Within all the rightly harsh press about “physician burnout” there is a bright spot called predictive analytics. I will get to that in a moment. Let me set the stage with a few words about physician burnout.

Physician burnout driven by non-clinical mandates

Practicing medicine has always been a hard job, but the gratification from helping patients plus the societal support and respect for doctors made the job worthy and worthwhile. But in the past 10 to 20 years, a few trends have threatened the sanctity and stability of our profession. The economic mandate pressuring physicians to go faster and see more patients is frankly the most corrosive of these elements.

Not far behind is the burgeoning demand on doctors’ time to do non-clinical tasks such as fill out prior auths or struggle with pharmacy step-care requirements or fill out disability or FMLA paperwork, and the like. Time spent on non-clinical work is extremely frustrating to doctors.

The growth of team-based care as an antidote

During this period, we have also witnessed the move from one doctor-one patient care, in the exam room, to team-based care wherein the doctor is joined often by a nurse practitioner, a nurse, a social worker, a patient tech and an administrative aid. This team may also be backed up by a quality officer whose sole job is to ride herd on contractually required quality measures, such as vaccinations and cancer screenings.

This trend towards team-based care has in general had a salutary effect on the work life of the doctor. Though this shift required major adjustments for most doctors, no one I know wants to go back to the old way. Team-based care more often than not helps doctors work more at the top of their license, and results in a more gratifying work experience.

Selecting the right patients for care management with predictive analytics

One element of team-based care is now usually called “care management”. Care managers are usually nurses who each assist a cohort of 50 to 200 patients who tend to have complex medical and psychosocial problems, and who can benefit from a telephonic outreach program.

How do we select the correct patients for a care manager? This is where predictive analytics comes in. Prior to the development of predictive analytics, we simply asked doctors to refer patients who they thought they needed help with and would benefit from care management. As you can imagine, this was at best a hit or miss method, often resulting in “non-actionable” patients getting on the care management roster.

For the purposes of this discussion, I am defining “non-actionable” as the nurse being unable to alter the trajectory of a patient’s self-care and/or their illness, and thus not positively affecting their overall medical utilization.

Success in care management is defined by me as a reduction in ER visits, hospitalizations, readmissions and a reduction in the total cost of care (TCOC). It is critical to keep the measurement goals as simple as possible given the complexity of measuring multiple processes in a human population.

Predictive analytics versus traditional approaches to population stratification

Predictive analytics are a tool that can assist providers in identifying the truly “actionable patients” as defined above. Traditional approaches to patient stratification had simply identified the highest utilizing and most costly patients. Though this may make some intuitive sense as a strategy, it ends up selecting a lot of patients who are not actionable. Examples of this include patients with metastatic cancer (who may benefit from hospice at some point) or patients with end-stage renal disease (ESRD). Patients with HIV on very expensive drugs might also fall in the expensive but not actionable category.

How do predictive analytics work?

So how do predictive analytics work? There are a wide range of methodologies out there, but generally speaking, predictive analytics apply statistical tools to data (often “Big Data”, although that’s not a requirement) to make predictions about the real world.

Sometimes, the models generated by those statistical tools can be refined through a “training” process called Machine Learning, in which the algorithm is repeatedly tested against its own outputs and adjusted to improve the performance. As analogy, imagine that there is a “payment” with each prediction made about an outcome, and a “cost” associated with each mistake the model makes in that prediction; as the algorithm runs against larger training datasets, it will attempt to maximize its “profit” by optimizing predictions it tries to make while minimizing the errors.

This process can be applied to care management by training against outcomes that represent appropriateness for enrollment. As the algorithm attempts to optimize appropriateness, it will take advantage of available datasets, which can include demographic data, such as the patient’s age and sex; claims data, such as inpatient admits; clinical data, such as lab results and surgical history; and socio-economic, such as housing insecurity and neighborhood education levels. Combining these data creates a potentially more holistic view of an individual patient, allowing a machine learning model to identify opportunities to engage with patients who might not otherwise be obvious candidates for care management.

The value of more accurate identification of patients for care management translates into savings for the ACO by avoiding the misfires of taking on the wrong patients and needing to then move them off the roster, which is both time consuming and expensive.

So back to the beginning of this article: Powerful information technology powers predictive analytics which provide the foundation for accurate identification of the right patients for care management. Care management benefits the doctor by taking work off his or her plate and allowing her to focus on the work she trained for! Modern care management powered by predictive analytics is a necessary and powerful element in the journey from fee for service medicine to value based care.

For a more comprehensive treatment of this topic, I encourage you to check out a talk I presented at the NEHIMSS 2019 Spring Conference. You can learn more here, or access the recording using the form below.

October 16, 2019