JMIR Publications

ABSTRACT

Background:

Medication reconciliation (MedRec) has the potential to improve patient safety by enhancing the continuity of medication information. MedRec involves three core tasks: the creation of a best possible medication history (BPMH), the identification of medication discrepancies among medication lists, and the resolution of medication discrepancies. While artificial intelligence (AI) has the potential to improve MedRec, the ways in which AI has been used to facilitate MedRec tasks and their constituent subtasks, as well as the level of automation achieved, have not been identified.

Objective:

This scoping review aimed to map how AI has been applied to MedRec tasks and subtasks, and to assess the extent of automation achieved.

Methods:

We searched MEDLINE, Embase, Web of Science, IEEE Xplore, and Compendex for studies that used AI to support a MedRec task or subtask, excluding entirely rule-based tools. After screening 2,345 unique records, we conducted forward and backward citation searching of studies included at the full-text stage, identifying an additional 795 unique records. We used a four-stage model of human information processing as a structural lens to guide our considerations of automation.

Results:

A total of 94 studies were included. All studies addressed subtasks related to the creation of a BPMH, while only two studies also addressed the identification of discrepancies. The highest stage of information processing automated was information analysis, though most only automated information acquisition steps. Most studies used free-text clinical notes from the electronic health record, though a significant proportion used images of pills or other medication-related items. Studies using text-based data employed a variety of machine learning methods (e.g., recurrent neural networks, conditional random fields, support vector machines, transformers), while those that leveraged images typically used convolutional neural networks. Most studies used publicly available data from benchmarking shared tasks (e.g., n2c2 2022) and were strictly model development studies, with only one usability study.

Conclusions:

Current applications of AI to automate MedRec tasks are preliminary, with most work focusing on extraction of medication information and limited to proof-of-concept model development. Future work should consider addressing infrastructural barriers to the AI-based automation of MedRec tasks (e.g., data incompleteness in sources of medication information) and exploring approaches to automate discrepancy resolution. Beyond developing models, there is also a need to implement them in tools and evaluate them in real-world contexts.