Evidence of compensatory neural hyperactivity in a subgroup of chemotherapy-treated breast cancer survivors and its association with brain aging.

Chemotherapy-related cognitive impairment (CRCI) remains poorly understood in terms of the mechanisms of cognitive decline. Neural hyperactivity has been reported on average in cancer survivors, but it is unclear which patients demonstrate this neurophenotype, limiting precision medicine in this population. We evaluated a retrospective sample of 80 breast cancer survivors and 80 non-cancer controls, age 35-73, for which we had previously identified and validated three data-driven, biological subgroups (biotypes) of CRCI. We measured neural activity using the z-normalized percent amplitude of fluctuation from resting state functional magnetic resonance imaging (MRI). We tested established, quantitative criteria to determine if hyperactivity can accurately be considered compensatory. We also calculated brain age gap by applying a previously validated algorithm to anatomic MRI. We found that neural activity differed across the three CRCI biotypes and controls (F = 13.5, p < 0.001), with Biotype 2 demonstrating significant hyperactivity compared to the other groups (p < 0.004, corrected), primarily in prefrontal regions. Alternatively, Biotypes 1 and 3 demonstrated significant hypoactivity (p < 0.02, corrected). Hyperactivity in Biotype 2 met several of the criteria to be considered compensatory. However, we also found a positive relationship between neural activity and brain age gap in these patients (r = 0.45, p = 0.042). Our results indicated that neural hyperactivity is specific to a subgroup of breast cancer survivors and, while it seems to support preserved cognitive function, it could also increase the risk of accelerated brain aging. These findings could inform future neuromodulatory interventions with respect to the risks and benefits of up or downregulation of neural activity.

Long-Term Engagement of Patients With Advanced Cancer: Results From the EPAC Randomized Clinical Trial.

Importance: The long-term effect of interventions that assist patients with establishing their end-of-life care preferences among patients with cancer remain relatively unknown. Objective: To evaluate the association of a long-term intervention of a lay health worker-led advance care planning intervention among patients with advanced stages of cancer with overall survival and end-of-life health care use and costs. Design, Setting, and Participants: This follow-up study of the EPAC randomized clinical trial conducted between August 2013 and February 2015 used data from 9.4 years after the first patient was enrolled with a data cut-off date of February 1, 2023. Overall, 213 participants with stage 3 or 4 or recurrent cancer in the US Veterans Affairs Palo Alto Health Care System were included. Interventions: A 6-month lay health worker-led education and support intervention to assist patients with establishing their end-of-life preferences vs usual care. Main Outcomes and Measures: The outcomes of interest were overall survival, risk of death, restricted mean survival time, and palliative care, hospice, and acute care use in the final 30 days before death for participants who died. Results: Among 213 participants randomized and included in the intention-to-treat analysis, the mean (SD) age was 69.3 (9.1) years; 211 (99.1%) were male, 2 (0.90%) were female. There were no demographic or clinical characteristic imbalances at enrollment. As of February 1, 2023, 188 had died. The intervention group had a 25% reduction in risk of death (hazard ratio, 0.75; 95% CI, 0.56-0.98); more palliative care (44 [50.0%] vs 35 [35.0%]) and hospice use (64 [72.7%] vs 53 [53.0%]); and lower emergency department use (20 [22.7%] vs 47 [47.0%]), hospitalizations (17 [19.3%] vs 46 [46.0%]), and median (IQR) total health care costs (median [IQR], $1637 [$383-$9026] vs $18 520 [$4790-$50 729]) than control group participants. Conclusions and Relevance: The effects of the lay health worker-led intervention remain durable, with nearly complete follow-up, supporting integration into routine cancer care. Trial Registration: ClinicalTrials.gov Identifier: NCT02966509.

Computing the Cost of Care per Patient per Day for Patients With Metastatic Neuroendocrine Neoplasms.

PURPOSE: Patients with well-differentiated, low-grade metastatic neuroendocrine neoplasms (NENs) usually have a long median survival and require complex, expensive care over many years at multidisciplinary centers. The cost burden for patients and institutions serves as a barrier to care. Understanding the drivers of these costs and whether intense monitoring adds value will help to optimize value-based care. METHODS: We adapted the cost of care per patient per day (CCPD) validated methodology to measure cost while accounting for varying follow-up duration. We queried the Stanford NEN Database, which aggregates data from the electronic health record and other electronic sources, to study patients with metastatic NENs receiving regular care at Stanford. Current Procedural Terminology codes for services incurred during the monitoring period for each patient were mapped to the corresponding cost conversion factor and date in the Medicare fee schedule. RESULTS: Two hundred two patients between 2010 and 2017 were studied with a mean CCPD of $119.11 in US dollars (USD); NEN-specific systemic therapy made up 55% of this cost. Somatostatin analogs were the costliest systemic therapy. Systemic therapy was the driver of cost differences among patients with various primary tumor types, stage of disease, tumor differentiation and grade, and functional hormone status. Patients in the most expensive CCPD group did not have a significant survival benefit (P = .66). CONCLUSION: The CCPD methodology was effective in studying cancer care value in NENs. Systemic therapy, specifically somatostatin analogs, was the primary driver of cost, and intense monitoring and higher-cost care did not improve survival outcomes.

Real-World Impact of Prophylactic Growth Factor Use on Timing of Febrile Neutropenia and Infection After High-Risk Chemotherapy.

BACKGROUND: Prophylactic growth-factor therapy with granulocyte colony-stimulating factor (G-CSF) reduces the risk of febrile neutropenia (FN) in patients with breast cancer initiating myelosuppressive chemotherapy. However, little is known about the protective benefit early in the chemotherapy cycle. METHODS: To assess the relationship between G-CSF prophylaxis and incidence of FN/infection in week 1 versus beyond week 1 of the first chemotherapy cycle, a retrospective study was conducted using Medicare claims from 2005 through 2020 among patients with breast cancer initiating high-risk chemotherapy. Two cohorts were compared based on G-CSF prophylaxis within 3 days following chemotherapy initiation. The primary outcome was FN or infection, defined as hospitalization with neutropenia, fever, or infection diagnosis. Secondary outcomes were a stricter definition of FN and infection-related hospitalization. Unadjusted and regression-adjusted proportions of patients experiencing each outcome during week 1 versus beyond week 1 of the first chemotherapy cycle were compared. RESULTS: Of 78,810 patients meeting all inclusion criteria (>98% female; mean age, 69 years), 79% initiated TC (docetaxel/cyclophosphamide), 14% TCH (docetaxel/carboplatin/trastuzumab), and 7% TAC (docetaxel/doxorubicin/cyclophosphamide). Among patients receiving G-CSF (74%), incidence of first-cycle FN/infection was lower compared with patients not receiving G-CSF (overall, 6% vs 13%; TAC, 12% vs 19%; TC, 6% vs 12%; TCH, 5% vs 15%). However, patients who received G-CSF were generally more likely to experience FN/infection in week 1 (adjusted odds ratio [aOR], 1.24 for all; 1.73 for TAC; 1.35 for TC; and 0.76 for TCH). Results were similar for strictly defined FN (overall aOR, 1.29 for week 1 and 0.12 for beyond week 1) and infection-related hospitalization (overall aOR, 1.33 for week 1 and 0.27 for beyond week 1). CONCLUSIONS: Overall, the rates of chemotherapy-related FN and infection in week 1 of the first chemotherapy cycle are similar for patients receiving and not receiving G-CSF, suggesting continued risk in week 1 despite prophylactic G-CSF.

Grade 4 Neutropenia Frequency as a Binary Risk Predictor for Adverse Clinical Consequences of Chemotherapy­Induced Neutropenia: A Meta-analysis.

Neutropenia is the major toxicity of myelosuppressive cancer chemotherapy. Grade 4 neutropenia (Gr4N) is a measure of chemotherapy-induced neutropenia (CIN) severity. We conducted a meta-analysis of CIN data. Gr4N incidence was significantly correlated with febrile neutropenia (FN), days of severe neutropenia (DSN), and nadir absolute neutrophil count (ANC), which are all important predictors of morbidity. With a Gr4N threshold of 65%, both FN and DSN were below levels for low risk of adverse CIN outcomes. Gr4N was highly predictive for adverse CIN outcomes, and a 65% threshold demarcated low vs. high risk for FN and other adverse CIN outcomes.

Efficacy of Plinabulin vs Pegfilgrastim for Prevention of Docetaxel-Induced Neutropenia in Patients With Solid Tumors: A Randomized Clinical Trial.

Importance: Prevention of chemotherapy-induced neutropenia (CIN) and its clinical consequences is an unmet need for which plinabulin, a selective immunomodulating microtubule-binding agent, is being tested. Objective: To demonstrate noninferiority between plinabulin and pegfilgrastim for days of severe neutropenia in cycle 1 in patients with solid tumors treated with docetaxel. Design, Setting, and Participants: The Plinabulin vs Pegfilgrastim for the Prevention of Docetaxel-Induced Neutropenia in Patients With Solid Tumors (PROTECTIVE-1) double-blind phase 3 randomized clinical trial was performed in multiple centers in China, Russia, Ukraine, and the US. Participants included patients with breast, prostate, or non-small cell lung cancer treated with single-agent docetaxel chemotherapy. Data were collected from June 1, 2018, to January 31, 2019. The database was locked on February 18, 2021. Data analysis was based on intention to treat and safety and performed from October 5, 2018, to February 23, 2021. Interventions: Plinabulin, 40 mg, plus placebo or pegfilgrastim, 6 mg, plus placebo. Main Outcomes and Measures: The primary end point was day of severe neutropenia in cycle 1. Additional end points included clinical consequences of CIN (febrile neutropenia, hospitalizations, infections, antibiotic use, and modifications of chemotherapy dose), patient-reported outcomes for bone pain score, markers for immune suppression (neutrophil-to-lymphocyte ratio [NLR] of >5), immature neutrophils (band, promyelocyte, and myelocyte counts >0), and safety. Results: Among the 105 patients included in the analysis (65 [6.19%] women; median age, 59 [range, 31-81] years), the primary end point was met within a noninferiority margin of 0.65 days, with a mean difference of 0.52 days (98.52% CI, 0.40-0.65 days). Grade 4 neutropenia frequency in cycle 1 was not significantly different. Plinabulin had earlier onset of action with less grade 4 neutropenia in week 1 of cycle 1. Plinabulin had fewer adverse clinical consequences with rates of febrile neutropenia (0 of 52 vs 1 of 53 [1.9%]), infections (4 of 52 [7.7%] vs 8 of 53 [15.1%]), chemotherapy dose delay of more than 7 days (2 of 52 [3.8%] vs 3 of 53 [5.7%]), and permanent chemotherapy discontinuation (7 of 52 [13.5%] vs 14 of 53 [26.4%]). Patients receiving plinabulin had significantly less bone pain (difference, -0.67 [95% CI, -1.17 to -0.16]; P = .01) and a better immunosuppressive profile (NLR >5 at day 8, 2 of 52 [3.8%] vs 24 of 51 [46.0%]; P < .001). Plinabulin was well tolerated, with comparable safety to pegfilgrastim. Conclusions and Relevance: Plinabulin has comparable efficacy to pegfilgrastim for the prevention of CIN, with better safety and a better immunosuppressive profile. Plinabulin's same-day dosing compared with pegfilgrastim's next-day dosing offers distinct advantages, including reducing use of health care services. Trial Registration: ClinicalTrials.gov Identifier: NCT03102606.

Machine Learning Applied to Electronic Health Records: Identification of Chemotherapy Patients at High Risk for Preventable Emergency Department Visits and Hospital Admissions.

PURPOSE: Acute care use (ACU) is a major driver of oncologic costs and is penalized by a Centers for Medicare & Medicaid Services quality measure, OP-35. Targeted interventions reduce preventable ACU; however, identifying which patients might benefit remains challenging. Prior predictive models have made use of a limited subset of the data in the electronic health record (EHR). We aimed to predict risk of preventable ACU after starting chemotherapy using machine learning (ML) algorithms trained on comprehensive EHR data. METHODS: Chemotherapy patients treated at an academic institution and affiliated community care sites between January 2013 and July 2019 who met inclusion criteria for OP-35 were identified. Preventable ACU was defined using OP-35 criteria. Structured EHR data generated before chemotherapy treatment were obtained. ML models were trained to predict risk for ACU after starting chemotherapy using 80% of the cohort. The remaining 20% were used to test model performance by the area under the receiver operator curve. RESULTS: Eight thousand four hundred thirty-nine patients were included, of whom 35% had preventable ACU within 180 days of starting chemotherapy. Our primary model classified patients at risk for preventable ACU with an area under the receiver operator curve of 0.783 (95% CI, 0.761 to 0.806). Performance was better for identifying admissions than emergency department visits. Key variables included prior hospitalizations, cancer stage, race, laboratory values, and a diagnosis of depression. Analyses showed limited benefit from including patient-reported outcome data and indicated inequities in outcomes and risk modeling for Black and Medicaid patients. CONCLUSION: Dense EHR data can identify patients at risk for ACU using ML with promising accuracy. These models have potential to improve cancer care outcomes, patient experience, and costs by allowing for targeted, preventative interventions.

Spotlight on International Quality: COVID-19 and Its Impact on Quality Improvement in Cancer Care.

This report from ASCO's International Quality Steering Group summarizes early learnings on how the COVID-19 pandemic and its stresses have disproportionately affected cancer care delivery and its delivery systems across the world. This article shares perspectives from eight different countries, including Austria, Brazil, Ghana, Honduras, Ireland, the Philippines, South Africa, and the United Arab Emirates, which provide insight to their unique issues, challenges, and barriers to quality improvement in cancer care during the pandemic. These perspectives shed light on some key recommendations applicable on a global scale and focus on access to care, importance of expanding and developing new treatments for both COVID-19 and cancer, access to telemedicine, collecting and using COVID-19 and cancer registry data, establishing measures and guidelines to further enhance quality of care, and expanding communication among governments, health care systems, and health care providers. The impact of the COVID-19 pandemic on cancer care and quality improvement has been and will continue to be felt across the globe, but this report aims to share these experiences and learnings and to assist ASCO's international members and our global fight against the pandemic and cancer.

Health management via telemedicine: Learning from the COVID-19 experience.

At the onset of the COVID-19 (coronavirus disease 2019) pandemic, telemedicine was rapidly implemented to protect patients and healthcare providers from infection. It is unlikely that care delivery will fully return to the pre-COVID form. Telemedicine offers many opportunities to improve care efficiency, accessibility, and patient outcomes, but many challenges exist related to technology interoperability, the digital divide, and usability. We propose that telemedicine evolve to support continuity of care throughout the patient journey, including multidisciplinary care teams and the seamless integration of data into the clinical workflow to support a learning healthcare system. Importantly, evidence is needed to support this paradigm shift in care delivery to ensure the quality and efficacy of care delivered via telemedicine. Here, we highlight gaps and opportunities that need to be addressed by the biomedical informatics community to move forward with safe and effective healthcare delivery via telemedicine.

Headache outcomes of a sleep behavioral intervention in breast cancer survivors: Secondary analysis of a randomized clinical trial.

BACKGROUND: Breast cancer survivors often have persisting headache. In a secondary analysis of the Brief Behavioral Therapy for Cancer-Related Insomnia (BBT-CI) clinical trial (ClinicalTrials.gov identifier NCT02165839), the authors examined the effects of BBT-CI on headache outcomes in patients with breast cancer. METHODS: Patients with breast cancer who were receiving chemotherapy were randomly assigned to receive either the BBT-CI intervention or the Healthy EAting Education Learning for healthy sleep (HEAL) control intervention, and both were delivered over 6 weeks by trained staff. Headache outcomes and heart rate variability (HRV) were measured at baseline, 6 weeks, 6 months, and 12 months. Mixed-effects models were used to examine longitudinal headache outcomes in the groups according to the intention to treat. Principal component analysis and agglomerative hierarchical clustering were conducted to reduce 16 variables for data-driven phenotyping. RESULTS: Patients in the BBT-CI arm (n = 73) exhibited a significant reduction in headache burden over time (P = .02; effect size [Cohen d] = 0.43), whereas the reduction was not significant among those in the HEAL arm (n = 66). The first principal component was positively loaded by headache, sleep, fatigue, and nausea/vomiting and was negatively loaded by cognitive, physical, and emotional functioning. Agglomerative hierarchical clustering revealed 3 natural clusters. Cluster I (n = 58) featured the highest burden of headache, insomnia, and nausea/vomiting; cluster II (n = 50) featured the lowest HRV despite a low burden of headache and insomnia; and cluster III (n = 31) showed an inverse relation between HRV and headache-insomnia, signifying autonomic dysfunction. CONCLUSIONS: BBT-CI is efficacious in reducing headache burden in breast cancer survivors. Patient phenotyping demonstrates a headache type featuring sleep disturbance, nausea/vomiting, and low physical functioning-revealing similarities to migraine. LAY SUMMARY: Breast cancer survivors often have persisting headache symptoms. In patients with cancer, treatment of chronic headache disorders using daily medications may be challenging because of drug interactions with chemotherapy and other cancer therapies as well as patients' reluctance to add more drugs to their medicine list. Headache and sleep disorders are closely related to each other. This study demonstrates that a sleep behavioral therapy reduced headache burden in breast cancer survivors. In addition, the majority of headache sufferers had a headache type with similarities to migraine-featuring sleep disturbance, nausea/vomiting, and low physical functioning.

Diverse patient trajectories during cytotoxic chemotherapy: Capturing longitudinal patient-reported outcomes.

BACKGROUND: High-value cancer care balances effective treatment with preservation of quality of life. Chemotherapy is known to affect patients' physical and psychological well-being negatively. Patient-reported outcomes (PROs) provide a means to monitor declines in a patients' well-being during treatment. METHODS: We identified 741 oncology patients undergoing chemotherapy in our electronic health record (EHR) system who completed Patient-Reported Outcomes Measurement Information System (PROMIS) surveys during treatment at a comprehensive cancer center, 2013-2018. PROMIS surveys were collected before, during, and after chemotherapy treatment. Linear mixed-effects models were performed to identify predictors of physical and mental health scores over time. A k-mean cluster analysis was used to group patient PROMIS score trajectories. RESULTS: Mean global physical health (GPH) scores were 48.7 (SD 9.3), 47.7 (8.8), and 48.6 (8.9) and global mental health (GMH) scores were 50.4 (8.6), 49.5 (8.8), and 50.6 (9.1) before, during, and after chemotherapy, respectively. Asian race, Hispanic ethnicity, public insurance, anxiety/depression, stage III cancer, and palliative care were predictors of GPH and GMH decline. The treatment time period was also a predictor of both GPH and GMH decline relative to pre-treatment. Trajectory clustering identified four distinct PRO clusters associated with chemotherapy treatment. CONCLUSIONS: Patient-reported outcomes are increasingly used to help monitor cancer treatment and are now a part of care reimbursement. This study leveraged routinely collected PROMIS surveys linked to EHRs to identify novel patient trajectories of physical and mental well-being in oncology patients undergoing chemotherapy and potential predictors. Supportive care interventions in high-risk populations identified by our study may optimize resource deployment. NOVELTY AND IMPACT: This study leveraged routinely collected patient-reported outcome (PROMIS) surveys linked to electronic health records to characterize oncology patients' quality of life during chemotherapy. Important clinical and demographic predictors of declines in quality of life were identified and four novel trajectories to guide personalized interventions and support. This work highlights the utility of monitoring patient-reported outcomes not only before and after, but during chemotherapy to help advert adverse patient outcomes and improve treatment adherence.

Treatment and Monitoring Variability in US Metastatic Breast Cancer Care.

PURPOSE: Treatment and monitoring options for patients with metastatic breast cancer (MBC) are increasing, but little is known about variability in care. We sought to improve understanding of MBC care and its correlates by analyzing real-world claims data using a search engine with a novel query language to enable temporal electronic phenotyping. METHODS: Using the Advanced Cohort Engine, we identified 6,180 women who met criteria for having estrogen receptor-positive, human epidermal growth factor receptor 2-negative MBC from IBM MarketScan US insurance claims (2007-2014). We characterized treatment, monitoring, and hospice usage, along with clinical and nonclinical factors affecting care. RESULTS: We observed wide variability in treatment modality and monitoring across patients and geography. Most women received first-recorded therapy with endocrine (67%) versus chemotherapy, underwent more computed tomography (CT) (76%) than positron emission tomography-CT, and were monitored using tumor markers (58%). Nearly half (46%) met criteria for aggressive disease, which were associated with receiving chemotherapy first, monitoring primarily with CT, and more frequent imaging. Older age was associated with endocrine therapy first, less frequent imaging, and less use of tumor markers. After controlling for clinical factors, care strategies varied significantly by nonclinical factors (median regional income with first-recorded therapy and imaging type, geographic region with these and with imaging frequency and use of tumor markers; P < .0001). CONCLUSION: Variability in US MBC care is explained by patient and disease factors and by nonclinical factors such as geographic region, suggesting that treatment decisions are influenced by local practice patterns and/or resources. A search engine designed to express complex electronic phenotypes from longitudinal patient records enables the identification of variability in patient care, helping to define disparities and areas for improvement.

Limited English Proficiency and Disparities in Health Care Engagement Among Patients With Breast Cancer.

PURPOSE: Race and ethnicity have been shown to affect quality of cancer care, and patients with low English proficiency (LEP) have increased risk for serious adverse events. We sought to assess the impact of primary language on health care engagement as indicated by clinical trial screening and engagement, use of genetic counseling, and communication via an electronic patient portal. METHODS: Clinical and demographic data on patients with breast cancer diagnosed and treated from 2013 to 2018 within the Stanford University Health Care system were compiled via linkage of electronic health records, an internal clinical trial database, and the California Cancer Registry. Logistic and linear regression models were used to evaluate for association of clinical trial engagement and patient portal message rates with primary language group. RESULTS: Patients with LEP had significantly lower rates of clinical trial engagement compared with their English-speaking counterparts (adjusted odds ratio [OR], 0.29; 95% CI, 0.16 to 0.51). Use of genetic counseling was similar between language groups. Rates of patient portal messaging did not differ between English-speaking and LEP groups on multivariable analysis; however, patients with LEP were less likely to have a portal account (adjusted OR, 0.89; 95% CI, 0.83 to 0.96). Among LEP subgroups, Spanish speakers were significantly less likely to engage with the patient portal compared with English speakers (estimated difference in monthly rate: OR, 0.43; 95% CI, 0.24 to 0.77). CONCLUSION: We found that patients with LEP had lower rates of clinical trial engagement and odds of electronic patient portal enrollment. Interventions designed to overcome language and cultural barriers are essential to optimize the experience of patients with LEP.

Distress Screening Through Patient-Reported Outcomes Measurement Information System (PROMIS) at an Academic Cancer Center and Network Site: Implementation of a Hybrid Model.

PURPOSE: Cancer care guidelines recommend regular distress screening of patients, with approximately one in three patients with cancer experiencing significant distress. However, the implementation of such programs is variable and inconsistent. We sought to assess the feasibility of implementing a hybrid electronic and paper screening tool for distress in all patients coming to a large academic cancer center and an associated integrated network site. METHODS: Patients at an academic cancer center (Stanford Cancer Center) and its associated integrated network site received either an electronic or on-paper modified Patient-Reported Outcomes Measurement Information System-Global Health questionnaire, to assess overall health and distress. We used the Reach, Effectiveness, Adoption, Implementation, and Maintenance implementation framework to test and report on the feasibility of using this questionnaire. Iterative workflow changes were made to implement the questionnaire throughout the healthcare system, including processes to integrate with existing electronic health records. RESULTS: From June 2015 to December 2017, 53,954 questionnaires representing 26,242 patients were collected. Approximately 30% of the questionnaires were completed before the visit on an electronic patient portal. The number of patients meeting the positive screen threshold remained around 40% throughout the study period. Following assessment, there were 3,763 referrals to cancer supportive services. Of note, those with a positive screen were more likely to have a referral to supportive care (odds ratio, 6.4; 95% CI, 5.8 to 6.9; P < .0001). CONCLUSION: The hybrid electronic and on-paper use of a commonly available patient-reported outcome tool, Patient-Reported Outcomes Measurement Information System-Global Health, as a large-scale distress screening method, is feasible at a large integrated cancer center.

Development and Use of Natural Language Processing for Identification of Distant Cancer Recurrence and Sites of Distant Recurrence Using Unstructured Electronic Health Record Data.

PURPOSE: Large-scale analysis of real-world evidence is often limited to structured data fields that do not contain reliable information on recurrence status and disease sites. In this report, we describe a natural language processing (NLP) framework that uses data from free-text, unstructured reports to classify recurrence status and sites of recurrence for patients with breast and hepatocellular carcinomas (HCC). METHODS: Using two cohorts of breast cancer and HCC cases, we validated the ability of a previously developed NLP model to distinguish between no recurrence, local recurrence, and distant recurrence, based on clinician notes, radiology reports, and pathology reports compared with manual curation. A second NLP model was trained and validated to identify sites of recurrence. We compared the ability of each NLP model to identify the presence, timing, and site of recurrence, when compared against manual chart review and International Classification of Diseases coding. RESULTS: A total of 1,273 patients were included in the development and validation of the two models. The NLP model for recurrence detects distant recurrence with an area under the curve of 0.98 (95% CI, 0.96 to 0.99) and 0.95 (95% CI, 0.88 to 0.98) in breast and HCC cohorts, respectively. The mean accuracy of the NLP model for detecting any site of distant recurrence was 0.9 for breast cancer and 0.83 for HCC. The NLP model for recurrence identified a larger proportion of patients with distant recurrence in a breast cancer database (11.1%) compared with International Classification of Diseases coding (2.31%). CONCLUSION: We developed two NLP models to identify distant cancer recurrence, timing of recurrence, and sites of recurrence based on unstructured electronic health record data. These models can be used to perform large-scale retrospective studies in oncology.