Augmenting Quality Assurance Measures in Treatment Review with Machine Learning in Radiation Oncology.

Purpose: Pretreatment quality assurance (QA) of treatment plans often requires a high cognitive workload and considerable time expenditure. This study explores the use of machine learning to classify pretreatment chart check QA for a given radiation plan as difficult or less difficult, thereby alerting the physicists to increase scrutiny on difficult plans. Methods and Materials: Pretreatment QA data were collected for 973 cases between July 2018 and October 2020. The outcome variable, a degree of difficulty, was collected as a subjective rating by physicists who performed the pretreatment chart checks. Potential features were identified based on clinical relevance, contribution to plan complexity, and QA metrics. Five machine learning models were developed: support vector machine, random forest classifier, adaboost classifier, decision tree classifier, and neural network. These were incorporated into a voting classifier, where at least 2 algorithms needed to predict a case as difficult for it to be classified as such. Sensitivity analyses were conducted to evaluate feature importance. Results: The voting classifier achieved an overall accuracy of 77.4% on the test set, with 76.5% accuracy on difficult cases and 78.4% accuracy on less difficult cases. Sensitivity analysis showed features associated with plan complexity (number of fractions, dose per monitor unit, number of planning structures, and number of image sets) and clinical relevance (patient age) were sensitive across at least 3 algorithms. Conclusions: This approach can be used to equitably allocate plans to physicists rather than randomly allocate them, potentially improving pretreatment chart check effectiveness by reducing errors propagating downstream.

Prospective 5-Year Analysis of the United States Radiation Oncology Job Market Using the ASTRO Career Center Website.

PURPOSE: We provide 5-year results of prospectively collected radiation oncology (RO) job opportunities and a longitudinal assessment of RO graduate numbers within the United States. METHODS AND MATERIALS: Full-time domestic RO job opportunities were collected and categorized using the American Society for Radiation Oncology (ASTRO) Career Center from July 1, 2016 to June 30, 2021. A chi-square test was used to compare regional job availability by city size and position type. The corresponding number of graduating United States (US) RO residents (2017-2021) was collected. US census and Medicare database resources were used as comparators for population and workforce estimates. Pearson's correlation coefficients were used to examine changes in data over time and a 2-tailed t test was used to assess for statistical significance. RESULTS: Over the 5-year study period, 819 unique job offers were posted, compared with 935 RO graduates (0.88 total jobs-to-graduates ratio). Most jobs were nonacademic (57.6%), located in populated areas >1 million (57.1%; median: 1.57M), with the largest proportion of jobs seen in the South region (32.4%). One-third of academic jobs were located at satellites. Regional differences were seen between academic versus nonacademic job availability (P < .01), with the highest proportion of academic jobs seen in the Northeast (60.3%) and the lowest in the Midwest (34.5%). Differences between regions were also observed for jobs in areas >1 million versus ≤1 million (P < .01), with the most jobs in areas >1 million seen in the West (64.6%) and the least in the South (51.3%). Regional job availability over time did not differ by position type (academic vs nonacademic) or population area size (P = .11 and P = .27, respectively). Annual graduate numbers increased with time (P = .02), with the highest percentage of graduates trained in the South (30.8%). Regional distribution of jobs versus graduates significantly differed (P < .01) with the lowest jobs-to-graduates ratio observed in the Northeast (0.67) and highest ratio in the West (1.07). Regional RO workforce estimates based on the 4336 radiation oncologists who were Medicare providers in 2020 were compared with total jobs and graduates by region with no difference observed between the distributions of the workforce and jobs (P = .39), but comparisons between the workforce and graduates were proportionally different (P < .01). The number of total jobs (vs graduates) per 10 million population in the Northeast, South, Midwest, and West were 30.2 (45.1), 21.0 (22.7), 30.6 (33.4), and 22.6 (21.2), respectively. CONCLUSIONS: This multiyear quantitative assessment of the RO job market and graduates identified fewer job opportunities than graduates overall in most regions, most notably in the Northeast. Regional differences were seen between available job type (academic vs nonacademic) and population size (>1 million vs ≤1 million). The findings are worrisome for trainee oversupply and geographic maldistribution. The number and distribution of RO trainees and residency programs across the US should be evaluated to minimize job market imbalance for future graduates, promote workforce stability, and continue to meet the future societal needs of patients with cancer.

Feature Engineering for Interpretable Machine Learning for Quality Assurance in Radiation Oncology.

Chart checking is a time intensive process with high cognitive workload for physicists. Previous studies have partially automated and standardized chart checking, but limited studies implement data-driven approaches to reduce cognitive workload for quality assurance processes. This study aims to evaluate feature selection methods to improve the interpretability and transparency of machine learning models in predicting the degree of difficulty for a pretreatment physics chart check. We compare chi-square, mutual information, feature importance thresholding, and greedy feature selection for four different classifiers. Random forest has the highest performance with SMOTE oversampling using mutual information for feature selection (accuracy 84.0%, AUC 87.0%, precision 80.0%, recall 80.0%). This study demonstrates that feature selection methods can improve model interpretability and transparency.

Demographics of ASTRO Student Members and Potential Implications for Future U.S. Radiation Oncology Workforce Diversity.

PURPOSE: The radiation oncology workforce in the United States is comparatively less diverse than the U.S. population and U.S. medical school graduates. Workforce diversity correlates with higher quality care and outcomes. The purpose of this study was to determine whether student members of the American Society for Radiation Oncology (ASTRO) are any more diverse than resident members-in-training using the recently established medical student membership category. METHODS AND MATERIALS: Self-reported sex, race and Hispanic ethnicity, medical school, and degree(s) earned for all medical students (n = 268) and members-in-training (n = 713) were collected from the ASTRO membership database. International members were excluded. The χ2 test was used to assess for differences between subgroups. RESULTS: Compared with members-in-training, student members were more likely to be female (40.0% vs 31.5%, P = .032), black or African American (10.7% vs 4.8%, P = .009), candidates for or holders of a DO rather than MD degree (5.2% vs 1.5%, P = .002), and from a U.S. medical school that is not affiliated with a radiation oncology residency program (30.5% vs 20.9%, P = .001). There was no significant difference in self-reported Hispanic ethnicity (7.3% vs 5.4%, P = .356). There were no indigenous members in either category assessed. CONCLUSIONS: Medical student members of ASTRO are more diverse in terms of black race, female sex, and osteopathic training, though not in terms of Hispanic ethnicity or nonmultiracial indigenous background, than the members-in-training. Longitudinal engagement with these students and assessment of the factors leading to specialty retention versus attrition may increase diversity, equity, and inclusion in radiation oncology.

Impact of rescheduling hydrocodone-combination products in an urban Texas county healthcare system.

OBJECTIVE: The objective of this study was to examine the rescheduling of hydrocodone-combination products (HCPs) and associated changes in prescriber patterns in an urban county healthcare system in Texas. METHODS: Pharmacy data were obtained electronically for tramadol, hydrocodone-acetaminophen, and acetaminophen-codeine from 180 days before and after the schedule change on October 6, 2014. x2 and t tests were used to calculate the significance of changes between the medications over the studied time. RESULTS: Hydrocodone-acetaminophen saw a decline in dispense events and pills dispensed of 80.2 and 67.9 percent, respectively, in the immediate 30-day period following the scheduling change with a total decrease of 80.8 and 67.5 percent, respectively, in the 180-day period. Acetaminophen-codeine dispense events and total pills dispensed increased by 302.3 and 288.9 percent, respectively, in the immediate 30-day period while 180-day results experienced an increase of 215.1 and 209.8 percent, respectively. There were no major changes with tramadol. Additionally, an increase of 69.5 percent in pills per dispense event of hydroco-done-acetaminophen was noted in the 180-day period following the schedule change. CONCLUSION: The scheduling change of HCPs is associated with an immediate decrease in hydrocodone-acetaminophen use at our institution while a simultaneous rise in acetaminophen-codeine products was observed.

Machine Learning Applications in Head and Neck Radiation Oncology: Lessons From Open-Source Radiomics Challenges.

Radiomics leverages existing image datasets to provide non-visible data extraction via image post-processing, with the aim of identifying prognostic, and predictive imaging features at a sub-region of interest level. However, the application of radiomics is hampered by several challenges such as lack of image acquisition/analysis method standardization, impeding generalizability. As of yet, radiomics remains intriguing, but not clinically validated. We aimed to test the feasibility of a non-custom-constructed platform for disseminating existing large, standardized databases across institutions for promoting radiomics studies. Hence, University of Texas MD Anderson Cancer Center organized two public radiomics challenges in head and neck radiation oncology domain. This was done in conjunction with MICCAI 2016 satellite symposium using Kaggle-in-Class, a machine-learning and predictive analytics platform. We drew on clinical data matched to radiomics data derived from diagnostic contrast-enhanced computed tomography (CECT) images in a dataset of 315 patients with oropharyngeal cancer. Contestants were tasked to develop models for (i) classifying patients according to their human papillomavirus status, or (ii) predicting local tumor recurrence, following radiotherapy. Data were split into training, and test sets. Seventeen teams from various professional domains participated in one or both of the challenges. This review paper was based on the contestants' feedback; provided by 8 contestants only (47%). Six contestants (75%) incorporated extracted radiomics features into their predictive model building, either alone (n = 5; 62.5%), as was the case with the winner of the "HPV" challenge, or in conjunction with matched clinical attributes (n = 2; 25%). Only 23% of contestants, notably, including the winner of the "local recurrence" challenge, built their model relying solely on clinical data. In addition to the value of the integration of machine learning into clinical decision-making, our experience sheds light on challenges in sharing and directing existing datasets toward clinical applications of radiomics, including hyper-dimensionality of the clinical/imaging data attributes. Our experience may help guide researchers to create a framework for sharing and reuse of already published data that we believe will ultimately accelerate the pace of clinical applications of radiomics; both in challenge or clinical settings.