A blood-based lung cancer risk model may classify more patients who develop lung cancer within 1 year as eligible for low-dose computed tomography screening compared with current US Preventive Services Task Force criteria or the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial model in an international validation study.
Researchers developed and validated the protein-based Integrative Analysis of Lung Cancer Risk and Etiology (INTEGRAL)-Risk model as a prescreening tool to estimate short-term lung cancer risk using age, smoking exposure variables, and 13 circulating protein biomarkers.
The case-cohort study included 3,695 participants with a smoking history from the Lung Cancer Cohort Consortium. The researchers recruited participants from the United States, Europe, Asia, and Australia between 1985 and 2009, with follow-up for lung cancer and other outcomes through 2021. They included 2,305 randomly sampled participants and 1,390 participants diagnosed with lung cancer within 3 years following blood collection.
After statistical weighting, the study represented 323,570 participants, 57% of whom were female, with a median age of 60 years. Plasma or serum samples were assayed with the INTEGRAL protein panel in 2022.
The model was trained in seven predefined cohorts and tested in seven independent cohorts. In the testing set, the INTEGRAL-Risk model had an area under the receiver operating characteristic curve of 0.88 for predicting lung cancer within 1 year compared with 0.79 for the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCOm2012) model. Performance declined over longer prediction horizons, with values of 0.84 at 2 years and 0.81 at 3 years.
At a risk threshold chosen to match the specificity of 2021 US Preventive Services Task Force (USPSTF) criteria, the INTEGRAL-Risk model captured 85% of lung cancer cases occurring within 1 year compared with 70% with PLCOm2012 and 63% with USPSTF 2021 criteria. The quasi-number needed to screen to classify one lung cancer case as eligible was 215 with the INTEGRAL-Risk model vs 262 with PLCOm2012 and 290 with USPSTF 2021 criteria.
At a threshold matching the specificity of 2013 USPSTF criteria, the model captured 74% of 1-year lung cancer cases compared with 55% using PLCOm2012 and 49% using USPSTF 2013 criteria.
Subgroup analyses showed higher 1-year discrimination with INTEGRAL-Risk compared with PLCOm2012 among Asian, non-Hispanic Black, and non-Hispanic White participants. The model’s area under the curve was 0.88 vs 0.78 among Asian participants, 0.90 vs 0.71 among non-Hispanic Black participants, and 0.88 vs 0.81 among non-Hispanic White participants. Other racial and ethnic groups were too small for stratified analysis.
The model also improved discrimination among participants who were ineligible for screening under USPSTF 2021 criteria, with an area under the curve of 0.85 vs 0.72 for PLCOm2012. However, the model underpredicted risk among Asian and non-Hispanic Black participants, and the researchers wrote that recalibration should be considered prior to clinical implementation.
The INTEGRAL-Risk model’s discriminative advantage diminished over longer follow-up periods. In analyses using postdiagnosis survival time as a proxy because stage data were limited, the INTEGRAL-Risk model had higher discrimination than PLCOm2012 among participants with lung cancer who survived less than 6 months or 6 months to less than 2 years. The models performed similarly among participants who survived 2 years or longer.
The researchers highlighted that the INTEGRAL-Risk model was designed to refine eligibility for low-dose computed tomography rather than replace it or serve as a standalone screening test. The researchers wrote that adequately powered prospective studies are needed prior to broader implementation in screening programs.
Study limitations included the lack of assessment of lung cancer mortality, false-positive results, screening risks, cost-effectiveness, implementation feasibility, or acceptability. The researchers could not assess whether the INTEGRAL-Risk model differentiated lung cancers with different driver sequence variants because those data were unavailable.
“Compared with questionnaire-based approaches, the INTEGRAL-Risk model using 13 circulating protein biomarkers improved short-term lung cancer risk discrimination at 1 year, with demonstrable improvements in sensitivity for the detection of incident lung cancer,” wrote lead study author Hana Zahed, PhD, of the Early Detection, Prevention, and Infections Branch at the International Agency for Research on Cancer in France, and colleagues. “This model has potential to improve selection of high-risk individuals who are most likely to benefit from lung cancer screening,” they concluded.
Full disclosures can be found in the published study. The study was funded by the National Cancer Institute, the Cancer Research Foundation of Northern Sweden, and a grant from the Swedish Department of Health Ministry. Additional support came in part from the Intramural Research Program of the National Institutes of Health.
Source: JAMA
