Deep learning to refine the identification of high-quality clinical research articles from the biomedical literature: Performance evaluation.

BACKGROUND: Identifying practice-ready evidence-based journal articles in medicine is a challenge due to the sheer volume of biomedical research publications. Newer approaches to support evidence discovery apply deep learning techniques to improve the efficiency and accuracy of classifying sound evidence. OBJECTIVE: To determine how well deep learning models using variants of Bidirectional Encoder Representations from Transformers (BERT) identify high-quality evidence with high clinical relevance from the biomedical literature for consideration in clinical practice. METHODS: We fine-tuned variations of BERT models (BERTBASE, BioBERT, BlueBERT, and PubMedBERT) and compared their performance in classifying articles based on methodological quality criteria. The dataset used for fine-tuning models included titles and abstracts of >160,000 PubMed records from 2012 to 2020 that were of interest to human health which had been manually labeled based on meeting established critical appraisal criteria for methodological rigor. The data was randomly divided into 80:10:10 sets for training, validating, and testing. In addition to using the full unbalanced set, the training data was randomly undersampled into four balanced datasets to assess performance and select the best performing model. For each of the four sets, one model that maintained sensitivity (recall) at ≥99% was selected and were ensembled. The best performing model was evaluated in a prospective, blinded test and applied to an established reference standard, the Clinical Hedges dataset. RESULTS: In training, three of the four selected best performing models were trained using BioBERTBASE. The ensembled model did not boost performance compared with the best individual model. Hence a solo BioBERT-based model (named DL-PLUS) was selected for further testing as it was computationally more efficient. The model had high recall (>99%) and 60% to 77% specificity in a prospective evaluation conducted with blinded research associates and saved >60% of the work required to identify high quality articles. CONCLUSIONS: Deep learning using pretrained language models and a large dataset of classified articles produced models with improved specificity while maintaining >99% recall. The resulting DL-PLUS model identifies high-quality, clinically relevant articles from PubMed at the time of publication. The model improves the efficiency of a literature surveillance program, which allows for faster dissemination of appraised research.

A Deep Learning Approach to Refine the Identification of High-Quality Clinical Research Articles From the Biomedical Literature: Protocol for Algorithm Development and Validation.

BACKGROUND: A barrier to practicing evidence-based medicine is the rapidly increasing body of biomedical literature. Use of method terms to limit the search can help reduce the burden of screening articles for clinical relevance; however, such terms are limited by their partial dependence on indexing terms and usually produce low precision, especially when high sensitivity is required. Machine learning has been applied to the identification of high-quality literature with the potential to achieve high precision without sacrificing sensitivity. The use of artificial intelligence has shown promise to improve the efficiency of identifying sound evidence. OBJECTIVE: The primary objective of this research is to derive and validate deep learning machine models using iterations of Bidirectional Encoder Representations from Transformers (BERT) to retrieve high-quality, high-relevance evidence for clinical consideration from the biomedical literature. METHODS: Using the HuggingFace Transformers library, we will experiment with variations of BERT models, including BERT, BioBERT, BlueBERT, and PubMedBERT, to determine which have the best performance in article identification based on quality criteria. Our experiments will utilize a large data set of over 150,000 PubMed citations from 2012 to 2020 that have been manually labeled based on their methodological rigor for clinical use. We will evaluate and report on the performance of the classifiers in categorizing articles based on their likelihood of meeting quality criteria. We will report fine-tuning hyperparameters for each model, as well as their performance metrics, including recall (sensitivity), specificity, precision, accuracy, F-score, the number of articles that need to be read before finding one that is positive (meets criteria), and classification probability scores. RESULTS: Initial model development is underway, with further development planned for early 2022. Performance testing is expected to star in February 2022. Results will be published in 2022. CONCLUSIONS: The experiments will aim to improve the precision of retrieving high-quality articles by applying a machine learning classifier to PubMed searching. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): DERR1-10.2196/29398.

Machine Learning Approaches to Retrieve High-Quality, Clinically Relevant Evidence From the Biomedical Literature: Systematic Review.

BACKGROUND: The rapid growth of the biomedical literature makes identifying strong evidence a time-consuming task. Applying machine learning to the process could be a viable solution that limits effort while maintaining accuracy. OBJECTIVE: The goal of the research was to summarize the nature and comparative performance of machine learning approaches that have been applied to retrieve high-quality evidence for clinical consideration from the biomedical literature. METHODS: We conducted a systematic review of studies that applied machine learning techniques to identify high-quality clinical articles in the biomedical literature. Multiple databases were searched to July 2020. Extracted data focused on the applied machine learning model, steps in the development of the models, and model performance. RESULTS: From 3918 retrieved studies, 10 met our inclusion criteria. All followed a supervised machine learning approach and applied, from a limited range of options, a high-quality standard for the training of their model. The results show that machine learning can achieve a sensitivity of 95% while maintaining a high precision of 86%. CONCLUSIONS: Machine learning approaches perform well in retrieving high-quality clinical studies. Performance may improve by applying more sophisticated approaches such as active learning and unsupervised machine learning approaches.

Optimizing a literature surveillance strategy to retrieve sound overall prognosis and risk assessment model papers.

OBJECTIVE: Our aim was to develop an efficient search strategy for prognostic studies and clinical prediction guides (CPGs), optimally balancing sensitivity and precision while independent of MeSH terms, as relying on them may miss the most current literature. MATERIALS AND METHODS: We combined 2 Hedges-based search strategies, modified to remove MeSH terms for overall prognostic studies and CPGs, and ran the search on 269 journals. We read abstracts from a random subset of retrieved references until ≥ 20 per journal were reviewed and classified them as positive when fulfilling standardized quality criteria, thereby assembling a standard dataset used to calibrate the search strategy. We determined performance characteristics of our new search strategy against the Hedges standard and performance characteristics of published search strategies against the standard dataset. RESULTS: Our search strategy retrieved 16 089 references from 269 journals during our study period. One hundred fifty-four journals yielded ≥ 20 references and ≥ 1 prognostic study or CPG. Against the Hedges standard, the new search strategy had sensitivity/specificity/precision/accuracy of 84%/80%/2%/80%, respectively. Existing published strategies tested against our standard dataset had sensitivities of 36%-94% and precision of 5%-10%. DISCUSSION: We developed a new search strategy to identify overall prognosis studies and CPGs independent of MeSH terms. These studies are important for medical decision-making, as they identify specific populations and individuals who may benefit from interventions. CONCLUSION: Our results may benefit literature surveillance and clinical guideline efforts, as our search strategy performs as well as published search strategies while capturing literature at the time of publication.

Defining the publication source of high-quality evidence in urology: an analysis of EvidenceUpdates.

OBJECTIVES: To determine the publication sources of urology articles within EvidenceUpdates, a second-order peer review system of the medical literature designed to identify high-quality articles to support up-to-date and evidence-based clinical decisions. MATERIALS AND METHODS: Using administrator-level access, all EvidenceUpdates citations from 2005 to 2014 were downloaded from the topics 'Surgery-Urology' and 'Oncology-Genitourinary'. Data fields accessed included PubMed unique reference identifier, study title, abstract, journal and date of publication, as well as clinical relevance and newsworthiness ratings as determined by discipline-specific physician raters. The citations were then coded by clinical topic (oncology, voiding dysfunction, erectile dysfunction/infertility, infection/inflammation, stones/endourology/laparoscopy, trauma/reconstruction, transplant, or other), journal category (general medical journal, oncology journal, urology journal, non-urology specialty journal, Cochrane review, or other), and study design (randomised controlled trial [RCT], systematic review, observational study, or other). Articles that were perceived to be misclassified and/or of no direct interest to urologists were excluded. Descriptive statistics using proportions and 95% confidence intervals, as well as means and standard deviations (SDs) were used to characterise the overall data cohort and to analyse trends over time. RESULTS: We identified 731 unique citations classified under either 'Surgery-Urology' or 'Oncology-Genitourinary' for analysis after exclusions. Between 2005 and 2014, the most common topics were oncology (48.6%, 355 articles) and voiding dysfunction (21.8%, 159). Within the topic of oncology, prostate cancer contributed over half the studies (54.6%, n = 194). The most common study types were RCTs (42.3%, 309 articles) and systematic reviews (39.6%, 290). Systematic reviews had a nearly fourfold relative increase within less than a decade. The largest proportion of studies relevant to urology were published in general oncology journals (20.0%, n = 146), followed by the Cochrane Library (19.3%, n = 141) and general medical journals (17.2%, n = 126). Urology-specific journals contributed to only approximately one-tenth of EvidenceUpdates alerts (9.4%, n = 69), with the highest contribution occurring during the 2013/2014 period. For clinical relevance and newsworthiness scores (each graded on scales of 1-7), urology journals scored the highest in clinical relevance with a mean (SD) of 5.9 (0.75) and general medical journals scored highest for newsworthiness at 5.3 (0.94). On average, RCTs scored highest both for clinical relevance and newsworthiness with mean (SD) scores of 5.71 (0.81) and 5.22 (0.91), respectively. CONCLUSION: A large number of high-quality, clinically relevant, and newsworthy peer-reviewed urology publications are published outside of traditional urology journals. This requires urologists to implement well-defined strategies to stay abreast of current best evidence.

Increasing the quantity and quality of searching for current best evidence to answer clinical questions: protocol and intervention design of the MacPLUS FS Factorial Randomized Controlled Trials.

BACKGROUND & AIMS: Finding current best evidence for clinical decisions remains challenging. With 3,000 new studies published every day, no single evidence-based resource provides all answers or is sufficiently updated. McMaster Premium LiteratUre Service--Federated Search (MacPLUS FS) addresses this issue by looking in multiple high quality resources simultaneously and displaying results in a one-page pyramid with the most clinically useful at the top. Yet, additional logistical and educational barriers need to be addressed to enhance point-of-care evidence retrieval. This trial seeks to test three innovative interventions, among clinicians registered to MacPLUS FS, to increase the quantity and quality of searching for current best evidence to answer clinical questions. METHODS & DESIGN: In a user-centered approach, we designed three interventions embedded in MacPLUS FS: (A) a web-based Clinical Question Recorder; (B) an Evidence Retrieval Coach composed of eight short educational videos; (C) an Audit, Feedback and Gamification approach to evidence retrieval, based on the allocation of 'badges' and 'reputation scores.' We will conduct a randomized factorial controlled trial among all the 904 eligible medical doctors currently registered to MacPLUS FS at the hospitals affiliated with McMaster University, Canada. Postgraduate trainees (n=429) and clinical faculty/staff (n=475) will be randomized to each of the three following interventions in a factorial design (AxBxC). Utilization will be continuously recorded through clinicians' accounts that track logins and usage, down to the level of individual keystrokes. The primary outcome is the rate of searches per month per user during the six months of follow-up. Secondary outcomes, measured through the validated Impact Assessment Method questionnaire, include: utility of answers found (meeting clinicians' information needs), use (application in practice), and perceived usefulness on patient outcomes. DISCUSSION: Built on effective models for the point-of-care teaching, these interventions approach evidence retrieval as a clinical skill. If effective, they may offer the opportunity to enhance it for a large audience, at low cost, providing better access to relevant evidence across many top EBM resources in parallel. TRIAL REGISTRATION: ClinicalTrials.Gov NCT02038439.

McMaster PLUS: a cluster randomized clinical trial of an intervention to accelerate clinical use of evidence-based information from digital libraries.

BACKGROUND: Physicians have difficulty keeping up with new evidence from medical research. METHODS: We developed the McMaster Premium LiteratUre Service (PLUS), an internet-based addition to an existing digital library, which delivered quality- and relevance-rated medical literature to physicians, matched to their clinical disciplines. We evaluated PLUS in a cluster-randomized trial of 203 participating physicians in Northern Ontario, comparing a Full-Serve version (that included alerts to new articles and a cumulative database of alerts) with a Self-Serve version (that included a passive guide to evidence-based literature). Utilization of the service was the primary trial end-point. RESULTS: Mean logins to the library rose by 0.77 logins/month/user (95% CI 0.43, 1.11) in the Full-Serve group compared with the Self-Serve group. The proportion of Full-Serve participants who utilized the service during each month of the study period showed a sustained increase during the intervention period, with a relative increase of 57% (95% CI 12, 123) compared with the Self-Serve group. There were no differences in these proportions during the baseline period, and following the crossover of the Self-Serve group to Full-Serve, the Self-Serve group's usage became indistinguishable from that of the Full-Serve group (relative difference 4.4 (95% CI -23.7, 43.0). Also during the intervention and crossover periods, measures of self-reported usefulness did not show a difference between the 2 groups. CONCLUSION: A quality- and relevance-rated online literature service increased the utilization of evidence-based information from a digital library by practicing physicians.