bioRxiv Science⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/504878

Generating real-world tumor burden endpoints from electronic health record data: Comparison of RECIST, radiology-anchored, and clinician-anchored approaches for abstracting real-world progression in non-small cell lung cancer

Griffith, S. D.·Tucker, M.·Bowser, B.·Calkins, G.·Chang, C.-h.·Guardino, E.·Khozin, S.·Kraut, J.·Miksad, R.·You, P.·Schrag, D.·Abernethy, A. P.
View original source ↗Export metadata ↓

Abstract

Real-world evidence derived from electronic health records (EHRs) is increasingly recognized as a supplement to evidence generated from traditional clinical trials. In oncology, tumor-based Response Evaluation Criteria in Solid Tumors (RECIST) endpoints are collected in clinical trials. The best approach for collecting similar endpoints from EHRs remains unknown. We evaluated the feasibility of a traditional RECIST-based methodology to assess EHR-derived real-world progression (rwP) and explored non-RECIST-based approaches. In this retrospective study, cohorts were randomly selected from Flatiron Healths database of patient-level EHR data in advanced non-small cell lung cancer. A RECIST-based approach was tested for feasibility (N=26). Three non-RECIST abstraction approaches were tested for feasibility, reliability, and validity (N=200): (1) radiology-anchored, (2) clinician-anchored, and (3) combined. RECIST-based cancer progression could be ascertained from the EHRs of 23% of patients (6/26). In 87% of patients (173/200), at least one rwP event was identified using both the radiology- and clinician-anchored approaches. rwP dates matched 90% of the time. In 72% of patients (124/173), the first clinician-anchored rwP event was accompanied by a downstream event (e.g., treatment change); the association was slightly lower for the radiology-anchored approach (67%; 121/180). Median overall survival (OS) was 17 months (95% confidence interval [CI]: 14, 19). Median real-world progression-free survival (rwPFS) was 5.5 (95% CI: 4.6, 6.3) and 4.9 months (95% CI: 4.2, 5.6) for clinician-anchored and radiology-anchored approaches, respectively. Correlations between rwPFS and OS were similar across approaches (Spearmans rho: 0.65-0.66). Abstractors preferred the clinician-anchored approach as it provided more comprehensive context. RECIST cannot adequately assess cancer progression in EHR-derived data due to missing data and lack of clarity in radiology reports. We found a clinician-anchored approach supported by radiology report data to be the optimal, and most practical, method for characterizing tumor-based endpoints from EHR-sourced data.

Source connections

Explore related subjects

bioinformaticsPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Griffith, S. D., Tucker, M., Bowser, B., Calkins, G., Chang, C.-h., Guardino, E., Khozin, S., Kraut, J., Miksad, R., You, P., Schrag, D., Abernethy, A. P.. 2019-01-03. Generating real-world tumor burden endpoints from electronic health record data: Comparison of RECIST, radiology-anchored, and clinician-anchored approaches for abstracting real-world progression in non-small cell lung cancer. https://doi.org/10.1101/504878

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Deducing splicing events from transcript sets with splicelogic

Long-read RNA sequencing captures a greater proportion of full-length transcripts, enabling more accurate isoform identification, discovery of novel isoforms, and more precise differential transcript usage analysis. However, comparing transcripts, whether between conditions or between sets such as novel and annotated isoforms, yields complex isoform changes that are hard to manipulate, summarize, and visualize. This is compounded by vertical integration of some tools, where downstream analysis depends on the specific output of a particular upstream tool. To address this gap, we have developed splicelogic, an R/Bioconductor package that makes transcript-set comparisons biologically interpretable by turning them into discrete splicing events, such as exon skipping, alternative splice sites, intron retention, and alternative transcription start and end sites. We designed splicelogic to integrate easily with existing RNA-seq workflows and tools. The splicelogic functions are modular, readable, and extensible, making it easy for users to adapt core logic to their own custom scientific questions or novel splicing definitions. By obtaining interpretable splicing events from transcript changes, splicelogic enables downstream analyses, including annotation of RNA-binding protein recognition sites, alongside summary statistics and visualization. The splicelogic package is publicly available and distributed through Bioconductor.

bioinformatics↗
bioRxiv2026

Contrastive Learning Improves Generalization on M2OR Olfactory Binding Dataset

Olfaction-based biosensors stand to have profound impacts in fields such as medical diagnostics, national security, and drug discovery. A promising mode of vapor phase biodetection mimics volatile organic compound-olfactory receptor (VOC-OR) interactions with target-specific protein or peptide probes; however, lack of experimental coverage remains a core bottleneck. Current computational approaches aim to improve prediction of VOC-OR binding to reduce experimental burden, and current machine learning approaches fail to demonstrate a level of generalization that addresses current engineering challenges. Here, using the Molecular to Olfactory Dataset (M2OR) and Insect Olfactory Binding Protein Database (iOBPdb) datasets, data partitioning methods and their performance are evaluated on a neural network and contrastive learning framework on such VOC-OR binding datasets, varying: model training, architecture, and structure of the input data distribution. It was found that holding out certain chemical or proteomic motifs from the training data diminished model performance with a two layer neural network model. The model was able to generalize to state of the art performance (0.88 AUCROC) on every data split by applying a contrastive learning objective defined by shared molecular constituents and a stochastic sampler. This suggests the contrastive objective provides a strong inductive bias for neural networks to predict novel VOC-OR pairs. These findings highlight that model performance on pairwise biological interaction datasets can be inflated through identity leakage from constituent overlap, and emphasize the need for more robust evaluation protocols. Furthermore, we show contrastive regularization mitigates dataset degeneracy through enforcement of compatibility in the models latent geometry that may extend to other binder target-datasets where identity leakage confounds evaluation.

bioinformatics↗
bioRxiv2026

Plate design confounds donor variation and model evaluation in single-cell perturbation screens

Single-cell screens of immune cells from many donors are used to study why individuals respond differently and to benchmark response models. Each donor-perturbation combination typically occupies one well, controls are grouped, and donors share one layout, so technical effects can look biological. Using a 90-cytokine screen, two compound screens, four cohorts and simulations, we show that well effects explain most replicate-control variation in abundant cell types, controls carry a shared offset, and identical control wells differ by hundreds of differentially expressed genes at a standard threshold, about as many as two receptor-negative proteins do. Negative controls, design-aware estimators and perturbation-specific evaluation change the interpretation of four recent analyses. A deep-learning model that scores below a perturbation-agnostic predictor under standard evaluation outperforms a simple baseline on perturbation-specific differences in two of three cell types (significantly in one with stricter controls), but predicts less well the component shared with inert wells.

bioinformatics↗