Evaluating the Completeness of Thyroid Nodule Ultrasound Reporting Using Natural Language Processing.

OBJECTIVE

Complete thyroid ultrasound reporting allows appropriate risk stratification and management of patients with thyroid nodules. Our objective was to assess the completeness of thyroid ultrasound reports based on documentation of key nodule features recommended by clinical guidelines.

METHODS

We conducted a retrospective review of patients who underwent thyroid ultrasound between 2013 and 2020, including those with at least one nodule larger than 1 cm. Using a rule-based natural language processing (NLP) pipeline, key thyroid nodule features were extracted from the reports. At the nodule level reports were classified as complete if all guideline-recommended characteristics were documented.

RESULTS

Out of 8,855 thyroid ultrasound reports reviewed, 5,792 (65%) documented at least one thyroid nodule ≥ 1 cm with 9,108 nodules evaluated. At the nodule level, 71% of reports included composition, 50% reported echogenicity, 25% noted shape, 35% described margins, and 40% described echogenic foci. Cervical lymph node assessment was included in 67% of reports. Overall, only 1988 out of 9,108 nodules greater than 1 cm (22%) had reports considered complete based on the inclusion of all key features. However, there was improvement in reporting completeness over time: 0% from 2013 to 2017, increasing to 16% in 2018, 65% in 2019, and 73% in 2020.

CONCLUSION

There was variability in the frequency of reported thyroid features that influence decision-making, with shape being less commonly reported. Completeness of ultrasound reporting improved over time, with 7 out of 10 nodule reports including all guideline-recommended features by 2020.