Edwin Ostrin

MD, PhD

University of Texas MD Anderson Cancer Center

Assistant Professor

PLCO (Learn more about this study)

2024-0139

Sep 9, 2025

Biomarker panels to predict lung cancer in those who have never smoked

Lung cancer is the leading cause of cancer mortality. While tobacco smoking continues to be the largest risk factor for developing lung cancer, smoking rates are falling along with lung cancer mortality and incidence. However, early detection by low-dose CT based screening is limited to those at highest risk, who have smoked over 20 pack-years, are still smoking or who have quit within the last 15 years, and are over the age of 50. Approximately half of new diagnoses of lung cancer occur in those ineligible for screening, and as smoking rates fall, these are becoming a profound, under-researched clinical problem. For instance, even in those who have never smoked, lung cancer would be the fifth-leading cause of cancer death. To date, there is no widely-accepted screening or early detection strategy for detecting lung cancer in lower risk individuals.
Our lab group has extensive experience with developing and validating biomarker panels, including in samples from the PLCO trial. We have recently shown that a 4-protein biomarker panel, consisting of an immunoassay to pro-surfactant protein B, CA125, CEA, and CYFRA21-1, can improve on clinical risk models and better identify those who may benefit from screening than current eligibility criteria in the US. We have also shown that this same panel can improve lung cancer mortality prediction and help to distinguish benign from malignant nodules. In PLCO samples, we have also discovered and validated a 4-metabolite panel, comprised of the oncometabolites diacetylspermine, creatine riboside, the carbohydrate antigen N-acetyllactosamine, and hypoxanthine, that complements the 4MP performance. In lower risk samples from the PLCO (those with 10-20 pack-year history of smoking), as well as lower risk samples from the Physicians Health Study, the 4MP and 4MetP continue to show near-equivalent performance to those in higher risk strata. We have also recently tested the 4MP and 4MetP in samples collected from individuals with lung cancer and no smoking history. A combination of the 4MP and 4MetP performed admirably in this cohort, showing an area under the curve of the receiver operating characteristic (ROC) of 0.82 (95% CI 0.68-0.90). We thus hypothesize that a composite panel, consisting of the 4MP, 4MetP, combined with clinical risk models, may identify individuals with no smoking history who have a risk that may justify CT-based screening.
The PLCO intervention arm included 115 individuals with no smoking history who developed lung cancer. We wish to obtain these samples, along with 10-fold matched controls, for assembly of additional validation cohorts. We will also obtain longitudinal samples and data for these individuals, as we have shown value in longitudinal trends of the 4MP. We will measure the 4MP, 4MetP, and calculate clinical risk scores including the PLCOall2014 score. We will conduct ROC analyses, measuring specificity and sensitivity at pre-determined outpoints equivalent to current screening guidelines. Using a predetermined set-aside validation set, we will fully validate a combination of the 4MP, 4MetP, and clinical risk scores, specifically looking for those with a >1% 6-year risk of lung cancer.

- Aim 1: To evaluate the predictive performance of the 4MP and 4MetP for 1-year risk prediction of lung cancer among participants in the PLCO cohort without a
smoking history. In the PLCO, there were 154 individuals in the intervention arm with no history of smoking that developed lung cancer. We are requesting the most proximate blood samples to lung cancer diagnosis for these individuals, along with 10 matched controls for a nested cohort study. We will measure the 4MP using a well-validated multiplex immunoassay. The 4MetP is evaluated using mass spectrometry. For each of these, we will conduct receiver operating curve analysis, including measurement of the sensitivity, specificity, positive predictive value, negative predictive value. The panels will be evaluated alone and in combination, as well as being combined with clinical risk models including the PLCOm2012 and PLCOall2014 models. We will divide the sample set into a training and validation cohort to conduct linear regression and machine learning approaches for construction of an optimal composite biomarker panel for those no smoking history. Previously, we have used a variety of survival models to evaluate the ability of the 4MP to predict lung cancer-specific death and shown that the 4MP can predict lung cancer death even when accounting for comorbidities and competing causes of mortality. Such an approach has the potential to reduce overdiagnosis, which is a major concern when considering broadening screening to those with lower risk. We will take a similar approach to evaluate the performance of the 4MP and 4MetP to predict lung cancer and lung cancer death in those with no smoking history. Using cumulative incidence models, we will calculate subdistributional and cause-specific hazard ratios for lung cancer death predicted by the 4MP, the 4MetP, and clinical risk models.

- Aim 2: To assess whether an algorithm that consider repeat biomarker testing improves sensitivity and lead-time detection of lung cancer among participants who have never smoked in the PLCO cohort. We will also request all longitudinally collected samples for the 10:1 control:case cohort described above. We will evaluate the performance of using a single threshold versus using a parametric empirical Bayes algorithm to develop a dynamic, individualized threshold to indicate lung cancer risk. As above, we will conduct sensitivity, specificity, positive predictive value, and negative predictive value assessments of each of these algorithms.