Follow Us

Part 1: Whole Population Segmentation Models

This report briefly outlines the different approaches to population segmentation in healthcare, including their advantages and disadvantages when used in the context of outcomes based commissioning.1

Back to PHM and Segmentation

Why Segment?

Health and care systems have historically categorised populations by the services utilised at a point in time. For example, people accessing care from their GP are delivered a different ‘bundle’ of services to those attending A&E, or receiving acute inpatient care, or rehabilitation from a chronic condition. This approach has several adverse effects:

  • Care may be inconsistent, duplicative or incomplete
  • Providers are only reimbursed for services delivered at specific locations at a specific point in time, even if there are better, more sustainable ways to achieve the same or better results
  • There is little incentive for integration that may result in savings across entire care pathways
  • People may not be seen by providers with the appropriate skill level, or in settings that have appropriate ancillary services. This ‘over-generalisation’ may result in unnecessary appointments, delayed care and inconvenience to patients and service users
  • Care may also be over-specialised, with services delivered in acute settings when they would more appropriately be delivered in the community

Segmentation aims to categorise the population according to health status, health care needs and priorities. This approach recognises that groups of people share characteristics that influence the way they interact with health and care services. To optimise health outcomes, service user experience, efficiency and care costs, care delivery systems should respond to the needs of different population segments in different ways.

Segmentation has been widely used in other industries, particularly those that are customer-facing eg. hospitality or the car industry. This enables the development of offerings that meet the needs of a specific customer segment perfectly, rather than meeting a minimum standard for all segments. The benefits of applying a similar approach to health care are considerable:

  • People’s needs are truly put in the centre of the system. Care can only be tailored to a segment after truly understanding the health needs and priorities of that population
  • Health and care outcomes are optimised by providing the right services, from the right provider, at the right time. This improves efficiency, overall experience, and provides timely and appropriate screening, diagnosis and treatment, where necessary
  • Services have the potential to be more integrated, if collaboration, information sharing, and pooled funding can be realised across all services in a segment
  • Providers can develop the specialist expertise that allows them to respond optimally to the needs and preferences of a specific population segment

Segmentation does not negate the need to tailor care for individual patients and service users. Variation will exist (eg. in risk factors, social determinants of health) even if most people have majority of their needs well met. Similarly, there may be value in stratifying patients by need, complexity, and acuity, within specific, clearly defined, segments. Despite this, individuals within a subgroup often share the same basic health and care requirements that care can broadly be organised around.

Ways to Segment

There are many distinct approaches to population segmentation. Approaches with potential application for outcomes measurement generally need to consider three main dimensions: the purpose segmentation is serving, the method or approach utilised, and a description of the variables which are used to group people. To evaluate different segmentation approaches, the appropriate combination of those three dimensions need to be present, for the purposes of outcomes based commissioning.

a) Why: intended purpose for segmentation?
Population segmentation can be undertaken for a range of different purposes. Existing evidence describes three distinct levels of integration from a care organisation perspective: ‘macro-‘, ‘meso-‘, and ‘micro-level’ integration.2,3 These serve as a framework to describe the general purposes of segmentation. These may lend themselves differently to specific purposes, such as improving direct patient care, outcomes measurement, establishing a financial envelope and/or capitated budget, health and social marketing, and so-on.

Care integration has recently been at the top of the agenda for many health and care systems globally. In reviewing evidence on integration, segmentation experts describe a distinction at three levels of care integration: macro-, meso- and micro-. The table 1 below describes these broad concepts and their application for the main specific purposes of segmentation in health systems.

b) What: method used to identify segments?
Different methods can be used when identifying cohorts in any segmentation exercise. Their application to segmentation in outcomes-based commissioning is evaluated here, including the potential for confusion when using risk stratification for segmentation purposes.

Vuik et al (2016) describe how segmentation can support the three different population strategies described above; ‘macro-‘, ‘meso-‘, and ‘micro-level’ integration, assisting with the identification of target populations, and providing helpful insights. While the analysis has emerged from the perspective of care integration design, it provides a useful framework to describe the different segmentation methods that could be used in the context of outcomes based commissioning.

Purpose vs Method
It is important to highlight the main difference between purpose and method of segmentation, as these two dimensions can be easily confused. The method selected for segmentation is generally based on its purpose i.e. if macro-level integration is required, a whole population method is usually the most appropriate. However, it is possible, for instance, to apply a whole population method for the purposes of identifying subpopulations. Delaware’s State Health Care Innovation Plan is an example of that approach, where the whole population is segmented into four age groups and 5 disease groups. Subsequently priority subgroups of interest were selected based on cost and the potential for intervention.2

Regardless of the segmentation method selected, or the level of care integration intended, identification of meaningful groups who share similar needs is required if relevant outcomes are to be defined and measured. Whilst some outcome measures can be universally applicable to nearly all people (e.g. quality of life), most people will have fundamentally different health and life circumstances, needs, and care expectations. Therefore, many outcomes which really matter to people, or which are unique to a specific population need to be described around groups of people with broadly similar needs.

Risk stratification tools, or stratification approaches generally, can provide useful insights while still allowing for meaningful outcomes measurement when applied within already defined segments. However, if applied to inadequately defined groups, risk stratification will generally only allow the identification of people with similar magnitude of needs, rather than similar types of needs. This therefore makes risk stratification in isolation largely ineffective for the purposes of outcomes measurement. Due to the common ambiguity between the definitions and purposes of risk stratification compared with segmentation, this paper dedicates a section to addressing some common misconceptions for the purposes of outcomes measurement. When selecting a cohort of interest, defining the precise inclusion/exclusion criteria for this segment is key in order for outcomes to be appropriately defined and measured i.e. what is meant by ‘Older People with Frailty’? In the next section, the different characteristics that can be used when segmenting populations are considered.

c) How: variables used to define segments?
There are a number of characteristics which can used to identify particular (potentially high-risk) subgroups, such as age or life stage, lifestyle, income, deprivation, condition, and so-on. These are briefly outlined below.

The set of characteristics or variables utilised by different segmentation models does not always take into account the basic health requirements shared within a cohort. Without these, they will have limited use when applied to outcomes definition and measurement.

The North West London (NWL) Whole Systems Integrated Care (WSIC) Programme analysed four broad categories of “primary organising characteristics” which can be used to group populations. Each have their pros and cons: type of condition and age, social and demographic factors, utilisation risk (risk stratification) and behaviour.13 These are useful conceptual groupings, and this report has reviewed their analysis to inform evaluation of the different segmentation models. However, when evaluating segmentation approaches specifically for outcomes based commissioning purposes, this report proposes a slightly different grouping of defining characteristics. These are largely aligned with the World Health Organisation
(WHO) determinants of health14.

  • Person-Centred: variables/characteristics that are inherent to the individual, and fairly constant at any point in time. They can be:
    • Health-specific i.e. condition and age. Although age could be classified as a person specific characteristic (see below), it plays a significant role in development of health conditions and health-specific requirements, especially in the older population. The term condition can be
      interpreted as either a specific health condition (i.e. presence of diabetes) or a health status (i.e. presence or absence of a long term condition, or healthy).
    • Person-specific i.e. gender, ethnicity, personality traits.
  • Social and Economic Related: these are factors related to the individual’s environment, and are
    known to have significant effects on health outcomes15

    • Income, education, social isolation, employment, housing.
  • Behavioural: behaviour and lifestyle are considered major determinants of health.16,17 In this context, the term behaviour is defined by personal lifestyle such as mobility and habits i.e. smoking and drinking. Although behavioural characteristics could be classified as person-centred, they are fluid and subject to continuous change, therefore sitting outside of our definition of person-centred. Behaviour towards the health system i.e. patient activation, is considered a system-focused variable in this context, while personality traits would be qualified as a person-specific characteristic
  • System-Focused: these are characteristics that define how a person utilises/interacts with the health system.
    • Activity, number of hospital admissions, costs, access to services, patient activation, behaviour towards one’s health.

In general, segmentation models that utilise health-specific, person-centred variables as determinants for groupings are a better fit for health and care outcomes definition and measurement. Take a common variable being used to define ‘patients’ – the presence of an unspecified long-term condition (LTC). A person who lives with diabetes might be concerned about avoiding complications such as a heart attack, strokes and amputations, while a person living with rheumatoid arthritis might be concerned about things like mobility, pain management, and potential disease progression. Although common outcomes can be identified for both groups – i.e. avoidance of complications – they also have fundamentally different outcomes expectations, which can only be identified once the defining variable is further refined. For example, utilising specific ‘condition type’ – e.g. people with coronary heart disease – as a common characteristic to identify a cohort is generally helpful from a clinical outcomes measurement perspective.

However, ambiguity can arise if the defined condition is too generic e.g. people who have stomach pain. Meanwhile social and economic variables may be more helpful from a social and personal outcomes measurement perspective.

Table 3 summarises the advantages and disadvantages of using different variables to define a health segment for the purposes of outcomes based commissioning:


Segmentation vs. Stratification

The terms segmentation and stratification are often used interchangeably in healthcare. At times, stratification is described as a form of segmentation.2 NHS England and the Department of Health agree on the following definition:

“Segmentation is grouping the local population by what kind of care they need as well as how often they might need it. Risk stratification means understanding who, within each segment, has the greatest risk of having a significant health event or is at most risk of deterioration”.18

As set out, outcomes can only be meaningfully defined and accurately measured when applied to groups of people with similar needs, as opposed to groups of people with similar costs. Within the context of outcomes based commissioning, the term ‘similar needs’ is defined by homogeneous health status, and/or healthcare needs. This includes the relevant clinical and preventative care needs, around which services are organised. Integration of services around people with similar needs enables outcomes, costs and processes to be measured for homogeneous groups of people. Where segmentation seeks to categorise populations according to health status, needs and priorities, alignment with outcomes definitions and measurement is possible – depending on the variable used for grouping, as discussed in the next section.

Limitations of Risk Stratification Models for Outcomes Based Commissioning
Risk stratification seeks to identify people who are most at risk of deterioration, or at risk of a significant care event, therefore generally failing to provide clear criteria for defining common desired outcomes – or end results of care – within each risk strata. Risk stratification without segmentation tends to group people by care usage, or cost to the system, which is not the same as a homogeneous set of needs. For example, young people with learning disabilities could be in the same group as older people with frailty, in terms of risk of care usage, cost, or care event, but often these two groups don’t share a common set of needs, or similar outcomes. Defining cohorts using risk stratification approaches generally allows an understanding of people’s needs from the system as a whole (and their associated costs), as opposed to the identification of their needs as individuals, or as groups of individuals.

Other important considerations are:

  • Cohorts identified through risk stratification alone may become out of date too quickly for the purposes of outcomes based commissioning; a 2013 report by Kent and Medway Public Health Observatory suggests that approximately 30% of patients move out of the very complex risk band (0.5% of the population) within one month; 50% after five months and 80% after one year.19
  • Many risk stratification methodologies rely on using diagnosis codes or Read codes to capture information about the risk factors and probability of future unplanned admission. This requires well-coded, linked datasets, with appropriate information governance in place to extract the most value.

Although stratification has limited use for the purposes of segmentation in outcomes-based commissioning, it can be usefully applied to defined segments. It can enable a better understanding of subpopulations within defined segments, as well as providing insights into how outcomes can be improved, once defined.

Brief Review of Whole Population Segmentation Models

The Bridges to Health Model (and accountable care approaches generally), focuses on a whole-population (and ‘macro-level’) approach to care integration, and other segmentation models that follow similar approaches are considered here.20

There are numerous examples of areas in the UK that have developed outcomes frameworks at a ‘meso-level’ of integration, focusing on specific cohorts of interest for which outcomes based contracts have been established. Sheffield and Bedfordshire have established Musculoskeletal Care outcomes based contracts,21, 22 while several North Central London CCGs have developed programmes aimed at populations like: People with Diabetes and Older People Living with Frailty.6, 7 The Care Pathway Framework23 developed by Health Dialog is an example of mixed population segmentation and risk stratification, where nine stages of illness are described: the first three stages (‘well’, ‘well at risk’, and ‘pre-diagnostic’) apply to the whole population. The remaining stages (‘condition onset’, ‘early progressive’, ‘late progressive’, ‘critical’, ‘sentinel event’ and ‘recovery’) are applicable to specific condition pathways (ie. cancer, COPD, diabetes, etc.). As they vary significantly from whole population approaches, they have not been reviewed for the purposes of this report.

As far as risk stratification tools are concerned, those that aim to stratify whole populations, and are more closely-related to segmented, whole population approaches have been reviewed. There are countless stratification tools that could be described/reviewed, such as the PARR++ algorithms12 and the Combined Predictive Model.12 However, as those have not been explicitly intended to stratify whole populations specifically for the purposes of outcomes based approaches, they have been excluded from further analysis here.

Finally, material that has been made available regarding whole population and vanguard, and/or New Models of Care programmes, has been reviewed and considered in the background research for this report.

Although segmentation and stratification approaches used in many areas have been reviewed and analysed, only areas that have very clear, well-developed approaches to whole-population segmentation, with appropriate definition of segments have been described further in this report.


In summary, an ideal whole population segmentation model for outcomes based commissioning should meet the following criteria:

  • Each segment should be broadly homogeneous, with common health prospects and priorities that can be addressed through careful system planning. Therefore, the variables utilised to describe segments should be key determinants of health needs.
  • Each segment should be sufficiently distinct, with unique health and health service delivery needs.
  • The set of population segments must include every person, acknowledging that individuals will move between segments, as their health needs change.
  • The number of segments must be limited in order to deliver accessible integrated services for the defined populations. Within segments, populations may be further sub segmented or stratified in response to specialised health care needs.
  • At any one time, everyone best matches one distinct segment, but over time people move through segments. Therefore, the variables utilise to describe the segments should be fairly constant.
  • Each segment’s definitions should be sufficiently precise to allow a baseline population number to be determined, assuming access to the appropriate dataset.

The following two tables briefly describe the main segmentation and stratification models that are most relevant to this analysis.





There are a number of terms used in this paper – definitions of these are set out below for reference:


  1. Bersani J, Dunbar-Rees R, McGough R. Contracting for outcomes. Outcomes Based Healthcare and Capsticks. Oct 2014.
  2. Vuik S, Mayer , Darzi, A. Patient segmentation analysis offers significant benefits for integrated care and support. Health Affairs 35, no.5, 2016;769-775, doi: 10.1377/hlthaff.2015.1311.
  3. The King’s Fund. Integrated care. 2011. (accessed 23/08/2016).
  4. North West London Whole System Integrated Care. (accessed 23/08/2016).
  5. McKinsey & Company. Population Segmentation for Integrated Care. Health Service Journal, Nov 2014. (accessed 23/08/2016).
  6. Price S, Lissauer R. Value based commissioning in north central london. (accessed 23/08/2016).
  7. Islington Clinical Commissioning Group. (accessed on 23/08/2016).
  8. Porter M, Lee T. The strategy that will fix healthcare. Harvard Business Review, Oct 2013; 91.10:50-70
  9. Shelton P, Sager M, Schraeder C. The community assessment risk screen (CARS): identifying elderly persons at risk for hospitalization or emergency department visit. The American Journal of Managed Care, Aug 2000; 6(8):925-33.
  10. British Columbia Ministry of Health. The health system matrix 6.1:understanding the health care needs of the british columbia population through population segmentation. Feb 2015.
  11. Clegg A, Young J, Iliffe S, et al. Frailty in elderly people. Lancet 2013; 381: 752–62.
  12. King’s Fund, Health Dialog, University of New York. Combined predictive model. December 2006. http://www. (accessed on 23/08/2016).
  13. Whole Systems Integrated Care. What are the different ways we can group? (accessed on 23/08/2016).
  14. World Health Organisation. The determinants of health. (accessed on 23/08/2016).
  15. Wilkinson M, Marmot M. Social determinants of health: the solid facts. World Health Organisation. 2003; 2nd Edition.
  16. Whitehead M, Dahlgren G, Gilson L. Developing the policy response to inequities in Health: a global perspective. in: Challenging inequities in health care: from ethics to action. New York: Oxford University Press; 2001:309-322.
  17. The King’s Fund. Broader Determinants of Health. broader-determinants-health (access on 23/08/2016).
  18. NHS England. ‘How to guide’: the BCF technical toolkit, section one: population segmentation, risk stratifcation and information governance. 2014. content/uploads/2014/09/1-seg-strat.pdf (accessed on 23/08/2016).
  19. NHS England. Using case fnding and risk stratifcation: a key service component for personalised care and support planning. Jan 2015. pdf (accessed on 23/08/2016).
  20. Lynn J, Straube BM, Bell KM, et al. Using population segmentation to provide better health care for all: the “Bridges to Health” model. The Milbank Quarterly. 2007;85(2):185-208. doi:10.1111/j.1468-0009.2007.00483.x
  21. Circle Health. (accessed on 23/08/2016).
  22. Moving Together. (accessed on 23/08/2016).
  23. Health Dialog. A clinical approach to population stratifcation analytics: dispelling the myths of traditional risk segmentation models used in population health management programs. Jun 2015. default/fles/whitepapers/A-Clinical-Approach-to-Population-Stratifcation-Analytics.pdf (accessed on 23/08/2016).
  24. Dr Foster Quality Investigator. (accessed on 23/08/2016).
  25. Choose Health Delaware. Delaware’s State Health Care Innovation Plan. Dec 2013. cmmi/fles/choosehealthplan.pdf (accessed on 23/08/2016)
  26. 3M Health Information Systems. 3M™ clinical risk groups: measuring risk, managing care. Jul 2016. http:// paper.pdf (accessed on 23/08/2016).
  27. Abrams M, Kleiman R, Schneider, et al. Overview of segmentation of high-need, high-cost patient population. National Academy of Medicine, Jan 2016. pdf (accessed on 23/08/2016).
  28. Department of Health. The healthy foundations lifestages segmentation. HIPD Social Marketing and Health Related Behaviour. Jul 2011. (accessed on 23/08/2016).