The Role of Predictive Analytics in Reducing Readmissions

Dr. Peter Edelstein, M.D.

Dr. Peter Edelstein, M.D., Chief Medical Officer, MEDai

Before health care reform’s implementation of regulations to penalize hospitals for 30-day readmissions, there were no incentives to avoid readmission, right? Well, no.

Avoiding readmission has always been a goal of health care providers, patients, and caregivers for one simple reason—quality of care. Patients don’t want to be readmitted, and their physicians, nurses, and caregivers don’t want them readmitted. We want our patients to do well.

One way to avoid readmissions and improve quality of care is to keep patients in the hospital until they are healed. But this is costly and can increase the risk of hospital-acquired infections. Thus, decades ago, in an effort to reduce costs, the industry focused on ways to drive down length of stay (LOS); as it did so, the rate of readmission for the same complaint rose. Readmission and LOS are the bookends of quality care. Somewhere in between these two bookends is the optimal amount of care a patient should receive from both quality and cost-efficiency perspectives.

Reducing Readmissions: A Difficult Quality Metric to Manage

Of all the in-hospital quality metrics hospitals currently focus on—among them, reducing sepsis, mortality, LOS, decubitus ulcers, and cardiac arrest—the hardest to accomplish is reducing 30-day readmissions. Hospitals, payers, and caregivers have spent years and millions of dollars creating programs specific to their patient populations to try to reduce readmissions. But there have been very few successes, and those successes have been minor.

Why? There are differing opinions, but one of the reasons may be that attempting to reduce readmissions while patients are hospitalized means, by definition, that you are working within a controlled environment. The reality is that 70% of a patient’s readmission risk may relate to factors in the patient’s environment outside the hospital.

A patient may be homeless or jobless, have limited financial means to purchase medications, have no access to a car or bus to see the physician for follow-up, and/or have no caregiver. Without plans in place to address these issues, the patient is at increased risk of being readmitted for the same or a related problem.

Thus, to a large extent and in some populations, the factors impacting 30-day readmission cannot be directly controlled while the patient is in the hospital.

Role of Predictive Analytics in Reducing Readmissions

For outcome metrics such as reducing in-hospital mortality, in-hospital sepsis, in-hospital transfer to the intensive care unit, or in-hospital decubitus ulcers, predictive analytics can help to identify patients at risk on a daily basis, stratify their risks each day, and allow providers to develop interventions to reduce those risks in-hospital. But readmission is the unique outcome that is greatly impacted by what happens in the patient’s world outside of the hospital.

Although predictive analytics cannot guarantee a reduction in readmission rates directly, predictive analytics can greatly assist health care providers who have developed outreach programs for their communities and populations through improved resource utilization. Predictive analytics can identify at-risk patients, stratify their risk level, and demonstrate risk trending, enabling providers to assign the right patients to the outpatient clinic, and reducing resources and, potentially, FTEs without negatively impacting readmissions, allowing healthcare systems to apply resources to those  patients who are truly at risk.

Predictive analytics is tremendously powerful in improving resource utilization for patients at high risk. It can help eliminate the shotgun approach in which every patient with heart failure, pneumonia, or acute myocardial infarction receives intensive inpatient and outpatient case management and clinical services. The shotgun approach is expensive and creates burnout among healthcare providers.

Stratifying risk of readmission means that predictive analytics tools must be able to include socioeconomic data—such as marital status, homeless status, financial status, and drug or alcoholism, among other variables—in the electronic medical record (EMR). Hospital personnel and vendors of predictive analytics tools must work together to identify those variables that are important to each particular hospital population, which varies from county to county, and city to city. The knowledge repository for each patient population lies with the hospital personnel—the doctors, nurses, case managers, C-suite, and medical records personnel—who can identify the variables that impact their own populations.

New Tool on Horizon

In early 2013, Elsevier MEDai will field-test a new predictive analytics tool called Pinpoint Review 2. This iPad-based real-time solution will be used by a case manager, nurse, or hospitalist at the patient bedside. It pulls clinical and administrative (ie, socioeconomic) information from the patient’s EMR each time the record is updated and provides risk stratification for five outcomes:

1. Risk of mortality during hospitalization

2. Risk of sepsis during hospitalization

3. Risk of transferring to the Intensive Care Unit during hospitalization

4. Risk of excessive LOS during hospitalization

5. Risk of 30-day readmission after discharge

Care providers can see on a scale of 1 to 5 which individual hospitalized patients are trending one way or another for each of these five risks. Of the five risks predicted by this tool, four are driven by medical information. The 30-day readmissions calculation is more hospital population specific, relying also on socioeconomic data important for the hospital’s particular patient population. In this way, the tool will become a formidable ally in targeting resource utilization to those patients who are at greater risk for 30-day readmissions.


Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>