The most costly 1% of patients often have the most complex needs.
At Kaiser Permanente in Southern California, 1% of our members account for 29% of our total health care costs. In 2015, this translated to about $4 billion in aggregate, for an average of $98,000 per member, annually. The concentration of health spending at Kaiser Permanente is in line with national data, contributing to an increasing focus on improvement efforts for patients with complex needs. To the extent that better management of such patients can improve their health and avoid the need for resource-intensive treatments, the more value we deliver.
Redesigning Care Delivery to Match Patient Risk
The U.S. health care system can do better for patients with the most complex needs. For example, disease-specific case management — which has worked for many patients — may not be the best approach for a patient with multiple chronic conditions. Because each condition is managed separately, this model may not take into consideration the total disease burden carried by the patient. The most complex patients often require a different approach.
At Kaiser Permanente, we envision a nimble and flexible care delivery system that can quickly adapt resource intensity to match patient need. This involves using risk stratification to develop flexible population segments, studying the resulting patient groups, and building care models that align with each segment. Leading-edge systems around the nation are working toward this goal.
For lower-risk patients, this model that relies heavily on automated, electronic and self-service resources may be sufficient. At integrated delivery systems such as Kaiser Permanente, where we are accountable for the health of a defined population, we should increase services through these light-touch models for our young and healthy members, many of whom may have no contact with the medical system for several years.
As patient risk increases, so should the intensity of the care. Higher-tiered models add layers of lay and licensed providers working on medical and social needs alike, followed by physicians embedded in interdisciplinary teams, and — for the highest risk patients — heavily resourced and high-intensity interdependent care teams.
Proactive Patient Identification Using Predictive Analytics
A recent commentary by the leaders of five foundations called for a concentrated effort to better characterize the highest-risk segment of the population and to redesign care to meet their needs. We are actively engaged in this work, by studying the needs of our population and identifying clinically cohesive subgroups. Rather than focusing on the patients who were most costly last year, we have created an algorithm to predict who will be our high-cost patients next year. Our algorithm is a novel approach, designed to be flexible and customizable, and built using existing tools in our data-rich environment.
Although our algorithm does not perfectly capture future spending risk, it does help us focus our effort on those patients who are on a trajectory toward extreme spending, rather than those whose historical high spending is already resolving. We are focusing on the population predicted to be in the top 1% of spending this year. Few of them were also in the top 1% last year, but mean per-member spending last year for this cohort was still far above the overall population average — $60,000 compared to $2,500.
The predicted top 1% population is not homogenous. Using Latent Class Analysis modeling, we identified nine subgroups of patients in this cohort. Interestingly, eight of the nine subgroups can be grouped into four pairs that are clinically similar but have divergent historical spending and utilization. The figure below shows the nine subgroups with clinically similar pairs grouped together. For example, 18% of all predicted high utilizers have kidney disease. Care for half of them cost $39,000 on average per member last year, and for the other half, more than twice as much.
Using design thinking principles and small-scale tests of change, we are actively testing new care models matched to specific patient segments. Our pilot tests are designed to produce rapid evidence and can be scaled and spread throughout our system, if proven beneficial. At the same time, we are expanding flexible, low-intensity models of care for patients at the other end of the cost spectrum.
There are very real challenges for providers and health plans to maintain a viable business model while caring for patients with complex needs. Kaiser Permanente is “leaning in,” by seeking opportunities to deliver more of the high-value care that patients need and want — care that is coordinated, empathetic, and patient-centered, and that allows patients to stay at home.
Because of our integrated delivery model and use of analytics, Kaiser Permanente has a unique ability to undertake structured tests of new care models across the continuum, at scale, and in a coordinated strategic fashion. Our pilot program to identify social non-medical needs and to fill those gaps by making community referrals is one significant step “in” — and we plan to go further. Incredible things can happen when care and coverage work hand in hand.
Jason Jones, PhD, developed the predictive algorithm described in this article. Ernest Shen, PhD, collaborated on the Latent Class Analysis. Janet Lee, MS, Dana Barnes, MPH, and Stephen Zuniga, PhD, were vital partners in data assembly and visualization. Margo Gordon, PhD, Kati Traunweiser, MBA, Lynn Garofalo, DPPD, MHA, Annet Arakelian, PharmD, and Artair Rogers, MS, helped lead this work at Kaiser Permanente Southern California. We also thank Jeffrey Brenner, MD, for his advice and insight.
This post originally appeared in NEJM Catalyst on December 1, 2016.