Independent clustering of real cases

‘Independent clustering of real cases’ is a method to identify these ‘unconscious consensual patterns of BCCs’ without being polluted by sterile expert discussions. It is based on the ‘perceived similarity’ between cases and the natural ability of the human brain to pattern recognition. The more experts accept that two cases can be classified together; independently of the personal factors they take into account to analyze the two cases, the more these two cases are likely to belong to the same pattern of DTT‐BCCs. In addition, step by step, this is a way to define consensual patterns of BCC that best approach the real unconscious classification of experts.

Practically, each expert was presented a series of patient cases and was asked to group them into clusters, with the objective to put in a same cluster all the BCC cases which he/she considered close according to the criteria he/she considered as relevant. Each expert had to do his/her clustering alone, ‘blinded’ from the other experts. Clustering was unsupervised, leaving each expert to choose his/her own criteria for his/her own partition of the cases. There were only two restrictions: (i) the clustering had to be useful for the practical classification of these cases, leaving to each expert his/her own definition of ‘usefulness’; and (ii) the final number of clusters had to be between 2 and 10, in order to end up with an operational classification, which would not be the case if there were too many groups. A mathematical model was used to find the convergence between the different individual clustering patterns, to find a limited number of clusters best representing consensus. We previously tested this strategy on the cases from BOLT study ² with eight experts in three teams and confirmed the feasibility (presented at 14th EADO meeting Marseille 2015). According to the French law, Institutional Review Board approval was not required in the setting of this non‐interventional study.