Background: No simple classification system has emerged for ‘advanced basal cell carcinomas’, and more generally for all difficult-to-treat BCCs (DTT-BCCs), due to the heterogeneity of situations, TNM inappropriateness to BCCs, and different approaches of different specialists. Objective: To generate an operational classification, using the unconscious ability of experts to simplify the great heterogeneity of the clinical situations into a few relevant groups, which drive their treatment decisions. Method: Non-supervised independent and blinded clustering of real clinical cases of DTT-BCCs was used. Fourteen international experts from different specialties independently partitioned 199 patient cases considered ‘difficult to treat’ into as many clusters they want (≤10), choosing their own criteria for partitioning. Convergences and divergences between the individual partitions were analyzed using the similarity matrix, K-mean approach, and average silhouette method. Results: There was a rather consensual clustering of cases, regardless of the specialty and nationality of the experts. Mathematical analysis showed that consensus between experts was best represented by a partition of DTT-BCCs into five clusters, easily recognized a posteriori as five clear-cut patterns of clinical situations. The concept of ‘locally advanced’ did not appear consistent between experts. Conclusion: Although convergence between experts was not granted, this experiment shows that clinicians dealing with BCCs all tend to work by a similar pattern recognition based on the overall analysis of the situation. This study thus provides the first consensual classification of DTT-BCCs. This experimental approach using mathematical analysis of independent and blinded clustering of cases by experts can probably be applied to many other situations in dermatology and oncology.

Position statement on classification of basal cell carcinomas. Part 1: unsupervised clustering of experts as a way to build an operational classification of advanced basal cell carcinoma based on pattern recognition

Fargnoli M. C.;
2021

Abstract

Background: No simple classification system has emerged for ‘advanced basal cell carcinomas’, and more generally for all difficult-to-treat BCCs (DTT-BCCs), due to the heterogeneity of situations, TNM inappropriateness to BCCs, and different approaches of different specialists. Objective: To generate an operational classification, using the unconscious ability of experts to simplify the great heterogeneity of the clinical situations into a few relevant groups, which drive their treatment decisions. Method: Non-supervised independent and blinded clustering of real clinical cases of DTT-BCCs was used. Fourteen international experts from different specialties independently partitioned 199 patient cases considered ‘difficult to treat’ into as many clusters they want (≤10), choosing their own criteria for partitioning. Convergences and divergences between the individual partitions were analyzed using the similarity matrix, K-mean approach, and average silhouette method. Results: There was a rather consensual clustering of cases, regardless of the specialty and nationality of the experts. Mathematical analysis showed that consensus between experts was best represented by a partition of DTT-BCCs into five clusters, easily recognized a posteriori as five clear-cut patterns of clinical situations. The concept of ‘locally advanced’ did not appear consistent between experts. Conclusion: Although convergence between experts was not granted, this experiment shows that clinicians dealing with BCCs all tend to work by a similar pattern recognition based on the overall analysis of the situation. This study thus provides the first consensual classification of DTT-BCCs. This experimental approach using mathematical analysis of independent and blinded clustering of cases by experts can probably be applied to many other situations in dermatology and oncology.
File in questo prodotto:
File Dimensione Formato  
Grob JEADV 2021_part 1_Position statement on classification of BCC_unsupervised clustering of experts.pdf

accesso aperto

Tipologia: Documento in Post-print
Licenza: Creative commons
Dimensione 2.68 MB
Formato Adobe PDF
2.68 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11697/182748
Citazioni
  • ???jsp.display-item.citation.pmc??? 1
  • Scopus 3
  • ???jsp.display-item.citation.isi??? 2
social impact