Abstract
Purpose :
The aim of the present study is to build an informàtic software to support clínical decisions in diabetic retinopathy (DR) screening, based on our diabetes mellitus (DM) population data.
Methods :
The estimated diabetic population in our area is about 147228 people. From it a sample of 2,323 patients was taken to build de CDSS. It was divided in two grups: a training set of 1,212 patients and a testing set of 1,111 patients. The clínical decision suport system (CDSS) is based on a fuzzy random forest integrated by fuzzy decision trees. A fuzzy decision tree is a hierarchical data structure which, depending on the values reached on those attributes related to the DR risk factors, classifies patients on diferent categories and levels. Each node of the tree is an attribute, and each branch of the node is related to a possible value of the attribute. The leaves of the tree link patients to a particular category, depending on the presence of diabetic retinopathy.
Results :
The CDDS was built from 200 trees in the forest and 3 variables at each node. Accuracy of the CDSS was 80.76%,, sensitivity 80.67%% and specificity 85.96%. Applied variables were: current age, sex, DM duration and treatment, arterial hypertension, body mass index, HbA1c, estimated glomerular filtration rate and microalbuminuria.
Conclusions :
Some studies concluded that one screening test every 3 years was cost-effective, but they did not consider some risk factors which could be important in retinopathy developing, and probably in some cases this frequency could not be appropriate.
We propose a random forest test with fuzzy rules to build a personalized CDSS, offering an individualized plan according to their risk for each patient. Despite more tests are needed to validate this system, this tool could be incorporated for the assesment in the screening diabetic retinopathy programs in the future to improve the screening models quality.
Our program has been patented under the name RETIPROGRAM (i-DEPOT evidencie number 069047) and currently is being validated with the electronic clínical records from our area with 547228 diabetic patients.
This is a 2020 ARVO Annual Meeting abstract.