The intensifying rates of opioid overuse and drug misuse require effective drug prescription scrutinizing tools. Despite the available prescription drug monitoring programs (PDMPs) in countries like the US, how should indispensable PDMPs be devised to help mitigate drug overprescription (Rx)? Zhang et al. propose an active hybrid prescription neural network (RxNet) model for recognizing refill overprescribing.
The authors argue that it is difficult to: (1) model the associations between the various prescription and dispensing records of physicians and patients; (2) simulate the deviation in the active and growing interactions among asymmetrical prescription refills and dispensing for diverse medical situations; and (3) encapsulate the decision-making patterns of overprescribed drugs. Consequently, they present RxNet for patterning the PDMP data to forecast overprescribing for patients at high risk of opioid or drug abuse.
The RxNet model consists of a dynamic heterogeneous graph used to reflect the interactions among various prescription and dispensing records, a periodic neural network used to ascertain the impacts of prescription drug records, a mechanism used to acquire abnormal refill behaviors and automatically update the appropriate prescription records of patients, and a dosing-adaptive convolution network used to pull out and recompute the medicating patterns of patients for identifying drug overprescriptions.
Experiments were performed using Ohio PDMP data, from 2016, to assess the effectiveness of the RxNet model. The data contained nearly three million prescriptions by over forty thousand physicians for almost three hundred thousand patients. Experimental results reveal that the RxNet method outperformed well-known neutral network models, such as the temporal graph network and temporal graph attention layer, when used to forecast overprescribing.
The authors present a reliable model for the risk analysis of drug misuse/abuse. Practitioners in healthcare and medicine should read this insightful paper.