Population Segmentation for Population Health Management

Questions of scale and scope

There is no simple numerical answer to the question of how large a population segment should be. Clearly, a cohort containing only one or two individuals might feel ill-defined and inappropriate as the target for a specific innovation or care programme; equally, a very large segment, covering most of the overall population, could be imprecise for many purposes, and not really actionable, while including significant variation.

Examining approaches to integrated care in 2011, Professor Chris Ham and Natasha Curry produced a King’s Fund report which used a three-tier model:

  • The macro-level at which providers, either together or with commissioners, deliver integrated care across the full spectrum of services to the populations they serve: examples include Kaiser Permanente, the Veterans Health Administration and integrated medical groups in the United States.
  • The meso-level at which providers, either together or with commissioners, deliver integrated care for a particular care group of people with the same disease or conditions: examples include care for older people, mental health, disease management programmes and managed clinical networks.
  • The micro-level at which providers, either together or with commissioners, deliver integrated care for individual service users and their carers through care co-ordination, care planning and other approaches. (See page 3 of the King’s Fund report referenced above.)

Other commentators have used a similar model to structure population segmentation approaches.

Thus, typically, a macro-level approach will segment the whole population served into a few broad groups. No individuals are missed — every person in the community is captured in exactly one segment. The National Association of Primary Care (NAPC) segmentation model is a good example of this, and indeed the Johns Hopkins HealthCare ACG® System also provides a macro-level segmentation.

The NAPC model, principally the work of Dr Steve Laitner and Dr Mark Davies, breaks down the population into nine core areas as shown below.

NAPC 9-box model

To illuminate these, other groups can be layered onto the nine-box model to illustrate its use, and a flexible risk assessment applied.

Further developed example of NAPC segmentation model

Quite large groups arising in such a model could still be addressed as a target cohort for a care intervention. For example, a preventive ‘active living’ health education campaign might target a large population segment of generally healthy adults and their families with advice and guidance. However, a simple segment of the ‘healthy’, defined in terms of clinical conditions and utilisation, can hide significant variation, and should be qualified with a deeper understanding of environmental and social factors to assist with better programme targeting, and the use of more nuanced communications methodologies.

Generally, whole population models are easy to understand, and it is straightforward to provide example patients who fall into such broad classifications. As a class, however, they may not necessarily provide the mechanisms for more targeted intervention which benefit smaller groups of patients with specific needs. Macro-level whole population models can be quite detailed, and break the population down into relatively small segments.

In the Sollis version and example analysis supporting the NAPC model, the risk models have been further developed, classifying patients as higher or lower risk within a segment and reflecting the characteristics of each segment, and a more granular age band has been applied. Each segment holds a patient count and an indicator of care costs.

Alternative Sollis example of NAPC segmentation model

A further more detailed analysis is available showing the breakdown of costs and activity per provider for each patient segment; this provides a rich, if high-level, management view of the care being delivered to each patient group.

Sollis Extension of NAPC whole pop seg model

Meso-level approaches do not attempt to segment the whole population, but instead single out specific cohorts with needs that may be better addressed by tailored care programmes. Such cohorts may be sub-segments of the macro–defined groups; in the ‘active living’ campaign example above, a harder-to-reach group defined by deprivation and other social factors might be the subject of a more targeted intervention.

Alternatively, a meso-level segment might cross the boundaries of several population segments defined at a macro-level.

Meso level segmentation crossing boundaries

A meso-level segment might also be defined and supported successfully in an area where wider macro-level techniques were not being used. The Slough CCG programme is an example of this.

Historically, meso-level segments have often been identified for interventions which will have more immediate impact; the Slough case study began to show appreciable benefits within a few months of its initiation. By comparison, macro-level segmentation is frequently used only as an introductory management visualisation, leading on to far more nuanced and analytically-based segmentation analyses, developing a richer, more detailed understanding of the needs of specific populations. This typically results in further meso-level segmentation to identify patient cohorts for specific initiatives.

Micro level approaches target individuals with specific needs or risks. An example might be individuals with a higher risk of contracting specific conditions, or who are identified as approaching the end of their lives, for whom specific community-based palliative care services may be appropriate.

A key characteristic distinguishing macro- from other forms of population segmentation is that of coverage. Macro-level segmentation covers the whole population, and each individual resides in exactly one segment. By comparison, while arguably in the case of meso- and micro-level segmentation approaches, a group of ‘patients not in this segment’ will exist by implication, and technically result in a similar ‘coverage of the whole,’ this negatively-defined segment would usually fail the criterion of utility which we established earlier.

It can be argued that, when it comes to questions of segmentation, there is little distinction between the meso and micro-levels. Whatever the scale of a segment, it will be used to identify patients who may be candidates for care intervention. And those care interventions and programmes will require planning and resourcing. A micro-level segment may thus simply be seen as a specific type of very small meso-level segment. Alternatively, some care interventions, such as social prescribing, take care to tailor the support given to match an individual patient’s personal wants and needs. This is often referred to as care personalisation rather than a specific very narrow micro-level segmentation.

Usually, however, one or more meso-level segments are defined from a range of characteristics, which identify patients who are ‘candidates for social prescribing services’ (which include a range of more granular initiatives). In the majority of the discussion which follows, we therefore focus on macro and meso segmentation approaches; recognising in so doing that the latter in most cases also subsumes micro-level approaches.

Further, to emphasise the difference between the approaches, we usually refer to macro-level approaches as ‘whole population segmentation’, and meso-/micro- level approaches as ‘targeted population segmentation’.Finally, if the purpose of segmentation is to identify patients for particular care initiatives, whether macro-, meso- or micro-level, the size of the population segment and the resources or capacity of the associated service offered need to be carefully aligned. Therefore, the process of segmentation may be iterative, to ensure a group of an appropriate size can be identified that matches the scale of service to be offered, and that the resources assigned to the programme match the anticipated benefits.

Population Segmentation White Paper

This is an extract from our new extensive introductory guide, Population Segmentation — An introduction to methods for identifying and supporting multiple diverse patient groups. Enter your email address and we’ll send you a link to download the complete PDF.

Tick this box if you would like us to contact you when we publish new white papers or case studies in future.

Your privacy is important, therefore we will never pass on your details to anyone else. Please review our privacy policy before subscribing.

Leave a Comment