What is the clinical evidence from this study?

Study design: Cohort. Population: Chronic obstructive pulmonary disease (COPD) (n=5485). Intervention: Multi-omic prediction models (proteomic and metabolomic scores) vs. Baseline covariates alone. Primary outcome: Five-year change in volume-noise-bias-adjusted (VNB) lung density (ΔR² 0.015, p=0.002).

What question did this study set out to answer?

This research aims to evaluate multi-omic models for predicting emphysema progression using lung density metrics.

May 20, 2026

C101-04 A Multi-omic Score Predicting Emphysema Progression in COPDgene

Key Result

Integrating proteomic and metabolomic data into a multi-omic model modestly improved the prediction of five-year CT-defined emphysema progression compared to baseline covariates alone (ΔR² = 0.015, p = 0.002).

Key Points

This research aims to evaluate multi-omic models for predicting emphysema progression using lung density metrics.
Analyzed 5,485 COPDGene participants with plasma proteomic and metabolomic data linked to CT scans.
Utilized multiview cooperative learning and elastic net regression to construct prediction models.
Participants were split into training (80%) and testing (20%) sets with model tuning via 10-fold cross-validation.
The combined multi-omic model achieved the highest R² of 0.243 for predicting lung density decline (p = 0.002).
Integration of proteomic and metabolomic data increased explained variance from 0.228 to 0.240 (p = 0.005).
The final model included 348 features including proteins and metabolites enriched in key biological pathways.

Study Design

Type

Cohort (n=5,485)

Multicenter

Yes

Structured PICO

Does a multi-omic score improve the prediction of five-year emphysema progression in patients with COPD?

Population

5,485 COPDGene participants with overlapping plasma proteomic (SomaScan v4.1) and metabolomic (Metabolon) data

Intervention

Multi-omic prediction models integrating proteomic and metabolomic data

Comparator

Baseline clinical covariates alone or single-omic models

Outcome

Five-year change in volume-noise-bias-adjusted (VNB) lung density (emphysema progression)surrogate

Integrating proteomic and metabolomic data modestly improves the prediction of five-year CT-defined emphysema progression in COPD patients.

Main Result

Effect estimate: ΔR² 0.015

Absolute Event Rate: 0.243% vs 0.228%

p-value: p=0.002

Abstract

Abstract Introduction Chronic obstructive pulmonary disease (COPD) is a heterogeneous disorder driven by complex genetic, molecular, and environmental factors. Longitudinal change in lung density, measured by computed tomography (CT), provides a quantitative marker of emphysema progression. We used volume-noise-bias-adjusted (VNB) lung density, a CT-derived metric correcting for inspiratory volume, image noise, and scanner bias to yield estimates of five-year structural change. Omic risk scores (ORSs) from individual molecular layers have shown promise for cross-sectional COPD phenotypes, but their value in predicting long-term change in lung density remains uncertain. Here, we developed and evaluated multi-omic models to predict five-year emphysema progression in COPDGene. Methods We analyzed 5,485 COPDGene participants with overlapping plasma proteomic (SomaScan v4.1) and metabolomic (Metabolon) data collected at Visit 2, linked to CT scans at Visits 2 and 3 (∼5 years apart). We constructed multi-omic prediction models using multiview cooperative learning (Ding et al), which applies elastic net regression across omic layers. Participants were split 80:20 into training and testing sets, with model tuning via 10-fold cross-validation. ORSs were generated for each omic layer individually and jointly to predict five-year change in VNB lung density (negative change = density loss). Association models were adjusted for age, sex, race, height, BMI, pack-years, smoking status, scanner model, clinical center, days between visits, and baseline (Visit 2) lung density. Results Integrating proteomic and metabolomic data modestly improved prediction of five-year lung density decline compared with single-omic models. In the test set, adding the proteomic score to baseline covariates increased explained variance (R²) from 0.228 to 0.240 (ΔR² = 0.013, p = 0.005), while the metabolomic score increased R² to 0.235 (ΔR² = 0.007, p = 0.042). The combined multi-omic model achieved the highest performance (R² = 0.243, ΔR² = 0.015, p = 0.002) and was associated with greater longitudinal loss of lung density (p = 2.1 × 10−³), consistent with accelerated emphysema progression. The final model selected 348 features (261 proteins, 87 metabolites), enriched for pathways including complement activation, PI3K signaling, FGFR1 activation, and extracellular-matrix remodeling. Conclusions Multi-omic integration improved prediction of five-year CT-defined emphysema progression, reflecting coordinated immune, matrix, and exposure-related processes underlying structural lung injury. These interpretable multi-omic scores advance a systems-biology framework for identifying individuals at risk for rapid structural progression of COPD. External validation in COPD-enriched and population-based cohorts is in progress. This abstract is funded by: This work was supported by NHLBI grants U01 HL089897 and U01 HL089856 and by NIH contract 75N92023D00011.

Bookmark