Cluster analysis to phenotype sarcoidosis using British Thoracic Registry data

Introduction & Aims: Clinical phenotyping in sarcoidosis can help define subpopulations with similar characteristics. Importantly, phenotyping may differentiate disease course. Using the BTS registry data we aim to organize patients into subgroups based on similarities to identify potential unbiased clinical phenotypes and compare radiological descriptors as additional variables.

Methods: A Two-step cluster analysis was performed comparing Scadding Stage (SS) and CT descriptors (CTD) as additional variables. Missing data was handled using median or modal imputation for core variables. Visual inspection of dendrograms from Hierarchical analysis (Ward’s method) suggested at least 3 Clusters.

Results: 774 individual records were available. Breathlessness (42%), cough (38%) and fatigue (24%) were commonly reported. 2 Step Cluster analysis identified 4 clusters for SS and CTD. Total missing data accounted for 17.2%, 64% had CXR stage recorded 75% had CT descriptors. 14 variables were included in final analysis. Predictor importance for each model is presented Figure 1. SS had much stronger influence on cluster designation than CTD. CT descriptors afforded greater influence from other variables which is possibly more reflective of the heterogeneity of this disease.

Conclusion: This cluster analysis indicated symptoms were most predictive which may lead to a treatable trait management strategy. Interpretation is limited by missing data.