Taxonomy-Based Automation of Prior Approval Using Clinical Guidelines

Proceedings of Recent Advances in Natural Language Processing, pages 598–607 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_065 598 Taxonomy-Based Automation of Prior Approval using Clinical Guidelines Saranya Krishnamoorthy, Ayush Singh inQbator AI at eviCore Healthcare Evernorth Health Services firstname.lastname@evicore.com Abstract Performing prior authorization on patients in a medical facility is a time-consuming and chal- lenging task for insurance companies. Automat- ing the clinical decisions that lead to autho- rization can reduce the time that staff spend executing such procedures. To better facili- tate such critical decision making, we present an automated approach to predict one of the challenging tasks in the process called primary indicator prediction, which is the outcome of this procedure. The proposed solution is to cre- ate a taxonomy to capture the main categories in primary indicators. Our approach involves an important step of selecting what is known as the “primary indicator” – one of the several heuristics based on clinical guidelines that are published and publicly available. A taxonomy- based PI classification system was created to help in the recognition of PIs from free text in electronic health records (EHRs). This tax- onomy includes comprehensive explanations of each PI, as well as examples of free text that could be used to detect each PI. The major contribution of this work is to introduce a taxon- omy created by three professional nurses with many years of experience. We experiment with several state-of-the-art supervised and unsuper- vised techniques with a focus on prior approval for spinal imaging. The results indicate that the proposed taxonomy is capable of increasing the performance of unsupervised approaches by up to 10 F1 points. Further, in the supervised setting, we achieve an F1 score of 0.61 using a conventional technique based on term fre- quency–inverse document frequency that out- performs other deep-learning approaches. 1 Introduction Real-world applications in the Natural Language Processing (NLP) domain are known to perform better when the language models that support them are trained and fine-tuned on the domain in ques- tion (Gu et al., 2021; Rojas et al., 2022; Zhou et al., 2022; Naseem et al., 2022). One domain where this idea is applicable at a high level is the healthcare domain. Applications herein must adhere to the domain-specific vocabulary and guidelines. Predic- tion tasks require large amounts of sensitive data that contain information about patients and other details about the facilities that provide treatment. While the data sensitivity and protection challenges alone can be considered overwhelming due to the caveats of anonymization and privacy efforts, other atypical challenges based on knowledge and repre- sentation add to the complexity of NLP solutions in healthcare. Knowledge from overworked staff, such as nurses and physicians, is critical to obtaining high-quality corpora to train NLP models. Due to the lack of time, medical personnel are often unwilling to par- ticipate in annotation tasks to transfer knowledge (Ishikawa, 2022; Fiałek, 2022; Aycock, 2022; Mi- ley, 2022). Furthermore, when staff can participate in annotation tasks, facilities are usually unwill- ing to release annotations for public …

Content truncated to 500 words