Differential diagnosis of dyspepsia

Download the Paper

Description: Assesses arguments for and against 5 different diagnoses for patients presenting with dyspepsia

Publet Introduction:

A very early demonstration of the potential value of computers in diagnosis was described by Mr. Tim de Dombal and colleagues at St. James Infirmary in Leeds UK (de Dombal et al, 1972). In one study of patients with upper gastro-intestinal symptoms de Dombal and his colleagues (Horrocks et al, 1995) collected detailed histories of several hundred patients presenting to a surgical unit in a general hospital, and once a definitive clinical diagnosis had been made they recorded the frequency of particular symptoms and signs for 5 diagnosis categories (duodenal ulcer, gastric ulcer, cholecystitis, gastric cancer and "functional" conditions). For subsequent patients these frequencies were treated as estimators of conditional probabilities in order to compute the posterior probabilty of each of the 5 conditions. Results compared well with the diagnostic accuracy of clinicians making the diagnosis for a series of new patients.

Fox et al (1980) reimplemented the Bayesian diagnosis system with data supplied by Mr. de Dombal, together with a rule-based diagnosis system which did not make use of the quantitative probability data. The two systems had very similar diagnostic accuracy overall, and the error matrices of the two diagnosis methods were also very similar. The rule-based system has been reimplemented here as a demonstration of one way in which OpenClinical tools can be used to develop such differential diagnosis applications. However the original Leeds data on which the system is based are so old that this should be regarded as only a technical demonstration.


  Information
Guideline objectives

Demonstrate use of argumentation in a simple differential diagnosis of patients with upper GI symptoms

Target setting OpenClinical training
Target users OpenClinical members
Overview
Provenance 7.No provenance has been assigned (default value)
Management
  • Author: J Fox, Oxford University
  • Release date:
  • Status: Draft - Under Development
  • History: Original data compiled for bayesian diagnosis by de Dombal, and later compared with a rule based method by Fox (see sources)
Safety case OpenClinical level 4 (high risk). The data on which the system is based are extremely old and this should only be used for technical training.
Sources

Horrocks JC, de Dombal FT. "Computer-aided diagnosis of dyspepsia". Am J Dig Dis. 1975 May;20(5):397-406.

Experience with computer-aided diagnosis of "dyspepsia" in a consecutive prospective series of 212 patients coming to surgery is described. Analysis is concentrated upon 122 patients who presented to an outpatient clinic de novo for diagnosis. During their first (outpatient) hospital contact, a firm diagnosis was made in just over half of these patients (though where made, it was usually correct). After full investigation, the diagnostic accuracy (prior to operation) was 92.6%. Using data elicited solely from the house surgeon's interview at the time of admission, the computer's overall diagnostic accuracy was 87.7%. The cost of each new computer diagnosis was around 25 new pence ($0.60). and the time taken was about 5 minutes. In a further small series designed to discriminate between organic and functional dyspepsia, the computer correctly assigned all but 1 of 23 patients with organic disease to the correct disease category. However, almost half of 33 patients with x-ray negative dyspepsia were predicted by the computer to have organic lesions. Time alone will tell whether the computer is a better early predictor of eventual organic disease than currently available radiologic methods.

Fox J, Barber D, Bardhan KD. "Alternatives to Bayes? A quantitative comparison with rule-based diagnostic inference" Methods of Information in Medicine 1980 Oct;19(4):210-5.

de Dombal FT, "Computers, diagnoses and patients with acute abdominal pain" (editorial) Arch Emerg Med. 1992 September; 9(3): 267–270.

 

References