Predicting pain score in different doses of ondansetron using a powerful machine learning method in upper limb orthopedic patients

Predicting pain scores in patients can help identify the appropriate analgesic dose. In this research,two different doses of ondansetron (۲ and ۴ mg) in combination with lidocaine were comparedwith the lidocaine group alone. In addition to comparing the three groups, different inputparameters were used to model the pain score using new machine-learning methods. The twoconventional training algorithms, Levenberg-Marquardt (LM) and Scaled Conjugate Gradient(SCG) were coded in MATLAB software to model the pain score. After comparing the pain scoresin the three groups, the third group with a dose of ۴ mg had a lower pain score at different times.The linear regression results between the output of the developed artificial neural network and thecorresponding targets show a high correlation coefficient.Background and aims: Simple and inexpensive methods of intravenous regional anesthesia areused in many orthopedic surgeries. Predicting the pain score at different times helps to estimatethe appropriate analgesic doses. The purpose of this article is to predict the pain score using differentdoses of ondansetron.Method: ۹۰ patients with upper limb injuries underwent intravenous regional anesthesia. Patientswere divided into three different groups of ۳۰ people, including group ۱ of lidocaine alone, group۲ of lidocaine with ۲ mg of ondansetron, and group ۳ of lidocaine with ۴ mg of ondansetron.Then, the pain score was evaluated before injection, ۵ and ۱۰ minutes after injection, when drainingthe tourniquet, and ۵ to ۶۰ minutes after draining the tourniquet in ۵-minute intervals using avisual analog scale that was taught to the patient before the operation. In order to model, an artificialneural network with LM and SCG optimizers was used. The variables of age, sex, systolicand diastolic blood pressure, heart rate, and pain at the mentioned times were used as inputs ofthe artificial neural network. ۷۰%, ۱۵%, and ۱۵% of the input data were used for training, testing,and validation, respectively. The number of layers and hidden neurons in each layer for the LMoptimizer are ۲ and ۸, respectively. For the SCG optimizer, these values are ۱ and ۱۰.Results: In the third group, the results showed less pain score than the other two groups. Thecorrelation coefficient (R-value) of the model developed using machine learning in all ۳ groupsof data is higher than ۸۰%. To check the developed model more accurately, one person from eachgroup of ۳۰ patients was used to estimate the pain score. The root mean square error value for all۳ patients was below ۲% and the results show the accuracy of the SCG training method for higherdoses of ondansetron.Conclusion: The statistical results of modeling and estimating the pain score of orthopedic patientsusing machine learning show the high accuracy of this method, which can help to predictthe appropriate doses of medication in the next few minutes. The pain score estimated at differenttimes, in addition to showing the effectiveness of ondansetron, indicates the need to prescribepainkillers.

کلیدواژه ها:

Machine learning ، Ondansetron ، Orthopedic Patients ، Artificial Intelligence ، Pain