Predictive Analytics Drives Patient Engagement and Improves Care

September/October 2013

Analytics

Predictive Analytics Drives Patient Engagement and Improves Care

 

Despite the best efforts of clinicians around the country, healthcare delivery is still largely a cottage industry. Just like the old family-run corner store, or the artist down the street who makes jewelry to sell at local craft fairs, isolated teams of wonderfully talented and committed individuals have for many years done the best they can to provide rescue care. Rescue care is reactive care, characterized by heroic efforts to respond to managing illness during face-to-face interactions and only when the patient has presented for care.

Each physician’s office and hospital has tried hard to optimize around a world that has now moved on and, much like cottage industries of the past, it’s time to find ways to scale up healthcare to focus beyond the bricks and mortar of today’s institutions. In fact, as many other industries have begun to learn, the real challenge will be how best to industrialize healthcare without losing the deep, trusted personal relationships that were so characteristic of the corner store.

Many healthcare professionals see the term “industrialization” and cringe. It smacks of Big Brother and the depersonalized efficiency of the manufacturing industry. However, when described in terms of Total Quality Improvement, Lean or Six-Sigma, most are much more willing to accept the idea of a virtuous cycle of continuous improvement that is characteristic of industrialization. Industrialization requires data, and healthcare organizations have embarked on a data gathering frenzy over the past decades. Recent advances in digital technology, coupled with incentive payments for adoption of certified clinical software technology, continues to accelerate that trend. However, the data party is nearly over, and it’s time to start to clean up. There is more than enough data, but it now needs to be managed as an essential organizational asset that can do more than automate workflow. It’s time to put data to work to reduce inefficiency, improve quality at both the point of care and beyond, and to provide proactive guidance based on deep, data-driven insights, rather than gut feel and intuition.

So, they say that first step in the 12-step program is acknowledging that you have a problem. Yes, you have a data problem, and although you’ve made progress on providing quality reports and departmental dashboards, it’s time to embark on that 12-step plan that improves your organization’s analytic maturity. Don’t just pay lip service to patient-centered care; put your data to use to make it happen. Take some of the best that the corner store had to offer and find ways to better understand your customers by proactively gathering personal needs, biases, preferences, and likely responses to enable “personalized care” and incorporate those insights into processes, operations and workflows to anticipate and make better decisions around care and drive improved engagement.

Using Analytics to Predict Patients at High Risk of Readmissions Researchers know that hospital readmissions within 30 days of discharge are a big reason why healthcare costs remain stubbornly high. Insurers were unable to figure out how to predict who was likely to be readmitted and intervene before the person landed back in the hospital. But now they can. Blue Cross and Blue Shield of North Carolina (BCBSNC) has built a model to more accurately predict hospital readmissions and deploy nurse case managers to help patients deemed most at risk. By looking at all causes for readmissions and looking at the data from the authorization stage, the model correctly beat chance by 400% in identifying at-risk patients. The Costs of Readmission About 13 percent of inpatients account for the majority of hospital costs and much of those costs are related to readmission. Hospitals, doctors and insurers have tried different approaches to reduce readmissions – from increasing discharge education to assigning case managers to follow up with patients. Like many insurers, BCBSNC used length-of-stay data or diagnosis to flag patients for post-hospital outreach from nurse case managers. But the readmission rate wasn’t budging. “Our nurse case managers told us we were flagging patients that didn’t require our help, while they were hearing anecdotally of other patients who needed intervention,’’ said Daryl Wansink, BCBSNC director of health economics. Digging into the Data BCBSNC wanted to come up with a better way to identify patients at risk for readmission and reach them quickly—in some cases while they were still in the hospital. But much of the detail about a hospitalization doesn’t make its way to the insurer until that 30-day window has passed. The goal: Predict the potential for readmission by looking at the data supplied at the time of authorization and admission so staff can engage with the patients and providers before discharge. To accurately predict and intervene, BCBSNC needed to analyze all its data on hospital readmissions to find patterns that could provide a much more nuanced flagging system. The insurer discovered 50 candidate predictors—details like whether a patient has diabetes or lives alone—that are factors in calculating readmission potential. The model doesn’t simply flag anyone with one of these predictors—that was a part of the crude method the insurer previously used. Instead, it looks at what a patient is being admitted for and compares that against the candidate predictors. For instance, a diabetic widower might be at very high risk of readmission following an in-patient stay for heart attack—but at hardly any risk for readmission for treatment of a stomach bug. In addition, the model is constantly rerun and enhanced as new information becomes available. Saving Money and Helping Patients While BCBSNC wants to manage medical expenses and keep premiums reasonable, the readmission project is mainly about enhancing the quality of life. As soon as BCBSNC receives admission notice on one of its clients, the prediction software calculates a readmission risk. From there, nurse managers contact not only the patient, but the hospital and physician practice, too, to begin planning for a successful discharge. “We want to do everything we can to keep the discharged patient out of the hospital, keep him healthy and keep his quality of life high,’’ said Wansink. The insurer takes a very collaborative approach toward using the information and shares it with its network providers on a daily basis. This transparent approach, and quality emphasis, has encouraged many providers to sign contracts with BCBSNC that tie reimbursement to their own efforts to reduce readmissions. Ease of Use Makes Project a Success BCBSNC uses analytics software from SAS to build the prediction models, which offers advanced statistical options like neural networks and decision trees. The analytics enable BCBS to find relationships that aren’t readily apparent when running standard regression models and it cuts down on the amount of work spent churning through data to look for relationships. Ultimately, the insurer wants its efforts to be productive for its members. By using insights provided by predictive models, BCBS can empower its nurse case managers to make a difference in people’s lives.

Moving Toward Accountable Systems of Care

Although it’s too early to say whether some of the emerging ACOs or bundled payment models will prove successful, there is growing sense that value-based payment is here to stay. Payment models that incentivize improved coordination of care and patient outcomes will thrive on data and, it won’t be good enough to analyze past performance; leading organizations will use predictive analytics to help reduce errors, improve outcomes, and control costs. Let’s look at how predictive analytics can improve organizational foresight by using the example of hospital readmissions.

Avoiding Patient Readmissions
At SAS, we focus on helping organizations use broad and deep data sets to identify patients at higher risk of readmission as well as to provide recommendations on which patient-specific interventions are most likely to reduce that risk. By revealing and personalizing individuals’ clinical risks factors, providers are able to understand and manage critical drivers of readmission risk as well as understand the anticipated near- and long-term impact of intervention. Integrated care delivery networks are better able to tailor intervention strategies to each individual’s specific needs, and deliver targeted interventions at scale using a variety of approaches and technologies.
Providers are now starting to leverage advanced analytic models to act as a safety net to recommend and prioritize “just in time” insights and recommendations at the point-of-care, as well as between visits, to facilitate both pre-discharge planning as well as make data-driven decisions about most effective personalized patient-level interventions during critical first days and weeks following discharge.

The Best of Both Worlds?
As we continue on the journey to industrializing healthcare, let’s use the data we already have more effectively to guide care, but let’s also make sure that we don’t let data de-personalize care delivery. We should learn from other consumer-centric industries that have learned that analytics can be the glue that allows for the best of both worlds—one that will allow us to deliver high value care at scale, while at the same time focusing on the personal circumstances, needs, and motivations of the individual.

Graham Hughes serves as chief medical officer on the SAS Center for Health Analytics and Insights team. He joined SAS in 2011, bringing to the organization more than 20 years of experience in developing and delivering innovative healthcare information technology (IT) products and services. Prior to joining SAS, Graham spent six years working as vice president of product strategy and chief medical informatics officer at GE Healthcare IT, leading a customer-facing advanced technologies innovation team, as well as spearheading the annual strategic planning process. He was the primary physician leader driving GE’s knowledge platform strategy and associated products in collaboration with Intermountain Healthcare and Mayo Clinic. He may be contacted at graham.hughes@sas.com.