Analysis of Artificial Intelligence-Based Automated Phenotyping of Coronary Artery Disease Diagnoses for Patients who Received Radiotherapy for Thoracic Malignancies

PURPOSE/OBJECTIVE(S)

Coronary artery disease (CAD) is a critical risk factor and toxicity endpoint in thoracic radiation oncology. The recent advent of artificial intelligence (AI) algorithms - specifically natural language processing (NLP) - allows for automated extraction of this data from electronic health records (EHR) more efficiently and at scale. We report on the application and performance characteristics of an EHR-based automated phenotyping tool to a cohort of lung cancer patients receiving radiation therapy (RT).

MATERIALS/METHODS

We applied to this radiation oncology use-case an AI-based disease phenotyping tool developed by other researchers for phenotyping CAD from the EHR. The CAD model was trained on 191,187 patients with at least one ICD code and validated with 158 gold-standard patients (Sensitivity: 73%, Specificity: 93%, PPV: 90%). NLP disease definitions were determined by UMLS and were employed with ICD codes through PheNorm classification method to create the phenotype algorithm (Yue et al., J Am Med Inform Assoc. 2018 Jan 1; 25(1):54-60). We applied the CAD phenotyping tool to two independent cohorts of RT treated lung cancer patients: 1) an annotated set of 677 patients who received RT from 1998 to 2014 and were studied in prior cardiac toxicity research; 2) an independent set of 209 patients who received RT from 1998 to 2018 and were annotated separately.

RESULTS

Only 7.7% of the AI training population had a CAD diagnosis code, compared to 32.5% and 65.0% in the 677 patient and 209 patient cohorts, respectively. Validation of the AI predicted CAD is shown in the table below. Performance on the 209 patient cohort best matched the AI training set. The 677-patient cohort differed from the 209-patient cohort in that its CAD definition included patients with radiographic coronary artery calcifications, which accounted for the difference in cohort sensitivities, F-scores, and NPVs.

CONCLUSION

The AI calculated CAD was somewhat accurate in detecting whether a patient had CAD and was very accurate in confirming no CAD. The differences in sensitivity and NPV reflect differences in defining CAD, but this study demonstrates the potential for automated AI-phenotyping to detect pre-treatment risk factors and cardiac toxicity events from the EHR.

Association of Parental Status and Gender with Burden of Multidisciplinary Tumor Boards

PURPOSE/OBJECTIVE(S)

Tumor boards are an integral part of the management of patients with cancer. However, there is limited data investigating the burden of tumor boards on physicians. Our objective was to determine what physician-related, and tumor board-related factors associate with higher burden.

MATERIALS/METHODS

Tumor board start times were collected by email from 22 National Cancer Institute-designated cancer centers and/or U.S. World and News Report Top 40 hospitals for cancer. Tumor board burden was assessed by a cross-sectional convenience survey posted on social media and by email to Cedars-Sinai Medical Center cancer physicians between March 3, 2022, and April 3, 2022. Tumor board burden was measured on a 4-point scale (1, not at all; 2, slightly; 3, moderately; 4, very burdensome). Univariable and multivariable analyses were performed using a probabilistic index model.

RESULTS

The timing of 392 tumor boards was collected from 22 institutions. The most common tumor board start time was at or before 0730 (24.6%). Surveys were completed by 111 physicians, of which 52.3% identified as women and 42.3% as men. Reported specialties were radiation oncology (39.6%), medical oncology (18.0%), surgery (15.3%), radiology (12.6%), and pathology (9.9%). On average, 41.4% attended ≥3 hours/week total of tumor boards and 1-2 hours/week of early/late tumor boards (defined as starting before 0800 or 1700 or after). Overall, 37.8% reported tumor boards were at least moderately burdensome. On multivariable analysis, radiology/pathology specialty (probability 0.68; 95% confidence interval [CI], 0.54-0.79; p = 0.015), attending ≥3 hours/week of tumor boards (probability 0.68; 95% CI, 0.58-0.76; p<.001), and having ≥2 children (probability 0.65; 95% CI, 0.52-0.77; p = 0.029), were associated with higher burden. Early/late tumor boards were frequently considered burdensome (20.7% moderately, 29.7% very burdensome). On multivariable analysis, identifying as a woman (probability 0.69; 95% CI, 0.57-0.78; p = 0.003) and having children (probability 0.75; 95% CI, 0.62-0.84; p<.001) remained associated with a higher level of burden from early/late tumor boards. Further, parents frequently reported that early/late tumor boards negatively affected childcare (55.8%), feeding and/or sleep logistics (33.8%), and overall family dynamics (63.7%).

CONCLUSION

Identifying as a woman and having children were associated with a higher level of burden from early/late tumor boards. The negative impact of early/late tumor boards on overall family dynamics, including children feeding, sleeping, and childcare logistics, was commonly reported by parents. Having ≥2 children, attending ≥3 hours/week of tumor boards, and radiology/pathology specialty were associated with a higher level of burden overall. Future strategies should aim to decrease burden, particularly the disparate impact on parents and women.

Variation in Left Anterior Descending Coronary Artery Accumulated Dose Estimates during Breast Cancer Radiotherapy

PURPOSE/OBJECTIVE(S)

Left anterior descending (LAD) coronary artery RT dose has been associated with the risk of coronary ischemic events in patients with breast cancer treated with radiotherapy (RT). However, consensus dose constraints commonly utilize mean heart dose, which has been shown to be an inadequate surrogate for LAD dose. Given the LAD is adjacent to the steep dose gradients of the left breast/chest wall, we hypothesize that variations in patient positioning or depth of breath hold may contribute to significant deviations in daily LAD dose exposure compared to predicted. Our objective was to investigate variations in accumulated LAD dose in patients with left-sided breast cancer treated with RT.

MATERIALS/METHODS

Retrospective analysis of 10 consecutive patients with left-sided breast cancer treated between 2019-2022 with volumetric modulated arc therapy (VMAT) in the supine position. RT was delivered using daily cone beam computed tomography (CBCT) image guidance with deep-inspiratory breath hold technique and an optical surface monitoring system. Daily CBCT scans were individually registered to each planning CT based on daily positional changes. The LAD was manually segmented and transformed to each CBCT. Daily fractional dose was calculated (mean, volume receiving 15 Gy [V15 Gy], V30 Gy, and max) and summed to produce an accumulated dose which was compared to the predicted dose. Significant deviations in accumulated dose were defined as at least ±15% from predicted.

RESULTS

The RT targets included breast/chest wall only (n = 1), supraclavicular nodes (n = 8), and/or internal mammary chain (n = 5). All plans were prescribed to 50 Gy in 25 fractions. The median predicted mean heart dose was 5.1 Gy. Overall, there were no significant differences between the median predicted vs. accumulated LAD doses: mean 10.4 vs. 10.2 Gy, V15 Gy 21% vs. 25%, V30 Gy 0% vs. 1%, max 24.5 vs 25.7 Gy (all p>.05). However, there was a subset of patients (n = 5, 50%) with significant deviations in accumulated vs. predicted dose (at least ±15%). For LAD mean, n = 2 had higher accumulated vs. predicted doses (16.6 vs. 11.6 Gy; 14.6 vs. 12.8 Gy), while n = 3 had lower (21.8 vs. 26.5 Gy; 2.7 vs. 4.2 Gy; 13.9 vs. 16.5 Gy). For LAD V15 Gy, n = 3 had higher accumulated vs. predicted doses (43 vs. 35%; 51% vs. 25%; 14% vs. 9%), while n = 2 had lower (49% vs. 61%; 37% vs. 56%). For LAD V30 Gy, n = 4 had higher accumulated vs. predicted doses (12% vs. 5%; 5% vs. 0%; 4% vs. 0%; 4% vs. 2%), while n = 1 had lower (32% vs. 44%).

CONCLUSION

Daily setup differences, including the extent of inspiratory breath hold, may contribute to deviations in accumulated LAD dose (more than 15% of predicted) in approximately half of patients and were more pronounced in V15 and V30 Gy metrics. The potential for significant LAD dose uncertainty in a clinically meaningful subset of patients should be recognized and warrants further analysis in an expanded cohort.

Deployment of a Deep Learning Automated Cardiac Sub-Structure Contouring Algorithm to Measure Coronary Dose Exposure Trends in Lung Cancer Radiation Therapy

PURPOSE/OBJECTIVE(S)

Radiation dose exposure to the left anterior descending coronary artery (LAD) is associated with major adverse cardiac events (MACE) and all-cause mortality (ACM) after radiotherapy (RT) for locally-advanced non-small cell lung cancer (LA-NSCLC). Segmentation of cardiac sub-structures has not been routinely performed historically, and the purpose of the study was to deploy a deep learning (DL)-based automated cardiac sub-structure contouring algorithm to calculate LAD radiation dose exposure and understand temporal trends in a large population.

MATERIALS/METHODS

We previously trained a DL algorithm to segment cardiac substructures, using research contours from 560 LA-NSCLC patients (pts), and validated the LAD auto-segmentation algorithm geometrically, dosimetrically and with outcomes. In this study, we deployed the DL algorithm to auto-segment the LAD and heart on planning CT scans in an independent cohort of 1708 consecutive LA-NSCLC pts treated at an academic center and 4 affiliated community centers from 2014-21. We calculated the LAD volume receiving 15 Gy (V15), and mean heart dose (MHD) using the DL contours and the treated plan. We analyzed the distribution of key dosimetric values for heart (LAD V15, MHD) and lung (V20, mean lung dose [MLD], V5) over time and used the Mann-Kendall statistical test for trend.

RESULTS

The median prescribed dose was 50 Gy. 63% of patients were treated with intensity-modulated radiation therapy (IMRT) and 37% 3D-conformal (Table), with increasing IMRT use over time. The median lung V20, V5, MLD, and MHD did not significantly change over time (p>0.05). The median LAD V15 decreased significantly (p<0.05 for trend) in 2020-1, with significantly more cases achieving a V15<10% compared to prior years (p<0.05). The LAD distribution shift coincides with the 2020 publication of a study showing an association of V15<10% with decreased risk of MACE/ACM.

CONCLUSION

DL-based auto-segmentation of cardiac substructures can be deployed at large-scale to study trends in dose exposure to newly identified critical organs-at-risk such as the LAD at the population level. Lowering LAD dose did not impact other cardiopulmonary constraints, but further work is needed to study impact on outcomes.

Non-Redundant Mechanisms of Immune Resistance to Radiotherapy Converge on Innate Immunity

PURPOSE/OBJECTIVE(S)

Despite evidence of preclinical synergy between radiotherapy (RT) and immune checkpoint blockade (ICB), randomized trials of RT/ICB have demonstrated limited benefit in solid tumors. We performed single-cell RNA sequencing (scRNA-seq) and CITE-seq (cellular indexing of transcriptomes and epitopes) to address the discordance between preclinical and clinical data. We hypothesized that multiple orthogonal inhibitory immune pathways restrain the local and systemic efficacy of RT beyond T-cell oriented immune checkpoints.

MATERIALS/METHODS

We used the EO771 syngeneic murine model of breast cancer to characterize the immune tumor microenvironment following RT with or without ICB. RT (16 Gy x 1) was delivered using the X-RAD SmART platform with CT image guidance. Neutralizing antibodies (anti-PD-1/Ly6G/Gr-1/CD47) were delivered by intraperitoneal injections. scRNA-seq analysis were performed by Seurat and BBrowser (BioTuring).

RESULTS

We found that adaptive ICB (anti-PD-1) reprogrammed the immune response to RT by promoting an M1-like interferon-primed state (ISG15, CXCL10) in tumor associated macrophages (TAMs) and by increasing the late recruitment of intratumoral neutrophils. Given that neutrophils may drive resistance to RT in other models, we evaluated the effect of intratumoral neutrophil depletion using anti-Ly6G or anti-Gr-1 on the antitumor efficacy of RT/ICB. Both neutrophil depletion strategies led to enhanced tumor control and improved survival in advanced EO771 tumors compared to RT/ICB alone (P<0.001). In parallel to this approach, we found that TAMs upregulated several innate immune checkpoints including SIRPα in response to RT. Disruption of the SIRPα-CD47 interaction by anti-CD47 antibodies similarly enhanced the antitumor efficacy of RT/ICB by improving tumor control and survival (P<0.001). Using scRNA-seq and unbiased clustering, we found that anti-CD47 eliminated an entire cluster of chronically inflamed TAMs, characterized by pro-inflammatory markers (IL1A, NOS2) and chemokines (CCL3, CXCL1/2/3). Anti-CD47 also reduced intratumoral neutrophils by eliminating a cluster of pathologically activated neutrophils, termed myeloid-derived suppressor cells (PMN-MDSCs) that expressed several markers of ferroptosis (TFRC, PTGS2, SLC3A2). Consistent with the potent immunosuppressive capacity of PMN-MDSCs, we found that anti-CD47 increased tumor-infiltrating lymphocytes including central memory TCF7+ T cells and CD19+ B cells. Lastly, by inference and analysis of cell-cell communication (CellChat), we found that anti-CD47 strengthened the interactions between TAMs and CD8+ T cells compared to RT/ICB alone.

CONCLUSION

Our data collectively indicate that resistance to RT/ICB in the EO771 model Is driven by innate immune cells including neutrophils and chronically inflamed TAMs. Targeted disruption of the CD47-SIRPα axis is a promising approach to overcoming immune resistance by reprogramming TAMs and eliminating PMN-MDSCs.

Left Anterior Descending Coronary Artery-Optimized Knowledge-Based Radiotherapy Planning in Patients with Locally Advanced Non-Small Cell Lung Cancer

PURPOSE/OBJECTIVE(S)

The left anterior descending (LAD) coronary artery volume receiving 15 Gy (V15 Gy) has been associated with major adverse cardiac events and mortality in patients with locally advanced non-small cell lung cancer (LA-NSCLC) treated with radiotherapy (RT). However, coronary arteries are not routinely contoured as standard-of-care and there is limited data describing coronary-sparing planning techniques. Our objective was to evaluate whether LAD-sparing could be improved compared to real-world plans using a commercial knowledge-based planning (KBP) software without clinically unacceptable tradeoffs to other critical organs at risk (OAR).

MATERIALS/METHODS

Retrospective replanning analysis was conducted on 31 RT plans developed for patients with LA-NSCLC. Each case was manually replanned with volumetric modulated arc radiotherapy (VMAT) optimized for LAD V15 Gy sparing with a prescription dose of 60 Gy. The resulting plans were used to create an LAD-optimized KBP model. The model included both KBP-generated objectives and additional point objectives to meet LAD (V15 Gy < 10%), lung (V5 Gy < 60%, V20 < 30%, mean < 17 Gy), heart (mean < 20 Gy), spinal cord (D0.03cc < 45 Gy), esophagus (V60 Gy < 17%, mean < 34 Gy), and planning target volume (PTV; D95% = 100%). The LAD-optimized model was deployed to replan 33 additional VMAT plans for LA-NSCLC with prescription dose 60 Gy, delivered between 2018 and 2022. Dose constraints or objectives were only unmet with KBP when not met in the original plan. Descriptive statistics were used to compare dosimetric parameters between original and LAD-optimized KBP plans.

RESULTS

For the 33 plans identified for LAD-optimized KBP replanning, the original plans had a mean LAD V15 Gy of 17.9% (standard error [SE], 3.9%), while the LAD-optimized KBP plans had a significantly lower mean LAD V15 Gy of 5.9% (SE, 1.9%; p = 0.031). There was no significant difference in mean heart dose between the original and LAD-optimized KBP plans (8.4 Gy [SE 6.6 Gy] vs. 7.3 Gy [SE 5.8 Gy]; p = 0.48). Of these 33 KBP plans, 42% (n = 14) had mean reductions in LAD V15 Gy from 40.5% (SE 4.3%) to 10.5% (SE 3.5%; p = 0.0002), with 9/14 now able to meet < 10%, without affecting other critical OAR objectives (p > 0.5 for lungs, esophagus, spinal cord). 45% (n = 15) originally had an LAD V15 Gy < 1%, which were unchanged using KBP (p > 0.3). Four plans (12%) had higher LAD V15 Gy in the KBP vs. original plan (2/4 still meeting < 10%), due to the model prioritizing other OARs to a greater degree.

CONCLUSION

Improvements in LAD-sparing was achievable in more than 40% of plans without apparent clinically significant tradeoffs in other critical OARs and without corresponding change in MHD. This result underscores the importance of standard implementation of LAD contouring and dose constraints, particularly given the discordance between whole heart and LAD dose exposure.

Association of Left Anterior Descending Coronary Artery Calcium Progression and Radiation Dose with Major Adverse Cardiac Events in Breast Cancer

PURPOSE/OBJECTIVE(S)

Coronary artery calcium (CAC) is associated with increased risk of major adverse cardiac events (MACE). Accelerated CAC progression has been observed in patients with breast cancer after radiotherapy (RT) and there is a relationship between left anterior descending (LAD) coronary artery RT dose and the risk of coronary events. However, there is lack of consensus on LAD dose constraints for breast RT and limited data on the extent and impact of CAC progression. Our objective was to evaluate the association of LAD dose exposure and CAC progression with the risk of MACE in patients with breast cancer following RT.

MATERIALS/METHODS

Retrospective analysis of 181 patients with breast cancer treated with RT between 2008 and 2019. CAC was manually measured on RT planning and follow-up CTs (with at least one-year interval) using the Agatston method. Coronary arteries were segmented using a deep learning-based automated algorithm and dosimetric parameters collected. MACE cumulative incidence was estimated, and Fine and Gray regressions performed, accounting for non-cardiac death as a competing risk.

RESULTS

The median follow-up following RT was 70 months (interquartile range [IQR], 53-86). The median age was 63 years (IQR, 53-72), 43% had hypertension, 40% hyperlipidemia, 8% coronary heart disease (CHD). Most had pathologic stage I-II disease (76%). RT was targeted to breast/chest wall only in 60% and included regional nodes in 40% (internal mammary chain in 4%). The most common dose/fractionation was 48-50.4 Gy/25-28 fractions (67%) and 42.6-42.7 Gy/16 fractions (30%). At the time of RT, 68 (38%) had at least moderate CAC burden (CAC >100; statin-therapy indicated), but only 29 (43%) were on statin therapy. At a median interval of 44 months (IQR, 26-63), 55% (n = 84) had CAC progression, with a median increase of 52%/year (IQR, 18-193). The median time to MACE was 68 months (IQR, 53-85), with a 5-year cumulative incidence of 7.3% (15 MACE overall). Accounting for age and CHD, there was an increased risk of MACE with LAD CAC progression (subdistribution hazard ratio [SHR] 1.02/10 CAC points; 95% confidence interval [CI] 1.01 = 1.03; p = .007) and the volume of LAD receiving 15 Gy (LAD V15 Gy; SHR 1.03/%; 95% CI, 1.01-1.06; p = .004). There was no association between mean heart dose, chemotherapy, or Her2 therapy exposure and MACE (p>.05).

CONCLUSION

LAD CAC progression and LAD V15 Gy dose exposure were associated with an increased risk of MACE following RT. Accelerated CAC progression was commonly observed, however most patients were under-optimized for cardiovascular (CV) risk, with less than half of statin-eligible patients with at least moderate CAC burden on statin therapy. Together, these data support more aggressive cardiac risk mitigation approaches, including guidelines-based CV risk factor modification and optimized sparing of LAD radiation dose.

Association of Left Ventricular Myocardium Dose Exposure with Heart Failure in Lung Cancer Radiotherapy

PURPOSE/OBJECTIVE(S)

The volume of left anterior descending (LAD) coronary artery dose receiving 15 Gy (LAD V15 Gy) has been associated with major adverse cardiac events (MACE; a composite endpoint of unstable angina, heart failure hospitalization/urgent visit, myocardial infarction, coronary revascularization, and cardiac death) and mortality in patients with locally advanced non-small cell lung cancer (LA-NSCLC) following radiotherapy (RT). With growing evidence supporting specific cardiac substructure dose exposure with distinct pathophysiologic cardiac endpoints, our objective was to assess whether RT dose to the left ventricular myocardium (LVM) was associated with the specific cardiac endpoints of heart failure (HF) vs. overall MACE.

MATERIALS/METHODS

Retrospective analysis of 701 patients with LA-NSCLC treated with RT between 2003 and 2014. LAD was segmented manually, while LVM (excluding blood pool) was segmented using an open-source deep learning-based automated algorithm (TotalSegmentator; Wasserthal et al, 2022) and dosimetric parameters (mean, max, and VX Gy in 5 Gy increments) collected. HF was defined as any grade ≥3 Common Terminology Criteria for Adverse Event (CTCAE) HF or cardiomyopathy. Receiver operating curve and cut-point analyses estimating HF were performed. Fine and Gray and Cox regressions, adjusting for preexisting cardiovascular (CV) risk factors and other prognostic factors, were calculated.

RESULTS

Of 701 patients, the median age was 65 years (interquartile range [IQR], 57-73), 51% were men. At a median follow up of 20 months (IQR, 8-45) following RT, there were 64 HF events. The median time to HF was 20 months (IQR, 9-43), with a 2-year cumulative incidence of 5.2%. The optimal cut-point for the LVM dose variable with the highest C-index predicting HF was LVM V10 Gy ≥10% (.58). By LVM V10 Gy cut point ≥10% vs <10%, the 2-year HF cumulative incidence was 7.2% vs 4.0% (p = .005). Adjusting for baseline CV risk, there was an increased risk of HF with LVM V10 Gy ≥10% (vs. <10%) (subdistribution hazard ratio [SHR] 2.81; 95% confidence interval [CI], 1.02-7.70; p = .045), but not LAD V15 Gy ≥10% (p>.3). Conversely, adjusting for baseline CV risk, there was an increased risk of MACE with LAD V15 Gy ≥10% (SHR = 19.70; 95% CI = 2.57-151.35; p = 0.004), but not LVM V10 Gy ≥10% (p>0.3). On all-cause mortality analysis, adjusting for CV and other prognostic factors, LAD V15 Gy ≥10% (HR = 1.4; 95% CI = 1.02-1.80; p = 0.036), but not LVM ≥10% (p>0.5), remained significantly associated with mortality.

CONCLUSION

Left ventricular myocardium V10 ≥10%, is associated with the specific endpoint of HF, but not overall MACE, while LAD V15 ≥10% remained strongly associated with MACE and all-cause mortality, but not HF. These findings underscore the need to study the interaction between RT dose to specific substructures (i.e., myocardium) and pathophysiologically relevant cardiac endpoints (i.e., HF).

Feasibility and Acceptability of ChatGPT Generated Radiology Report Summaries for Cancer Patients

PURPOSE/OBJECTIVE(S)

Patients now have direct access to their diagnostic imaging reports. However, they can include complex terminology that can be difficult for patients to understand. ChatGPT (OpenAI, San Francisco, CA) is an artificial intelligence (AI) text-generating model that can simplify complex text and generate human-like responses. We assessed ChatGPT's ability to generate summarized MRI reports for patients with prostate cancer and evaluated physician satisfaction with providing patients with an AI-summarized report.

MATERIALS/METHODS

We used ChatGPT to summarize five prostate cancer MRI reports performed at our institution from 2021-2022. Using a standard prompt, we asked ChatGPT to summarize the full MRI reports into a patient letter at a 6th grade reading level. To account for variability in text output, we generated three different summarized reports per unique MRI report. Full MRI and summarized reports were assessed for readability using Flesch-Kincaid Grade Level (FK) score. Radiation oncologists at our institution were asked to evaluate the summarized reports with an anonymous questionnaire. Physicians were shown two full MRI reports and three summarized versions for each full report. For each summarized report, physicians were asked six questions assessing the following: factual correctness, ease of understanding, completeness, potential for harm, overall quality, and likelihood they would send the report to a patient. Qualitative responses were given on a 1-5 Likert-type scale.

RESULTS

A total of 15 summarized reports were generated from five full MRI reports using ChatGPT. The median FK score for the full MRI reports and summarized reports was 9.6 vs. 5.0, (p<0.05), respectively. 12 radiation oncologists responded to our questionnaire with experience levels of: resident (25%), attending <5 years (33%), attending 5-10 years (17%), and attending >10 years (25%). The mean [SD] rating across all six summarized reports for each of the questions were: factual correctness (4.0 [0.6], understanding 4.0 [0.7]), completeness (4.1 [0.5]), potential for harm (3.5 [0.9]), overall quality (3.4 [0.9]), and likelihood to send to patient (3.1 [1.1]). 89%, 78%, and 93% of respondents answered agree or strongly agree for correctness, ease of understanding, and completeness of the summarized reports. 51%, 53%, and 46% of respondents answered agree or strongly agree for potential for harm, overall quality, and likelihood to send to patient.

CONCLUSION

ChatGPT was able to summarize prostate MRI reports at a reading level appropriate for patients. Physicians were likely to be satisfied with the summarized reports with respect to factual correctness, ease of understanding, and completeness. They were less likely to be satisfied with respect to potential for harm, overall quality, and likelihood to send to patients. Further research is needed to optimize ChatGPT's ability to summarize radiology reports for patients and understand what factors influence physician trust in AI-summarized reports.

How medical students can bring about curricular change

Traditionally, medical school committees have been charged with curricular improvement and modification, while medical students have had little or no involvement in reform efforts. However, medical students can sometimes be ahead of faculty in recognizing new topics that need to be covered, and their energy, commitment, and vision can be a very important impetus for curricular change. In 1995-96, as part of a general curricular restructuring effort, faculty at Dartmouth Medical School began to design and offer new electives in innovative topics, with the idea that electives might become part of the required curriculum if the material presented in them were deemed to be "core." Students were invited to organize their own electives if a topic in which they were interested was not being covered. The authors (two were second-year medical students and the third was their faculty sponsor) developed an elective in women's health. This paper describes the development and implementation of this elective, and the process by which the course was later made part of the required curriculum at Dartmouth. The success of the authors' efforts highlights the crucial role students can play in reforming medical curricula.