Early cancer detection significantly improves patient outcomes by enabling earlier treatment and reducing the need for aggressive therapies like chemotherapy and radiotherapy. Current screening methods are limited by their focus on specific cancer types and high costs, leaving many cancers undetected. This study aimed to develop a rapid, cost-effective liquid biopsy for detecting multiple cancer types using FTIR spectroscopy and machine learning. The test analyzed the complete biochemical profile of blood samples, capturing both tumor-derived and immune-derived information, which could improve early cancer detection compared to single-biomarker approaches that often miss early-stage cancers with minimal genetic material shedding. The non-specific nature of early cancer symptoms often leads to delayed diagnosis, making a pan-cancer detection method particularly useful for patients with unclear symptoms. This novel approach is designed to address these gaps in current cancer diagnostics.

The introduction cites existing research emphasizing the importance of early cancer detection and the limitations of current methods. Studies are referenced highlighting the success rate of surgical resection for early-stage cancers versus late-stage cancers, the challenges of widespread cancer screening due to low prevalence and high cost, and the limitations of existing liquid biopsy approaches focusing on single biomarkers. The authors point to the need for a rapid, cost-effective test capable of detecting multiple cancer types, particularly for patients with non-specific symptoms.

The study employed the Dxcovr® Cancer Liquid Biopsy, which utilizes FTIR spectroscopy to analyze blood serum samples. A large cohort of 2092 patients was studied, including 1542 with confirmed cancer diagnoses across eight cancer types (brain, breast, colorectal, kidney, lung, ovarian, pancreatic, and prostate) and 550 non-cancer controls (91 asymptomatic, 459 symptomatic). Machine learning algorithms were trained using a nested cross-validation strategy to prevent data leakage and minimize bias. The 70:30 split of training and testing data was repeated five times. Model hyperparameters were optimized to achieve desired sensitivity and specificity. Organ-specific classifiers were developed for each cancer type, comparing them against symptomatic non-cancer controls, with separate analyses for ovarian (versus female controls) and prostate (versus male controls) cancers. A pooled analysis classified 'any cancer' versus non-cancer patients. Feature importance analysis was conducted to identify the spectral regions most important for classification.

High AUC values (0.76-0.91) were obtained for individual cancer types. For the 'any cancer' classification against symptomatic non-cancer controls, the sensitivity-tuned model achieved 99% sensitivity (59% specificity), while the specificity-tuned model achieved 99% specificity (57% sensitivity). Specifically, the specificity-tuned model detected 64% of Stage I cancers. High detection rates were observed across all cancer stages for both sensitivity- and specificity-tuned models in the pooled analysis. Feature importance analysis identified specific wavenumber regions associated with different biomolecules (proteins, lipids, nucleic acids) as being critical for cancer classification. The Amide II band (around 1530 cm⁻¹) was particularly important in the overall cancer vs. non-cancer classification. Organ-specific analyses revealed that different wavenumber regions were important for different cancer types.

The study's results demonstrate the potential of this spectroscopic liquid biopsy for early cancer detection across multiple cancer types. The ability to tune the test for either sensitivity or specificity provides flexibility to adapt to varying clinical needs and healthcare system requirements. The use of symptomatic non-cancer controls provides a more realistic assessment of test performance in a real-world clinical setting compared to studies that use only healthy controls. The high AUC values and detection rates for early-stage cancers suggest that the Dxcovr® test could have a significant impact on improving early cancer diagnosis, which could lead to improved patient outcomes. The discussion also acknowledges that the test does not provide information to guide treatment but that it can be effectively combined with other tests like DNA-based tests.

The Dxcovr® Cancer Liquid Biopsy shows promise as a low-cost, rapid, and easily integrated blood test for detecting multiple cancer types, with a particular strength in detecting early-stage disease. Further research with larger cohorts and cancer-specific control groups is needed to validate these findings and optimize the test's performance. Future studies should also explore the potential of combining this test with other diagnostic methods to further enhance diagnostic accuracy and improve cancer management.

The study's limitations include the use of a single symptomatic non-cancer control group for all cancer types, which could potentially affect the accuracy of sensitivity and specificity measurements. A larger, more diverse patient cohort would strengthen the study’s results. Additionally, while the test demonstrates the potential for early detection, it does not provide information regarding specific cancer types or guide treatment decisions. Future research should investigate the use of cancer-specific control groups and explore potential for combining this test with other orthogonal tests like cell-free DNA analysis.