Blogs

The push for new ways to use big data is impacting Medicare and Medicaid plans. Artificial Intelligence is poised to enter the health care arena, and CMS has proposed the AI Health Outcomes Challenge as a contest to incentivize big data specialists to engage the Healthcare Commons. The idea is that tools such as machine learning and Bayesian neural networks can identify actionable structures in data to assist the CMS Innovation Center in testing innovative payment and service delivery models. Part of the contest is to make broad based data available from the Medicare Claims Dataset. The idea of making more CMS data available to the provider community is gaining currency.

At this early stage in using AI at this scale, there is a need to remind payors and providers alike, of the need for broad based population analytics. Broad based does not only mean large transactional datasets in the millions of beneficiaries, but also other diverse data from the medical, social, financial, and scientific spheres that collectively can be termed the health care commons. This large scope does justice to the power of AI. For example, the objective of reducing unplanned admissions to hospitals and SNF's, lowering ER utilization, reducing imaging costs, and many other types of transactions, requires more in depth and complex analysis, than simply counting beneficiary transactions.

As part of posing an issue to be addressed, there must be an understanding of context of each issue. There is no issue that exists in isolation, each has an impact of multiple parts of the healthcare commons. The health care world is stratified, the context has many layers, and these layers span the scale from large populations to local provider groups to the single provider and most meaningfully, the patient.

Each scale (layer) generates its own type of information or metadata, and has its own needs for analysis of this information. AI must be deployed accordingly. The largest CMS scale needs information to formulate policy. The smaller scales are local and regional, and reflect presence (or absence) of provider networks. Smaller yet is the provider networks where localized "small data" is important, such as use of evidence based medicine, admission rates, cost of care, and all the transactions. The smallest scale can be as small as N of 1, the patient. At this smallest scale, the healthcare commons has an ethical dimension.

An interesting and emerging subgroup of AI is in the application of critical realism, which offers tools that are a step beyond statistical probabilities and is appropriate for N of 1, the single patient. Having the potential to empower the "scale of holism", the complexity of the single patient, that as a provider knows is realistic (e.g. realism) for a patient who has known and unknown features of health such as psychosocial, medical, preferences, disease pathophysiology, and more. Realism accepts all variables, or dimensions in realist terminology. The analytic aspect of N of 1 is most interesting in that the patient can be treated as an open system (realism again) facilitating discovery of why the patient does what he or she does and what can be done about it, even if it had not been tried before in this patient. Looking back though all relevant data at the patient level suggests the potential (not the probability!) of an action to benefit a patient out of some number of choices. This selection of choices (action spectrum), based on real time patient assessment, matched to the population level of open sources of data is termed retrodiction (realist method).

This is an outline of a learning tool for Tenncare, whereby access to large datasets matched to the single patient can open the door achieving objectives of all levels of the healthcare commons.