JCO Clinical Cancer Informatics最新文献

Purpose: Although the International Neuroblastoma Risk Group Data Commons (INRGdc) has enabled seminal large cohort studies, the research is limited by the lack of real-world, electronic health record (EHR) treatment data. To address this limitation, we evaluated the feasibility of extracting treatment data directly from EHRs using the REDCap Clinical Data Interoperability Services (CDIS) module for future submission to the INRGdc.

Methods: Patients enrolled on the Children's Oncology Group neuroblastoma biology study ANBL00B1 (ClinicalTrials.gov identifier: NCT00904241) who received care at the University of Chicago (UChicago) or the Vanderbilt University Medical Center (VUMC) after the go-live dates for the Fast Healthcare Interoperability Resources (FHIR)-compliant EHRs were identified. Antineoplastic drug orders were extracted using the CDIS module. To validate the CDIS output, antineoplastic agents extracted through FHIR were compared with those queried through EHR relational databases (UChicago's Clinical Research Data Warehouse and VUMC's Epic Clarity database) and manual chart review.

Results: The analytic cohort consisted of 41 patients at UChicago and 32 VUMC patients. Antineoplastic drug orders were identified in the extracted EHR records of 39 (95.1%) UChicago patients and 26 (81.3%) VUMC patients. Manual chart review confirmed that patients with missing (n = 8) or discontinued (n = 1) orders in the CDIS output did not receive antineoplastic agents during the timeframe of the study. More than 99% of the antineoplastic drug orders in the EHR relational databases were identified in the corresponding CDIS output.

Conclusion: Our results demonstrate the feasibility of extracting EHR treatment data with high fidelity using HL7-FHIR via REDCap CDIS for future submission to the INRGdc.

This new editorial discusses the promise and challenges of successful integration of natural language processing methods into electronic health records for timely, robust, and fair oncology pharmacovigilance.

Purpose: Real-world evidence (RWE)-derived from analysis of real-world data (RWD)-has the potential to guide personalized treatment decisions. However, because of potential confounding, generating valid RWE is challenging. This study demonstrates how to responsibly generate RWE for treatment decisions. We validate our approach by demonstrating that we can uncover an existing adjuvant chemotherapy (ACT) guideline for stage II and III colon cancer (CC)-which came about using both data from randomized controlled trials and expert consensus-solely using RWD.

Methods: Data from the population-based Netherlands Cancer Registry from a total of 27,056 patients with stage II and III CC who underwent curative surgery were analyzed to estimate the overall survival (OS) benefit of ACT. Focusing on 5-year OS, the benefit of ACT was estimated for each patient using G-computation methods by adjusting for patient and tumor characteristics and estimated propensity score. Subsequently, on the basis of these estimates, an ACT decision tree was constructed.

Results: The constructed decision tree corresponds to the current Dutch guideline: patients with stage III or stage II with T stage 4 should receive surgery and ACT, whereas patients with stage II with T stage 3 should only receive surgery. Interestingly, we do not find sufficient RWE to conclude against ACT for stage II with T stage 4 and microsatellite instability-high (MSI-H), a recent addition to the current guideline.

Conclusion: RWE, if used carefully, can provide a valuable addition to our construction of evidence on clinical decision making and therefore ultimately affect treatment guidelines. Next to validating the ACT decisions advised in the current Dutch guideline, this paper suggests additional attention should be paid to MSI-H in future iterations of the guideline.

Purpose: Natural language understanding (NLU) may be particularly well equipped for enhanced data capture from the electronic health record given its examination of both content-driven and context-driven extraction.

Methods: We developed and applied a NLU model to examine rates of pathological node positivity (pN+) and rates of lymphedema to determine whether omission of routine axillary staging could be extended to younger patients with estrogen receptor-positive (ER+)/cN0 disease.

Results: We found that rates of pN+ and arm lymphedema were similar between patients age 55-69 years and ≥70 years, with rates of lymphedema exceeding rates of pN+ for clinical stage T1c and smaller disease.

Conclusion: Data from our NLU model suggest that omission of sentinel lymph node biopsy might be extended beyond Choosing Wisely recommendations, limited to those older than 70 years and to all postmenopausal women with early-stage ER+/cN0 disease. These data support the recently reported SOUND trial results and provide additional granularity to facilitate surgical de-escalation.

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.

Purpose: We present and validate a rule-based algorithm for the detection of moderate to severe liver-related immune-related adverse events (irAEs) in a real-world patient cohort. The algorithm can be applied to studies of irAEs in large data sets.

Methods: We developed a set of criteria to define hepatic irAEs. The criteria include: the temporality of elevated laboratory measurements in the first 2-14 weeks of immune checkpoint inhibitor (ICI) treatment, steroid intervention within 2 weeks of the onset of elevated laboratory measurements, and intervention with a duration of at least 2 weeks. These criteria are based on the kinetics of patients who experienced moderate to severe hepatotoxicity (Common Terminology Criteria for Adverse Events grades 2-4). We applied these criteria to a retrospective cohort of 682 patients diagnosed with hepatocellular carcinoma and treated with ICI. All patients were required to have baseline laboratory measurements before and after the initiation of ICI.

Results: A set of 63 equally sampled patients were reviewed by two blinded, clinical adjudicators. Disagreements were reviewed and consensus was taken to be the ground truth. Of these, 25 patients with irAEs were identified, 16 were determined to be hepatic irAEs, 36 patients were nonadverse events, and two patients were of indeterminant status. Reviewers agreed in 44 of 63 patients, including 19 patients with irAEs (0.70 concordance, Fleiss' kappa: 0.43). By comparison, the algorithm achieved a sensitivity and specificity of identifying hepatic irAEs of 0.63 and 0.81, respectively, with a test efficiency (percent correctly classified) of 0.78 and outcome-weighted F1 score of 0.74.

Conclusion: The algorithm achieves greater concordance with the ground truth than either individual clinical adjudicator for the detection of irAEs.

Purpose: The emergence of large real-world clinical databases and tools to mine electronic medical records has allowed for an unprecedented look at large data sets with clinical and epidemiologic correlates. In clinical cancer genetics, real-world databases allow for the investigation of prevalence and effectiveness of prevention strategies and targeted treatments and for the identification of barriers to better outcomes. However, real-world data sets have inherent biases and problems (eg, selection bias, incomplete data, measurement error) that may hamper adequate analysis and affect statistical power.

Methods: Here, we leverage a real-world clinical data set from a large health network for patients with breast cancer tested for variants in BRCA1 and BRCA2 (N = 12,423). We conducted data cleaning and harmonization, cross-referenced with publicly available databases, performed variant reassessment and functional assays, and used functional data to inform a variant's clinical significance applying American College of Medical Geneticists and the Association of Molecular Pathology guidelines.

Results: In the cohort, White and Black patients were over-represented, whereas non-White Hispanic and Asian patients were under-represented. Incorrect or missing variant designations were the most significant contributor to data loss. While manual curation corrected many incorrect designations, a sizable fraction of patient carriers remained with incorrect or missing variant designations. Despite the large number of patients with clinical significance not reported, original reported clinical significance assessments were accurate. Reassessment of variants in which clinical significance was not reported led to a marked improvement in data quality.

Conclusion: We identify the most common issues with BRCA1 and BRCA2 testing data entry and suggest approaches to minimize data loss and keep interpretation of clinical significance of variants up to date.

Purpose: To evaluate natural language processing (NLP) methods to infer metastatic sites from radiology reports.

Methods: A set of 4,522 computed tomography (CT) reports of 550 patients with 14 types of cancer was used to fine-tune four clinical large language models (LLMs) for multilabel classification of metastatic sites. We also developed an NLP information extraction (IE) system (on the basis of named entity recognition, assertion status detection, and relation extraction) for comparison. Model performances were measured by F1 scores on test and three external validation sets. The best model was used to facilitate analysis of metastatic frequencies in a cohort study of 6,555 patients with 53,838 CT reports.

Results: The RadBERT, BioBERT, GatorTron-base, and GatorTron-medium LLMs achieved F1 scores of 0.84, 0.87, 0.89, and 0.91, respectively, on the test set. The IE system performed best, achieving an F1 score of 0.93. F1 scores of the IE system by individual cancer type ranged from 0.89 to 0.96. The IE system attained F1 scores of 0.89, 0.83, and 0.81, respectively, on external validation sets including additional cancer types, positron emission tomography-CT ,and magnetic resonance imaging scans, respectively. In our cohort study, we found that for colorectal cancer, liver-only metastases were higher in de novo stage IV versus recurrent patients (29.7% v 12.2%; P < .001). Conversely, lung-only metastases were more frequent in recurrent versus de novo stage IV patients (17.2% v 7.3%; P < .001).

Conclusion: We developed an IE system that accurately infers metastatic sites in multiple primary cancers from radiology reports. It has explainable methods and performs better than some clinical LLMs. The inferred metastatic phenotypes could enhance cancer research databases and clinical trial matching, and identify potential patients for oligometastatic interventions.

Purpose: Immune checkpoint inhibitors (ICIs) have revolutionized cancer treatment, yet their use is associated with immune-related adverse events (irAEs). Estimating the prevalence and patient impact of these irAEs in the real-world data setting is critical for characterizing the benefit/risk profile of ICI therapies beyond the clinical trial population. Diagnosis codes, such as International Classification of Diseases codes, do not comprehensively illustrate a patient's care journey and offer no insight into drug-irAE causality. This study aims to capture the relationship between ICIs and irAEs more accurately by using augmented curation (AC), a natural language processing-based innovation, on unstructured data in electronic health records.

Methods: In a cohort of 9,290 patients treated with ICIs at Mayo Clinic from 2005 to 2021, we compared the prevalence of irAEs using diagnosis codes and AC models, which classify drug-irAE pairs in clinical notes with implied textual causality. Four illustrative irAEs with high patient impact-myocarditis, encephalitis, pneumonitis, and severe cutaneous adverse reactions, abbreviated as MEPS-were analyzed using corticosteroid administration and ICI discontinuation as proxies of severity.

Results: For MEPS, only 70% (n = 118) of patients found by AC were also identified by diagnosis codes. Using AC models, patients with MEPS received corticosteroids for their respective irAE 82% of the time and permanently discontinued the ICI because of the irAE 35.9% (n = 115) of the time.

Conclusion: Overall, AC models enabled more accurate identification and assessment of patient impact of ICI-induced irAEs not found using diagnosis codes, demonstrating a novel and more efficient strategy to assess real-world clinical outcomes in patients treated with ICIs.

Purpose: Adverse effects of chemotherapy often require hospital admissions or treatment management. Identifying factors contributing to unplanned hospital utilization may improve health care quality and patients' well-being. This study aimed to assess if patient-reported outcome measures (PROMs) improve performance of machine learning (ML) models predicting hospital admissions, triage events (contacting helpline or attending hospital), and changes to chemotherapy.

Materials and methods: Clinical trial data were used and contained responses to three PROMs (European Organisation for Research and Treatment of Cancer Core Quality of Life Questionnaire [QLQ-C30], EuroQol Five-Dimensional Visual Analogue Scale [EQ-5D], and Functional Assessment of Cancer Therapy-General [FACT-G]) and clinical information on 508 participants undergoing chemotherapy. Six feature sets (with following variables: [1] all available; [2] clinical; [3] PROMs; [4] clinical and QLQ-C30; [5] clinical and EQ-5D; [6] clinical and FACT-G) were applied in six ML models (logistic regression [LR], decision tree, adaptive boosting, random forest [RF], support vector machines [SVMs], and neural network) to predict admissions, triage events, and chemotherapy changes.

Results: The comprehensive analysis of predictive performances of the six ML models for each feature set in three different methods for handling class imbalance indicated that PROMs improved predictions of all outcomes. RF and SVMs had the highest performance for predicting admissions and changes to chemotherapy in balanced data sets, and LR in imbalanced data set. Balancing data led to the best performance compared with imbalanced data set or data set with balanced train set only.

Conclusion: These results endorsed the view that ML can be applied on PROM data to predict hospital utilization and chemotherapy management. If further explored, this study may contribute to health care planning and treatment personalization. Rigorous comparison of model performance affected by different imbalanced data handling methods shows best practice in ML research.