Cluster analysis to determine headache types.

Cluster analysis was used to separate 726 headache patients into clusters of patients with similar symptoms. This was done to answer two questions: what "naturally occurring' groups of patients can be found? And how do these groups correspond to traditional headache types? When only two clusters were required, the best two clusters were tension and migraine-like. However, eight clusters could also be distinguished, and the migraine group then became very small. The clusters were tested for clinical interpretability by having 12 physicians name and prescribe treatment for the clusters. The suggested treatment was similar to what patients had actually received in about 2/3 of the cases but was rather different for the remainder. Further, the outcomes associated with different treatments appeared to vary by cluster. This suggests that the current method for classifying and treating headaches may not be optimal.