Regression models for average hazard.

Limitations of using the traditional Cox's hazard ratio for summarizing the magnitude of the treatment effect on time-to-event outcomes have been widely discussed, and alternative measures that do not have such limitations are gaining attention. One of the alternative methods recently proposed, in a simple 2-sample comparison setting, uses the average hazard with survival weight (AH), which can be interpreted as the general censoring-free person-time incidence rate on a given time window. In this paper, we propose a new regression analysis approach for the AH with a truncation time τ. We investigate 3 versions of AH regression analysis, assuming (1) independent censoring, (2) group-specific censoring, and (3) covariate-dependent censoring. The proposed AH regression methods are closely related to robust Poisson regression. While the new approach needs to require a truncation time τ explicitly, it can be more robust than Poisson regression in the presence of censoring. With the AH regression approach, one can summarize the between-group treatment difference in both absolute difference and relative terms, adjusting for covariates that are associated with the outcome. This property will increase the likelihood that the treatment effect magnitude is correctly interpreted. The AH regression approach can be a useful alternative to the traditional Cox's hazard ratio approach for estimating and reporting the magnitude of the treatment effect on time-to-event outcomes.

Empirical evaluation of language modeling to ascertain cancer outcomes from clinical text reports.

BACKGROUND: Longitudinal data on key cancer outcomes for clinical research, such as response to treatment and disease progression, are not captured in standard cancer registry reporting. Manual extraction of such outcomes from unstructured electronic health records is a slow, resource-intensive process. Natural language processing (NLP) methods can accelerate outcome annotation, but they require substantial labeled data. Transfer learning based on language modeling, particularly using the Transformer architecture, has achieved improvements in NLP performance. However, there has been no systematic evaluation of NLP model training strategies on the extraction of cancer outcomes from unstructured text. RESULTS: We evaluated the performance of nine NLP models at the two tasks of identifying cancer response and cancer progression within imaging reports at a single academic center among patients with non-small cell lung cancer. We trained the classification models under different conditions, including training sample size, classification architecture, and language model pre-training. The training involved a labeled dataset of 14,218 imaging reports for 1112 patients with lung cancer. A subset of models was based on a pre-trained language model, DFCI-ImagingBERT, created by further pre-training a BERT-based model using an unlabeled dataset of 662,579 reports from 27,483 patients with cancer from our center. A classifier based on our DFCI-ImagingBERT, trained on more than 200 patients, achieved the best results in most experiments; however, these results were marginally better than simpler "bag of words" or convolutional neural network models. CONCLUSION: When developing AI models to extract outcomes from imaging reports for clinical cancer research, if computational resources are plentiful but labeled training data are limited, large language models can be used for zero- or few-shot learning to achieve reasonable performance. When computational resources are more limited but labeled training data are readily available, even simple machine learning architectures can achieve good performance for such tasks.

Patient and proxy reports regarding the experience of treatment decision-making in cancer care.

OBJECTIVE: Shared decision-making, including the elicitation of patient preferences regarding treatment decisions, is considered part of high-quality cancer care. However, patients may not be able to self-report due to illness, and therefore proxy reports may be used. We sought to determine the difference between proxy and patient reports about patient decisions and preferences among patients who received or were scheduled for chemotherapy using data from a large, population-based survey of patients with incident lung or colorectal cancer. METHODS: Of 3573 patients who received or were scheduled for chemotherapy, 3108 self-reported and 465 had proxies reporting on their behalf about preferred and actual decision roles regarding this treatment. Preferred and actual decision roles were assessed using the Control Preferences Scale, and categorized as shared, patient-controlled, or doctor-controlled. Multivariable logistic regression models were used to assess the association between patient and proxy responses and whether preferences were met. The models adjusted for sociodemographic and clinical variables and patient/proxy-reported health status. RESULTS: Sixty-three percent of all respondents reported actual roles in decisions that matched their preferred roles (role attainment). Proxies and patients were similarly likely to report role attainment (65% vs 63%). In adjusted analyses, proxies were more likely report role attainment (OR = 1.27, 95%CI = 1.02-1.59), but this difference was smaller if health variables were excluded from the model (OR = 1.14, 95%CI = 0.92-1.41). CONCLUSION: Most patients' preferences for treatment participation were met. Surveys from proxies appear to yield small differences on the reports of attainment of preferred treatment decision-making roles in cancer care vs surveys from patients.

Pre-existing autoimmune disease and the risk of immune-related adverse events among patients receiving checkpoint inhibitors for cancer.

INTRODUCTION: Patients with pre-existing autoimmune diseases have been excluded from clinical trials of immune checkpoint inhibitors (ICIs) for cancer. Real-world evidence is necessary to understand ICI safety in this population. METHODS: Patients treated with ICIs from 2011 to 2017 were identified using data from a large health insurer. Outcomes included time to (1) any hospitalization; (2) any hospitalization with an irAE diagnosis; and (3) outpatient corticosteroid treatment. The key exposure was pre-existing autoimmune disease, ascertained within 12 months before starting ICI treatment, and defined either by strict criteria (one inpatient or two outpatient claims at least 30 days apart) or relaxed criteria only (any claim, without meeting strict criteria). RESULTS: Of 4438 ICI-treated patients, pre-existing autoimmune disease was present among 179 (4%) by strict criteria, and another 283 (6%) by relaxed criteria only. In multivariable models, pre-existing autoimmune disease by strict criteria was not associated with all-cause hospitalization (HR 1.27, 95% CI 0.998-1.62), but it was associated with hospitalization with an irAE diagnosis (HR 1.81, 95% CI 1.21-2.71) and with corticosteroid treatment (HR 1.93, 95% CI 1.35-2.76). Similarly, pre-existing autoimmune disease by relaxed criteria only was not associated with all-cause hospitalization (HR 1.11, 95% CI 0.91-1.34), but was associated with hospitalization with an irAE diagnosis (HR 1.46, 95% CI 1.06-2.01) and corticosteroid treatment (HR 1.46, 95% CI 1.13-1.88). CONCLUSION: Pre-existing autoimmune disease was not associated with time to any hospitalization after initiating ICI therapy, but it was associated with a modest increase in hospitalizations with irAE diagnoses and with corticosteroid treatment.

Assessment of baseline symptom burden in treatment-naïve patients with lung cancer: an observational study.

BACKGROUND: Patients with newly diagnosed lung cancer who have not yet begun treatment may already be experiencing major symptoms produced by their disease. Understanding the symptomatic effects of cancer treatment requires knowledge of pretreatment symptoms (both severity and interference with daily activities). We assessed pretreatment symptom severity, interference, and quality of life (QOL) in treatment-naïve patients with lung cancer and report factors that correlated with symptom severity. METHODS: This was a retrospective analysis of data collected at initial intake. Symptoms/interference were rated on the MD Anderson Symptom Inventory (MDASI) between 30 days prediagnosis and 45 days postdiagnosis. We examined symptom severity by disease stage and differences in severity by histology. Linear regression analyses identified significant predictors of severe pain and dyspnea. RESULTS: Of 460 eligible patients, 256 (62%) had adenocarcinoma, 30 (7%) had small cell carcinoma, and 100 (24%) had squamous cell carcinoma; > 30% reported moderate-to-severe (rated ≥ 5, 0-10 scale) pretreatment symptoms. The most-severe were fatigue, disturbed sleep, distress, pain, dyspnea, sadness, and drowsiness. Symptoms affected work, enjoyment of life, and general activity (interference) and physical well-being (QOL) the most. Patients with advanced disease (n = 289, 63%) had more-severe symptoms. Cancer stage was associated with pain severity; both histology and cancer stage were associated with severe dyspnea. CONCLUSION: One third of lung cancer patients were symptomatic at initial presentation. Quantification of pretreatment symptom burden can inform patient-specific palliative therapy and differentiate disease-related symptoms from treatment-related toxicities. Poorly controlled symptoms could negatively affect treatment adherence and therapeutic outcomes.

Hospitalization by cytotoxic chemotherapy regimen among older women with stage IV breast cancer.

BACKGROUND: Older patients with cancer are at risk for increased side effects of treatment. Our goal was to inform treatment for older patients by analyzing the relationship between chemotherapy regimen and hospitalization among older women receiving palliative cytotoxic chemotherapy for breast cancer. METHOD: We identified women aged 66-99 years with stage IV de novo breast cancer diagnosed between 2010 and 2013 who received any of the 10 most common cytotoxic chemotherapy-containing regimens in the Surveillance, Epidemiology, and End Results (SEER)-Medicare database. The primary outcome was hospitalization or death within 30 days of starting a new line of chemotherapy. Generalized linear mixed effects models with patient-specific random effects were used for multivariable analysis of the association between chemotherapy regimen and this outcome. Additional covariates included number of prior lines of therapy; time since diagnosis; hormone receptor and HER2 status; sites of metastatic disease; and age, race, and marital status. The unit of analysis was each new line of chemotherapy. RESULTS: Of 972 lines of chemotherapy initiated among 693 patients, 188 (19%) were followed by hospitalization or death within 30 days. After adjustment, there was significant variation in this outcome by chemotherapy regimen (P = .03); compared with capecitabine, hospitalization/death rates were higher with cyclophosphamide + docetaxel (odds ratio [OR], 2.71; 95% confidence interval [CI], 1.31-5.59), cyclophosphamide + doxorubicin (OR, 2.45; 95% CI, 1.19-5.03), docetaxel (OR, 2.49; 95% CI, 1.19-5.21), and gemcitabine (OR, 3.51; 95% CI, 1.72-7.19). CONCLUSION: Treatment regimen was associated with significant variation in 30-day hospitalization or death among older women receiving cytotoxic chemotherapy for stage IV de novo breast cancer.

Rates of BRCA1/2 mutation testing among young survivors of breast cancer.

Guidelines in the United States recommend consideration of testing for mutations in the BRCA1 and BRCA2 genes for women diagnosed with breast cancer under age 45. Identification of mutations among survivors has implications for secondary prevention and familial risk reduction. Although only 10 % of breast cancers are diagnosed under age 45, there are approximately 2.8 million breast cancer survivors in the United States, such that the young survivor population likely numbers in the hundreds of thousands. However, little is known about genetic testing rates in this population. We assessed trends in BRCA1/2 testing among breast cancer survivors who were under age 45 at diagnosis and were treated from 2005 to 2012. Using insurance claims from a national database (MarketScan), we identified incident breast cancer cases among (1) women aged ≤40 and (2) women aged 41-45. We measured BRCA1/2 testing using Kaplan-Meier analysis and Cox proportional hazards models. Among 26,985 patients analyzed, BRCA1/2 testing rates increased with each year of diagnosis from 2005 to 2012 (P < 0.001). However, among women treated in earlier years, testing rates did not approach those of patients treated later, even after extended follow-up (median time from surgery to testing among patients treated in 2005, not reached; median time to testing among patients treated in 2012, 0.2 months for women aged ≤40 and 1.0 month for women aged 41-45). Women aged 41-45 had lower rates than women aged ≤40 throughout the analysis period (P < 0.001 for each year). BRCA1/2 testing rates among young women with incident breast cancer increased substantially in the last decade. However, most survivors treated in earlier years have never been tested. Our results demonstrate a need to better incorporate genetic counseling into survivorship and primary care for this population.

Extraction and Imputation of Eastern Cooperative Oncology Group Performance Status From Unstructured Oncology Notes Using Language Models.

PURPOSE: Eastern Cooperative Oncology Group (ECOG) performance status (PS) is a key clinical variable for cancer treatment and research, but it is usually only recorded in unstructured form in the electronic health record. We investigated whether natural language processing (NLP) models can impute ECOG PS using unstructured note text. MATERIALS AND METHODS: Medical oncology notes were identified from all patients with cancer at our center from 1997 to 2023 and divided at the patient level into training (approximately 80%), tuning/validation (approximately 10%), and test (approximately 10%) sets. Regular expressions were used to extract explicitly documented PS. Extracted PS labels were used to train NLP models to impute ECOG PS (0-1 v 2-4) from the remainder of the notes (with regular expression-extracted PS documentation removed). We assessed associations between imputed PS and overall survival (OS). RESULTS: ECOG PS was extracted using regular expressions from 495,862 notes, corresponding to 79,698 patients. A Transformer-based Longformer model imputed PS with high discrimination (test set area under the receiver operating characteristic curve 0.95, area under the precision-recall curve 0.73). Imputed poor PS was associated with worse OS, including among notes with no explicit documentation of PS detected (OS hazard ratio, 11.9; 95% CI, 11.1 to 12.8). CONCLUSION: NLP models can be used to impute performance status from unstructured oncologist notes at scale. This may aid the annotation of oncology data sets for clinical outcomes research and cancer care delivery.

Artificial Intelligence in Oncology: Current Landscape, Challenges, and Future Directions.

Artificial intelligence (AI) in oncology is advancing beyond algorithm development to integration into clinical practice. This review describes the current state of the field, with a specific focus on clinical integration. AI applications are structured according to cancer type and clinical domain, focusing on the four most common cancers and tasks of detection, diagnosis, and treatment. These applications encompass various data modalities, including imaging, genomics, and medical records. We conclude with a summary of existing challenges, evolving solutions, and potential future directions for the field. SIGNIFICANCE: AI is increasingly being applied to all aspects of oncology, where several applications are maturing beyond research and development to direct clinical integration. This review summarizes the current state of the field through the lens of clinical translation along the clinical care continuum. Emerging areas are also highlighted, along with common challenges, evolving solutions, and potential future directions for the field.

Perspectives of Oncologists on the Ethical Implications of Using Artificial Intelligence for Cancer Care.

Importance: Artificial intelligence (AI) tools are rapidly integrating into cancer care. Understanding stakeholder views on ethical issues associated with the implementation of AI in oncology is critical to optimal deployment. Objective: To evaluate oncologists' views on the ethical domains of the use of AI in clinical care, including familiarity, predictions, explainability (the ability to explain how a result was determined), bias, deference, and responsibilities. Design, Setting, and Participants: This cross-sectional, population-based survey study was conducted from November 15, 2022, to July 31, 2023, among 204 US-based oncologists identified using the National Plan & Provider Enumeration System. Main Outcomes and Measures: The primary outcome was response to a question asking whether participants agreed or disagreed that patients need to provide informed consent for AI model use during cancer treatment decisions. Results: Of 387 surveys, 204 were completed (response rate, 52.7%). Participants represented 37 states, 120 (63.7%) identified as male, 128 (62.7%) as non-Hispanic White, and 60 (29.4%) were from academic practices; 95 (46.6%) had received some education on AI use in health care, and 45.3% (92 of 203) reported familiarity with clinical decision models. Most participants (84.8% [173 of 204]) reported that AI-based clinical decision models needed to be explainable by oncologists to be used in the clinic; 23.0% (47 of 204) stated they also needed to be explainable by patients. Patient consent for AI model use during treatment decisions was supported by 81.4% of participants (166 of 204). When presented with a scenario in which an AI decision model selected a different treatment regimen than the oncologist planned to recommend, the most common response was to present both options and let the patient decide (36.8% [75 of 204]); respondents from academic settings were more likely than those from other settings to let the patient decide (OR, 2.56; 95% CI, 1.19-5.51). Most respondents (90.7% [185 of 204]) reported that AI developers were responsible for the medico-legal problems associated with AI use. Some agreed that this responsibility was shared by physicians (47.1% [96 of 204]) or hospitals (43.1% [88 of 204]). Finally, most respondents (76.5% [156 of 204]) agreed that oncologists should protect patients from biased AI tools, but only 27.9% (57 of 204) were confident in their ability to identify poorly representative AI models. Conclusions and Relevance: In this cross-sectional survey study, few oncologists reported that patients needed to understand AI models, but most agreed that patients should consent to their use, and many tasked patients with choosing between physician- and AI-recommended treatment regimens. These findings suggest that the implementation of AI in oncology must include rigorous assessments of its effect on care decisions as well as decisional responsibility when problems related to AI use arise.

Effects of KRAS Genetic Interactions on Outcomes in Cancers of the Lung, Pancreas, and Colorectum.

BACKGROUND: KRAS is among the most commonly mutated oncogenes in cancer, and previous studies have shown associations with survival in many cancer contexts. Evidence from both clinical observations and mouse experiments further suggests that these associations are allele- and tissue-specific. These findings motivate using clinical data to understand gene interactions and clinical covariates within different alleles and tissues. METHODS: We analyze genomic and clinical data from the AACR Project GENIE Biopharma Collaborative for samples from lung, colorectal, and pancreatic cancers. For each of these cancer types, we report epidemiological associations for different KRAS alleles, apply principal component analysis (PCA) to discover groups of genes co-mutated with KRAS, and identify distinct clusters of patient profiles with implications for survival. RESULTS: KRAS mutations were associated with inferior survival in lung, colon, and pancreas, although the specific mutations implicated varied by disease. Tissue- and allele-specific associations with smoking, sex, age, and race were found. Tissue-specific genetic interactions with KRAS were identified by PCA, which were clustered to produce five, four, and two patient profiles in lung, colon, and pancreas. Membership in these profiles was associated with survival in all three cancer types. CONCLUSIONS: KRAS mutations have tissue- and allele-specific associations with inferior survival, clinical covariates, and genetic interactions. IMPACT: Our results provide greater insight into the tissue- and allele-specific associations with KRAS mutations and identify clusters of patients that are associated with survival and clinical attributes from combinations of genetic interactions with KRAS mutations.

Biomarker Inference and the Timing of Next-Generation Sequencing in a Multi-Institutional, Cross-Cancer Clinicogenomic Data Set.

PURPOSE: Observational clinicogenomic data sets, consisting of tumor next-generation sequencing (NGS) data linked to clinical records, are commonly used for cancer research. However, in real-world practice, oncologists frequently request NGS in search of treatment options for progressive cancer. The extent and impact of this dynamic on analysis of clinicogenomic research data are not well understood. METHODS: We analyzed clinicogenomic data for patients with non-small cell lung, colorectal, breast, prostate, pancreatic, or urothelial cancers in the American Association for Cancer Research Biopharmaceutical Consortium cohort. Associations between baseline and time-varying clinical characteristics and time from diagnosis to NGS were measured. To explore the impact of informative cohort entry on biomarker inference, statistical interactions between selected biomarkers and time to NGS with respect to overall survival were calculated. RESULTS: Among 7,182 patients, time from diagnosis to NGS varied significantly by clinical factors, including cancer type, calendar year of sequencing, institution, and age and stage at diagnosis. NGS rates also varied significantly by dynamic clinical status variables; in an adjusted model, compared with patients with stable disease at any given time after diagnosis, patients with progressive disease by imaging or oncologist assessment had higher NGS rates (hazard ratio for NGS, 1.61 [95% CI, 1.45 to 1.78] and 2.32 [95% CI, 2.01 to 2.67], respectively). Statistical interactions between selected biomarkers and time to NGS with respect to survival, potentially indicating biased biomarker inference results, were explored. CONCLUSION: To evaluate the appropriateness of a data set for a particular research question, it is crucial to measure associations between dynamic cancer status and the timing of NGS, as well as to evaluate interactions involving biomarkers of interest and NGS timing with respect to survival outcomes.

Identifying Oncology Clinical Trial Candidates Using Artificial Intelligence Predictions of Treatment Change: A Pilot Implementation Study.

PURPOSE: Precision oncology clinical trials often struggle to accrue, partly because it is difficult to find potentially eligible patients at moments when they need new treatment. We piloted deployment of artificial intelligence tools to identify such patients at a large academic cancer center. PATIENTS AND METHODS: Neural networks that process radiology reports to identify patients likely to start new systemic therapy were applied prospectively for patients with solid tumors that had undergone next-generation sequencing at our center. Model output was linked to the MatchMiner tool, which matches patients to trials using tumor genomics. Reports listing genomically matched patients, sorted by probability of treatment change, were provided weekly to an oncology nurse navigator (ONN) coordinating recruitment to nine early-phase trials. The ONN contacted treating oncologists when patients likely to change treatment appeared potentially trial-eligible. RESULTS: Within weekly reports to the ONN, 60,199 patient-trial matches were generated for 2,150 patients on the basis of genomics alone. Of these, 3,168 patient-trial matches (5%) corresponding to 525 patients were flagged for ONN review by our model, representing a 95% reduction in review compared with manual review of all patient-trial matches weekly. After ONN review for potential eligibility, treating oncologists for 74 patients were contacted. Common reasons for not contacting treating oncologists included cases where patients had already decided to continue current treatment (21%); the trial had no slots (14%); or the patient was ineligible on ONN review (12%). Of 74 patients whose oncologists were contacted, 10 (14%) had a consult regarding a trial and five (7%) enrolled. CONCLUSION: This approach facilitated identification of potential patients for clinical trials in real time, but further work to improve accrual must address the many other barriers to trial enrollment in precision oncology research.

Collaborative Large Language Models for Automated Data Extraction in Living Systematic Reviews.

Objective: Data extraction from the published literature is the most laborious step in conducting living systematic reviews (LSRs). We aim to build a generalizable, automated data extraction workflow leveraging large language models (LLMs) that mimics the real-world two-reviewer process. Materials and Methods: A dataset of 10 clinical trials (22 publications) from a published LSR was used, focusing on 23 variables related to trial, population, and outcomes data. The dataset was split into prompt development (n=5) and held-out test sets (n=17). GPT-4-turbo and Claude-3-Opus were used for data extraction. Responses from the two LLMs were compared for concordance. In instances with discordance, original responses from each LLM were provided to the other LLM for cross-critique. Evaluation metrics, including accuracy, were used to assess performance against the manually curated gold standard. Results: In the prompt development set, 110 (96%) responses were concordant, achieving an accuracy of 0.99 against the gold standard. In the test set, 342 (87%) responses were concordant. The accuracy of the concordant responses was 0.94. The accuracy of the discordant responses was 0.41 for GPT-4-turbo and 0.50 for Claude-3-Opus. Of the 49 discordant responses, 25 (51%) became concordant after cross-critique, with an increase in accuracy to 0.76. Discussion: Concordant responses by the LLMs are likely to be accurate. In instances of discordant responses, cross-critique can further increase the accuracy. Conclusion: Large language models, when simulated in a collaborative, two-reviewer workflow, can extract data with reasonable performance, enabling truly 'living' systematic reviews.

Genomic Characterization and Clinical Outcomes of Patients with Peritoneal Metastases from the AACR GENIE Biopharma Collaborative Colorectal Cancer Registry.

Peritoneal metastases (PM) are common in metastatic colorectal cancer (mCRC). We aimed to characterize patients with mCRC and PM from a clinical and molecular perspective using the American Association of Cancer Research Genomics Evidence Neoplasia Information Exchange (GENIE) Biopharma Collaborative (BPC) registry. Patients' tumor samples underwent targeted next-generation sequencing. Clinical characteristics and treatment outcomes were collected retrospectively. Overall survival (OS) from advanced disease and progression-free survival (PFS) from start of cancer-directed drug regimen were estimated and adjusted for the left truncation bias. A total of 1,281 patients were analyzed, 244 (19%) had PM at time of advanced disease. PM were associated with female sex [OR: 1.67; 95% confidence interval (CI): 1.11-2.54; P = 0.014] and higher histologic grade (OR: 1.72; 95% CI: 1.08-2.71; P = 0.022), while rectal primary tumors were less frequent in patients with PM (OR: 0.51; 95% CI: 0.29-0.88; P < 0.001). APC occurred less frequently in patients with PM (N = 151, 64% vs. N = 788, 79%) while MED12 alterations occurred more frequently in patients with PM (N = 20, 10% vs. N = 32, 4%); differences in MED12 were not significant when restricting to oncogenic and likely oncogenic variants according to OncoKB. Patients with PM had worse OS (HR: 1.45; 95% CI: 1.16-1.81) after adjustment for independently significant clinical and genomic predictors. PFS from initiation of first-line treatment did not differ by presence of PM. In conclusion, PM were more frequent in females and right-sided primary tumors. Differences in frequencies of MED12 and APC alterations were identified between patients with and without PM. PM were associated with shorter OS but not with PFS from first-line treatment. SIGNIFICANCE: Utilizing the GENIE BPC registry, this study found that PM in patients with colorectal cancer occur more frequently in females and right-sided primary tumors and are associated with worse OS. In addition, we found a lower frequency of APC alterations and a higher frequency in MED12 alterations in patients with PM.

Enhancing Precision in Detecting Severe Immune-Related Adverse Events: Comparative Analysis of Large Language Models and International Classification of Disease Codes in Patient Records.

PURPOSE: Current approaches to accurately identify immune-related adverse events (irAEs) in large retrospective studies are limited. Large language models (LLMs) offer a potential solution to this challenge, given their high performance in natural language comprehension tasks. Therefore, we investigated the use of an LLM to identify irAEs among hospitalized patients, comparing its performance with manual adjudication and International Classification of Disease (ICD) codes. METHODS: Hospital admissions of patients receiving immune checkpoint inhibitor (ICI) therapy at a single institution from February 5, 2011, to September 5, 2023, were individually reviewed and adjudicated for the presence of irAEs. ICD codes and an LLM with retrieval-augmented generation were applied to detect frequent irAEs (ICI-induced colitis, hepatitis, and pneumonitis) and the most fatal irAE (ICI-myocarditis) from electronic health records. The performance between ICD codes and LLM was compared via sensitivity and specificity with an α = .05, relative to the gold standard of manual adjudication. External validation was performed using a data set of hospital admissions from June 1, 2018, to May 31, 2019, from a second institution. RESULTS: Of the 7,555 admissions for patients on ICI therapy in the initial cohort, 2.0% were adjudicated to be due to ICI-colitis, 1.1% ICI-hepatitis, 0.7% ICI-pneumonitis, and 0.8% ICI-myocarditis. The LLM demonstrated higher sensitivity than ICD codes (94.7% v 68.7%), achieving significance for ICI-hepatitis (P < .001), myocarditis (P < .001), and pneumonitis (P = .003) while yielding similar specificities (93.7% v 92.4%). The LLM spent an average of 9.53 seconds/chart in comparison with an estimated 15 minutes for adjudication. In the validation cohort (N = 1,270), the mean LLM sensitivity and specificity were 98.1% and 95.7%, respectively. CONCLUSION: LLMs are a useful tool for the detection of irAEs, outperforming ICD codes in sensitivity and adjudication in efficiency.

Elucidating Analytic Bias Due to Informative Cohort Entry in Cancer Clinico-genomic Datasets.

BACKGROUND: Oncologists often order genomic testing to inform treatment for worsening cancer. The resulting correlation between genomic testing timing and prognosis, or "informative entry," can bias observational clinico-genomic research. The efficacy of existing approaches to this problem in clinico-genomic cohorts is poorly understood. METHODS: We simulated clinico-genomic cohorts followed from an index date to death. Subgroups in each cohort who underwent genomic testing before death were "observed." We varied data generation parameters under four scenarios: (i) independent testing and survival times; (ii) correlated testing and survival times for all patients; (iii) correlated testing and survival times for a subset of patients; and (iv) testing and mortality exclusively following progression events. We examined the behavior of conditional Kendall tau (Tc) statistics, Cox entry time coefficients, and biases in overall survival (OS) estimation and biomarker inference across scenarios. RESULTS: Scenario #1 yielded null Tc and Cox entry time coefficients and unbiased OS inference. Scenario #2 yielded positive Tc, negative Cox entry time coefficients, underestimated OS, and biomarker associations biased toward the null. Scenario #3 yielded negative Tc, positive Cox entry time coefficients, and underestimated OS, but biomarker estimates were less biased. Scenario #4 yielded null Tc and Cox entry time coefficients, underestimated OS, and biased biomarker estimates. Transformation and copula modeling did not provide unbiased results. CONCLUSIONS: Approaches to informative clinico-genomic cohort entry, including Tc and Cox entry time statistics, are sensitive to heterogeneity in genotyping and survival time distributions. IMPACT: Novel methods are needed for unbiased inference using observational clinico-genomic data.

Utilizing Electronic Health Records (EHR) and Tumor Panel Sequencing to Demystify Prognosis of Cancer of Unknown Primary (CUP) patients.

Cancer of unknown primary (CUP) is a type of cancer that cannot be traced back to its original site and accounts for 3-5% of all cancers. It does not have established targeted therapies, leading to poor outcomes. We developed OncoNPC, a machine learning classifier trained on targeted next-generation sequencing data from 34,567 tumors from three institutions. OncoNPC achieved a weighted F1 score of 0.94 for high confidence predictions on known cancer types (65% of held-out samples). When applied to 971 CUP tumors from patients treated at the Dana-Farber Cancer Institute, OncoNPC identified actionable molecular alterations in 23% of the tumors. Furthermore, OncoNPC identified CUP subtypes with significantly higher polygenic germline risk for the predicted cancer type and significantly different survival outcomes, supporting its validity. Importantly, CUP patients who received first palliative intent treatments concordant with their OncoNPC-predicted cancer sites had significantly better outcomes (H.R. 0.348, 95% C.I. 0.210 - 0.570, p-value 2.32 × 10-5). OncoNPC thus provides evidence of distinct CUP subtypes and offers the potential for clinical decision support for managing patients with CUP.

Practice-Level Variation in Molecular Testing and Use of Targeted Therapy for Patients With Non-Small Cell Lung Cancer and Colorectal Cancer.

Importance: All patients with newly diagnosed non-small cell lung cancer (NSCLC) and colorectal cancer (CRC) should receive molecular testing to identify those who can benefit from targeted therapies. However, many patients do not receive recommended testing and targeted therapies. Objective: To compare rates of molecular testing and targeted therapy use by practice type and across practices. Design, Setting, and Participants: This cross-sectional study used 100% Medicare fee-for-service data from 2015 through 2019 to identify beneficiaries with new metastatic NSCLC or CRC diagnoses receiving systemic therapy and to assign patients to oncology practices. Hierarchical linear models were used to characterize variation by practice type and across practices. Data analysis was conducted from June 2019 to October 2022. Exposures: Oncology practice providing care. Outcomes: Primary outcomes were rates of molecular testing and targeted therapy use for patients with NSCLC and CRC. Secondary outcomes were rates of multigene testing for NSCLC and CRC. Results: There were 106 228 Medicare beneficiaries with incident NSCLC (31 521 [29.7%] aged 65-69 years; 50 348 [47.4%] female patients; 2269 [2.1%] Asian, 8282 [7.8%] Black, and 91 215 [85.9%] White patients) and 39 512 beneficiaries with incident CRC (14 045 [35.5%] aged 65-69 years; 17 518 [44.3%] female patients; 896 [2.3%] Asian, 3521 [8.9%] Black, and 32 753 [82.9%] White patients) between 2015 and 2019. Among these beneficiaries, 18 435 (12.9%) were treated at National Cancer Institute (NCI)-designated centers, 8187 (5.6%) were treated at other academic centers, and 94 329 (64.7%) were treated at independent oncology practices. Molecular testing rates increased from 74% to 85% for NSCLC and 45% to 65% for CRC. First-line targeted therapy use decreased from 12% to 8% among patients with NSCLC and was constant at 5% for patients with CRC. For NSCLC, molecular testing rates were similar across practice types while rates of multigene panel use (13.2%) and targeted therapy use (16.6%) were highest at NCI-designated cancer centers. For CRC, molecular testing rates were 3.8 (95% CI: 1.2-6.5), 3.3 (95% CI, 0.4-6.1), and 12.2 (95% CI, 9.1-15.3) percentage points lower at hospital-owned practices, large independent practices, and small independent practices, respectively. Rates of targeted therapy use for CRC were similar across practice types. After adjusting for patient characteristics, there was moderate variation in molecular testing and targeted therapy use across oncology practices. Conclusions and Relevance: In this cross-sectional study of Medicare beneficiaries, molecular testing rates for NSCLC and CRC increased in recent years but remained lower than recommended levels. Rates of targeted therapy use decreased for NSCLC and remained stable for CRC. Variation across practices suggests that where a patient was treated may have affected access to recommended testing and efficacious treatments.

Machine learning for genetics-based classification and treatment response prediction in cancer of unknown primary.

Cancer of unknown primary (CUP) is a type of cancer that cannot be traced back to its primary site and accounts for 3-5% of all cancers. Established targeted therapies are lacking for CUP, leading to generally poor outcomes. We developed OncoNPC, a machine-learning classifier trained on targeted next-generation sequencing (NGS) data from 36,445 tumors across 22 cancer types from three institutions. Oncology NGS-based primary cancer-type classifier (OncoNPC) achieved a weighted F1 score of 0.942 for high confidence predictions ([Formula: see text]) on held-out tumor samples, which made up 65.2% of all the held-out samples. When applied to 971 CUP tumors collected at the Dana-Farber Cancer Institute, OncoNPC predicted primary cancer types with high confidence in 41.2% of the tumors. OncoNPC also identified CUP subgroups with significantly higher polygenic germline risk for the predicted cancer types and with significantly different survival outcomes. Notably, patients with CUP who received first palliative intent treatments concordant with their OncoNPC-predicted cancers had significantly better outcomes (hazard ratio (HR) = 0.348; 95% confidence interval (CI) = 0.210-0.570; P = [Formula: see text]). Furthermore, OncoNPC enabled a 2.2-fold increase in patients with CUP who could have received genomically guided therapies. OncoNPC thus provides evidence of distinct CUP subgroups and offers the potential for clinical decision support for managing patients with CUP.