Patient Complexity and Population Stratification Strategies

Kathy Schneider, PhD and Miriam Isola, DrPH CPHIMS

The presence of chronic conditions has become epidemic with over 117 million people, or nearly half of the US population, suffering from a chronic condition [1].  The high prevalence of chronic disease among the Medicare population has been well documented [2, 1] and is of particular concern because many people over 65 suffer from not one, but multiple chronic conditions [1].  As the healthcare industry works toward improving quality and patient experience as well as managing healthcare costs, efforts to understand per capita service use and expenditures among complex patient groups, and how to manage costs are needed [1].

In a published study [3], we analyzed Medicare claims to determine the prevalence, utilization, and Medicare program costs for some common and high cost chronic conditions in the Medicare fee-for-service (FFS) population in 2005.  Six high frequency and high cost chronic conditions were selected for study: diabetes, chronic obstructive pulmonary disease (COPD), heart failure, cancer, chronic kidney disease (CKD), and depression.  The care settings commonly used for treating the conditions, as well as the per capita use and average per beneficiary Medicare payments by medical condition, were examined.  Costs were defined as total Medicare payment, or the sum of all FFS claim payment amounts, per beneficiary for 2005. 

We found that fifty percent of Medicare FFS beneficiaries were receiving care for one or more of these chronic conditions.  The highest prevalence was observed for diabetes, with nearly one-fourth of the Medicare FFS study cohort receiving treatment for this condition (24.3 percent). 

As displayed in Table 1, about half of Medicare FFS beneficiaries studied had none of the six chronic conditions (50.7 percent).  Twenty-nine percent of beneficiaries were receiving care for only one of these six chronic conditions, 12.7 percent were receiving care for two of the conditions and 7.6 percent were receiving care for three or more of the conditions (Table 1).  The annual Medicare payment amounts for a beneficiary with only one of the chronic conditions was $7,172.  For those with two conditions, payment jumped to $14,931, and for those with three or more conditions, the annual Medicare payments per beneficiary were $32,498.  Utilization within each care setting increased as the number of chronic conditions increased.  The presence of even a single chronic condition escalated the use of services in every setting. 

The highest proportion of beneficiaries with multiple chronic conditions was observed for CKD (see Figure 1).  Almost 33 percent of beneficiaries with CKD had one of the other conditions, and nearly 50 percent had two or more other chronic conditions.  The most common co-occurring conditions were HF (52.9% of those with CKD) and diabetes (51% of those with CKD; data not shown).  For diabetes, depression, and cancer, however, beneficiaries were more often diagnosed with only that condition (e.g., 47.3 percent had only diabetes).

   Source:    Schneider KM, O’Donnell BE, and Dean D.  The Prevalence of Multiple Chronic Conditions in the Medicare Population.  Health and Quality of Life and Outcomes . 2009, 7:82.

Source: Schneider KM, O’Donnell BE, and Dean D.  The Prevalence of Multiple Chronic Conditions in the Medicare Population. Health and Quality of Life and Outcomes. 2009, 7:82.

 

Measuring Chronic Conditions is Not Sufficient for Risk Scores

The prevalence of multiple chronic conditions is significant, and must factor into strategies for effectively managing care for patient populations.  However, the presence of chronic disease is not the only meaningful way to risk stratify this particular patient population.  There is a great body of emerging evidence about treating complex patients.  In a recent study published in Medical Care [4], we demonstrated that it is useful to go beyond conventional comorbidity measures (i.e., classifying or counting the number of chronic conditions for measuring patient health status).  These conventional approaches are limited in terms of understanding the impact of disease severity or identifying patients who may have diminished functional status that requires additional assistance to maintain their level of functioning, or perhaps be near the end of life.  We defined and identified function-related indicators (FRIs) from administrative data, and found that this information significantly explained which patients were at highest risk of mortality within 12 months [4].  By employing more comprehensive examination of patient complexity, including FRIs, we better capture heterogeneity of patients and inform more customized care management approaches. 

An actionable risk scoring or stratification strategy must be guided by the objectives for the stratification effort.  The first step is to determine how your organization will use this information: is a risk score needed to identify the sickest and most vulnerable patients for interventions, to develop registries of patients with particular conditions so they may be effectively monitored, or is it to identify patients where there is the greatest opportunity for reducing future costs? It is important to consider the data that will be needed to meet your objective, which often goes beyond inpatient clinical data. Social determinants of health account for about 70 percent of health outcomes, yet are often overlooked in risk segmentation strategies [6]. Some clinical, behavioral and social risk factors, if addressed appropriately, have the potential to dramatically reduce avoidable hospitalizations and emergency care [5]. Therefore a stratification that includes clinical, functional and behavioral/social risk factors has the ability to alter a disease trajectory thereby “bending” the cost curve.

Technological solutions such as risk stratification tools, or analytic tools embedded in the electronic medical record, often seek to identify patients with particular chronic diseases or a high risk score [7].  There is a plethora of scoring strategies and tools for risk scoring available, many of which are designed to accomplish a particular objective and may not be useful for your particular organizational goals.  For example, some tools fall short of being able to examine population health since the scores require data on patients who have been hospitalized – which includes only a subset of an organization’s population.  Models that target both care and cost management are needed for healthcare organizations and benefit patients. Organizations may want to customize their risk stratification model to make it fit better with their goals and organizational strategy.

Risk Scores Require Strategy

Sum-IT Health Analytics, LLC recommends moving beyond a “one-size-fits-all” risk score, and using a more strategic analytic approach to comprehensively identify potential population care and coordination needs.  Although the idea of using a number or score to classify patients is appealing, it over-simplifies the task of configuring care delivery to address the needs of the population beyond those that have experienced a hospitalization.  A scoring approach that assesses the disease burden of the underlying population and addresses population complexity would ideally include information beyond hospitalization data and identify people at risk for a variety of potentially avoidable adverse outcomes as early as possible, such as uncontrolled diabetes.  By identifying complex patients early and providing appropriate routine care, urgent and emergent care may be avoided.

At the practice level, patient population stratification is an important step in proactively managing the health of the population as providers consider the different goals that may be achieved through various care coordination strategies.  For example, a relatively young and healthy population could benefit from routine preventive services and ongoing health and wellness programs; an otherwise healthy population with a single chronic condition may benefit more from intensive disease management than a population with many indicators of diminished functional capacity, multiple chronic conditions, and advanced age.  Identifying patients who are terminally ill or experiencing a serious decline in health due to advanced age can trigger important discussions with them and their family regarding aggressiveness of treatment, symptomatic support and possibly a broad view of palliative care. Additionally, asking patients and populations about root causes of poor health, such as whether they have enough food or heat in their home for the winter [8], allows for recognition of the need for particular social and community resources that could benefit the population.    Practices can use their data to evaluate their evidence-based disease management practices for all of their patients, including those with multiple chronic conditions and FRIs, and coordinate care so that patients are treated in the most appropriate care setting, preventing hospitalizations and the need for emergency care to the extent possible.

Risk Scores Require Action

Caring for your population doesn’t begin with a risk score; it begins with a clear objective for the scoring strategy.  The risk score cannot be the sole driver in caring for the population, but rather it is a key metric to use for population segmentation and refining strategies that ultimately lead to the appropriate plan of care.  A customized risk model that includes clinical and social/behavioral risk factors may be desirable to meet the organization’s objectives for population stratification.  Then, once the risk model has been appropriately designed and tuned to the organization’s data, a plan must be developed to address how to act on this risk scoring information.  Anyone at the organization who relies on the scores should have a basic understanding of what the score means – and what it doesn’t mean.  For example, a score that identifies people with historically high hospital spending is not ideal for selecting patients for a practice-based diabetes registry, since many who could benefit would be missed.  Furthermore, a risk score is just a number –these numbers must be interpreted to evaluate the overall health of the patient population and then you need to decide what constitutes a high or low score.  The answer is not the same for every provider.  Where you choose to draw the line has important implications in terms of what proportion of your population is stratified into the various levels and/or programs.

Conclusions

As organizations begin to work with predictive models to assess the risk of their populations, they need to take a strategic approach to analytics.  Steps for doing this include:

  • First, consider how to customize your models to address specific objectives such as gaps in quality or provision of services.  This allows you to use the data to guide interventions and makes the data actionable. 

  • Next, include FRIs, social and behavioral risk factors in the analytic model along with clinical data.  Using only clinical data will not provide a full picture of population risk.

  • Finally, factor in multiple chronic conditions as they contribute important data points related to overall disease burden and cost.

    The resultant individual risk scores must be evaluated in terms of how they look within the broader population; this information will inform strategic planning and re-configuration of services.

    Although organizations must avoid taking a simplistic approach to risk stratification, that doesn’t mean the process will be difficult – if you have the right data partners.  Sum-IT Health Analytics can help providers work through their strategic objectives for risk stratification so that useful risk stratification models can be constructed.  Sum-IT also helps providers implement strategies for interpreting the risk scores and intervening with the population.  Ideally, risk scores will provide insight in terms of the health of the population, how to best configure services for the population, and then ultimately individualize care plans.

 

The authors would like to thank June Wilwert for sharing insights on actionable risk scores.

Contact Sum-IT authors at:

 

References

  1. Ward BW, Schiller JS, and Goodman RA: Multiple Chronic Conditions Among US Adults: A 2012 Update. Prev Chornic Dis 2014, 11:130389. DOI: http://dx.doi.org/10.5888/pcd11.13038

  2. Wolff J, Starfield B, Anderson G: Prevalence, Expenditures, and Complications of Multiple Chronic Conditions in the Elderly. Archives of Internal Medicine 2002, 162:2269-2276.

  3. Schneider KM, O’Donnell BE, and Dean D.  The Prevalence of Multiple Chronic Conditions in the Medicare Population. Health and Quality of Life and Outcomes. 2009, 7:82.

  4. Chrischilles EA,  Schneider KM, Wilwert J, Lessman G, O’Donnell B, et al. “Beyond comorbidity: Expanding the definition and measurement of complexity among older adults using administrative claims data.” Medical Care. 2014; 52(3), S75-84.

  5. Jackson C and DuBard, A. “It’s All about Impactability! Optimizing Targeting for Care Management of Complex Patients.” Community Care of North Carolina. Data Brief. November 2, 2015. Issue 4.

  6. Monegain, B. “Chilmark: Risk calculations heave to change with value-based care.” Healthcare IT News, June 16, 2016.

  7. LACE Index – to identify patients at risk for readmission (see, for example: http://www.besler.com/lace-risk-score/); The Rothman Index (for example, http://www.perahealth.com/); Providers may have analytic tools embedded in their EMR to examine patient risk scores. 

  8. Health Leads.  “Social Needs Screening Toolkit.” Downloaded from https://healthleadsusa.org/what-we-do/solutions/.