e14062 Background: Real world evidence generated from electronic health records (EHRs) is playing an increasing role in health care decisions. It has been recognized as an essential element to assess cancer outcomes in real-world settings. Automatically abstracting outcomes from notes is becoming a fundamental challenge in medical informatics. In this study, we aim to develop a system to automatically abstract outcomes (Progression, Response, Stable Disease) from notes in lung cancer. Methods: A lung cancer cohort (n = 5,003) was obtained from the Mount Sinai Data Warehouse. The progress, pathology and radiology notes of patients were used. We integrated various techniques of Natural Language Processing (NLP) and Artificial Intelligence (AI) and developed a system to automatically abstract outcomes. The corresponding images, biopsies and lines of treatments (LOTs) were abstracted as attributes of outcomes. This system includes four information models: 1. Customized NLP annotator model: preprocessor, section detector, sentence splitter, named entity recognition, relation detector; CRF and LSTM methods were applied to recognize entities and relations. 2. Clinical Outcome container model: biopsy evidence extractor, lines of treatment detector, image evidence extractor, clinical outcome event recognizer, date detector, and temporal reasoning; Domain-specific rules were crafted to automatically infer outcomes. 3. Document Summarizer; 4. Longitudinal Outcome Summarizer. Results: To evaluate the outcomes abstracted, we curated a subset (n = 792) from patient cohort for which LOTs were available. About 61% of the outcomes identified were supported by radiologic images (time window = ±14 days) or biopsy pathology results (time window = ±100 days). In 91% (720/792) of patients, Progression was abstracted within a time window of 90 days prior to first-line treatment. Also, 72% of the Progression events identified were accompanied by a downstream event (e.g., treatment change or death). We randomly selected 250 outcomes for manual curation, and 197 outcomes were assessed to be correct (precision = 79%). Moreover, our automated abstraction system improved human abstractor efficiency to curate outcomes, reducing curation time per patient by 90%. Conclusions: We have demonstrated the feasibility and effectiveness of NLP and AI approaches to abstract outcomes from lung cancer EHR data. It promises to automatically abstract outcomes and other clinical entities from notes across all cancers.
e19512 Background: Targeted therapy is an important treatment for chronic lymphocytic leukemia (CLL). However, optimal strategies for deploying small molecule inhibitors or antibody therapies in the real world are not well understood, largely due to a lack of outcomes data. We implemented a novel temporal phenotyping algorithm pipeline to derive lines of therapy (LOT) and disease progression in CLL patients. Here, the CLL treatment pattern and time to the next treatment (TTNT) were analyzed in real-world data (RWD) using patient electronic health records. Methods: We identified a CLL cohort with LOT from the Mount Sinai Data Warehouse (2003-2020). Each LOT consisted of either a single agent or combinations defined by NCCN CLL guidelines. We developed a natural language processing (NLP)-based temporal phenotyping approach to automatically identify the number of lines and therapeutic regimens. The sequence of treatment and time interval for each patient were derived from the systematic treatment data. Time to event analysis and multivariate (i.e., age, gender, race, other treatment patterns) Cox proportional hazard (CoxPH) models were used to analyze the patterns and predictors of TTNT. Results: Four hundred eleven CLL patients received 1 to 7 LOTs. Ibrutinib was the predominant 1 st LOT (40.8% of patients) followed by anti-CD20-based antibody therapies and chemotherapy in 30.6 and 19.2% of patients, respectively, followed by Acalabrutinib, Venetoclax, and Idelalisib in 3.4, 2.7, and 0.7% of patients, respectively (Table 1). The 2 nd to 5 th LOT showed the same or similar trends. We next analyzed the TTNT in the 1 st line of each therapeutic class. Acalabrutinib resulted in a longer median TTNT than Ibrutinib. Both Acalabrutinib and Ibrutinib showed longer TTNT compared to Venetoclax (median TTNTs were 742 and 598 vs. 373 days: HR = 0.23, p=0.015 and HR = 0.48, p=0.03, respectively). In addition, patients with age equal to or older than 65 showed longer TNNT (HR=0.16, p=0.016). Conclusions: Our result shows the potential of RWD usage in clinical decision making as real-world evidence reported here is consistent with results derived from clinical trial data. Linking this study to genetic data and other covariates affecting treatment outcomes may provide additional insights into the optimal sequences of the targeted therapies in CLL. Table 1: Therapeutic class and patient numbers (%) in each line.[Table: see text]
The consistent and persuasive evidence illustrating the influence of social determinants on health has prompted a growing realization throughout the health care sector that enhancing health and health equity will likely depend, at least to some extent, on addressing detrimental social determinants. However, detailed social determinants of health (SDoH) information is often buried within clinical narrative text in electronic health records (EHRs), necessitating natural language processing (NLP) methods to automatically extract these details. Most current NLP efforts for SDoH extraction have been limited, investigating on limited types of SDoH elements, deriving data from a single institution, focusing on specific patient cohorts or note types, with reduced focus on generalizability. This study aims to address these issues by creating cross-institutional corpora spanning different note types and healthcare systems, and developing and evaluating the generalizability of classification models, including novel large language models (LLMs), for detecting SDoH factors from diverse types of notes from four institutions: Harris County Psychiatric Center, University of Texas Physician Practice, Beth Israel Deaconess Medical Center, and Mayo Clinic. Four corpora of deidentified clinical notes were annotated with 21 SDoH factors at two levels: level 1 with SDoH factor types only and level 2 with SDoH factors along with associated values. Three traditional classification algorithms (XGBoost, TextCNN, Sentence BERT) and an instruction tuned LLM-based approach (LLaMA) were developed to identify multiple SDoH factors. Substantial variation was noted in SDoH documentation practices and label distributions based on patient cohorts, note types, and hospitals. The LLM achieved top performance with micro-averaged F1 scores over 0.9 on level 1 annotated corpora and an F1 over 0.84 on level 2 annotated corpora. While models performed well when trained and tested on individual datasets, cross-dataset generalization highlighted remaining obstacles. To foster collaboration, access to partial annotated corpora and models trained by merging all annotated datasets will be made available on the PhysioNet repository.
Ground-glass opacities (GGOs) appearing in computed tomography (CT) scans may indicate potential lung malignancy. Proper management of GGOs based on their features can prevent the development of lung cancer. Electronic health records are rich sources of information on GGO nodules and their granular features, but most of the valuable information is embedded in unstructured clinical notes.
FHIR standard is designed to enable interoperability and integration with the newest and adopted technologies by the industry. This chapter presents a number of blueprints for the design and development of FHIR servers that enable the integration between HIT systems with m-health applications via FHIR. Each blueprint is based on the location that FHIR servers can be placed with respect to the components of the m-health application (UI, API, server) or a HIT system in order to define and design the necessary infrastructure to facilitate the exchange of information via FHIR. To demonstrate the feasibility of the work, this chapter utilizes the Connecticut concussion tracker (CT2) m-health application as a proof-of-concept prototype that fully illustrates the blueprints of the design and development steps that are involved. The blueprints can be applied to any m-health application and are informative and instructional for medical stakeholders, researchers, and developers.
e18747 Background: Accurate longitudinal cancer treatments are vital for establishing primary endpoints such as outcome as well as for the investigation of adverse events. However, many longitudinal therapeutic regimens are not well captured in structured electronic health records (EHRs). Thus, their recognition in unstructured data such as clinical notes is critical to gain an accurate description of the real-world patient treatment journey. Here, we demonstrate a scalable approach to extract high-quality longitudinal cancer treatments from lung cancer patients' clinical notes using a Bidirectional Long Short Term Memory (BiLSTM) and Conditional Random Fields (CRF) based natural language processing (NLP) pipeline. Methods: The lung cancer (LC) cohort of 4,698 patients was curated from the Mount Sinai Healthcare system (2003-2020). Two domain experts developed a structured framework of entities and semantics that captured treatment and its temporality. The framework included therapy type (chemotherapy, targeted therapy, immunotherapy, etc.), status (on, off, hold, planned, etc.) and temporal reasoning entities and relations (admin_date, duration, etc.) We pre-annotated 149 FDA-approved cancer drugs and longitudinal timelines of treatment on the training corpus. A NLP pipeline was implemented with BiLSTM-CRF-based deep learning models to train and then apply the resulting models to the clinical notes of LC cohort. A postprocessor was developed to subsequently post-coordinate and refine the output. We performed both cross-evaluation and independent evaluation to assess the pipeline performance. Results: We applied the NLP pipeline to the 853,755 clinical notes, and identified 1,155 distinct entities for 194 cancer generic drugs, including 74 chemotherapy drugs, 21 immunotherapy drugs, and 99 targeted therapy drugs. We identified chemotherapy, immunotherapy, or targeted therapy data for 3,509 patients in the LC cohort from the clinical notes. Compared to only 2,395 patients with cancer treatments in structured EHR, this pipeline identified cancer treatments from notes for additional 2,303 patients who did not have any available cancer treatment data in the structured EHR. Our evaluation schema indicates that the longitudinal cancer drug recognition pipeline delivers strong performance (named entity recognization for drugs and temporal: F1 = 95%; drug-temporal relation recognition: F1 = 90%). Conclusions: We developed a high-performance BiLSTM-CRF based NLP pipeline to recognize longitudinal cancer treatments. The pipeline recovers and encodes as twice as many patients with cancer treatments compared with structured EHR. Our study indicates deep NLP with temporal reasoning could substantially accelerate the extraction of treatment profiles at scale. The pipeline is adjustable and can be applied across different cancers.
BACKGROUND Ground-glass opacities (GGOs) appearing in computed tomography (CT) scans may indicate potential lung malignancy. Proper management of GGOs based on their features can prevent the development of lung cancer. Electronic health records are rich sources of information on GGO nodules and their granular features, but most of the valuable information is embedded in unstructured clinical notes. OBJECTIVE We aimed to develop, test, and validate a deep learning–based natural language processing (NLP) tool that automatically extracts GGO features to inform the longitudinal trajectory of GGO status from large-scale radiology notes. METHODS We developed a bidirectional long short-term memory with a conditional random field–based deep-learning NLP pipeline to extract GGO and granular features of GGO retrospectively from radiology notes of 13,216 lung cancer patients. We evaluated the pipeline with quality assessments and analyzed cohort characterization of the distribution of nodule features longitudinally to assess changes in size and solidity over time. RESULTS Our NLP pipeline built on the GGO ontology we developed achieved between 95% and 100% precision, 89% and 100% recall, and 92% and 100% <i>F</i><sub>1</sub>-scores on different GGO features. We deployed this GGO NLP model to extract and structure comprehensive characteristics of GGOs from 29,496 radiology notes of 4521 lung cancer patients. Longitudinal analysis revealed that size increased in 16.8% (240/1424) of patients, decreased in 14.6% (208/1424), and remained unchanged in 68.5% (976/1424) in their last note compared to the first note. Among 1127 patients who had longitudinal radiology notes of GGO status, 815 (72.3%) were reported to have stable status, and 259 (23%) had increased/progressed status in the subsequent notes. CONCLUSIONS Our deep learning–based NLP pipeline can automatically extract granular GGO features at scale from electronic health records when this information is documented in radiology notes and help inform the natural history of GGO. This will open the way for a new paradigm in lung cancer prevention and early detection.
6592 Background: Clinical trial phenotyping is the process of extracting clinical features and patient characteristics from eligibility criteria. Phenotyping is a crucial step that precedes automated cohort identification from patient electronic health records (EHRs) against trial criteria. We establish a clinical trial phenotyping pipeline to transform clinical trial eligibility criteria into computable criteria and enable high throughput cohort selection in EHRs. Methods: Formalized clinical trial criteria attributes were acquired from a natural-language processing (NLP)-assisted approach. We implemented a clinical trial phenotyping pipeline that included three components: First, a rule-based knowledge engineering component was introduced to annotate the trial attributes into a computable and customizable granularity from EHRs. The second component involved normalizing annotated attributes using standard terminologies and pre-defined reference tables. Third, a knowledge base of computable criteria attributes was built to match patients to clinical trials. We evaluated the pipeline performance by independent manual review. The inter-rater agreement of the annotation was measured on a random sample of the knowledge base. The accuracy of the pipeline was evaluated on a subset of randomly selected matched patients for a subset of randomly selected attributes. Results: Our pipeline phenotyped 2954 clinical trials from five cancer types including Non-Small Cell Lung Cancer, Small Cell Lung Cancer, Prostate Cancer, Breast Cancer, and Multiple Myeloma. We built a knowledge base of 256 computable attributes that included comorbidities, comorbidity-related treatment, previous lines of therapy, laboratory tests, and performance such as ECOG and Karnofsky score. Among 256 attributes, 132 attributes were encoded using standard terminologies and 124 attributes were normalized to customized concepts. The inter-rater agreement of the annotation measured by Cohen’s Kappa coefficient was 0.83. We applied the knowledge base to our EHRs and efficiently identified 33258 potential subjects for cancer clinical trials. Our evaluation on the patient matching indicated the F1 score was 0.94. Conclusions: We established a clinical trial phenotyping pipeline and built a knowledge base of computable criteria attributes that enabled efficient screening of EHRs for patients meeting clinical trial eligibility criteria, providing an automated way to efficiently and accurately identify clinical trial cohorts. The application of this knowledge base to patient matching from EHR data across different institutes demonstrates its generalization capability. Taken together, this knowledge base will be particularly valuable in computer-assisted clinical trial subject selection and clinical trial protocol design in cancer studies based on real-world evidence.
Adenocarcinoma of the rectum accounts for approximately 5 percent of new cancer cases and for an estimated 31 percent of new cases of colorectal cancer. Although surgical resection remains the treatment of choice for almost all cases of this disease, a significant number of patients will develop locoregional recurrence after surgery if no other treatment is given. Relatively modest doses of radiation may improve long-term survival and can prevent locoregional relapse after surgery in many high-risk patients.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)