A neurofuzzy model to identify insignificant prostate cancer: development and validation using the ProtecT randomised controlled trial


Year:

Session type:

James Catto1, Maysam Abbod5, Derek Linkens1, Derek Rosario1, Jenny Donovan4, David Neal3, Freddie Hamdy2

1University of Sheffield, UK, 2University of Oxford, UK, 3University of Cambridge, UK, 4University of Bristol, UK, 5Brunel University, London, UK

Abstract

Background
The use of serum PSA has dramatically increased the incidence of prostate cancer and lead to a disease-migration towards low-risk tumours. Currently around 50% of those diagnosed are low-risk with a serum PSA<10ng/ul, Stage

Method
From the radical prostatectomy arm of the ProtecT study we studied cases whose pre-operative parameters suggested a potentially insignificant tumour: Inclusion criteria were Serum PSA <20ng/ml, Gleason grade≤3, Stage≤T2, <50% positive biopsy cores, <20mm total cancer in all cores and more than 40mm benign tissue. Insignificant cancer was defined in prostatectomy specimens as: Gleason score≤6, Volume≤0.5cc and Stage≤pT2. We divided the cases into 66% for NFM development and 33% for validation.

Results
In total, 235 patients were studied and insignificant cancer was present in 54 (21%). From 11 pre-operative parameters we developed various NFMs and tested them in a blinded manner using the validation cohort. The most accurate and parsimonious NFM used 7 parameters (various biopsy features, prostate volume and PSA) and in the validation cohort had a Specificity=0.9, Accuracy=0.8 and Concordance index=0.7. This NFM had a similar concordance (0.7) and a better specificity (0.2) to the best predictive nomogram in these cases.

Conclusion
Our NFM can identify significant prostate cancer with a 90% specificity. The NFM produces distinct and complementary predictions to a nomogram. In our population the use of this NFM would have avoided surgery in 22 men with insignificant cancers.

Share this abstractTweet about this on TwitterPrint this pageShare on FacebookEmail this to someoneShare on LinkedIn