healthcare provider, patient, and health system. Optimizing across the three of those is nuanced and requires different lenses on the same situation relative to each parties’ objectives and requirements. The patient wants the highest quality life, to define paths that can engage their family, and have confidence in the outcomes and compromises of each alternative. The provider is looking to assure that the best possible evidence has been brought to bear in that decision, to be partnered with the patient in the selection of a treatment strategy, and to be supportive of the patient and their family during the course of that treatment. The payer is looking to assure that every decision is supported by evidence and that expenditures bring real benefit in making a diagnosis accurate, avoiding non-response, getting a positive response, but further knowing that the patient will successfully complete the full program of treatment.

As we advance, AI, machine learning, and cognitive technologies will increasingly augment, but not replace, expertise. Progress will be made with a clear recognition of the technologies ready to solve specific problems today, how they are best integrated and deployed, and how we are going accelerate their improvement or replacement. Throughout the focus must be on the patient and the outcomes we are trying to realize – only then can the newest technology enabled and digital solutions bring new sources of meaningful value.    


Jeff Elton, PhD, is Managing Director and Global Lead of Predictive Health Intelligence at Accenture Life Sciences.

1https://www.mskcc.org/blog/msk-trains-ibm-watson-help-doctors-make-better-treatment-choices , http://www.wired.co.uk/article/ibm-watson-medical-doctor , and https://www.mdanderson.org/newsroom/2013/10/md-anderson--ibm-watson-work-together-to-fight-cancer.html
2http://blogs.sciencemag.org/pipeline/archives/2017/02/20/an-ibm-watson-collaboration-goes-under
3http://www.prweb.com/releases/2016/01/prweb13153347.htm
4Elton, Jeff and O’Riordan, Anne. Healthcare Disrupted: Next Generation Business Models and Strategies. John Wiley & Sons, 2016.

 


Since you're here...
... and value our content, you should sign-up to our newsletter. Sign up here

comments powered by Disqus