Study On Data-driven Background Infusion Rate Class Recommendation Model For Patient-Controlled Intravenous Analgesia

Opioid-based patient-controlled intravenous analgesia(PCIA)is widely used for acute postoperative pain management.In actual clinical applications of PCIA,it is important-but challenging-to find the best background infusion regimen;this determination largely depends on the anesthesiologist's experience.With the rapid development of electronic healthcare records and surgical information systems in hospitals,a large amount of surgical patient data has been accumulated,which has provided the possibility and support for the research of data-driven PCIA background infusion rate class recommendation model.Data-driven model can automatically recommend individualized background infusion rate classes for PCIA.To develop this recommendation model,this thesis introduced machine learning methods to learn and mine anesthesiologists' excellent decisions for PCIA,which have good acute postoperative pain management effects.Furthermore,the developed model was performed feature analysis and application evaluation.This thesis applied the developed recommendation model to a new dataset that did not overlap with the model development data and compared the acute postoperative pain management effects between the model's recommendations and the actual human anesthesiologists'decisions.The results show that the final model obtained an AUC of 0.92 for both the low and high dosage classes.When evaluating the application of the model,postoperative nausea and vomiting(PONY)incidence based on the final model recommendations was significantly lower in patients whose actual background infusion rate classes matched the recommendations.The incidences were 18%and 26%respectively(p<0.001).Moreover,there was no significant difference in acute postoperative pain incidences between two groups who matched and did not match the recommendations.We conclude that this thesis has developed a machine learning-based recommendation model that learned experience from anesthesiologists'excellent decisions concerning background infusion rate classes.In addition,the recommendations offered by the model can lower PONV incidence without reducing pain relief.