[Prediction of the effectiveness of the therapy of LUTS/BPH by Serenoa repens extracts]

For the treatment of LUTS/BPH is used a wide range of drugs that patients have to take for a long time. Therefore, it is important to develop methods for predicting long-term results of therapy. The purpose of this work is to evaluate the possibility to predict long-term results of drug therapy of LUTS/BPH using mathematical modeling on the example of treatment with Serenoa repens extract (ESR - Permixon).

For prediction using the methods of predictive analytics of the therapeutic ESR effect in the long term, materials from the open study "Clinical and biological long-term tolerance of a lipidosterolic extract of Serenoa repens (Permixon) in patients with symptomatic benign prostatic hypertrophy" (No. P0048 95 GP 401) were used. The study took place in 1995-1999 in 3 Moscow medical centers: Research Institute of Urology of the Ministry of Health of the Russian Federation, Urological Clinic of the Moscow Medical Academy named after Sechenov and the urology department of Moscow Clinical Hospital No 60. The study included 155 patients aged 52 to 87 years (65.3) who received the drug in 320 mg capsules per day for two years. The target indicators of the prognosis identified key clinical parameters: a decrease IPSS of>25% or>3 points and an increase in Qmax>25% at 12 and 24 months of treatment. When evaluating the results, a binary approach was used: improvement achieved (1), not achieved (0).

Using the methods of predictive analytics, mathematical models were built to predict the long-term results of treatment according to the most significant 7 initial criterias (predictors): IPSS; Qmax; average urine flow rate; urination volume, urination time, residual urine volume, prostate volume. For each target field and time interval, mathematical models were built using ensembles from 7 selected machine learning algorithms with the best predictive qualities: BNet; C5.0; SVM; KNN; NNet; CHAID; C&RT. Verification of models on internal randomized samples showed their high prognostic properties: sensitivity 82.4-99.0; specificity 75.0-96.1; AUC 0,864-0,965.

The potential for effective prediction by the methods of predictive analytics and data mining of the separated results of drug therapy of LUTS / BPH according to the main clinical criteria was demonstrated. It is necessary to continue training and testing the model with the inclusion of new clinical observations in the data set. This approach is applicable to the creation of similar models for predicting the effect of other drugs.