Early cancer detection is generally considered an effective strategy for reducing patient mortality. In particular, screening programs for lung cancer, the most common cause of death worldwide, have significantly reduced mortality rates. One screening method is called low dose computed tomography (LD-CT). This approach detects a large fraction of lung nodules, abnormal growths in lung tissue, which are usually benign. However, a fraction of these nodules are so-called indeterminate pulmonary nodules (IPNs) and have an intermediate cancer risk. They are called indeterminate because they may or may not be cancerous. To definitively detemine whether a particular IPN is cancerous, invasive biopsy procedures are necessary; thus, a better method of differentiating cancerous IPNs from benign ones—a method which does not involve invasive procedures—is desirable. To address this need, Dr. Paul Lampe, a professor in the Translational Research Program, and Dr. Paul Kinahan, a professor of Radiology at the University of Washington, developed a noninvasive approach to more accurately determine whether IPNs are malignant or not: the PSR (plasma, semantic, radiomic) risk prediction model. This approach extracts information using: 1.) Plasma samples for identification of biomarkers, 2.) Semantic features (common characteristics of a tumor using a structured reporting format) observed by radiologists in CT images and 3.) Radiomic features (identifies image texture and other features that are not apparent to the human eye) analyzed by algorithms. Their results describing the method were recently published in Cancers.
The participants of this study were enrolled in the Fred Hutchinson Lung Cancer Early Detection and Prevention Clinic (LCEDPC). They were divided into two cohorts: FH1 and FH2. The FH1 cohort included 69 subjects with non-small cell lung cancer (NSCLC, case group) and 66 with benign nodules (control group) and this cohort was matched by age, gender, and cigarette pack years. Meanwhile, the FH2 cohort included 71 subjects with NSCLC and 78 controls and was unmatched by age, gender, and pack years. The FH1 cohort was used to generate the PSR risk prediction model. The FH2 cohort was used to test the predicted model trained in FH1.
To begin with, the authors looked for upregulation of biomarkers in the plasma samples. The authors noticed that several proteins were upregulated in the case group of the FH1 cohort, and several other proteins exhibited glycan modifications. Further, the autoantibody-antigen complex analysis revealed higher levels of IgG and IgM in the case group of the FH1 cohort when compared with the control group. These findings were validated in the FH2 cohort. The results showed that a variety of biomarkers were elevated in the plasma of patients with malignant IPNs in two independent cohorts.
For the semantic features, an experienced thoracic radiologist, blinded to clinical and histologic findings, reviewed the CT images of all the participants. For the radiomic features, the authors used PyRadiomics, an open-source package for radiomic data analysis with pre- and post processing steps that were extensively evaluated to improve repeatability and reproducabilty.
Finally, the authors generated the PSR risk model prediction using a panel of nine biomarkers: five plasma markers (ALPL, TNFRSF8, WNT5B, RGL1-IgG, and WNT10A-IgG), three semantic features (smooth margin, spiculated margin, and part-solid nodule density), and one radiomic feature ("least axis length"). This model yielded an area under the curve (AUC) value of 0.98 for the case group of the FH1 cohort and an AUC of 0.85 for the case group of the FH2 cohort (AUC is a statistic model used to determine how good a model is at classifying between two groups, in this case, case and control group. AUC closer to 1, the better the model is at classifying IPNs). The authors then incorporated known clinical risk factors such as age, gender, and pack-years into their PSR model. The AUC of their risk prediction model improved to 0.90 and was more accurate than a well-characterized clinical risk prediction model from Mayo Clinic (AUC = 0.80).
The PSR risk model prediction demonstrates the promise of a noninvasive approach for assessing the risk of IPNs. Based on their risk prediction model, the authors aim to assign a high or low risk score to each individual with IPN. The high-risk prediction score would identify patients who require immediate diagnostic follow-up. Low-risk prediction scores would indicate patients who can be monitored noninvasively with repeated imaging.
This spotlighted research was supported by the National Institutes of Health.
Fred Hutch/University of Washington/Seattle Children's Cancer Consortium members Drs. Paul Lampe and Paul Kinahan contributed to this work.
Lastwika KJ, Wu W, Zhang Y, Ma N, Zečević M, Pipavath SNJ, Randolph TW, Houghton AM, Nair VS, Lampe PD, Kinahan PE. Multi-Omic Biomarkers Improve Indeterminate Pulmonary Nodule Malignancy Risk Assessment. Cancers (Basel). 2023 Jun 29;15(13):3418. doi: 10.3390/cancers15133418.w