I attended the Conference on Health IT and Analytics recently. The event sees participation from government health and human services (HHS) executives and includes discussions on the next big thing in HHS analytics. This year the focus was on leveraging AI (Artificial Intelligence)/Machine Learning (ML). Almost every person I spoke with mentioned that AI/ML-driven analytics is among the key initiatives that their agencies would like to explore in the near future.
This is not surprising. HHS agencies are looking for ways to arrest the escalating cost of healthcare and manage the impact of chronic diseases on the population and state resources. AI/ML-driven analytics can help.
AI/ML can enable agencies to become more proactive in the management of population health. Instead of reacting to an event, they can use these technologies to anticipate the likelihood of an event and act to prevent it.
HHS agencies have access to reams of data (both transactional and patient generated). However, getting accurate intelligence from this data, particularly the predictive and prescriptive insights, remains expensive, complex and time-consuming.
Advancements in data science and use of ML has helped cross-industry organizations create predictive models at a breakneck speed. HHS agencies can also use ML to solve the challenges outlined above and fast-track development of sophisticated analytical models to predict complex and unique public health scenarios such as:
- Forecasting potential risk movers within a population subset
- Predicting frequent service utilizers, particularly those who use high cost services like the emergency room and inpatient care
- Identifying segments of the population susceptible to opioid abuse
- Predicting high risk patients with a colonized bacterial condition that might pose a threat to other patients in an ICU
- Predicting events of shock based on vital sign time series, etc.
Without ML, developing predictive models to address situations like the above would probably take months and require investments in expensive data science resources. With the tools and solutions that exist today, such models can be built, tested and deployed in a couple of days.
While quite a few HHS agencies are experimenting with AI/ML technologies to navigate the next in analytics, a couple of questions remain unanswered. Are AI/ML technologies only used to achieve faster and smarter predictions? How can these technologies enable agencies to go beyond mathematical data equations and solve business problems?
Let us consider an example – identifying patients who are likely to suffer severe asthmatic attacks during a dust storm and may have to visit an ER.
While predictive models can help agencies identify WHO these high-risk patients may be, the real value will be realized when they can also generate recommendations on WHY and HOW to manage these patients. This next-step in AI/ML-driven insights is what I call “Next-Best-Actions.”
Such recommendations (Next-Best-Actions) obtained through augmented next level analytics will help the consumers of the information (for example, the care manager, provider etc.) proactively manage the patient population.
In the example above, the ML output can arm care managers and other stakeholders (e.g., the provider) with intelligence that:
- The patient’s condition is going to get worse (and not just because of the weather condition), and;
- The patient is likely to visit the ER because he/she ran out of the rescue inhaler or has not refilled the rescue inhaler in last six months and the dosage left is only good for next few days.
In addition to this intelligence, the models can also suggest the following set of proactive actions:
- Recommend that the provider send out an eRx refill of the rescue inhaler to the pharmacy nearest to the patient, and;
- Alert the care manager to contact the patient and to advise him/her to take the prescription
Recommendations such as these (i.e., Next-Best-Actions) can significantly improve care delivery, clinical outcomes and cost.
Therefore, when agencies look to use AI/ML for advanced analytics, they should look for solutions that not only generate predictive insights (WHO), but also generate Next-Best-Actions (WHAT, WHY, and HOW) to solve a complex health care problem end-to-end.