Skip to main content

Table 5 Segmentation outcome evaluations

From: A systematic review of the clinical application of data-driven population segmentation analysis

Number of segments (parsimony) No. of studies Examples
<=3 76 A population of PTSD patients was segmented based on symptoms: “High-Symptom”, “Dysphoric”, and “Threat” [91].
4–5 98 A group of children was divided into clusters of different patterns of sun protective behaviors: “Multiple protective behaviors”, “Clothing and shade”, “Pants only”, and “Low/inconsistent protective behaviors” [40].
6–9 55 An adult population was segmented by dietary patterns: “Traditional Irish”, “Continental”, “Unhealthy foods”, “Light-meal foods & low-fat milk”, “Healthy foods”, and “Wholemeal bread & dessert” [43].
> = 10 4 A female population was divided into 43 groups based on mammography status, access to care, health behaviors (e.g. smoking), health status etc. 44
Internal validation
 Yes 216 The optimal number of clusters was assessed using the Bayesian Information Criterion [92]
 No 0  
External validation
 Yes 138 Using risks of tonsillectomies and wheezing frequency to validate segmentation analysis based on symptoms of sleep disordered breathing [50]
 No 78  
Identifiability/Interpretability
 Yes 216 Segmentation analysis of dietary patterns derived clusters that are easily identified as “Alcohol cluster”, “Meat cluster”, “Healthy cluster”, and “Refined sugars cluster” [47]
 No 0  
Substantiality
 Yes 216 The smallest segment of a clustering analysis of asthma symptoms is composed of 15.8% of the population [93]
 No 0  
Stability
 Yes 10 A segmentation analysis of a asthma patient population with 10-year follow up showed the segments remain relatively stable 10 years apart (probability of cluster membership in the same asthma cluster at both times varied between 54 to 88%) [94]
 No 206  
Actionability/Accessibility
 Yes 216 A population is divided into segments with distinct sun protection behavioral patterns, for each of which future sun protection interventions tailored to specific subgroups can be designed and delivered to achieve meaningful behavioral changes [40]
 No 0