Anemia is a disease characterized by a decrease in the amount of healthy erythrocytes in the body of healthy erythrocytes, often occurs in patients with chronic kidney diseases who need to undergo routine hemodialysis. Accordingly, the erythrocyte-stimulating agents (ERYTHROPOIESIS-Stimulating Agents, ESA) and Iron Supplements, IS are introduced within the framework of this process. But at the same time complications may arise if the patients have changed the metabolism of iron or poor reaction to drugs. In addition, medications are usually expensive and lowered a difficult financial burden on public health or the patient himself. Thus, taking into account the fact that the number of such patients is currently growing, there is a great demand for additional support systems with "abilities" to making decisions. One option is to use artificial intelligence technology (Artificial Intelligence, AI), which seems to be a promising method, but requires a large data array and is not practical due to various patient health states.
In a recent study, the results of which were published at International Journal Of Medical Sciences, scientists from Japan nevertheless tried to solve the problem. They decided instead of making AI to study the complex physiology of the patient's body, use the prediction model based on solutions of experienced doctors. Associate Professor Toshiaki Ohara (Toshiaki Ohara) from Okayam University explains:
We are developing a system based on the principles that are used in the thinning process of experienced doctors. In the end, they do not calculate the detailed values of the life reactions in the patient's body when making a decision on the dosage, which means that prediction models based on biochemistry are not necessarily needed.
Scientists have prepared two data sets obtained in 2 hospitals - one to teach their model, and the other for testing and verifying its forecasts. At the same time, they recorded prescribed prescription prescriptions in two hospitals and considered the reaction to the two drug mentioned above applied during hemodialysis.
On their basis, an AI model was built, called the "Artificial Intelligence Anemia" (ARTIFICIAL-Intelligence-Supported Anemia Control System, Aisacs), which received a total of five input sources (four points of blood and anamnesis) and in The quality of the output chose the probability of the need for dosage for two drugs. In addition, to increase the efficiency of the process, they compensated for a time delay between blood testing and deciding on the dosage using "data adjustment" to bring the date of decision-making in accordance with the dates of the survey.
As a result, Aisacs showed high precision of the forecast with the correct classification (solutions corresponding to the conclusions of doctors) at 72% -87%. But even more interesting was that in some cases, Aisacs provided "clinically correct" classifications with even higher indicators (92% -97%). These were solutions that did not coincide with the diagnosis of doctors, but still considered correct from a medical point.