Data-Driven Patient Outreach: Financially Efficient or Morally Compromised?

I recently spoke with a clinical analytics vendor about one of their provider customer’s bizarre decisions – to no longer encourage a subset of their diabetic population to take the HbA1C test. The reason? These diabetics were covered under a new P4P payment contract, which no longer rewarded such testing.

I am not naïve to the fact that, since the beginning of the practice of medicine, patients have been treated according to different care standards…with payment to blame more often than not.

Now our payment models are evolving and population health management (PHM) has become a word du jour. In addition, the payer-provider line is blurring, and data is flowing more freely than ever between the two former adversaries. From a PHM-centric standpoint, new incentives now present themselves to treat similar patients differently based on these newly acquired datasets.

Different Payment, Different Data, Different Outreach

Take for example, massive healthcare systems such as Intermountain and Geisinger. These HCOs own their own health plans, but unlike Kaiser Permanente, also accept outside insurance. Patients can therefore be covered be under FFS, P4P contracts, risk-based contracts, as well as the HCO-specific payer plan. In addition, these HCOs have developed their own advanced care models that should, in theory, override the influences of payment on care. However, things soon get tricky with regards to PHM outreach:

• If Debbie the Diabetic is covered under FFS, she will get zero outreach or, if she is lucky, outreach based on her sophisticated HCO’s advanced care models — with the HCO acting against its own short term financial interests.
• If Debbie is covered under an all-upside P4P contract, she will get reminders in order for the HCO to tick off certain boxes in her payer’s P4P contract (process or outcomes-based), for example, to encourage her to come in for HbA1C testing, foot exams, etc. Perhaps there will be minimal effort invested in helping her control her blood glucose better.
• If Debbie is covered under an at-risk contract that incentivizes global cost reduction (MSSP ACO, ACO with potential downside, capitation), someone will be trying to figure out how to get her cost utilization down. Her claims data will be run through a risk assessment model, and her costs will be projected. If Debbie is sufficiently out of control and really racking up the costs, she will be assigned a high touch case manager.

Note: the important enabler within these different outreach models is different datasets and the analytics software that runs on top. P4P Registries run on top of billing or EHR data. Risk-based analytics runs on top of adjudicated claims data. Within each analytics system, care gaps and associated outreach are tailored toward entirely different payment models.

Giving up on “Hopeless” patients (because the data says so)

There are many more usecases ahead in which data can be leveraged to perform unequal PHM-outreach. Take, for example, the ability of a risk-bearing HCO to predict which patients will actually comply with outreach, and which can be assigned to the “Hopeless” bucket.

Every medical professional can recount various patient horror stories. The mentally ill woman who visits the ED every weekend even though there is nothing wrong with her physically. The morbidly obese diabetic who has just lost a foot but continues to eat himself into oblivion. Despite the repeated pleas of multiple case managers, these patients are unable or unwilling to modify their respective trajectories.

If there is absolutely nothing the HCO can do to change certain patients’ behavior (within a given budget), should the HCO cut its losses, ignore the Hopeless, and focus resources on other patients?

Shiny new predictive models could be used to this end. Think of the growing number of variables with predictive power beyond ICD, CPT, and charges: BMI, race, zipcode, past outreach responses, self-generated fitness data. Let’s enter morally shady territory and include variables sold by data scrapers: real estate transactions, courthouse documents, credit worthiness of facebook friends, etc.

Running Debbie the Diabetic through this predictive model, we might learn that there is a 1% chance that she will respond to automated outreach, and a 15% change that she will respond to high touch outreach. It may be financially efficient to focus care management resources on other patients, but what would Debbie’s doctor say about this?

The medical establishment isn’t exactly in love with unfamiliar predictive models that use sensitive data sources to tell them which patients will respond to outreach. Clinicians usually already know who these patients are within their panels — and believe in a more equal approach to outreach.

As providers continue to disrupt themselves via value-based reimbursement, PHM-outreach inequalities will be just one of the many contentious problem areas going forward…and the desires of risk managers and clinicians will continue to collide. In my conversations with vendors and end-users for the Clinical Analytics Market Report research, I for one was surprised at how nonchalant my interviewees were in discussing this morally ambiguous territory — and I have yet to hear the patient’s voice.