Four major strategies for managing high-risk populations 

Predicting risk - improving the approach to risk stratification 

Healthcare organisations have used risk stratification for decades, and the availability of data-driven tools for stratifying and segmenting people at scale in near-real time based on health risks has increased, with the tools becoming more mature and sophisticated over that time. 

For years, much of the focus has been applied to primary care but is now being applied to population health management in order to identify different at-risk cohorts that require different interventions. 

Much of the existing evidence for risk and approach is still at the population level, and progress is slowed by a historic reliance on paper and siloed information. The new age of interoperability is enabling the ability to standardise data and normalise records from multiple disparate health and care systems, utilise information from the citizen and all types of care providers. This combination is providing insight that has never been available at the speeds and reproducibility available today.  

Technology is evolving at pace, and with it, the availability of big data, growing AI and machine learning capabilities is enabling new insights that can be applied at scale across a population as well as to an individual citizen. Because of technology, we are closer to being able to understand individual risk and truly personalise health and care management plans than ever before.  

Having more information in real and near-real time provides the ability to move risk stratification into the clinician’s workflow to provide them the ability to utilise information as they carry out clinical assessments and interventions. Risk stratification is the initial step towards knowing their at-risk cohort, before assigning the most appropriate and timely resources to each person at a high-level of risk. This process helps care professionals determine which people to include in a care management programme and how to best target impactful interventions to these individuals.  

In the UK, groups like the Bradford Institute of Health Research are pioneering risk stratification tools such as the Electronic Frailty Index² to help recognise and diagnose frailty earlier, and better address the complex needs for this vulnerable group through individually targeted evidence-based pathways of care.    

The data used in calculating the relative health risks of each person is a major factor for determining the robustness of a risk-stratification process. This process requires a multitude of data sources that must be gathered and normalised. These sources could include cost and pharmacy data, electronic health records (EHRs) and other transactional system data on a person’s health history. 

The maturing of the digital landscape now allows for new data sets to be applied to risk stratification. Indeed, more recently, healthcare experts have agreed that the process of risk stratification needs to consider another data point when choosing people for inclusion in a high-risk population – impactability. In 2013, a quartet of researchers working to improve risk stratification for high-risk populations called for the use of impactability models alongside predictive models.  

According to the findings of Lewis et al.,³ population health management programmes are more likely to achieve their intended cost and outcome improvements if providers understand which people in a given population responded positively to care management interventions. 

“In response, many organisations have developed impactability models that seek to identify the subgroups of high-risk people who are most likely to engage with and respond to various preventive interventions,” they wrote. “This additional filter is intended to improve the cost-effectiveness of preventive programmes.” 

Speaking to Chris Delaney from Insignia Health, he highlights the importance of understanding levels of patient activation in enabling person-centred care. He advocates that individuals low in activation are often typically familiar with health failings, and much less likely to engage, even with support. These individuals stand to gain the most when supported appropriately. Chris advises to step back from the clinical or medical activation model (what you want a patient to do) and focus on behavioural activation and understand what an individual is capable of taking on. 

Risk stratification based on both predictive models and impactability, therefore, is most likely to determine the success of targeting high-risk populations. 

Choosing the right care management approach 

In 2016, the research of Hibbard et al.⁴ built on these findings and sought to define impactability more thoroughly by measuring a person’s self-management ability with PAMs as a means of predicting that person’s care utilisation and cost within the framework of a population health management programme, such as an integrated care system (ICS).  

“Current approaches fail to recognise that even though many high-risk patients have a heavy disease burden, they may also have well-developed self-management skills, while other high-risk patients may not.”  

- Hibbard et al.  

“Patients with the same risk level tend to be treated the same, regardless of their ability and willingness to manage their own conditions,” they asserted.

Using data from an academic health system, Hibbard et al. concluded that people with limited self-management capabilities were more likely to visit the emergency department and become hospitalised.

To address cost efficiency and any associated financial risk properly, the researchers called for future care models “to be increasingly nuanced and to take into consideration patients’ behavioural tendencies as well as their social and clinical profiles.” 

The research of Hibbard et al. expanded on impactability by measuring a person’s self-management capabilities as a means of predicting ambulatory care sensitive (ACS) utilisation and predicting the diagnosis of a new long-term condition.  

• Patients with the lowest activation score at baseline had a 62 percent greater likelihood of having an avoidable hospitalisation compared to the most activated group one year later (again, after controlling for baseline demographics and chronic conditions). Two years later, the difference between the least and most activated groups was 40 percent, while three years later the difference was still 30 percent.
  

  

  Figure 1: Ambulatory care sensitive (ACS) hospital use by baseline PAM level

   

• Patients at the lowest activation level at baseline were 25 percent more likely to develop a new chronic disease in the next calendar year compared to patients at the highest activation level. The same analysis two years after baseline showed a 31 percent difference between the lowest and highest activation groups. 
• Patients with the lowest activation were also found to be 51 percent more likely to have a new chronic condition diagnosis two years later compared the highest activation.

“Understanding the patient’s capability for self-management is a key part of understanding the risk of health declines and of avoidable utilisation.” 

– Hibbard et al.

“By stratifying populations by patient activation scores, healthcare delivery systems can identify and help those patients with limited self-management skills in time to prevent poor outcomes and unnecessary utilisation.”   

In 2018,v the Health Foundation published findings addressing the relationship between PAM levels of activation (there are four levels along an empirically derived 100-point scale) and A&E attendance and emergency admissions.⁵ They found that patients who were most able to manage their health conditions (Level 4) had 38 percent fewer emergency admissions than the patients who were least able to (Level 1). They also had 32 percent fewer attendances at A&E, were 32 percent less likely to attend A&E with a minor condition that could be better treated elsewhere and had 18 percent fewer general practice appointments.

If those currently least able to manage their conditions were better supported, so that they could manage their conditions as well as those most able, this could prevent 436,000 emergency admissions and 690,000 attendances at A&E, equal to seven percent and six percent respectively of the total in England each year. 

Based on the literature, achieving the goals of managing high-risk populations — improved health outcomes and reduced costs — begins by recognising the many factors that will determine whether the resources dedicated to this group will have the intended effect.  

Adjoining social determinants, such as economic stability and education, along with and behavioural profiles is generating the advent of much more interesting and diverse models for risk stratification and associated care management intervention. For example, the primary stratification could be based on disease burden and medical risk, with a secondary stratification based on citizens’ activation levels to make a behaviour change (for example using PAMs), but it has also been proven that doing a primary stratification based on activation first regardless of disease is an effective approach to addressing population risk and delivering outcomes. For example, PAM Level 1 is 62 percent more likely to have a hospitalisation compared to Level 4,⁶ and movement from Level 1 to Level 2 would prevent 333,000 ED admissions and 504,000 A&E attendances, equivalent to five percent of all admissions, six percent of A&E attendances in England last year.⁷  

Educating the workforce to complement technology 

Alongside technology, the importance of care professionals' understanding, particularly stakeholders involved in decision making, should not be overlooked. Equipped with knowledge of predictive models, processes, challenges, benefits and opportunities, it is likely that the project will deliver more value. Throughout the engagement process, consider:

• Building a general awareness towards the range of predictive models available and the context in which they fit in, for example, segmentation versus risk stratification. This is particularly applicable in the initial stages.
• Providing a clear understanding of the particular model being implemented in terms of its scope, purpose, utilisation, and, most importantly, the interpretation of model predictions.
• Set expectations about limitations of the selected model(s) and the extent to which its outputs can be translated to beneficial actions that do not lead to unintended consequences. For example, accuracy measures such as sensitivity, specificity and appropriate use need to be clearly communicated so that actions can be tailored based on certainty of predictions.

Using key health IT capabilities

To care for well-stratified, high-risk populations, members of the care team require comprehensive knowledge of each person’s health status. Health and care organisations have the responsibility of using new and existing health IT capabilities that enable a holistic view of the people in these groups.

To move forward, we need to use the power of technology to move from a reactive care system where clinicians care for patients who seek it once they are ill, to a proactive prevention and management system where the growing availability of data, via the longitudinal record, coupled with the advent of cloud computing allows for continuous and proactive surveillance.

It is no longer when clinicians review a patient’s chart when they are in front of them, rather an always-available system is monitoring for rising risk and notifies the right member of care team including the patient themselves.

These IT capabilities enable care teams to effectively manage high-risk populations: 

 1. Longitudinal record

The longitudinal record serves as the single source of truth regarding a person’s health and care status. Without a comprehensive electronic record, a population health management programme lacks a common reference point for managing the health risks associated with each person in the high-risk population. A complete electronic picture of a person which aggregates clinical, financial and socio-economic data from multiple sources is necessary to fully understand the opportunities for care interventions. 

2. Data analytics and modelling

Data analytics and modelling are essential to understanding a given population, its health needs and its utilisation rates. Analytics tools can help healthcare organisations identify opportunities for improvement, examine condition-specific encounters or events and isolate underlying trends and variables that may affect a given population. Predictive models deliver proactive recommendations and decision support into the care team’s workflow and into the citizen’s care plan. This combined knowledge enables care teams to make more informed decisions at the person and population level.

3. Longitudinal plan and care management plan

A longitudinal plan is a comprehensive care plan that is shared across the entire care team. This plan is a collection of instructions, goals, activities, educational resources, objectives and measurements housed in a single view, enabling all care team members to work from a single, interactive plan. It should also leverage predictive models to suggest care activities that might be appropriate according to evidenced based care pathways.

4. Chronic condition care pathways and wellness plans

These plans enable the care team to focus on eliminating gaps in care and promoting prevention, respectively. Disease care pathways enable the care team to focus on people with one or more chronic conditions or other specific conditions, and deliver the services needed to guide them to an optimal health outcome. Wellness plans ensure that both people at the highest risk for costly, unanticipated health events and those who are currently healthy receive timely screenings and other forms of preventive care.

 5. Care management and coordination system

People in high-risk populations will receive services across health and care settings. The healthcare service/system ultimately responsible for the health of a high-risk population needs systems for care management and care coordination that use the aggregated data across care venues and enable communication between care professionals and providers regardless of the electronic record system in use.

6. Prevention and condition management system

People with complex conditions often receive care from multiple venues of care and must be able to navigate their care networks efficiently. Referral management systems can optimise the care coordination process and ensure that people can be referred efficiently. The care management team can use referral data to identify individuals’ treatment timelines.

In conclusion

Delivering the Quadruple Aim will rely on the above factors, delivered within a long-term strategic approach. Indeed, as the NHS begins to incorporate a fifth aim – health inequalities –and appreciate the role of the social determinants of health, never has it been more exciting to be able to look intelligently and holistically at individuals and the wider population.

Effective partnerships with shared governance and financial arrangements will be key to delivering truly integrated services and value-based care. With these in place, and with intelligence to help organisations to understand their populations, they can then identify specific at-risk citizen cohorts, and engage them to improve health outcomes, utilisation and wellbeing.

1. Hasan, H., Population health managers meet three patient types central to your success. Advisory Board, October 2013.

2. Improvement Academy, Healthy Ageing, 2018.

3. Lewis, G. et al., How health systems could avert ‘triple fail’ events that are harmful, are costly, and result in poor patient satisfaction,’ Health Affairs, April 2013.

4. Hibbard, J. et al., Adding a measure of patient self-management capability to risk assessment can improve prediction of high costs. Health Affairs, March 2016.

5. Denny, S, Thorlby, R., Steventon, A., Reducing emergency admissions, Unlocking the potential of people to better manage their long term conditions. The Health Foundation, August 2018.

6. Hibbard, J. et al., Improving Population Health Management Strategies: Identifying Patients Who Are More Likely to Be Users of Avoidable Costly Care and Those More Likely to Develop a New Chronic Disease, Health Services Research, September 2016.

7. Health Foundation PAM Study Briefing: Reducing emergency admissions: unlocking the potential of people to better manage their long-term conditions. August 2018.

UK_Fl_001_2019_v3_ComponentsNecessaryForManagingHigh-riskPopulationsWhitePaper/July2019
© Cerner Corporation. All rights reserved. All other trademarks referenced herein are the property of their respective owners.