When William of Ockham meets Thomas Bayes: finding a few diagnoses among a great many symptoms

An optimized experimental protocol based on neuro-evolutionary algorithms application to the classification of dyspeptic patients and to the prediction of the effectiveness of their treatment

OBJECTIVE

This paper aims to present a specific optimized experimental protocol (EP) for classification and/or prediction problems. The neuro-evolutionary algorithms on which it is based and its application with two selected real cases are described in detail. The first application addresses the problem of classifying the functional (FD) or organic (OD) forms of dyspepsia; the second relates to the problem of predicting the 6-month follow-up outcome of dyspeptic patients treated by helicobacter pylori (HP) eradication therapy.

METHODS AND MATERIAL

The database built by the multicentre observational study, performed in Italy by the NUD-look Study Group, provided the material studied: a collection of data from 861 patients with previously uninvestigated dyspepsia, being referred for upper gastrointestinal endoscopy to 42 Italian Endoscopic Services. The proposed EP makes use of techniques based on advanced neuro-evolutionary systems (NESs) and is structured in phases and steps. The use of specific input selection (IS) and training and testing (T and T) techniques together with genetic doping (GenD) algorithm is described in detail, as well as the steps taken in the two benchmark and optimization protocol phases.

RESULTS

In terms of accuracy results, a value of 79.64% was achieved during optimization, with mean benchmark values of 64.90% for the linear discriminant analysis (LDA) and 68.15% for the multi layer perceptron (MLP), for the classification task. A value of 88.61% was achieved during optimization for the prediction task, with mean benchmark values of 49.32% for the LDA and 70.05% for the MLP.

CONCLUSIONS

The proposed EP has led to the construction of inductors that are viable and usable on medical data which is representative but highly not linear. In particular, for the classification problem, these new inductors may be effectively used on the basal examination data to support doctors in deciding whether to avoid endoscopic examinations; whereas, in the prediction problem, they may support doctors' decisions about the advisability of eradication therapy. In both cases the variables selected indicate the possibility of reducing the data collection effort and also of providing information that can be used for general investigations on symptom relevance.

Contribution of artificial neural networks to the classification and treatment of patients with uninvestigated dyspepsia

OBJECTIVES

To verify whether symptoms reported by patients with uninvestigated dyspepsia might be helpful in either classifying functional from organic dyspepsia (1st experiment), or recognising which Helicobacter pylori infected patients may benefit from eradication therapy (2nd experiment).

METHODS

We compared the performance of artificial neural networks and linear discriminant analysis in two experiments on a database including socio-demographic features, past medical history, alarming symptoms, and symptoms at presentation of 860 patients with uninvestigated dyspepsia enrolled in a large observational multi-centre Italian study.

RESULTS

In the 1st experiment, the best prediction for organic disease was given by the Sine Net model (specificity of 87.6% with 13 patients misclassified) and the best prediction for functional dyspepsia by the FF Bp model (sensitivity of 83.4% with 56 patients misclassified). The highest global accuracy of linear discriminant analysis was 65.1%, with 150 patients misclassified. In the 2nd experiment, the highest predictive performance was provided by the SelfDASn model: all infected patients who became symptom-free after successful eradicating treatment were correctly classified, whereas nine errors were made in forecasting patients who did not benefit from such a therapy. The highest global performance of linear discriminant analysis was 53.2%, with 37 patients misclassified.

CONCLUSIONS

In patients with uninvestigated dyspepsia, artificial neural networks might have potential for categorising those affected by either organic or functional dyspepsia, as well as for identifying all Helicobacter pylori infected dyspeptic patients who will benefit from eradication.

The end of diagnosis: when to stop testing

Defining the end of diagnosis and deciding when to stop further testing represent crucial features of many medical pursuits. The aim of this article is to analyze the principles that govern the decision to discontinue a chain of consecutive diagnostic tests. Using Bayes formula and threshold analysis, four common "stop rules" of medical diagnostics can be derived. The first rule relates to a lack of therapeutic consequence associated with continued testing. The second rule concerns empirical therapy, that is, the use of a benign, effective, and specific type of therapy to cap a diagnostic chain and use therapeutic success as confirmation for a diagnostic suspicion. The third rule states that the benefit of a suspected diagnosis should stay higher than the costs of a test invested in confirming its presence. Lastly, as the fourth rule, the cumulative test costs should not exceed the expected risks of a missed diagnosis. Although in clinical practice the ambiguities and variations among patients may compromise the calculation of exact stop values for each rule, knowledge of these underlying general principles may be sufficiently helpful in managing the individual patient.