System includes clinical alerts, diagnosis risk groups
A predictive modeling program is just the first step in identifying members for a comprehensive disease management program, says Michael Cousins, PhD, manager of health informatics for Health Management Corp. (HMC), based in Richmond, VA.
"A predictive model program is necessary, but it’s not sufficient to obtain positive outcomes unless it is coupled with a robust disease management program," adds Cousins, who is primarily responsible for the development and testing of the predictive models for HMC’s award-winning Healthy Returnssm System, which delivers targeted individualized interventions to members in the company’s disease management programs.
The system received an Innovation in Quality Care and Patient Safety in a Health System Award from the Healthcare Delivery Solutions Congress.
The heart of its Healthy Returnssm System is Health Management Corporation’s proprietary AccuStrat predictive model that uses medical claims information from its clients to identify members with chronic conditions and assign risk stratification for future use of health care resources.
Health Management Corp.’s clients are large employer groups, managed care organizations, and state Medicaid and Medicare organizations.
The AccuStrat predictive model is a hybrid among typical risk stratification models, according to Cousins.
It includes both clinical alert algorithms, which identify members who are not complying with disease management treatment regimes, and diagnosis risk groups, which tabulate the relative risk of the need for future health care interventions.
The AccuStrat predictive model takes hundreds of different elements and comes up with a single score that is used for the overall ranking. Both components, the clinical alert algorithm and the diagnosis risk algorithm, are combined to predict the risk of each individual, Cousins reports.
"A single risk stratification model can predict the severity of disease in an entire population, but it tends to break down if it’s used for medical management," he says.
The model uses 144 diagnosis risk groups and computes the relative risk of each disease, then assigns a risk score to the member, taking into account any clinical alert algorithms.
"Our goal is to find people before their care gets expensive. We’re looking for people who have conditions or events that indicate they will have future, more expensive events or conditions and higher utilization," Cousins says.
For instance, the predictive model identifies people who have diabetes and are at risk for an amputation by examining, among other things, pharmacy use and physician visit for issues related to circulation.
Diabetes, congestive heart failure, asthma, cardiac conditions, chronic obstructive pulmonary disease, and asthma are the primary conditions that HMC manages, along with more than 20 secondary conditions that typically are comorbidities. For instance, the company manages obesity and hypertension along with diabetes.
Once the members are risk stratified, the information is loaded into the clinical management tool, which includes overall risk scores, comorbidities, and a patient profile that includes information about prescription history, tests that are indicated but haven’t been done, and physician visits.
"Our system’s goal is to find everyone with a condition so we’re loose enough to find everybody but tight enough not to have a false positive," Cousins says.
HMC uses the clinical alert algorithm as a basis for intervention as well as to identify high-risk members.
For instance, the AccuStrat program breaks out members who have indicators of noncompliance, such as not refilling prescriptions in a timely manner or not having tests or procedures that are recommended in clinical protocols.
Members whose overall risk score falls below the threshold for individual disease management receive a letter reminding them of what they need to do to manage their disease. The company also sends a letter to the primary care physician, alerting him or her that the patient is not following the protocol.
"Every members identified with one of the chronic conditions receives a single risk score. If the score is above a certain amount, they are identified as high risk and go into the high-intensity disease management program," Cousins says.
The Healthy Returnssm System identifies members who are eligible for the disease management program and flags pertinent clinical information for the disease management nurse. For instance, if a member hasn’t refilled his or her prescription, or a diabetic hasn’t had a hemoglobin A1c test in more than a year, the information appears in red.
The new system has resulted in fewer inpatient admissions and shorter lengths of stay, he adds.
"We have been tracking the outcomes before and after using the AccuStrat predictive modeling system and have gotten improved outcomes. The only problem is that the results are confounded by the fact that we implemented our third-generation clinical management system at the same time," Cousins says.