Setoguchi, Soko

e13559 Background: Recent advances in targeted therapies have increased the survival and quality of life for non-small cell lung cancer (NSCLC) patients. Understanding clinical characteristics and predicting HER2 status may facilitate efficient use of next-generation sequencing (NGS) testing and promote precision medicine/targeted therapies. Methods: NSCLC patients ≤65 years old, diagnosed between 01/2011 and 06/2021 who have NGS test results were identified in from two National Cancer Institute designated cancer centers in the Northeastern region of the United States. Demographics, behavioral, and clinical characteristics were extracted from electronic health record (EHR) data and EHR-linked cancer registry. Prevalence prediction modelling of HER2/ERBB2 mutation was performed using machine learning (ML) methods including logistic regression (LR), random forest (RF), naïve Bayes (NB), extreme gradient boosting (XGBoost) and multi-layer perceptron (MLP). 10- fold cross validation and k-nearest neighbor method were used to impute missing values. Under-sampling, oversampling and feature selection methods were tried. Results: Among 761 NSCLC patients with NGS reports available, 353 (46%) were ≤65 years old. Among 353 (mean age 57, 58% female, 88% non-Hispanic, 69% white, 57% had metastasis, 58% prescribed chemotherapy) 25 (7%) had a HER2 mutation. Based on univariate significance tests with LR on the complete data: cerebrovascular disease, BMI between 27.5-29.9, KRAS, EGFR and MET were significant with odds ratios (ORs) ranging from 0.2-3.9. In our prediction models, BMI, cerebrovascular disease, weight loss, hepatic metastases, diabetes, congestive heart failure, and KRAS, EGFR, MET and RET alteration status were significant predictors in multivariate analyses, suggesting the potential usefulness of these features in predicting HER2 mutation among patients with NSCLC. Area under the ROC curve (AUC) varied across different models from 70.9% (RF), 70% (XGB and MLP), 69.5% (LR) to 68.2% (Naïve Bayes). The best performing model with RF yielding an AUC ~71% can provide relative high specificity and acceptable sensitivity depending on the cut off. Conclusions: Our prediction models yielded AUC around 70% with the best performing model’s AUC being 71% in a subgroup aged ≤65. After further validation, our prediction model may be used to identify patients appropriate for repeat NGS testing who did not have HER2 testing in first molecular profiling. Future work is needed to predict status in repeat testing with previously test negative patients who relapse or have progression of disease. Random forest (RF) Model performance across different cutoff for the predicted probability. [Table: see text]

Background: Administrative data such as Medicare data are commonly used for health services research and comparative effectiveness studies. These databases are readily available and a good source of long-term data on healthcare utilization and health outcomes in real-world settings. Outcomes are captured using diagnostic codes, but the evidence on the validity of these codes is scarce in contemporary Medicare data. Methods: We linked the Reasons for Geographic and Racial Differences in Stroke (REGARDS) study data using SSN, date of birth and sex to 2003-2009 Medicare claims. In addition to the events captured by REGARDS phone interviews, medical charts were pursued for Medicare-identified strokes not previously identified in REGRADS. Events were adjudicated by stroke specialists using the retrieved medical charts. Using all the adjudicated strokes as gold standard, we calculated the sensitivity, specificity, PPV and NPV of inpatient stroke algorithms with ICD-9-CM codes [430, 431, 433.x1, 434.x1, 436] in the primary discharge diagnosis. Results: We successfully linked 15,089 REGARDS participants with mean age of 69.3, 52% female and 37% Black. We adjudicated 457 strokes, of which 48 were identified by claims only and not by interviews. The tested algorithms had high specificity (99.8-100%), PPV (88.6-90.5%) and NPV (98.9-99.9%), but low sensitivity (58.6-67.4%) (Table). Conclusions: High specificity and PPV of our algorithms to identify strokes in contemporary Medicare populations support their use in etiological and comparative effectiveness studies. While our inpatient Medicare algorithms identified 12% extra cases of strokes, their usefulness for estimating stroke incidence or related healthcare utilizations is limited by their low sensitivity. Linking cohorts to Medicare data is feasible and should be considered to increase the completeness of follow-up. Further studies are needed to evaluate more sensitive Medicare algorithms.

Introduction: The relative performance of carotid artery stenting (CAS) and carotid endarterectomy (CEA) among Medicare beneficiaries in routine clinical practice has not been established. Objective: To compare the performance of CAS relative to CEA among Medicare beneficiaries. Methods: We linked Medicare data (2000-2009) to the Society for Vascular Surgery’s Vascular Registry (SVS-VR; 2005-2008) and to the NCDR® Carotid Artery Revascularization and Endarterectomy Registry (CARE; 2006-2008/9). Medicare patients aged ≥66 years undergoing CAS or CEA were followed from the procedure date for the outcomes of death, stroke/transient ischemic attack (TIA), peri-procedural myocardial infarction (MI) or a composite endpoint for these outcomes. We derived high-dimensional propensity scores using registry and Medicare data to control for patient-level factors and adjusted for provider-level factors including past-year CAS/CEA physician and hospital volume, hospital ownership, teaching affiliation, and hospital size in a Cox regression model comparing CAS to CEA. Results: Among 5,254 SVS-VR (1,999 CAS and 3,255 CEA) and 4,055 CARE (2,824 CAS and 1,231 CEA) patients, CAS patients were more often at higher surgical risk (SVS-VR: 96.7% vs. 44.5%; CARE: 71.3% vs. 44.7%) and had a higher comorbidity burden. Crude outcome risks for death and stroke/TIA were lower for CEA. Adjusting for patient-level factors drove estimates downwards for CAS relative to CEA but only after further adjustment for provider-level factors did the performance of CAS and CEA become comparable on all outcomes (figure 1). Conclusion: Performance of CAS and CEA among Medicare beneficiaries was comparable after accounting for patient- and provider-level factors, which is consistent with landmark trials. Further studies are needed to understand the role of heterogeneity in provider-level characteristics.