Ward, Kevin C.

PURPOSE Automated curation of breast cancer treatment data with minimal human involvement could accelerate the collection of statewide and nationwide evidence for patient management and assessing the effectiveness of treatment pathways. The primary challenges are the complexity and inconsistency of structured clinical data streams and accurate extraction of this information from free-text clinical narratives. MATERIALS AND METHODS We proposed a hybrid two-phase information extraction framework that combined a Unified Medical Language System parser (phase-1) with a fine-tuned large language model (LLM; phase-2) to extract longitudinal treatment timelines from time-stamped clinical notes. Our framework was developed through end-to-end joint learning as a question-answering model, where the model was trained to simultaneously answer five questions, each corresponding to a specific treatment. RESULTS We fine-tuned and internally validated the model on 26,692 patients with breast cancer (diagnosed between 2013 and 2020) receiving treatment at Mayo Clinic and externally validated the model on 162 randomly selected patients from Stanford Healthcare. Zero-shot LLM (out-of-the-box) had high specificity but low sensitivity, indicating that although these frameworks are useful for generic language understanding, they are lacking in terms of targeted clinical tasks. The proposed model achieved 0.942 average AUROC on the internal and 0.924 on the external data, demonstrating only marginal drop in performance when evaluated on external. The proposed model also achieved better trade-off between sensitivity (average: 79.2%) and specificity (average: 76.2%) compared with rule-based (average sensitivity: 70.5%, average specificity: 68.1%) and structured codes (average sensitivity: 64.1%, average specificity: 83.5%). CONCLUSION The proposed framework can extract temporal information about cancer treatments from various time-stamped clinic notes, regardless of the setting of treatment administration (inpatient or outpatient) or time frame. To support the cancer research community for such data curation and longitudinal analysis, we have packaged the code as a docker image, which needs minimal system reconfiguration and shared with an open-source academic license.

PURPOSE Patient-centered outcomes (PCOs) are pivotal in cancer treatment, as they directly reflect patients' quality of life. Although multiple studies suggest that factors affecting breast cancer–related morbidity and survival are influenced by treatment side effects and adherence to long-term treatment, such data are generally only available on a smaller scale or from a single center. The primary challenge with collecting these data is that the outcomes are captured as free text in clinical narratives written by clinicians. MATERIALS AND METHODS Given the complexity of PCO documentation in these narratives, computerized methods are necessary to unlock the wealth of information buried in unstructured text notes that often document PCOs. Inspired by the success of large language models (LLMs), we examined the adaptability of three LLMs, GPT-2, BioGPT, and PMC-LLaMA, on PCO tasks across three institutions, Mayo Clinic, Emory University Hospital, and Stanford University. We developed an open-source framework for fine-tuning LLM that can directly extract the five different categories of PCO from the clinic notes. RESULTS We found that these LLMs without fine-tuning (zero-shot) struggle with challenging PCO extraction tasks, displaying almost random performance, even with some task-specific examples (few-shot learning). The performance of our fine-tuned, task-specific models is notably superior compared with their non–fine-tuned LLM models. Moreover, the fine-tuned GPT-2 model has demonstrated a significantly better performance than the other two larger LLMs. CONCLUSION Our discovery indicates that although LLMs serve as effective general-purpose models for tasks across various domains, they require fine-tuning when applied to the clinician domain. Our proposed approach has the potential to lead more efficient, adaptable models for PCO information extraction, reducing reliance on extensive computational resources while still delivering superior performance for specific tasks.

Abstract Background: Racial disparities in breast cancer outcomes persist in the US, in which black women are more likely to die from breast cancer than white women. Despite levels of mammographic screening comparable to the US, Atlanta has a disproportionate burden of late-stage breast tumors and higher mortality rates, especially among black women. To better understand this disparity, we sought to identify tumor, treatment, and patient characteristics that may contribute to the observed differences in breast cancer mortality between black and white women in the metropolitan Atlanta area. Methods: In this study, patients were identified from the Georgia Cancer Registry. We included 4943 non-Hispanic white and 3580 non-Hispanic black women with an initial diagnosis of stage I-IV primary breast cancer between January 2010 and December 2014 in Atlanta. Cox proportional hazard regression was used to calculate the hazard ratios (HRs) and 95% confidence intervals (CIs) to compare black vs. white breast cancer mortality by tumor (stage, grade, ER status, and molecular subtype), treatment (surgery type, receipt of chemotherapy, radiation, hormone therapy, and trastuzumab), and patient (insurance, marital status, SES, and location of service) characteristics. We performed a mediation analysis to evaluate the contribution of subtype, stage, and socioeconomic status (SES) as mediators on the path between race and breast cancer mortality. Results: Compared with white breast cancer patients, black women were younger with tumors that were higher stage, higher grade, and more likely to be larger, node positive and triple negative. In fully adjusted models, black women diagnosed with Luminal A subtype were 58% more likely to die of breast cancer compared to their white counterparts (HR=1.58, 95%CI 1.31-2.00). Similarly, black women with private insurance were 60% more likely to die of breast cancer (HR=1.61, 95%CI 1.25-1.98), and black women in the highest SES group were more than twice as likely to die from breast cancer (HR=2.26, 95%CI 1.26-4.06) than white women with comparable SES. The smallest disparities in breast cancer mortality by race were observed among women without insurance, in the lowest SES index, or those diagnosed with triple-negative breast cancer. The mediation analysis showed 84.8% of the effect between race and breast cancer mortality was through stage, subtype, and SES. Conclusion: Our results indicate variation in racial disparities in breast cancer mortality by tumor and patient characteristics. Consistent with previous reports, our mediation results suggest that later stage and triple-negative subtype among black women are major contributors of the observed disparity. However, our results also shed new light on the disparity, suggesting that the largest disparities are observed among women with ER+ tumors amenable to adjuvant therapies and are most pronounced among women of high SES. More research is needed to understand the drivers of disparities in these treatable tumors. Citation Format: Lindsay J. Collin, Renjian Jiang, Kevin C. Ward, Keerthi Gogineni, Preeti D. Subhedar, Mark Sherman, Mia M. Gaudet, Carmen Radecki Breitkopf, Olivia D'Angelo, Sheryl Gabram-Mendola, Jolie Siegel, Rana Aneja, Anne Gaglioti, Lauren E. McCullough. Identification of factors contributing to breast cancer mortality disparities in the metropolitan Atlanta area [abstract]. In: Proceedings of the Eleventh AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2018 Nov 2-5; New Orleans, LA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2020;29(6 Suppl):Abstract nr C053.

e14608 Background: Beyond curative surgery, the optimal treatment for early stage pancreatic cancer is still a matter of debate. We conducted a comparative survival analysis between patients with resectable pancreatic cancer who received adjuvant treatment with either gemcitabine or 5-fluorouracil based chemotherapy and chemoradiation regimens. Methods: The Surveillance, Epidemiology and End Results (SEER)-Medicare database was used to identify patients with pancreatic cancer diagnosed from 1998 to 2005 who received curative surgery and adjuvant chemotherapy with either 5-fluorouracil or gemcitabine. These groups were subdivided by treatment with radiotherapy. Patients were followed until death, study endpoint or a maximum of five years after diagnosis. Results: 359 patients received 5-fluorouracil and 346 received gemcitabine. Compared to chemoradiation with 5-fluorouracil, outcomes for patients who received chemoradiation with gemcitabine did not differ. Patients who received gemcitabine without radiation had increased hazards (hazard ratio (HR) = 1.50 for high grade tumors (HGT), HR = 1.32 for low grade tumors (LGT)). However, outcomes of patients who received 5-fluorouracil without radiation varied with tumor grade. In LGT, patients had better outcomes with 5-fluorouracil when compared with chemoradiation with 5-fluorouracil (HR = 0.43). In HGT, the opposite was true (HR 2.10). Conclusions: Patients with low grade resectable pancreatic cancer may have better outcomes with 5-fluorouracil based chemotherapy without radiation when compared to 5-fluorouracil with radiation.

AbstractBackgroundRacial disparities in breast cancer (BC) outcomes persist where non-Hispanic black (NHB) women are more likely to die from BC than non-Hispanic white (NHW) women, and the extent of this disparity varies geographically. We evaluated tumor, treatment, and patient characteristics that contribute to racial differences in BC mortality in Atlanta, Georgia, where the disparity was previously characterized as especially large.MethodsWe identified 4943 NHW and 3580 NHB women in the Georgia Cancer Registry with stage I–IV BC diagnoses in Atlanta (2010–2014). We used Cox proportional hazard regression to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) comparing NHB vs NHW BC mortality by tumor, treatment, and patient characteristics on the additive and multiplicative scales. We additionally estimated the mediating effects of these characteristics on the association between race and BC mortality.ResultsAt diagnosis, NHB women were younger—with higher stage, node-positive, and triple-negative tumors relative to NHW women. In age-adjusted models, NHB women with luminal A disease had a 2.43 times higher rate of BC mortality compared to their NHW counterparts (95% CI = 1.99 to 2.97). High socioeconomic status (SES) NHB women had more than twice the mortality rates than their white counterparts (HR = 2.67, 95% CI = 1.65 to 4.33). Racial disparities among women without insurance, in the lowest SES index, or diagnosed with triple-negative BC were less pronounced.ConclusionsIn Atlanta, the largest racial disparities are observed in luminal tumors and most pronounced among women of high SES. More research is needed to understand drivers of disparities within these treatable features.