Related Experiment Video
Updated: May 28, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Locked-Window EQ-5D-5L (Index and VAS) Benchmarking in Sarcoma Care: Rule-Based Traffic-Light Classification Across
Isabel Gloor1, Beatrice Meier2, Jehona Rexhai2
1Department of Orthopaedics and Trauma, LUKS Sarcoma-IPU, University Teaching Hospital LUKS, Spitalstrasse, 6000 Lucerne, Switzerland.
Background:
Value-based sarcoma care requires outcome measures that reflect the patient perspective; however, many sarcoma episodes begin with near-normal function and undergo necessary morbidity for oncologic control, making simple "improvement" an unreliable proxy of value. In routine care, patient-reported outcome data are often irregular and incomplete, limiting benchmarking and learning across institutions. We therefore developed a rule-based EQ-5D-5L (index and VAS) traffic-light framework and evaluated its feasibility and benchmarking signal in two institutions.
Methods:
We performed a retrospective, two-institution cohort analysis of 729 malignant and intermediate episodes, defined using a prespecified histology behavior mapping. PROM evaluation was anchored to a hierarchical T0 (index surgery date; if unavailable, radiotherapy start date; if unavailable, systemic therapy start date where a valid and interpretable start date was available). EQ-5D-5L index and EQ-VAS were assigned to prespecified locked windows: baseline (-90 to +14 days preferred; +15 to +90 days fallback), 12 months (180-365 days; target 270), and 24 months (660-820 days; target 730). A rule-based traffic-light classification was applied at 12 and 24 months (RED if index < 0.75 or VAS < 50; GREEN if index ≥ 0.85 and VAS ≥ 70; otherwise YELLOW). PROM evaluability was defined as the availability of at least one valid EQ-5D-5L index and/or EQ-VAS value within each window.
Results:
PROM evaluability in locked windows was feasible but incomplete. Baseline PROMs were available for 107/729 episodes (14.7%), 12-month PROMs for 119/729 (16.3%), and 24-month PROMs for 84/729 (11.5%). At 12 months, evaluable episodes included 75 from Institution A and 44 from Institution B; at 24 months, 56 and 28, respectively. Traffic-light outputs showed heterogeneity at both timepoints and clearer cross-institution difference at 24 months than at 12 months. At 12 months, the distribution was predominantly GREEN in both institutions (Institution A: 73.3% GREEN, 9.3% YELLOW, 17.3% RED; Institution B: 65.9% GREEN, 18.2% YELLOW, 15.9% RED; p = 0.373). At 24 months, Institution A maintained a high GREEN proportion with a low RED fraction (76.8% GREEN, 17.9% YELLOW, 5.4% RED), whereas Institution B showed a lower GREEN proportion and higher YELLOW/RED fractions (50.0% GREEN, 25.0% YELLOW, 25.0% RED; p = 0.014). Absolute EQ-5D-5L medians remained high overall, but the follow-up distributions showed a broader lower tail in Institution B.
Conclusions:
A prespecified EQ-5D-5L (index and VAS) traffic-light framework anchored by hierarchical T0 and evaluated in locked windows yields interpretable patient-perspective benchmarking signals in real-world sarcoma care. The approach was operationally feasible within the evaluable subset and appeared more discriminative at 24 months than at 12 months, while incomplete PROM capture remains a major implementation limitation for representative and reliable network-scale benchmarking and learning.
