A few weeks ago, we hosted a webinar in which Dr. Hugo Bowne-Anderson, data scientist and educator at DataCamp, discussed AI in the healthcare industry with Arnaub Chatterjee, Senior Vice President at Medidata Solutions, previously Associate Partner in the pharma analytics at McKinsey & Company and Director of Data Science at Merck. Find the transcript of their discussion here or read on for an overview of the topics covered.
The impact of AI has been seen in drug discovery, claims administration, imaging and diagnostics, and risk screening
AI has dramatically changed many healthcare disciplines. There are a lot of different players in the space in terms of verticals, like pharmaceuticals and insurance companies, and entities, like hospital systems and Pharmacy Benefit Managers. Let’s take a look at some of the use cases for AI in healthcare.
About 50% of late-stage clinical trials fail because there are ineffective drug targets that fail to produce the desired therapeutic effect, and only 15% of drugs actually get approved. Within pharma, AI has improved drug discovery—the process by which potential medications are discovered—by streamlining the clinical trial operations process.
AI has helped with the claims administration process, including:
- Automating the process to check claim statuses
- Managing accounts receivable
- Streamlining workflows for reviewing claims and approving insurance checks
Imaging and diagnostics
AI-driven diagnostic tools are helping physicians to more accurately diagnose disease and make better clinical decisions. These tools have completely changed the face of fields like radiology and pathology. This space has been well-explored, and companies are racing to get the most imaging data.
Predictive models can identify patients that are at higher risk for a certain disease, helping physicians identify what they need to target and better treat. Facebook has leveraged AI for suicide risk screening, creating very accurate risk models.
The biggest wins for AI in healthcare have been incremental instead of leaps
Reaching milestones, like the goal of using AI to discover new drugs like a treatment for the novel coronavirus, isn’t possible without incremental change. A lot of the breakthroughs have been due to foundational work that doesn’t always make headlines, like establishing good training sets.
Imaging and diagnostics wins are due to amassing large amounts of data
Let’s use imaging and diagnostics as an example of this. There is a huge volume of data in different formats in imaging and diagnostics, from MRIs to CTs to PET scans. Hospitals are producing a staggering amount of petabytes of data per year in imaging—90% of healthcare data comes from medical imaging. We’re starting to see the linkages to the electronic medical records, and can use that to interpret what the images reveal and derive expected outcomes.
A recent example is Google Health’s work on a deep learning AI model in which they built test sets of breast screening mammograms and predicted with high accuracy which women were likely to have breast cancer. This AI model produced 10% fewer false negatives and 5% fewer false positives than the top board-certified radiologists in the U.S.
Models like this are incredibly powerful if they're deployed properly, and the good news is they’re increasingly becoming commonplace. Machine learning algorithms can now successfully differentiate between different types of cancers and produce accurate predictions, outperforming top pathologists, radiologists, and ophthalmologists. The next step for AI is taking that knowledge and translating it into improving clinical care.
AI requires ethical application
Tempering expectations in improving clinical outcomes
There’s a lot of excitement about AI applications for healthcare, and rightfully so—it’s improving millions of lives. But we should temper expectations. Slow and steady incremental improvements means AI won’t be replacing physicians anytime soon, and the chances that AI will create new drugs within five years are low due to the long process to develop drugs, test them clinically, and gain regulatory approval. What AI will continue to impact in the near term is improving processes along the patient journey to improve clinical outcomes.
Ethics in healthcare AI
In healthcare, as in any other field, there are many concerns around data privacy and ethics, particularly in regards to patient privacy. Companies will continue to adopt and iterate on their own ethical AI guidelines. And algorithmic bias in healthcare must be acknowledged and corrected for, because the impacts can be devastating for individuals and entire communities. Companies working with healthcare data have a responsibility to figure out what their blind spots are and ensure that their data is representative so that it can be used for good.