Method for assessing the need and type of surgical re-treatment in glaucoma using an artificial neural network

Abstract

Relevance. The problem of evaluating the choice of treatment method for patients with glaucoma is widely covered in the medical literature, but currently there are no publications on indications and decision making about reoperation in case of complete or partial ineffectiveness of the treatment. Purpose. To describe a method for assessing the need and type of reoperation in the surgical treatment of glaucoma using an artificial neural network. Material and methods. 7801 cases of observation and treatment of patients diagnosed with glaucoma were selected for 2018 –2020. The development and statistical analysis of the factors associated with repeated surgical operations was carried out, a model for predicting the probability of repeated operations was created, mathematical modeling of the onset of complications depending on the identified factors was carried out using the method of constructing classification trees. The resulting model was used to assess the need for surgical treatment in patient registered in clinic in 2019. Cases of specialized medical care have been identified, in which, with a high degree of probability, long-term compensation of the pathological course of glaucoma will not occur and repeated surgical treatment will be required. These cases were copied into a separate database for further work. The preferred reoperation method for patient management was selected using an artificial neural network. Results. A number of factors that are statistically significantly associated with reoperation have been identified, and an artificial neural network model has been created to predict the type of reoperation in glaucoma. Application of the obtained method on patients who received treatment at the clinic in 2019 –2021 years made it possible to predict the probability of reoperation within a three-year follow-up period in 5%, while among the patients the need for treatment was distributed as follows: microfistulizing deep sclerectomy with allodrainage – 88.37%, selective laser trabeculoplasty – 6.98%, transscleral cyclophotocoagulation – 4.65%. Conclusion. The obtained method allows predicting the probability of re-treatment of glaucoma and the type of surgery required. Considering the specificity and sensitivity of the obtained models, it is necessary to increase the number of observations, further clinical and organizational study of aspects of the provision of medical services in previously operated glaucoma, and an assessment of the clinical and economic effect of the introduction of this method. Keywords: glaucoma, artificial neural network, assessment method