J. R. T. Monson
Artificial neural networks to predict presence of significant pathology in patients presenting to routine colorectal clinics
Monson, J. R. T.; Maslekar, S.; Gardiner, A. B.; Monson, J. R. T.; Duthie, G. S.
A. B. Gardiner
J. R. T. Monson
G. S. Duthie
Aim Artificial neural networks (ANNs) are computer programs used to identify complex relations within data. Routine predictions of presence of colorectal pathology based on population statistics have little meaning for individual patient. This results in large number of unnecessary lower gastrointestinal endoscopies (LGEs - colonoscopies and flexible sigmoidoscopies). We aimed to develop a neural network algorithm that can accurately predict presence of significant pathology in patients attending routine outpatient clinics for gastrointestinal symptoms. Method Ethics approval was obtained and the study was monitored according to International Committee on Harmonisation -Good Clinical Practice (ICH-GCP) standards. Three-hundred patients undergoing LGE prospectively completed a specifically developed questionnaire, which included 40 variables based on clinical symptoms, signs, past-and family history. Complete data sets of 100 patients were used to train the ANN; the remaining data was used for internal validation. The primary output used was positive finding on LGE, including polyps, cancer, diverticular disease or colitis. For external validation, the ANN was applied to data from 50 patients in primary care and also compared with the predictions of four clinicians. Results Clear correlation between actual data value and ANN predictions were found (r = 0.931; P = 0.0001). The predictive accuracy of ANN was 95% in training group and 90% (95% CI 84-96) in the internal validation set and this was significantly higher than the clinical accuracy (75%). ANN also showed high accuracy in the external validation group (89%). Conclusion Artificial neural networks offer the possibility of personal prediction of outcome for individual patients presenting in clinics with colorectal symptoms, making it possible to make more appropriate requests for lower gastrointestinal endoscopy.
Maslekar, S., Gardiner, A. B., Monson, J. R. T., & Duthie, G. S. (2010). Artificial neural networks to predict presence of significant pathology in patients presenting to routine colorectal clinics. Colorectal Disease, 12(12), 1254-1259. https://doi.org/10.1111/j.1463-1318.2009.02005.x
|Journal Article Type||Article|
|Acceptance Date||May 16, 2009|
|Online Publication Date||Nov 11, 2010|
|Deposit Date||Nov 13, 2014|
|Peer Reviewed||Peer Reviewed|
|Keywords||Artificial neural networks; Lower gastrointestinal endoscopy; Flexible sigmoidoscopy; Colonoscopy|
This file is under embargo due to copyright reasons.
You might also like
Quality assurance in colonoscopy: role of endomucosal clips
Quality assurance in flexible sigmoidoscopy: Medical and nonmedical endoscopists
Anterior anal sphincter repair for fecal incontinence: Good longterm results are possible
Randomized clinical trial of Entonox ® versus midazolam-fentanyl sedation for colonoscopy