A DNA Methylation-based algorithm Improves Lung Cancer risk prediction in the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial.

Authors

Dawes K, Mills JA, Hoffman RM, Froehlich EM, deBlois K, Sieren JC, Williams C, Merkle S, Long JD, Beach SR, Philibert RA

Affiliations

• Behavioral Diagnostics, Coralville, IA 52241, USA.
• Department of Psychiatry, University of Iowa, Iowa City, IA, USA.
• Department of Internal Medicine, University of Iowa, Iowa City, IA 52242, USA.
• Department of Radiology, University of Iowa, Iowa City, IA 52242, USA.
• Information Management Services, Inc. Calverton, MD 20705, USA.
• Department of Psychiatry, University of Iowa, Iowa City, IA, USA; Department of Biostatistics, University of Iowa, Iowa City, IA 52242, USA.
• Department of Psychology, University of Georgia, Athens, GA 30602, USA.
• Behavioral Diagnostics, Coralville, IA 52241, USA; Department of Psychiatry, University of Iowa, Iowa City, IA, USA. Electronic address: robert-philibert@uiowa.edu.

Abstract

INTRODUCTION: Measuring DNA methylation levels at cg05575921 can improve prediction of lung cancer (LC) risk in a screening eligible population. However, these findings were based on a limited number of largely White study participants, with a history of heavy smoking (>30 pack years [PY]) and the cg05575921 based-metric was not directly compared to existing standards for LC prediction.] METHOD: We determined cg05575921 methylation levels in a nested case and control cohort featuring 1156 LC cases and 3039 controls, matched for age, sex, race and self-reported smoking status (current and former), who participated in the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial. We then constructed survival algorithms that tested whether adding cg05575921 methylation levels to a model consisting of PLCO_m2012and PY improved the prediction of time to lung cancer diagnosis as compared to the PLCO_m2012algorithm.

RESULTS: Models adding cg05575921 methylation levels to PLCO_m2012and PY significantly improved prediction of LC occurrence over 20-year follow up for subjects who report current or former smoking. In this set of subjects matched for age, sex and smoking status, a simple algorithm using cg05575921 and PY outperformed the PLCO_m2012in all smokers (20-yr area under the curve [AUC] 0.725 vs 0.689), ≥ 20 PY smokers (20-yr AUC 0.662 vs 0.634) and < 20 PY smokers (20-yr AUC 0.666 vs 0.549). Among participants with < 20 PY smoking histories, those with cg05575921 methylation levels < 80% were at 3.3-fold greater risk for LC than those with similar PY history but with cg05575921 methylation levels > 80%.

CONCLUSION: The use of cg05575921 methylation levels can improve the accuracy of LC risk prediction and may be particularly useful identifying persons with a < 20 PY history who are at elevated risk for LC.These findings require validation in an external screening population.