Response to repeat echoendoscopic celiac plexus neurolysis in pancreatic cancer patients: A machine learning approach.

BACKGROUND

/

Objectives

Efficacy of repeat echoendoscopic celiac plexus neurolysis is still unclear. Aim of the study was to assess the efficacy of repeat celiac plexus neurolysis and to build an artificial neural network model able to predict pain response.

METHODS

Data regarding 156 patients treated with repeat celiac plexus neurolysis between 2004 and 2019 were reviewed. Artificial neural network and logistic regression models were built to predict pain response after treatment. Performance of the models was expressed in terms of accuracy, positive predictive value, and positive likelihood ratio.

RESULTS

Median age was 62 years (range 39-86) and most patients were male (66%) with pre-procedural visual analogue score 7. Fifty-one patients (32.6%) experienced treatment response, of which 6 (3.8%) complete pain suppression. Median duration of pain relief was 6 (2-8) weeks. Tumoral stage, interval from initial to repeat treatment, response to initial neurolysis, and tumor progression between the two treatments resulted as significant predictors of pain response. The performance of the artificial neural network in predicting treatment response was higher than regression model (area under the curve: 0.94, 0.89-0.97 versus 0.85, 0.78-0.89; p < 0.001). Positive predictive value and positive likelihood ratio resulted 90.3% and 19.35, respectively. Classification error rate was 5.7% with the artificial neural network compared to 14.7% of regression model (p < 0.001). These findings were confirmed through ten-fold cross validation.

CONCLUSIONS

Pain response following repeat neurolysis is generally less pronounced than after initial treatment. Artificial neural network may help to identify those subjects likely to benefit from repeat neurolysis.