Rationale:
Retailers use segmentation methods to identify groups of distinct and homogeneous customers, tailoring their products and services to these groups. In healthcare, patient-centred integrated care also aims to design care models around the patient, but the use of data to support this is limited. Data-driven segmentation could be used to identify patients with similar care needs, who might benefit from integrated care initiatives.

Aim:
To define the potential role of data-driven population segmentation in designing patient-centred integrated care.

Methods:
Existing applications of segmentation in healthcare were explored through literature, case study and systematic reviews. Segmentation analyses were performed on a 300,000-patient database, containing primary and secondary care data. Methods included k-means cluster analysis, regression analysis, artificial neural networks and decision trees, in addition to descriptive and statistical analyses.

Results:
Several integrated care programmes apply segmentation, but their use of data-driven methods is limited. Nevertheless, there exist many healthcare studies that used cluster analysis to segment patient populations. Segmenting a whole population resulted in eight distinct care user segments, providing an evidence base for population health. Segmenting the subpopulation of patients with ACSC hospitalisations identified four different care utilisation patterns, each requiring different preventive interventions. Risk stratification is a segmentation method in itself, but descriptive segmentation can help to identify different groups within the high-risk population. Where no patient-level data is available, an a priori rule can be used to identify high-needs patients.

Conclusion:
Data-driven segmentation can play an important role in designing patient-centred integrated care. It can be used to describe different patient groups within a population, a subpopulation, or a high-risk population, and design integrated care interventions around the needs of each segment. It can also be used to predict which patients are in the target group for integrated care initiatives.