Open and reproducible thoracic CT phenotypic age for mortality risk in the National Lung Screening Trial

Principal Investigator

Name
Kin Meng Eric Fang

Degrees
M.D., M.Med., M.Eng.

Institution
National University Hospital (Singapore) Pte Ltd

Position Title
Consultant Radiologist and Clinical Director of Diagnostic Imaging

Email
eric_km_fang@nuhs.edu.sg

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
NLST-1530

Initial CDAS Request Approval
Oct 12, 2026

Title
Open and reproducible thoracic CT phenotypic age for mortality risk in the National Lung Screening Trial

Summary
Low-dose chest CT contains quantitative information about skeletal muscle, adipose tissue, vertebral attenuation and vascular calcification that is associated with mortality. Composite CT health indices have previously been evaluated in the National Lung Screening Trial (NLST), but reproducibility and transportability remain limited by heterogeneous feature definitions, technical failure, acquisition variability and incomplete access to analytical implementations. We will establish an independently implemented, reproducible thoracic CT phenotypic-age benchmark using publicly available NLST DICOM images, open analytical components and linked clinical and mortality data requested through the Cancer Data Access System (CDAS).

Our primary question is whether a prespecified panel of explainable thoracic CT biomarkers can be combined into a mortality-anchored phenotypic-age score that adds predictive information beyond chronological age and routinely available risk factors. The cohort will comprise participants randomised to the CT arm with a technically usable baseline T0 examination and adequate follow-up. One baseline axial series per participant will be selected using prespecified reconstruction and image-quality rules. Open-source tools will derive candidate measurements of skeletal muscle quantity and attenuation, thoracic adiposity, vertebral attenuation and vascular calcification. Only measurements meeting prespecified anatomical-coverage, technical-success and quality-control thresholds will enter modelling.

Time at risk will begin on the baseline CT date. All-cause mortality will be analysed using death_days; participants without recorded death will be censored using fup_days, with both referenced to scr_days0. A penalised survival model will be developed using nested resampling. Predicted mortality risk will be mapped to the equivalent age on a chronological-age reference model to produce a thoracic CT phenotypic-age estimate and age gap. Performance will be assessed using discrimination, calibration, Brier score or index of prediction accuracy, and incremental value beyond chronological age, sex, race or ethnicity, body mass index, smoking status and pack-years. Internal-external validation across masked NLST centres or study components will assess transportability. Prespecified subgroup and sensitivity analyses will examine sex, race or ethnicity, smoking status, acquisition characteristics, and exclusion of participants with lung cancer at baseline or diagnosed within two years.

We will release analysis code, prespecified feature definitions and final model parameters only to the extent permitted by the executed CDAS Data Transfer and Use Agreement and any required disclosure review. No participant-level CDAS data or participant-level derived outputs will be released. Restricted data will be held in an access-controlled institutional environment, used only for this approved project and accessed only by authorised personnel. We will not attempt to identify or contact participants, investigators or study personnel.

Aims

* Technical feasibility and reproducibility: Determine which prespecified thoracic CT biomarkers can be extracted reliably from heterogeneous baseline NLST low-dose CT examinations using open tools and explicit quality-control thresholds. Assess the effects of acquisition characteristics and quantify automated extraction success, missingness and quality-control failure across biomarker domains.

* Phenotypic-age construction: Develop an interpretable, mortality-anchored thoracic CT phenotypic-age score and corresponding age gap using all-cause mortality and follow-up data. Include only biomarkers meeting prespecified technical and anatomical criteria, and map modelled mortality risk to the equivalent age on a chronological-age reference model.

* Validation and incremental value: Evaluate calibration, discrimination and transportability using nested resampling and internal-external validation across masked NLST centers or study components. Determine incremental prognostic value beyond chronological age, sex, race or ethnicity, body mass index, smoking status and pack-years. Conduct prespecified subgroup and sensitivity analyses addressing acquisition characteristics and lung cancer diagnosed at baseline or within two years.

Collaborators

Kin Meng Eric Fang National University Hospital (Singapore) Pte Ltd