Top Rewards (with Risks) for Technology and Healthcare:

I have been working in the areas of healthcare and computer science since I was an undergrad. Technology and Healthcare met in my early years.

Recently, I have felt healthcare challenges (payment complexities, social determinants of health, outcomes management, accurate treatment, evidence-based/informed guidelines, ever-expanding curricula and training needs, iatrogenic risks, etcetera), with innovations in drug discovery and development, remote patient monitoring, genetic testing and genomics, quantified-self, N-of-One, and the like with the applications of current and promising technologies offered by machine learning, artificial intelligence (AI), virtual and augmented reality, deep learning, CRISPR, CAR-T, biologics and more. The sheer breadth and scope of applications are breathtaking.

Technology and Healthcare Innovation

A recent section of the Wall Street Journal offered some interesting technology and healthcare applications in perhaps some non-obvious ways. Such as healthcare providers fine-tuning their empathy and bedside manner via virtual simulations.

Another company is using Siri-like tools, wearables, and algorithms in order to learn dysfunctional patterns of communication through acoustic and linguistic data, along with concomitant physiological information between bickering couples in order to subsequently identify similar patterns and intervene to head off future arguments before that can start.

In graduate school, I learned early on in my initial psychopharmacology course that many of the pioneering meds that seem to be therapeutic did so without being completely understood as to why. (Ditto that but worse vis-à-vis trephining or lobotomies.)

Likewise in psychotherapy and its multitude of schools of treatment approaches, it’s only been recently that companies like Ieso Lab are taking a look (via clinical science, artificial intelligence, and software engineering) to understand the causes of mental illness and identify the “curative” aspects of psychotherapy. This is a valuable innovation with technology and healthcare.

Along those lines, Danton Char noted in a recent special report on artificial intelligence and medicine, “…there are many areas of medicine — my own field anesthesia, for example — where we don’t know how a therapy works.”

In AI, this is referred to as the “black box” problem. This raises the question of utilization of an algorithm without truly understanding how it works, but again, in medicine, this has rarely stopped us. Char notes that algorithms in healthcare don’t necessarily “…need to be transparent but they (do) need to be explainable or at least auditable.” I. Glenn Cohen concurs, “…we have a good reason to believe that AI in medicine will improve outcomes that it works (rather than)…how it works (emphasis added).”

Similarly, Ziad Obermeyer opines that “medicine uses a lot of things we don’t understand and can’t explain. So the last thing we want to do is hold algorithms to that standard. We want them to discover and help us learn about things we don’t understand. In fact that’s most of the upside.”

Nevertheless, we need to proceed with caution and awareness of a less visible gremlin in medical AI applications, “…Saeb and colleagues note that the standard approach to evaluate predictive algorithm accuracy is via cross-validation. However, not all are statistically meaningful. They noted that ‘…record-wise cross-validation often massively overestimates the prediction accuracy of the algorithms… (and) …that this erroneous method is used by almost half of the retrieved studies… (to) predict clinical outcomes.’”

Let’s Get a Little into the Weeds with a Quick Example

Technology and Healthcare Examined

I think it may be useful to demystify how AI can work in a medical setting, so here is one approach that friends and colleagues at Elder Research used to conduct research supported by the Michael J. Fox Foundation (and other benefactors) to use AI in the predictive diagnosis of Parkinson’s disease. Feel free to skip ahead if this kind of “sausage making” seems daunting, but I encourage you to give it a read.

They used clinical assessments, analytes from sources like plasma, cerebral spinal fluid, and saliva, genetic data, imaging results, and other medical findings derived from research programs, treatment clinics, and physician’s offices. The goal was to determine which tests/data offer the greatest value in predicting Parkinson’s onset. This is a quoted excerpt from their publicly available paper, which I encourage readers to check out. (Also, I have no economic relationship with Elder Research.)

“A proprietary clustering method was used to identify twelve clusters of patients based on the test results available for each patient. Prior to model selection, we used two feature selection algorithms to produce two lists of medical tests for each cluster. A random forest algorithm (called Boruta) was used to produce an all relevant tests list. Potentially useful for scientific research, this list included all tests the algorithm found useful for disease prediction. We then used a recursive elimination algorithm to generate a minimal test list, representing the smallest set of tests needed for accurate disease prediction. Each cluster was further divided into ten subgroups, and algorithms were trained on each set of 90% and validated on the remaining, held-out 10%. This 10-fold cross-validation allowed us to accurately score how well our algorithms performed (and) determine how often a test was chosen within each cluster, as well as between clusters, enabling us to rank the importance of each medical test for disease prediction.
“Several models were trained on each cluster and the most accurate model was selected. Target shuffling was then used to further validate the statistical accuracy of our model results… By target shuffling over 300 times, we confidently estimated the performance of our models (Figure 1).

You may find some of our other articles using data for healthcare interesting such as Improve Healthcare with Data Intelligent.