What is Machine Learning?
First, and quite important, what it is not. It is not a magic bag of pixie dust, nor an artificial intelligence. What it is, rather, a sophisticated-comprehensive math + computation. Likewise, it is cutting edge pattern identification’s and fruitful correlations for successful outcomes. Unlike traditional statistics, Machine Learning is an iterative testing of many millions of combinations of varied data points. Also, Machine Learning can learn many pathways accounting for variance that exists in human health, rather than, mapping to average outcomes. We expose lots of relevant information to various machine learning models by pre-processing clinical notes using NLP technologies. Machine learning will be a team sport requiring the integration of data science and medicine. With such, new players, and the training of doctors in data science, statistics, and behavioral science will be necessary.
Where does Machine Learning Fit In?
Data access and its relationships; content sourcing and context creation; algorithms building; and lastly, insights for decision making
Healthcare industry challenges and implications for machine learning, training and teaching:
(unlike in other industries where ML is becoming a reality, e.g. Retail, Airlines, Banking, Logistics)
In healthcare, we don’t measure outcomes because we don’t get paid, while machine learning relies on outcomes to ‘learn’ correlations to them. This means machine learning is likely to be limited to helping schedule more services, more effectively, versus, keeping people out of hospitals.
Challenges: Machine Learning for Healthcare Industry
- The healthcare delivery system remains very complicated, fragmented, thereby, difficult to navigate and understand. The system, likewise, lacks transparency, collaboration and accountability. Most people don’t have any sense of how the system works “at large” and continue to get lost in its process. Unfortunately, we, as patient-consumers, only grasp the mechanics of this delivery on a “piece by piece” basis. Getting to a holistic, convenient, and consumer “valued” experience, remains a big challenge.
- With diversity in populations, various stakeholder engagement’s, differential motives and desires, our health systems have great difficulty around alignment and true coordination in achieving the necessary goal of cost control & quality care.
- Healthcare regulations continue to not only take on a spirit of gradualness, but, remain cumbersome and difficult to track, thus, act on.
- Clinical care is “just a small component” in terms of Healthcare. Healthcare is a narrow definition of health. Now, health is moving to digital environments empowering patients, consumers and caregivers, leading to and creating, a “new healthcare economy” that will have implications, synergies, opportunities and threats.
- 90% of health data is generated outside the physician office environment; the clinical component substantiates the other 10 to 20% of this capture. Clinical data, the core for innovation, is fragmented, unstructured, hidden at times, and not interoperable. Likewise, and critically important, often untrusted.
- Clinical care delivery is siloed, volume driven, and care coordination is not yet fully incentivized.
- While basic EMR implementation is complete in most care delivery settings, it remains frustrating and unsatisfactory, causing physician burnout and confusions, by not delivering quality information at the point of care. The access to multiple “patient portals” for a consumer with varying degrees of chronic illness, isn’t a reality in terms of centralizing this complex data, nor, normalizing the content into an understandable context.
- Healthcare reimbursement based on clinical documentation from ICD and CPT codes continues to be very complex.
- We don’t have a healthcare “costing” system. Employers and patients don’t have access to pricing info (i.e. transparency) nor how their dollars are being spent by payers and providers.
- Individual health risk assessment, quality measurement, while burdensome, remains essential for ensuring outcomes and performance.