Sloan-Kettering Research Articles

Shareable artificial intelligence to extract cancer outcomes from electronic health records for precision oncology research.

Databases that link molecular data to clinical outcomes can inform precision cancer research into novel prognostic and predictive biomarkers. However, outside of clinical trials, cancer outcomes are typically recorded only in text form within electronic health records (EHRs). Artificial intelligence (AI) models have been trained to extract outcomes from individual EHRs. However, patient privacy restrictions have historically precluded dissemination of these models beyond the centers at which they were trained. In this study, the vulnerability of text classification models trained directly on protected health information to membership inference attacks is confirmed. A teacher-student distillation approach is applied to develop shareable models for annotating outcomes from imaging reports and medical oncologist notes. 'Teacher' models trained on EHR data from Dana-Farber Cancer Institute (DFCI) are used to label imaging reports and discharge summaries from the Medical Information Mart for Intensive Care (MIMIC)-IV dataset. 'Student' models are trained to use these MIMIC documents to predict the labels assigned by teacher models and sent to Memorial Sloan Kettering (MSK) for evaluation. The student models exhibit high discrimination across outcomes in both the DFCI and MSK test sets. Leveraging private labeling of public datasets to distill publishable clinical AI models from academic centers could facilitate deployment of machine learning to accelerate precision oncology research.

7076 Inhibiting Oxidosqualene Cyclase, An Enzyme In The Cholesterol Biosynthetic Pathway, Suppresses Lung Metastasis Of Triple-Negative Breast Cancer Cells

Abstract Disclosure: Y. Liang: None. S.M. Hyder: None. Triple-negative breast cancers (TNBC) lack ER, PR and Her2neu proteins that are commonly targeted in breast cancer therapies. As a consequence, women who suffer from TNBC have poor prognosis and limited treatment options, and are usually administered toxic, non-specific chemotherapeutic agents. Drug-resistance almost always occurs, leading to tumor metastasis, the main cause of patient death. New non-toxic therapeutic strategies to control metastasis of TNBC are urgently needed. Cholesterol is an essential structural and functional component of cell membranes that is necessary for the growth of both primary tumors and metastatic colonies that reside in distant organs. Therefore, inhibiting cholesterol production is an attractive therapeutic strategy. Treatment with statins, a class of cholesterol biosynthesis inhibitors that target HMGCoA-reductase, is associated with certain undesirable side effects; consequently, we targeted oxidosqualene cyclase (OSC), an enzyme that occurs downstream of HMGCoA-reductase in the cholesterol biosynthetic pathway, to disrupt tumor metastasis. Potent small molecule inhibitors of OSC have been identified. RO 48-8071 (4′-[6-(allylmethylamino)hexyloxy]-4-bromo-2′-fluorobenzophenone fumarate) (RO), has emerged as a useful chemotherapeutic agent for treating several forms of primary tumors. We developed a mouse model for studying lung metastasis using a MDA-MB-231 variant (LM2) TNBC cell line obtained from Dr. Massagué (Memorial Sloan-Kettering Cancer Center). These are aggressive cells, and injection of only 2 X 105 cells produces detectable lung metastatic colonies in approximately 4-6 weeks. We inoculated female nu/nu mice, 5-6 weeks old, with 2 X 105 MDA-MB-231/LM2 cells in 0.1 ml DMEM/F12 medium without serum via tail vein. Five days after tumor cell inoculation animals received 20 mg/kg (n=6), 40 mg/kg (n=7) RO (ip) or vehicle alone (n=7) every other day for two weeks. Animals were then sacrificed, and both lungs harvested and immediately fixed in Bouin’s fixative. Metastatic colonies on the lung surface were identified and counted under a dissecting microscope. Our study showed that RO effectively reduced lung metastasis without affecting animal weight. Compared with 56 + 7 colonies (Mean ± SEM) in controls given vehicle alone, RO significantly reduced lung metastatic colonies to 28 ± 5 in animals exposed to 20 mg/kg and 6 ± 1 in 40 mg/kg treated animals (p<0.05, ANOVA). Reduction in number of colonies in 40 mg/kg group was also significantly different from colonies counted in 20 mg/kg treated animals. Our study is the first to demonstrate that disruption of cholesterol biosynthesis in a model of human TNBC leads to reduced metastasis in lungs. Further studies will determine mechanisms through which RO suppresses TNBC metastasis to the lungs. Supported by Zalk Missouri Professor Endowment Funds. Presentation: 6/3/2024

7076 Inhibiting Oxidosqualene Cyclase, an Enzyme in the Cholesterol Biosynthetic Pathway, Suppresses Lung Metastasis of Triple-negative Breast Cancer Cells

Abstract Disclosure: Y. Liang: None. S.M. Hyder: None. Triple-negative breast cancers (TNBC) lack ER, PR and Her2neu proteins that are commonly targeted in breast cancer therapies. As a consequence, women who suffer from TNBC have poor prognosis and limited treatment options, and are usually administered toxic, non-specific chemotherapeutic agents. Drug-resistance almost always occurs, leading to tumor metastasis, the main cause of patient death. New non-toxic therapeutic strategies to control metastasis of TNBC are urgently needed. Cholesterol is an essential structural and functional component of cell membranes that is necessary for the growth of both primary tumors and metastatic colonies that reside in distant organs. Therefore, inhibiting cholesterol production is an attractive therapeutic strategy. Treatment with statins, a class of cholesterol biosynthesis inhibitors that target HMGCoA-reductase, is associated with certain undesirable side effects; consequently, we targeted oxidosqualene cyclase (OSC), an enzyme that occurs downstream of HMGCoA-reductase in the cholesterol biosynthetic pathway, to disrupt tumor metastasis. Potent small molecule inhibitors of OSC have been identified. RO 48-8071 (4′-[6-(allylmethylamino)hexyloxy]-4-bromo-2′-fluorobenzophenone fumarate) (RO), has emerged as a useful chemotherapeutic agent for treating several forms of primary tumors. We developed a mouse model for studying lung metastasis using a MDA-MB-231 variant (LM2) TNBC cell line obtained from Dr. Massagué (Memorial Sloan-Kettering Cancer Center). These are aggressive cells, and injection of only 2 X 10[5] cells produces detectable lung metastatic colonies in approximately 4-6 weeks. We inoculated female nu/nu mice, 5-6 weeks old, with 2 X 10[5] MDA-MB-231/LM2 cells in 0.1 ml DMEM/F12 medium without serum via tail vein. Five days after tumor cell inoculation animals received 20 mg/kg (n=6), 40 mg/kg (n=7) RO (ip) or vehicle alone (n=7) every other day for two weeks. Animals were then sacrificed, and both lungs harvested and immediately fixed in Bouin’s fixative. Metastatic colonies on the lung surface were identified and counted under a dissecting microscope. Our study showed that RO effectively reduced lung metastasis without affecting animal weight. Compared with 56 + 7 colonies (Mean ± SEM) in controls given vehicle alone, RO significantly reduced lung metastatic colonies to 28 ± 5 in animals exposed to 20 mg/kg and 6 ± 1 in 40 mg/kg treated animals (p<0.05, ANOVA). Reduction in number of colonies in 40 mg/kg group was also significantly different from colonies counted in 20 mg/kg treated animals. Our study is the first to demonstrate that disruption of cholesterol biosynthesis in a model of human TNBC leads to reduced metastasis in lungs. Further studies will determine mechanisms through which RO suppresses TNBC metastasis to the lungs. Supported by Zalk Missouri Professor Endowment Funds. Presentation: 6/3/2024

A Nomogram Using Imaging Features to Predict Ipsilateral Breast Tumor Recurrence After Breast-Conserving Surgery for Ductal Carcinoma In Situ.

To develop a nomogram that integrates clinical-pathologic and imaging variables to predict ipsilateral breast tumor recurrence (IBTR) in women with ductal carcinoma in situ (DCIS) treated with breast-conserving surgery (BCS). This retrospective study included consecutive women with DCIS who underwent BCS at two hospitals. Patients who underwent BCS between 2003 and 2016 in one hospital and between 2005 and 2013 in another were classified into development and validation cohorts, respectively. Twelve clinical-pathologic variables (age, family history, initial presentation, nuclear grade, necrosis, margin width, number of excisions, DCIS size, estrogen receptor, progesterone receptor, radiation therapy, and endocrine therapy) and six mammography and ultrasound variables (breast density, detection modality, mammography and ultrasound patterns, morphology and distribution of calcifications) were analyzed. A nomogram for predicting 10-year IBTR probabilities was constructed using the variables associated with IBTR identified from the Cox proportional hazard regression analysis in the development cohort. The performance of the developed nomogram was evaluated in the external validation cohort using a calibration plot and 10-year area under the receiver operating characteristic curve (AUROC) and compared with the Memorial Sloan-Kettering Cancer Center (MSKCC) nomogram. The development cohort included 702 women (median age [interquartile range], 50 [44-56] years), of whom 30 (4%) women experienced IBTR. The validation cohort included 182 women (48 [43-54] years), 18 (10%) of whom developed IBTR. A nomogram was constructed using three clinical-pathologic variables (age, margin, and use of adjuvant radiation therapy) and two mammographic variables (breast density and calcification morphology). The nomogram was appropriately calibrated and demonstrated a comparable 10-year AUROC to the MSKCC nomogram (0.73 vs. 0.66, P = 0.534) in the validation cohort. Our nomogram provided individualized risk estimates for women with DCIS treated with BCS, demonstrating a discriminative ability comparable to that of the MSKCC nomogram.

Histologic spectrum and outcome of Human papillomavirus (HPV)-associated oral cavity squamous cell carcinoma: a single center experience and a survey of The Cancer Genome Atlas (TGCA) cohort.

While high-risk human papillomavirus (HPV) serves as an essential pathogen and an important prognostic and predictive biomarker for oropharyngeal squamous cell carcinoma, it occurs at low frequency (2.2-6%) in oral cavity squamous cell carcinoma (OCSCC). To date, the pathologic features of HPV-associated OCSCC (HPV( +)-OCSCC) have been sparsely reported and its prognosis is not well-defined. We herein described detailed clinicopathologic features and outcomes of a retrospective series of 27 HPV( +)-OCSCC, including 13 from Memorial Sloan Kettering Cancer Center (MSKCC) and 14 from The Cancer Genomic Atlas program (TCGA). The frequency of HPV positivity in OCSCC was 0.7% in MSKCC cohort and 4.9% in TCGA cohort. Although HPV( +)-OCSCC was predominantly non-keratinizing (in 81%) with various degree of maturation, its histologic spectrum was expanded to include keratinizing subtype (19%), adenosquamous carcinoma (7%), and papillary architecture (subtype, 7%). HPV( +)-OCSCC predominantly affected male patients (male:female ratio = 12.5:1) and (ex) smokers (77%). It might occur in mandibular mucosa, floor of mouth, tongue, retromolar trigone, buccal mucosa, maxillary mucosa, or hard palate. In oral cavity, positivity of HPV by RNA in situ hybridization was required, and p16 immunohistochemistry alone was insufficient to confirm the HPV + status. The positive predictive value of p16 immunopositivity in detecting HPV infection was 68%. HPV-positivity did not appear to affect outcomes, including disease specific survival and progression free survival in OCSCC.

Paclitaxel, Ifosfamide, and Cisplatin as Initial Salvage Chemotherapy for Germ Cell Tumors: Long-Term Follow-Up and Outcomes for Favorable- and Unfavorable-Risk Disease.

Paclitaxel, ifosfamide, and cisplatin (TIP) is an established salvage regimen for germ cell tumors (GCT) on the basis of a phase II trial, but efficacy on a large patient cohort including patients with unfavorable risk features and long-term outcomes has not been reported. Herein, we report updated treatment efficacy and long-term follow-up with TIP. Patients with GCT who received TIP after cisplatin-based chemotherapy were eligible. Favorable response (complete response or partial response with negative tumor markers), overall survival (OS) and progression-free survival (PFS) rates, relapse, and toxicity were determined. Disease was reclassified according to the International Prognostic Factor Study Group (IPFSG) score. Of the 104 patients, 87 had favorable risk factors and 17 had at least one unfavorable factor by Memorial Sloan Kettering Cancer Center (MSKCC) criteria. Ten patients were treated for a second gonadal primary GCT. With a median follow-up of 8.9 years, the 5-year PFS and OS rates were 66% (95% CI, 55 to 74) and 69% (95% CI, 59 to 77), respectively. Among 87 patients with favorable-risk disease, 69 (79%) achieved a favorable response with 5-year PFS and OS rates of 67% (95% CI, 56 to 76) and 72% (95% CI, 61 to 80), respectively. Among 17 patients with MSKCC unfavorable-risk disease, 13 (76%) achieved a favorable response with 5-year PFS and OS rates of 59% (95% CI, 33 to 78) and 56% (95% CI, 28 to 76), respectively. After IPFSG reclassification, 5-year PFS and OS rates for patients with ≤intermediate-risk disease were 75% (95% CI, 50 to 89) and 73% (95% CI, 55 to 85), respectively. TIP is an effective second-line regimen for patients with GCT. Similar outcomes were observed in patients with favorable- and unfavorable-risk disease. The randomized TIGER trial (ClinicalTrials.gov identifier: NCT02375204) comparing TIP with high-dose chemotherapy will determine the optimal second-line treatment approach.

Prognosis of Poorly Differentiated Thyroid Carcinoma: A Systematic Review and Meta-Analysis.

Poorly differentiated thyroid carcinoma (PDTC) accounts for a small portion of thyroid carcinomas but contributes to a significant proportion of thyroid carcinoma-associated deaths. The clinicopathological prognostic factors and clinical outcomes of PDTC remain unclear. We aimed to evaluate the clinical outcomes of patients with PDTC after curative treatment. A comprehensive search was performed up to September 2023. We included studies investigating treatment outcomes in patients with PDTC who underwent initial surgery. The 5-year disease-free survival (DFS) and overall survival (OS) were extracted. In this meta-analysis, the enrolled PDTC histological criteria included 3rd, 4th, and 5th World Health Organization (WHO) and Memorial Sloan Kettering Cancer Center (MSKCC) classification. A random-effects model was used for the pooled proportion analysis. Meta-regression analysis was conducted to evaluate the prognostic factors. Twenty retrospective studies published between 2007 and 2023, including 1,294 patients, met all inclusion criteria. Studies that diagnosed PDTC based on various histological criteria including 3rd WHO (n=5), 4th WHO (n=12), 5th WHO (n=2), and MSKCC (n=1) were included. Overall, 5-year DFS and 5-year OS were 49.4% (95% confidence interval [CI], 42.3 to 56.4) and 73.8% (95% CI, 66.5 to 79.9), with moderate heterogeneity of 58% and 55%, respectively. In meta-regression analysis, extrathyroidal extension (ETE) was a prognostic factor for OS. The meta-analysis of DFS and OS in patients with PDTC show the moderate heterogeneity with a variety of histological criteria. ETE appears to have a significant impact on OS, regardless of histological criteria.

Forty years of human G-CSF: A short history in time.

Human G-CSF was identified in 1984 at Memorial Sloan-Kettering Cancer Centre, New York. Based on these findings, recombinant G-CSF was developed by Amgen, Thousand Oaks. In 1987, clinical trials began using recombinant G-CSF in cancer patients following chemotherapy to reduce the duration of neutropenia and in patients with congenital neutropenia (CN) to increase the number of neutrophils. It has changed the quality of life for many cancer patients and saved the lives of many patients with (CN).

Virtual Biostats Day: An Interactive Online Biostatistics Outreach Program Directed at High School Students in Groups Historically Underrepresented in STEM Careers

Persistent underrepresentation of Black and Hispanic Statistics degree holders relative to the U.S. population occurs at all levels in post-secondary education, contributing to the underrepresentation of Black and Hispanic Bio/Statisticians. Attempting to address this inequity before the undergraduate level, Memorial Sloan Kettering (MSK)’s Bridge to Biostats committee was established with the mission to increase awareness of and interest in the field of Biostatistics among high school students in groups underrepresented in STEM fields. We describe in this article an outreach program targeted to underrepresented students in New York City with innate aptitude and interest in STEM subjects, but minimal exposure to Statistics. Virtual Biostats Day is a 1–1/4 hr online exposure program that incorporates interactive elements exploring statistical concepts alongside clear descriptions of the field of Biostatistics and the roles and responsibilities of a Biostatistician. Students are also introduced to professional Biostatisticians who describe their journeys from high school to where they are today. This combination of didactic and active learning has engaged students, even in the online environment. We relate our experience in implementing and presenting our program to students and provide a complete toolkit for other Bio/Statistics departments or organizations interested in engaging in similar outreach work. Supplementary materials for this article are available online.

Adjuvant imatinib in high-risk resected gastrointestinal stromal tumors: Merely delaying the inevitable?

Patients with high-risk resected gastrointestinal stromal tumors (GIST) receiving adjuvant imatinib have improved recurrence-free survival (RFS), however whether a complete cytocidal effect exists is unknown. We investigated this using a normalized recurrence timeline measured from end of oncologic treatment (EOOT), defined as the later of resection or end of adjuvant therapy. We reviewed patients with resected high-risk GIST at our cancer center from 2003 to 2018. RFS (measured from resection and EOOT), overall survival (OS), and time to imatinib resistance (TTIR) were analyzed using Kaplan-Meier analysis and multivariable Cox proportional hazards modeling. The performance of the Memorial Sloan Kettering (MSK) GIST nomogram was assessed. We identified 86 patients with high-risk GIST with a median 106 months of postsurgical follow-up. One-third (n = 29; 34%) did not receive adjuvant imatinib, while 57 (66%) did for a median of 3 years. The MSK nomogram-predicted 5-year RFS for patients receiving adjuvant imatinib was similar to those who did not (29% vs. 31%, p = 0.64). When RFS was measured from EOOT, the MSK-predicted RFS was independently associated with EOOT RFS (hazard ratio 0.22, p = 0.02), while adjuvant imatinib receipt and duration were not. Neither receipt nor duration of adjuvant imatinib were associated with TTIR or OS (all p > 0.05). Treatment with adjuvant imatinib delays, but does not clearly impact ultimate recurrence, TTIR, or OS, suggesting many patients with high-risk GIST may receive adjuvant imatinib unnecessarily. Additional studies are needed to establish the benefit of adjuvant therapy versus initiating therapy at first radiographic recurrence.

Mucoepidermoid carcinoma: A comparative analysis of histological grading systems

: Salivary gland carcinomas comprise of only 3-5% of all head and neck malignancies. Mucoepidermoid carcinoma (MEC) is the most common malignant salivary gland tumor. The morphologic diversity of MEC can pose diagnostic challenges hence, various histological grading systems have been proposed based on the quantitative and qualitative analysis. This proves to be greatly consequential in the management and prognosis of patients with MEC.To compare histologic grading methods in MEC of major & minor salivary glands.20 histopathologically diagnosed cases of MEC (10 each major & minor salivary gland) will be analysed using following methods: Qualitative Method: 1) Modified Healey 2) Memorial Sloan-Kettering Cancer Center method. Quantitative method: 1) Armed force Institute of Pathology (AFIP) 2) Brandwein method Histological findings were evaluated. In our study AFIP grading system, 50% Cases were classified as low grade, 35% as intermediate grade and 15% as high graded. According to Brandwein grading system20% of cases were categorized as low, 35% cases as intermediate, 45% cases as high grade MEC. Modified Healey grading system of MEC showed 50%, 40%, 10% cases as low, intermediate and high grade MEC respectively. MSKCC grading system revealed as 55%, 30% and 15% cases as low, intermediate and high grade MEC respectively. Our finding indicated that MSKCC grading system was the most favourable histological grading system as percentage of agreement found to be 85%.Careful microscopic examination is the most important parameter in the grading of MEC. This meticulous microscopic examination emerges as the cornerstone in grading MEC. Both MSKCC and Modified Healey grading methods exhibits effectiveness in evaluating MEC. Our finding indicated that Memorial Slon Kettering Cancer Center (MSKCC) was the most favourable histological grading system as percentage of agreement found to be 85%. However, further longer studies are imperative to substantiate this finding and for establish of universally accepted grading system for Mucoepidermoid carcinoma.

Morbidity following robot-assisted surgery in a gynecological oncology setting: A cohort study.

The objective of the study was to provide a comprehensive description of perioperative morbidity associated with robot-assisted surgery (RAS) in a gynecological oncology setting in order to improve the preoperative counseling of women and support shared decision-making. All women scheduled for intended RAS between January 2015 and December 2022 were prospectively included in an electronic morbidity database for the analyses of perioperative complications. In total, 2225 women were included. Sixty-four patients (2.9%) experienced an intraoperative complication. Intraoperative complications were associated with a higher rate of conversion to laparotomy (15.6% vs. 1.8%, p < 0.001), a higher rate of major postoperative morbidity (9.3% vs. 2.4%, p < 0.001), and a higher rate of reoperation (9.3% vs. 1.7%, p < 0.001), compared to cases without intraoperative complications. Thirty-day postoperative morbidity was evaluated according to the Memorial Sloan-Kettering Cancer Center Surgical Secondary Events Grading System. Grade 3-5 events were considered major. A total of 57 patients (2.6%) experienced a major event after surgery, postoperative rupture of the vaginal vault being the most common complication requiring surgical intervention. Conversion to laparotomy occurred in 49 cases (2.2%) and was associated with higher intraoperative blood loss (300 mL vs. 25 mL, p < 0.001), a higher rate of postoperative major events (20.4% vs. 2.2%, p < 0.001), and a higher rate of reoperation (11.8% vs. 1.6%, p < 0.001). Our study demonstrates low rates of major perioperative morbidity and conversion to laparotomy after RAS performed by trained high-volume surgeons in a gynecological oncology setting.

Proposal for an optimised definition of adverse pathology (unfavourable histology) that predicts metastatic risk in prostatic adenocarcinoma independent of grade group and pathological stage.

Histological grading of prostate cancer is a powerful prognostic tool, but current criteria for grade assignment are not fully optimised. Our goal was to develop and test a simplified histological grading model, based heavily on large cribriform/intraductal carcinoma, with optimised sensitivity for predicting metastatic potential. Two separate non-overlapping cohorts were identified: a 419-patient post-radical prostatectomy cohort with long term clinical follow-up and a 209-patient post-radical prostatectomy cohort in which all patients had pathologically confirmed metastatic disease. All prostatectomies were re-reviewed for high-risk histological patterns of carcinoma termed 'unfavourable histology'. Unfavourable histology is defined by any classic Gleason pattern 5 component, any large cribriform morphology (> 0.25 mm) or intraductal carcinoma, complex intraluminal papillary architecture, grade 3 stromogenic carcinoma and complex anastomosing cord-like growth. For the outcome cohort, Kaplan-Meier analysis compared biochemical recurrence, metastasis and death between subjects with favourable and unfavourable histology, stratified by pathological stage and grade group. Multivariable Cox proportional hazards models evaluated adding unfavourable histology to the Memorial Sloan Kettering Cancer Center (MSKCC) post-prostatectomy nomogram and stratification by percentage of unfavourable histology. At 15 years unfavourable histology predicted biochemical recurrence, with sensitivity of 93% and specificity of 88%, metastatic disease at 100 and 48% and death at 100 and 46%. Grade group 2 prostate cancers with unfavourable histology were associated with metastasis independent of pathological stage, while those without had no risk. Histological models for prediction of metastasis based on only large cribriform/intraductal carcinoma or increasing diameter of cribriform size improved specificity, but with lower sensitivity. Multivariable Cox proportional hazards models demonstrated that unfavourable histology significantly improved discriminatory power of the MSKCC post-prostatectomy nomogram for biochemical failure (likelihood ratio test P < 0.001). In the retrospective review of a separate RP cohort in which all patients had confirmed metastatic disease, none had unequivocal favourable histology. Unfavourable histology at radical prostatectomy is associated with metastatic risk, predicted adverse outcomes better than current grading and staging systems and improved the MSKCC post-prostatectomy nomogram. Most importantly, unfavourable histology stratified grade group 2 prostate cancers into those with and without metastatic potential, independent of stage. While unfavourable histology is driven predominantly by large cribriform/intraductal carcinoma, the recognition and inclusion of other specific architectural patterns add to the sensitivity for predicting metastatic disease. Moreover, a simplified dichotomous model improves communication and could increase implementation.

A single-center retrospective analysis of allergies and survival outcomes in pancreatic adenocarcinoma.

e16299 Background: Survival from pancreatic adenocarcinoma (PDAC) remains poor, with a 5 year survival of 12.5%. A prior 2010 epidemiological study done at Memorial Sloan Kettering Cancer Center (MSKCC) found an association between self-reported allergies and improved survival in PDAC patients. In this retrospective study, we analyzed self-reported allergies and outcomes in both resected and unresected PDAC patients. Methods: Data including allergy status and clinical information were collected from an ongoing PDAC registry study at MSKCC. 1389 eligible PDAC patients were approached and 917 were enrolled between April 2004 and November 2017. 874 patients were included in our overall survival (OS) analysis using Kaplan-Meier methods. 43 patients were excluded due insufficient data or loss to follow up. Cox proportional hazards models were used to adjust for covariates. Results: Of the 874 patients analyzed, mean age at diagnosis was 63.5 years (standard deviation: 11.4), 52% were male, and race was 87% white, 6% black, 4% Asian, and 3% other race. 32.3% had surgery and 85% received chemotherapy. Median overall survival (OS) was 32.8 vs. 27.7 months [Hazard Ratio (HR) 0.95, p = 0.73] and 12.5 vs. 11.2 months [HR 1.04, p = 0.66] in for PDAC patients with allergies compared to those without in the resected and unresected cohorts, respectively. Advanced stage was associated with worse outcomes in both groups. Chemotherapy was significantly associated with improved survival only in the unresected group (HR 0.48, 95% CI 0.30-0.78, p = 0.030). Other established PDAC risk factors were also analyzed. Smoking history, diabetes, and family history were not associated with survival. Only obesity was significantly associated with worse outcomes in the unresected group (HR 1.3, 95% CI 1.01-1.67, p = 0.04). Conclusions: Contrary to previous findings, self-reported allergies were not significantly associated with improved survival in PDAC. In atopic patients, T helper type 2 cells are upregulated with allergen exposure and produce more IgE. Tumor-specific IgE has shown protective benefits in preclinical models but can be downregulated in the tumor microenvironment in PDAC. Further studies are needed to explore the relationship between allergies and PDAC and understand potential aspects of the immune system that can be harnessed to improve outcomes. One study limitation was that allergies were self-reported and not confirmed with allergy testing. We were also able to redemonstrate worse outcomes with more advanced stages and the benefit of chemo in the advanced setting. Further studies are needed to explore the mechanisms behind obesity and PDAC. [Table: see text]

Shareable artificial intelligence to extract cancer outcomes from electronic health records.

11000 Background: Clinical outcomes such as response, progression, and metastasis represent crucial data for observational cancer research, but outside of clinical trials, such outcomes are usually recorded only in unstructured notes in electronic health records (EHRs). Manual EHR annotation is too resource-intensive to scale to large datasets. Individual cancer centers have trained artificial intelligence natural language processing (AI/NLP) models to extract outcomes from their EHRs. However, due to concerns that models trained on protected health information (PHI) might encode private data, such models usually cannot be exported to other centers. Methods: EHR data from Dana-Farber Cancer Institute (DFCI) and Memorial-Sloan Kettering (MSK) collected through the AACR Project GENIE Biopharma Collaborative were used to train and evaluate Bidirectional Encoder Representations from Transformers (BERT)-based NLP models to extract outcomes from imaging reports and oncologist notes annotated with the Pathology, Radiology/Imaging, Signs/Symptoms, Medical oncologist, and bioMarkers (PRISSMM) framework. Document-level outcomes included the presence of active cancer; response; progression; and metastatic sites. ‘Teacher’ models trained on DFCI EHR data were used to label imaging reports and discharge summaries from the public MIMIC-IV dataset. ‘Student’ models trained to use MIMIC documents to predict teacher labels were transferred to MSK for evaluation. Results: Teacher models were trained at DFCI on 30,332 imaging reports for 2609 patients and 32,173 oncologist notes for 2917 patients with non-small cell lung, colorectal, breast, prostate, pancreatic, or urothelial cancer. The models were used to label 217,642 imaging reports and 141,377 discharge summaries from MIMIC for student model training. These DFCI-trained student models were evaluated at MSK, demonstrating high discrimination (AUROC &gt; 0.90) across outcomes. Conclusions: This privacy-preserving “teacher-student” AI/NLP framework could expedite linkage of genomic data to clinical outcomes across institutions, accelerating precision cancer research. [Table: see text]

Fibrolamellar carcinoma registry: Opportunities and challenges.

e16384 Background: Fibrolamellar carcinoma (FLC) is a rare form of primary liver cancer which affects young adults and adolescents without underlying liver disease or cirrhosis. The tumors are typically well circumscribed masses characterized pathologically by well differentiated polygonal hepatic cells with eosinophilic and granular cytoplasm surrounded by fibrous bands. DNAJB1 and PRKACA fusion is perceived as a dominant recurrent genetic alteration in FLC. FLC affects patients worldwide and a continued lack of an international database remains a knowledge gap. Further, there remains a poor understanding of FLC etiology, and major unmet need for more effective therapy. Methods: A patient-centered multi time-points, dynamic registry was developed by Memorial Sloan Kettering (MSK) in collaboration with Google and Maven Wave. The effort was supported by the Fibrolamellar Cancer Foundation (FCF). The study was opened worldwide except for the European Union pending the acquisition of the requiredGeneral Data Protection Regulation (GDPR) of data control and processing. The latter were assigned to Trinity College Dublin (TCD); a partner in this study. Participants were required to have a Google account to access the study. Demographic, clinical and response data were collected at baseline and at selected time points (monthly for up to 18 months). The aim of the registry was to provide a longitudinal understanding of the biography and natural history of the disease and outcomes. Results: Between May 2021 and December 2023, the study was broadcasted publicly worldwide through different venues including MSK, FCF, websites and Google engine. Only 14 patients responded. Table 1 summarizes the study sequence and patients’ enrollment. On average, the participants completed 41% of the monthly surveys. Four responders out of 5 patients were female, with a median age of 32 years (range 21-53). Conclusions: The registry provided a unique opportunity to access a broad global community of patients with FLC. Poor participation was key main limiting factor. Barriers include: (1) the presence of competing registries or studies for FLC, (2) technical limitations such as the exclusivity to access the registry to Google account. Follow-up emails may have been received in spam or junk folders, (3) complexities posed by the GDPR in Europe, and (4) patient-related factors including preoccupation with treatments and symptom management of their illness add to the disinterest in or inability to fill the monthly surveys. Unifying efforts and collaborations among caring physicians and scientists, medical centers, foundations, and patient advocacy groups across the world can help overcome the current challenges and accelerate the buildup of resources for research and practice improvement for the rare FLC. [Table: see text]

Body mass index (BMI) as a prelude to hepatocellular carcinoma (HCC) and metabolic dysfunction–associated steatohepatitis (MASH) risk factors in patients with no identified risk factor for developing hepatocellular carcinoma (HCC).

e16193 Background: Common risk factors for hepatocellular carcinoma (HCC) include viral hepatitis (hepatitis B and hepatitis C), alcoholic liver disease, and metabolic dysfunction-associated steatohepatitis (MASH). Non-risk factored HCC has been suggested where patients who do not have common risk factors develop HCC (Journal of Clinical Oncology Volume 35, Number 15_suppl, May 2017). This is described more commonly among young, females more than males. Is this a misdiagnosis or a truly non-factored risk? Methods: Patients diagnosed with histology confirmed HCC between years 2014 and 2017 with none of the known HCC risk factors, were assessed part of our previous effort add to more recent gathered data at Memorial Sloan Kettering (MSK). In an institutional IRB approved retrospective study, demographic and treatment data in addition to DNA sequencing were compiled and evaluated. Detailed clinical and genomic information was abstracted from the medical record. Pathology data was reviewed (AS). Results: A total of 428 patients diagnosed with HCC were identified, of whom 18 (4.2%) patients had no identifiable HCC risk factors. Fourteen patients were females (78%), and four were males (22%). Median age at diagnosis was 59 years (range 18-77 years). All 18 patients did not meet any of the clinical and/or radiologic alleged criteria for a diagnosis of MASH. The average body mass index (BMI) was 27.9 kg/m2 (16.7-54.4). Six patients (33%) had BMI greater than 30 kg/m2, and five (28%) patients had BMI ranging from 25 to 29.9 kg/m2. Three patients had diabetes mellitus. With the lack of non-tumorous liver cancer surrounding tissue, only three patients had mild steatosis documented on pathology. At the molecular level, few somatic mutations were detected including TERT and CTNNB1 in 37% of cases. No germline mutations were identified in 4 tested patients. Conclusions: Among 18 patients diagnosed with HCC with presumed no or unknown risk factors we previously reported, average BMI was 27.9 kg/m2. This is in the absence of any clinical and/or radiologic alleged criteria for a diagnosis of MASH. In the presence of subtle clinical and/or radiologic findings suggestive of MASH, evaluating patient’s BMI may be considered as an added variable for evaluation of patients with HCC. We clinicians need to be aware of increased BMI as a possible prelude for MASH and HCC. This is added to an invitation for acquired biopsies to include tumor and normal tissue to assess liver cirrhosis or fibrosis.

A pan-cancer analysis of SMARCA4 alterations and the unique clinicogenomic characteristics associated with SMARCA4 mutation types.

3073 Background: SWI/SNF complexes are regulators of cell lineage and alterations in these proteins occur in &gt;20% of solid tumors with SMARCA4being the most common. SMARCA4 mutations are associated with worse clinical outcomes in various cancer types but are also vulnerable to novel SMARCA2 inhibitors currently being evaluated in clinical trials. We examined SMARCA4 alterations across cancer types to understand their interplay with other genomic abnormalities and clinical consequences. Methods: We analyzed the Memorial Sloan Kettering (MSK) institutional cohort profiled by MSK-IMPACT tumor-normal sequencing. SMARCA4 mutations were annotated based on OncoKB and literature. FACETS resolved zygosity and genomic metrics of complexity including tumor mutational burden (TMB) and fraction genome altered (FGA). The pan-cancer landscape of somatic SMARCA4 mutations was mapped by both category (Class I vs II) and zygosity (mono- vs biallelic). DISCOVER was employed to evaluate mutational signatures and frequency-adjusted co-occurring and mutually exclusive mutations. We then extracted progression-free (PFS) and overall survival (OS) to systemic therapy using natural language processing. Results: We identified 3,385 tumors from 3,049 patients across cancer types with a total 3,965 oncogenic somatic SMARCA4 alterations. Copy number estimation was successful for 2,343 (69%). Non-small cell lung cancer (NSCLC) was the most abundant SMARCA4-altered histology ( n=493, 7.1% of all NSCLC), and the majority (87%) were biallelic. Other common histologies were colorectal ( n=163, 2.9%), bladder ( n=122, 5.5%), uterine ( n=88, 4.3%), pancreatic ( n=81, 3.2%) and esophagogastric ( n=92, 4%). SMARCA4biallelic mutated tumors had higher chromosomal instability compared to WT (FGA = 0.53 vs 0.47; p&lt;0.001), while monoallelic had lower (FGA = 0.3; p&lt;0.001). Monoallelic SMARCA4 altered tumors had a higher TMB (10.5) than both WT (4.4) and biallelic cancers (6.1; p&lt;0.001). Monoallelic and biallelic SMARCA4 cancers exhibited distinct mutational signatures in most tumor types. Patients with biallelic SMARCA4 mutations had worse outcomes with both immune checkpoint blockade (ICB) and platinum chemotherapy across in certain tumor types, whereas monoallelic uterine cancers were associated with improved outcomes with ICB. Conclusions: This pan-cancer clinicogenomic analysis demonstrates that monoallelic and biallelic SMARCA4-mutated tumors have distinguishable genomic features. In addition, biallelic SMARCA4 alterations correlated with poor outcomes across different cancer-types and classes of therapy, while improved outcomes to ICB in monoallelic SMARCA4alterations were potentially due to higher representation of MSI-H tumors in this subset. Our findings suggest SMARCA4 zygosity may be important biomarkers for ongoing clinical trials in SMARCA4-deficient tumors.

Deep-GenMut: Automated genetic mutation classification in oncology: A deep learning comparative study

Early cancer detection and treatment depend on the discovery of specific genes that cause cancer. The classification of genetic mutations was initially done manually. However, this process relies on pathologists and can be a time-consuming task. Therefore, to improve the precision of clinical interpretation, researchers have developed computational algorithms that leverage next-generation sequencing technologies for automated mutation analysis. This paper utilized four deep learning classification models with training collections of biomedical texts. These models comprise bidirectional encoder representations from transformers for Biomedical text mining (BioBERT), a specialized language model implemented for biological contexts. Impressive results in multiple tasks, including text classification, language inference, and question answering, can be obtained by simply adding an extra layer to the BioBERT model. Moreover, bidirectional encoder representations from transformers (BERT), long short-term memory (LSTM), and bidirectional LSTM (BiLSTM) have been leveraged to produce very good results in categorizing genetic mutations based on textual evidence. The dataset used in the work was created by Memorial Sloan Kettering Cancer Center (MSKCC), which contains several mutations. Furthermore, this dataset poses a major classification challenge in the Kaggle research prediction competitions. In carrying out the work, three challenges were identified: enormous text length, biased representation of the data, and repeated data instances. Based on the commonly used evaluation metrics, the experimental results show that the BioBERT model outperforms other models with an F1 score of 0.87 and 0.850 MCC, which can be considered as improved performance compared to similar results in the literature that have an F1 score of 0.70 achieved with the BERT model.

External validation of the Memorial Sloan Kettering Cancer Center preoperative nomogram predicting lymph node invasion in a cohort of high-grade prostate cancer patients.

Commonly used preoperative nomograms predicting clinical and pathological outcomes in prostate cancer (PCa) patients have not been yet validated in high-grade only PCapatients. Our objective is to perform an external validation of the Memorial Sloan Kettering Cancer Center (MSKCC) preoperative nomogram as a predictor of lymph node invasion (LNI) in a cohort of high-grade PCa patients. We included patients with high-grade PCa (Gleason ≥8) treated at our institution between 2011 and 2020 with radical prostatectomy and pelvic lymph node dissection without receiving neoadjuvant or adjuvant therapy. The area under the curve (AUC) of the receiver operator characteristic (ROC) was used to quantify the accuracy of the model to predict LNI. A calibration plot was used to evaluate the model's precision, and a decision curve analysiswas computed to evaluate the net benefit associated with its use. This study was approved by our institution's ethics board. A total of 242 patients with a median age of 66 (60-71) years were included. LNI was observed in 70 (29%) patients with a mean of 16 (median = 15; range = 2-42) resected nodes. The MSKCC nomogram discriminative accuracy, as evaluated by the AUC-ROC was 79.0% (CI: [0.727-0.853]). The MSKCC preoperative nomogram is a good predictor of LNI and a useful tool associated with net clinical benefit in this patient population.