Click here to

Session: Advances in Safety [Return to Session]

A Digital Peer to Automatically Detect Atypical Prescriptions in Radiotherapy

Q Li*, J Wright, R Hales, K Voong, T McNutt, The Johns Hopkins University School of Medicine, Baltimore, MD


TU-H-BRA-1 (Tuesday, 7/12/2022) 2:45 PM - 3:45 PM [Eastern Time (GMT-4)]

Ballroom A

Purpose: Prescribing appropriate radiation doses is crucial to patient safety in radiotherapy. It is essential to address the problem of prescription anomaly detection, even if the erroneous prescription is rare. The current quality assurance heavily depends on a manual and laborious peer review chart round to detect such errors. Physicians may not identify mistakes due to time constraints and caseload. A prescription anomaly detection tool assisting peer review chart rounds to detect the mistakes automatically is in great need.

Methods: We designed a novel prescription anomaly detection algorithm that utilizes historical data to predict anomalous cases. Such a tool can serve as a digital physician peer who will assist the peer-review process providing extra safety to the patients.We created two dissimilarity metrics in our primary model, R and F. R defining how far a new patient's prescription is from historical prescriptions. F represents how far away a patient's feature set is from the group with an identical or similar prescription. We flag prescription if either metric is greater than specific optimized cut-off values. We used thoracic cancer patients (n=2356) as an example and extracted seven features.

Results: Our testing f1 score is between 73%-94% for different treatment technique groups. We also independently validate our results by conducting a mock peer review with three thoracic specialists. Our model has a lower type II error rate than manual peer-review physicians.

Conclusion: Our model has many advantages over traditional machine learning algorithms, mainly that it does not suffer from class imbalance. It can also explain why it flags each case and separates prescription and non-prescription-related features without learning from the data.

Funding Support, Disclosures, and Conflict of Interest: This work was supported by SBIR Phase II contract 2035750 awarded by the National Science Foundation. Q.L.: US Provisional Patent Application 63/253,618; T.M.: Oncospace Inc.; R.V.: AstraZeneca J.W.: Honoraria for site visits for program accreditation through ASTRO's radiation oncology accreditation program, Vice-chair of clinical affairs and quality committee ASTRO.


Quality Assurance, Computer Software, Scatter


IM/TH- Informatics: Informatics in Therapy (general)

Contact Email