ABSTRACT
Background: The aim of this study was to validate a three-class sentiment classification model for clinical trial abstracts combining adversarial learning and the BioBERT language processing model as a tool to assess trends in biomedical literature in a clearly reproducible manner. We then assessed the model's performance for this application and compared it to previous models used for this task. Methods: Using 108 expert-annotated clinical trial abstracts and 2,000 unlabeled abstracts this study develops a three-class sentiment classification algorithm for clinical trial abstracts. The model uses a semi-supervised model based on the Bidirectional Encoder Representation from Transformers (BERT) model, a much more advanced and accurate method compared to previously used models based upon traditional machine learning methods. The prediction performance was compared to those previous studies. Results: The algorithm was found to have a classification accuracy of 91.3%, with a macro F1-Score of 0.92, significantly outperforming previous studies used to classify sentiment in clinical trial literature, while also making the sentiment classification finer grained with greater reproducibility. Conclusion: We demonstrate an easily applied sentiment classification model for clinical trial abstracts that significantly outperforms previous models with greater reproducibility and applicability to large-scale study of reporting trends.
ABSTRACT
Publication bias and the decreased publication of trials with negative or non-significant results is a well-recognized problem in human and veterinary medical publications. These biases may present an incomplete picture of evidence-based clinical care and negatively impact medical practices. The purpose of this study was to utilize a novel sentiment analysis tool as a quantitative measure for assessing clinical trial reporting trends in human and veterinary medical literature. Abstracts from 177,617 clinical trials in human medical journals and 8684 in veterinary medical journals published in the PubMed database from 1995 to 2020. Abstracts were analyzed using the GAN-BioBERT sentiment classifier for both general trends and percentage of neutral/negative publications. Sentiment was defined on a - 1 (highly negative) to 1 (highly positive) scale. Human-based clinical trial publications were less likely to feature positive findings (OR 0.87, P < 0.001) and more likely to include neutral findings (OR 1.18, P < 0.001) relative to veterinary clinical trials. No difference was found in reporting of negative sentiment trials (OR 1.007, P = 0.83). In both groups, the published sentiment of clinical trials increased over time. Using sentiment analysis to evaluate large publication datasets and compare publication trends within and between groups, this study is significant in its detection of significant publication differences between human and veterinary medicine clinical trials and a continued unbalanced positive sentiment in the published literature. The implications of this unbiased reporting have important clinical and research implications that require consideration.
Subject(s)
Sentiment Analysis , Animals , Humans , Publication BiasABSTRACT
INTRODUCTION: Sentiment analysis, by evaluating written wording and its context, is a growing tool used in computer science that can determine the level of support expressed in a body of text using artificial intelligence methodologies. The application of sentiment analysis to biomedical literature is a growing field and offers the potential to rapidly and economically explore large amounts of published research and characterize treatment efficacy. METHODS: We compared the results of sentiment analysis of 115 article abstracts analyzed in a recently published meta-analysis of peripheral nerve block usage in primary hip and knee arthroplasty to the conclusions drawn by the authors of the original meta-analysis. RESULTS: A moderately positive outlook supporting the utilization of regional anesthesia for hip and knee arthroplasty was found in the 115 articles that were included for analysis, with 46% expressing positive sentiment, 35% expressing neutral sentiment, and 19% of abstracts expressing negative sentiment. This was well aligned with the conclusions reached by a previous meta-analysis of the same articles. DISCUSSION: Sentiment analysis applied to the medical literature can rapidly evaluate large collections of published data and generate an impression of overall findings that are aligned with the findings of a traditional meta-analysis.