| | April 20189UTLOOK Healthcare Tech focus in a fee-for-service world. One example of this includes the opportunity around readmissions not within 30 days, but rather at 60 and 90 days. In a world where 30-day readmissions are a focus because of payer scrutiny, we like many identified this as a top priority. As we explored the data, we found that 60- and 90- day readmissions were far more of an opportunity with rates much higher than 30-day readmissions. Different from the acute care focus of 30-day readmissions, 60- and 90-day readmissions implicate coordination opportunities in post-acute setting and ambulatory care. We not only were focused on the wrong issue, but the work done to improve had limited value to the larger issue.For outpatient care management, the analytics quickly gave us an indication that there was an opportunity for a higher level of effectiveness. As we examined the `high cost' population, we realized that patients in that category changed quickly and often; as quickly as the claims data was refreshed (monthly or quarterly) a new cohort of high cost patients emerged. In literature, this phenomenon is termed regression to the mean and reflects the tendency of cost to spike briefly, but then return to a baseline level of spend after an event. This was an insight that did not resonate until we were able to witness in our own data. The implication was that for care management, using this approach to identify patients eligible for the intervention meant that we were too late. Claims data is inherently delayed given that it reflects spend that has already occurred compounded by the time needed to transfer from payer to provider. By the time a care manager had been dispatched to a patient, the event had occurred and the patient was back to baseline. For Advocate to realize a greater level of effectiveness, we would need to predict which patients were likely to incur a high cost event and mitigate.Once again, using our Hadoop data platform and leveraging more clinically focused as well as some historical claims information, we developed several different predictive models focused on events clinicians identified as preventable within 90 days by a telephonic care management intervention. The preventable events included hospital utilization (inpatient, observation, and emergency) for clinical care needs like an exacerbation of asthma or worsening of heart failure symptoms. Operationally, the goal was to flag these patients that were rising in risk of unnecessary utilization and for care managers to intervene quickly and resolve the risk. Analytically, the predictive models performed well-enough that we moved to pilot phase quickly with the heart failure model as a start. Results so far have been extremely promising. The pilot has yielded no hospital utilization for patients intervened on and although early, a randomized trial is being planned. There have been additional learnings around how best to align analytic insight with operational realities including:· Relying on the model alone for patient identification is not enough and an initial filter should be applied to gauge the patient's eligibility for care management (e.g., excluding SNF or hospice patients).· The models are very targeted toward a specific disease and allow for an immediate focus on the disease even though patients may have multiple co-morbid conditions. This is a different approach than the traditional approach of a lengthy initial assessment. · Related to the above, the predicted utilization was found to occur shortly after prediction and there is a need to act quickly to mitigate.As we continue to learn in this new industry paradigm focused on value, the need for deeper insight is clear. Data and analytics will be central in ensuring we can look beyond the surface to create that value for patients and providers. HTData and analytics will be central in ensuring we can look beyond the surface to create that value for patients and providers"
< Page 8 | Page 10 >