Purpose Quantifying the uptake of screening mammography at a population level is challenged by suboptimal surveillance data. Insurance claims represent a potentially useful data source; however, claims data can be limited by misclassification. Using a quantitative bias analysis framework, we demonstrate an approach for dealing with imperfect accuracy in an insurance claims database. Methods Insurance claims were obtained at the census tract level for screening mammography for the State of Delaware for 2019 and 2022. A measurement model was used to relate the observed uptake of mammography to the true uptake via two approaches: (1) the total number of medical claims and (2) the proportion of government-funded health insurance. Probabilistic Monte Carlo methods yielded a simulation distribution of screening mammograms accounting for misclassification and the associated uncertainty in misclassification parameters. Results During the year 2022 and based on the observed measure, the statewide average uptake was 172 mammograms per 1,000 women age 40 and above. This figure increased to a median of 176 mammograms per 1,000 women (95% simulation interval: 173 to 179) using the first adjustment approach and 322 mammograms per 1,000 women (95% simulation interval: 309 to 336) using the second adjustment approach. Conclusion When working with administrative data sets for measuring the uptake of screening mammography in a population, researchers and health care administrators should account for potential underreporting.
Purpose Treatment with chimeric antigen receptor T-cell (CAR-T) therapies has led to remission in patients with relapsed/refractory hematologic malignancies. However, CAR-T therapies are associated with severe toxicities. An electronic patient-reported outcome (ePRO) system for CAR-T therapies was developed on a digital electronic clinical outcome assessment platform. The system was designed to capture and monitor symptoms and side effects in CAR-T patients within routine clinical care. This study aimed to assess the usability of the ePRO system and use the findings to inform refinements that enhance user experience. Methods Participants were recruited from a previous study and a National Health Service center. In two rounds of online sessions, participants completed a series of tasks, semistructured interviews, cognitive debriefing, and the system usability scale (SUS). Using the Fit between Individual, Task, and Technology framework, we evaluated system usability. Sessions were audiorecorded, transcribed, and analyzed inductively and deductively to generate themes. Results Fifteen participants, age 18-79 years, evaluated the ePRO system. Qualitative analysis identified eight themes. For effectiveness, 82% of tasks completed error-free for the application and 70% for the clinical dashboard. For efficiency, tasks required an average 29 minutes to complete for the application and 20 minutes for the dashboard. SUS scores indicated excellent user satisfaction. Conclusion This study identified key usability strengths, challenges, and areas for refinement of the ePRO system. Quantitative results showed high usability and user satisfaction, supported by qualitative insights highlighting specific issues and advantages. In response, targeted improvements were made, including refining welcome text, enhancing font readability, and updating the user manual. Overall, these findings demonstrate that the ePRO system is usable and acceptable to both patients and clinicians for symptom and quality-of-life monitoring in routine CAR-T care.
Purpose Metastatic castration-resistant prostate cancer (mCRPC) shows marked clinical heterogeneity and poor long-term survival, highlighting the need for simple and reliable risk stratification tools beyond the limited usability of existing complex prognostic models. In this study, we developed and validated PROGRESS (Prostate Cancer Global Risk Evaluation and Stratification Score), a simplified prognostic score derived through machine learning‑guided feature selection distilled into a simple laboratory-based tool. Methods Baseline data from 2,035 patients with mCRPC (training set) were embedded into synthetic representations, reduced with UMAP, and clustered with DBSCAN to discover outcome-related subpopulations. Feature contribution analysis informed a multivariable model that yielded three independent predictors (prostate specific antigen [PSA], alkaline phosphatase [ALP], and AST) composing PROGRESS. External validation used two independent data sets (n = 1,239) across different therapeutic settings. Overall survival (OS) was analyzed with Kaplan-Meier and Cox regression; discrimination was assessed by ROC-AUC, AUC(t), and Harrel C-index. Results Unsupervised modeling identified two patient risk subpopulations with significantly different OS rates (median 27.4 v 17.7 months; hazard ratio [HR], 2.20 [95% CI, 1.91 to 2.54]; P < .00001). The three retained predictors (PSA, ALP, and AST) were combined into a composite score (PROGRESS) that showed strong discrimination in the training cohort (ROC-AUC 0.89; AUC(t) 0.60; Harrel C-index 0.69). Using a prespecified cutoff, patients classified by PROGRESS as increased risk had significantly shorter survival than low-risk patients (median 18.3 v 25.6 months; HR, 1.72 [95% CI, 1.50 to 1.97]; P < .0001). PROGRESS prognostic performance was consistent across all validation cohorts (all P < .0001). Conclusion By combining ML-based pattern discovery with classical modeling, PROGRESS provides a simple, objective, and readily implementable prognostic tool using three inexpensive routine laboratory measures. Its consistent performance across independent data sets supports clinical utility for individual risk stratification, trial enrichment, and early decision making. Prospective real-world validation is warranted.
Purpose Patients with cancer experience nutrition challenges, but access to registered dietitians (RDs) is limited. The purpose of this study was to determine whether a specialized Artificial Intelligence (AI) Nutrition Platform trained on curated high-quality evidence is equivalent in providing nutrition information to patients with cancer compared with oncology-credentialed RDs. Methods Twenty nutrition queries were developed by an expert panel, encompassing common cancer types, symptoms, and food allergies/preferences. These queries were administered to the AI Platform and to four nationally recruited human dietitians. Three expert reviewers with >10 years of cancer-specific nutrition experience blindly reviewed all query responses using a modified version of the Quality Assessment of Medical AI (mQAMAI) instrument, which assigns a score between 1 and 5 for accuracy, clarity, relevance, completeness, and usefulness (total quality score between 5 and 25). The primary outcome was mQAMAI total score. Equivalence was evaluated a priori using two one-sided tests (TOST). Secondary outcomes included readability and response time. Results The AI Nutrition Platform met equivalence criteria with human dietitians at ±3 to ±5 points (mean difference: 0.9 points; 90% CI, -0.27 to 2.15; TOST P < .05), while scoring higher on individual components: clarity (4.2 v 3.9; P = .001), relevance (4.1 v 3.9; P = .04), and usefulness (3.8 v 3.5; P = .02). AI responses were easier to read (Flesch-Kincaid grade level 7.2 v 9.8; P < .001) with faster response time (8.5 v 18.5 minutes per two queries; P < .001). Conclusion The AI Platform was equivalent to human dietitians for cancer nutrition information, delivering responses with greater clarity, relevance, usefulness, and readability in less time. This tool could broaden access to nutrition support for patients in settings where access to dietitians is limited.
