Name

Création d'une application de support clinique d'aide à la décision clinique [CDSS] pour le dépistage de la rétinopathie diabétique

The aim of the present study was to develop a Clinical Decision Support System [CDSS] which will help clinicians [Family doctors, endocrinologists and ophthalmologists] to estimate the risk of a DM patient developing diabetic retinopathy, to personalize a patient's screening needs and to schedule a follow-up visit anytime in the following 3 years [1].

To develop the CDSS, we selected a sample of 2,323 patients with T2DM diagnosed from 15,811 screened at our HCA. We studied the following risk factors for each patient: current age, sex, DM duration, arterial hypertension, DM treatment, body mass index, HbA1c, and microalbuminuria. Data preparation process aimed to transform numerical variables into linguistic ones by constructing a fuzzy set for each term and defining a fuzzy partition over the original numerical reference scale. A fuzzy random forest [FRF], which is an ensemble of fuzzy decision trees [fuzzy classifiers] was constructed with the data of the patients of the training set, using an adaptation of the FRF induction algorithm proposed by Yuan and Shaw. More explanations about the different measures that appear in this algorithm can be found in our group of study, Saleh E et al.

We made an empirical analysis using a 10-fold cross-validation on the training set, considering the following ranges of values: 100-200-300 trees, 1-2-3-4 randomly selected attributes in each node, and a leaf creation threshold between 0 and 1 [in 0.1 intervals]. The best results on this validation were obtained with 100 trees, 3 selected attributes per node and a high leaf creation threshold [0.8-1.0]. With these values of the parameters, the classification of the testing set was made with an accuracy of 80.29%, a sensitivity of 80.67% and a specificity of 80.18%.  The FRF model has been compared with the following classifiers: a) Logistic regression, b) K-Nearest neighbours. The best results were obtained with k=5 neighbours, c) A single decision tree, constructed with the classical ID3 algorithm, d) A Random Forest, in which variables have not been submitted to fuzzy rules.The best results are obtained with the two models based on Random Forests, in which levels of specificity and sensitivity are over 80%. When we apply fuzzy rules, sensitivity improves. Thus, FRF offer the best sensitivity, keeping a high level of specificity. The CDD system was tested on our own population which yielded reasonably good specificity of 84-85%, sensitivity a bit lower at between 75-78%, and accuracy between 79.90-81.74%. We also applied the classifier on the data of 28,344 patients from another health care area in Catalonia, and results obtained had a high specificity of between 87-89 %, but low sensitivity levels of between 44-50% and accuracy of between 78.55-80.81%, 

During the development of our CDSS, it specificity has always achieves high levels, 80.89% with random forest using fuzzy rules and 91.35% with logistic regression. We also observe that with higher specificity values we obtain lower sensitivity values. Only with random forest and fuzzy rules were we able to achieve a balance between specificity and sensitivity, so we believe that is the most suitable choice, with both good sensitivity and good specificity values.  Finally, in this study, we applied our CDSS to two populations, our own 15,811 patients and an outside population of 28,444 DM patients taken from another of our studies previously conducted in Spain. Results showed that CDSS maintains good specificity in 84-85% of our own population, and higher levels [87-89%] in the outside population.

We have developed a CDSs, which can help more effectively screen for diabetic retinopathy.