Automated categorization of virtual reality studies in cardiology based on the device usage: a bibliometric analysis (2010-2022).

Aims: Currently, virtual reality (VR) constitutes a vital aspect of digital health, necessitating an overview of study trends. We classified type A studies as those in which health care providers utilized VR devices and type B studies as those in which patients employed the devices. This study aimed to analyse the characteristics of each type of studies using natural language processing (NLP) methods.

Methods and results: Literature related to VR in cardiovascular research was searched in PubMed between 2010 and 2022. The characteristics of studies were analysed based on their classification as type A or type B. Abstracts of the studies were used as corpus for text mining. A binary logistic regression model was trained to automatically categorize the abstracts into the two study types. Classification performance was evaluated by accuracy, precision, recall, F-1 score, and c-statistics of the receiver operator curve (ROC) analysis. In total, 171 articles met the inclusion criteria, where 120 (70.2%) were type A studies and 51 (29.8%) were type B studies. Type A studies had a higher proportion of case reports than type B studies (18.3% vs. 3.9%, P = 0.01). As for abstract classification, the binary logistic regression model yielded 88% accuracy and an area under the ROC of 0.98. The words 'training', '3d', and 'simulation' were the most powerful determinants of type A studies, while the words 'patients', 'anxiety', and 'rehabilitation' were more indicative for type B studies.

Conclusions: NLP methods revealed the characteristics of the two types of VR-related research in cardiology.