Early cancer detection plays a pivotal role in improving the likelihood of survival by preventing tumor growth and metastasis. The National Cancer Plan released by the National Institutes of Health's National Cancer Institute in April 2023 has declared early cancer detection as one of its main goals (30).
The results of the study showed that the SpotitEarly screening test detects malignant lung, breast, colorectal, and prostate tumors in exhaled breath samples with 93.9% sensitivity and 94.3% specificity. Furthermore, the test demonstrates similar performance for early detection of cancer, with 94.8% sensitivity, a characteristic that sets it apart from many other screening tests which often struggle to maintain high sensitivity for early-stage cancer detection (31).
Other studies that used canines alone without AI for cancer detection in double-blind settings, reported a sensitivity range of 71–99% and a specificity range of 91–99% (14, 16–18) However, unlike the current study, they were conducted in non-commercialized settings, where the overall number of tested samples was small and the number of positive samples within a test set was fixed. Other studies that included varying numbers of cancer-positive samples within a test set - similar to this study - often exhibited limited performance (26, 32). Furthermore, in the current study the dogs were trained to detect several types of cancer in a single breath sample. The remarkable performance of SpotitEarly's test in this challenging scenario, surpassing that of previous studies, can be attributed to the innovative bio-hybrid approach. By integrating an AI layer atop the canines for decision-making, data analysis, and real-time monitoring, this approach not only enhances performance but also allows for fine-tuning the prediction algorithm to meet specific product requirements, particularly regarding the sensitivity and specificity optimization tradeoff. Moreover, the AI layer facilitates the integration of additional data sources beyond canine responses to samples, including patient demographics and medical information. Indeed, our analysis showed that performance outcomes decreased dramatically when AI was not used. Moreover, the simulated outcomes from the new bio-hybrid algorithm, which utilized more comprehensive data and a more adaptable model architecture, surpassed those of the currently used bio-hybrid algorithm.
The sensitivity of SpotitEarly’s tests for the detection of each cancer type was comparable to the sensitivity of gold standard screening tests, namely low-dose computed tomography (33) for lung cancer, mammography -- with or without --ultrasonography (34, 35) for breast cancer, fecal immunochemical tests (36) and colonoscopy (37) for colorectal cancer, and prostate-specific antigen for prostate cancer (38). Hence, it provides a multitype cancer screening test covering all four cancer types in a single test and may potentially demonstrate clinical utility for the test as an adjunct to current screening procedures. As the test is not invasive and the samples can be easily provided even in non-clinical settings, it could potentially increase compliance to cancer screening among the general population by serving as a preliminary screening step to gold standard screening. Furthermore, SpotitEarly's test yielded high-performance results for 17 cancer types that were not part of the canines' training. This observation supports the suggestion that various cancer types exhibit similar VOC patterns in addition to a specific VOC profile characteristic of specific tumors. This conclusion suggests the possibility of extending SpotitEarly's screening to detect additional types of cancer.
In addition to achieving promising accuracy results, the SpotitEarly test demonstrates notable efficiency in scaling compared to traditional gold standard screening methods. The rapid decision-making process by canines, occurring in less than a second, enables the testing of a vast number of samples within a short duration. Furthermore, the compact size of the testing environment allows for easy replication, facilitating laboratory scale-up. Additionally, scalability can be further enhanced by increasing the number of canines involved in the screening process.
Other tests for screening of multiple cancers are currently being developed, where the most prevalent test is based on liquid biopsy. Despite their innovative and appealing approach of testing dozens of cancer types in a single blood test, their sensitivity is low, particularly during the early stages when the detection's impact is most crucial. They fail to detect early cancers due to various limitations, including a low signal-to-noise ratio and the lack of distinct biomarkers for each type of cancer. (39). The Circulating Cell-Free Genome Atlas study using a multi-cancer detection test Galleri (GRAIL) was a validation study evaluating the role of targeted methylation-based multicancer early detection assay in a healthy population (NCT02889978). The results of the pre-specified sub-study, which included 4077 participants in an independent validation set which included 2823 participants with cancer and 1254 participants without cancer confirmed at one year of follow-up, reported specificity of 99.5% and sensitivity of 51.5% across all cancer types. The sensitivity for stage 1, 2, 3, and 4 cancers was 16.8%, 40.4%, 77.0%, and 90.1%, respectively. The false-positive rate of the test was 0.5% (40). The sensitivity of early-stage (I/II) cancer detection in the PATHFINDER study (NCT04241796), which analyzed data of 6621 adults aged ≥ 50 years without signs or symptoms of cancer who underwent testing with the GRAIL MCED was 12.7% (41) The SYMPLIFY study evaluated the GRAIL MCED test in participants with symptoms for potential gynecological, lung, or upper or lower gastrointestinal cancers. Out of 5,461 participants included in the evaluation, 368 were diagnosed with cancer, while a cancer signal was identified in 323 patients, of whom 244 were diagnosed with cancer. The reported sensitivity of the test was 66.3% (95% CI 61.2–71.1), and the specificity was 98.4% (95% CI 98.1–98.8%). Sensitivity increased with increasing age and cancer stage, from 24.2% in stage I to 95.3% in stage IV. (42). In comparison to these studies, the SpotitEarly test showed improved sensitivity and specificity of cancer detection, specifically at early stages.
The accuracy of the screening test in a commercial setting may be different that that obtained in a clinical trial given the exclusion of samples from participants with benign tumors, participants with active inflammatory disease, and participants who had smoked two hours prior to providing the breath sample. However, Sonoda et al. (16) reported that these parameters did not confound canine odor detection of early colorectal cancer. Breath odors associated with dietary factors, such as coffee, garlic, onion or other types of food could also interfere with detection accuracy. However, according to Ehmann et al.(17), lung cancer detection was independent of chronic obstructive pulmonary disease and the presence of tobacco smoke and food odors. Similarly, Biehl et al. (43) found no effect of consumption habits, nutrition, medications or concomitant diseases on canines’ ability to detect lung cancer in breath sample.
In conclusion, this study introduces a scalable, low cost multi-cancer screening method using a bio-hybrid approach of canines and AI that achieves high performance in early-stage detection using breath samples, within a setup that closely mirrors commercial-phase conditions. These results are expected to pave the way for the development of a new generation of cancer screening tests, enhancing cancer screening capabilities.