Joeky Senders

139 Comparing NLP methods sets can be assembled, labelled, and analyzed; however, the increase in efficiency can come at the cost of transparency. Lack of transparency incurs the risk of large-scale misinterpretations of automatically assembled data sets. Researchers should therefore balance the yield of automated medical text analysis against the risk and consequences of potential misclassification. Establishing standards for model evaluation, as well as a minimal threshold for model performance, might help in estimating and mitigating this risk. Although the heterogeneity across NLP endeavors in healthcare might limit the establishment of uniform standards, defining general guidelines that can be further specified at study-level can foster a safe and effective implementation of NLP in medical research and even clinical care. Conclusion The recent advent and widespread popularization of electronic medical records have led to an unprecedented volume of free-text clinical reports available for research purposes. Machine learning algorithms enable NLP techniques to learn from previously classified examples, thereby making it unnecessary to hard-code the rules for text analysis. Combining these techniques can therefore facilitate clinical research by optimizing the speed, accuracy, and consistency of clinical chart review. This study compares several NLP approaches for the classification of free-text radiology reports of brain metastasis patients, which can serve as a proof-of-concept and framework for NLP of electronic medical records. Supplementary material Supplementary Table S1 available online at: https: / /ascopubs.org/doi /10.1200/CCI .18.00138?ur l_ver=Z39.88-2003&rfr_ id=ori :rid:crossref.org &rfr_dat=cr_pub%20%200pubmed