[Use of neural networks in medicine: concerning dyspeptic pathology].

OBJECTIVES

Development and training of a neurone network that enables the patients who attend the clinic with symptoms of dyspepsia to be classified into two groups: those who very probably have peptic ulcer disease or gastro-oesophageal reflux (GOR) and those more likely to have functional or idiopathic dyspepsia. Results obtained with the neurone network and with other statistical classifiers were compared.

DESIGN

Retrospective study.

SETTING

Three urban primary care clinics. Participants. 81 patients with a diagnosis of dyspepsia, who underwent a digestive tract endoscopy and/ or oesophageal-gastro-duodenal meal, recorded in the clinical notes. Method. Face-to-face interview with a set questionnaire on the symptoms and risk factors of dyspepsia pathology. Data were analysed with determinist classifier, statistical classifier and neurone network based on a multi-layer perception.

RESULTS

The neurone network correctly classified 81% of patients, with negative predictor value of 90% and positive predictor value of 80%.

CONCLUSIONS

The neurone network provides very high accuracy rates in classifying patients on the basis of the presence or otherwise of determined symptoms. There was a tendency to distinguish negative diagnoses (functional or idiopathic dyspepsia) better than positive ones (peptic ulcer disease or GOR). Systematic use of neurone networks in primary care clinics would assist the doctor by increasing the accuracy of diagnostic and/or clinical decisions.