Leveraging multi-cancer blood tests to improve diagnostic efficiency for patients with nonspecific signs and symptoms

Most cancers (86% in the US) are not detected by recommended screening [1] and often not diagnosed until symptoms arise. Although certain symptoms are associated with a specific cancer type or organ system (e.g., breast lump, painless hematuria, painless jaundice, rectal bleeding) and have a high predictive value for cancer that facilitates a directed workup, most symptoms are nonspecific, with low predictive value [2]. As nonspecific symptoms, such as unexplained weight loss, fatigue and abdominal pain, do not precisely point to a particular disease or organ, achieving an efficient diagnostic pathway based on clinical presentation is hindered by lack of focality and direction to workups. Regardless of the underlying disease—cancer or not—delayed diagnosis can cause potential harm to patients [3]. Although nonspecific symptoms may be indicators of cancer, inefficient or undirected diagnostic workups can lead to an extended time to diagnosis, later-stage diagnosis, delayed treatment initiation and worse outcomes [4–6]. In this commentary, we highlight the negative consequences of inefficient diagnostic evaluations and consider novel approaches that could triage symptomatic patients based on cancer probability to optimize clinical decision making, which in turn may improve clinical outcomes and resource utilization. Specifically, we discuss the potential for blood-based multi-cancer early detection (MCED) tests to aid diagnostic evaluations of patients with nonspecific signs and symptoms.

1. The diagnostic dilemma of nonspecific symptoms for patients & physicians

Patients may delay seeking medical attention for nonspecific symptoms for several reasons. Patients may be unaware of the symptoms' association with cancer, afraid of the diagnosis, concerned about wasting their doctor's time or without access to care [4,7]. Furthermore, symptoms of other underlying medical conditions might overlap with cancer-associated symptoms (e.g., symptoms of irritable bowel syndrome can mask those of colon cancer), so patients with comorbid conditions may not seek timely care. Evaluation of United Kingdom (UK) public health awareness campaigns about signs and symptoms of cancer indicate that awareness campaigns may help increase the number of patients seeking medical care, but thus far, the impact on stage at diagnosis and survival is small [8].

Once patients decide to seek medical care, their nonspecific symptoms may not only indicate a wide range of cancers but also noncancer conditions [5]. Nonspecific and non-focal signs and symptoms pose a diagnostic dilemma for physicians, contributing to inefficient, lengthy, resource intensive and potentially unsafe diagnostic evaluations. Though the severity of a potential cancer diagnosis can drive an aggressive cancer-focused evaluation, nonspecific symptoms have a low (<5%) positive predictive value for any individual type of cancer [2,5] or other noncancer diseases, providing little direction to these evaluations. For most cancer types, physicians lack tools that can accurately predict the probability and nature of a potential cancer. Without such tools, fear of missing a cancer often drives potentially prolonged and unfocused testing that is out of proportion to the diagnostic yield and can result in physical, financial and psychological harms to the patient.

In fact, an evaluation of patients from England's National Cancer Diagnosis Audit found that delays occurred for almost one-quarter of patients with cancer, most commonly during the interval between when tests were requested versus performed [9]. These delays can increase the median time to diagnosis by approximately 2 months, with 10% of patients experiencing a delay longer than 5 months [9]. Even short diagnostic delays could risk cancers progressing to more advanced stages, which could negatively impact treatment eligibility and patient prognosis. A scenario worse than a delayed cancer diagnosis is a missed diagnosis, which comprised almost one-half of all primary care diagnostic errors in a US study of malpractice claims in the ambulatory care setting [10]. An important consequence of diagnostic delays is an escalation of symptoms to the point where the symptoms are severe, so the patient seeks treatment at emergency departments or urgent care facilities. Emergency presentations are costly and associated with reductions in short-term and long-term survival, as the cancers, once identified, are more likely to be advanced and treatment options are limited [4,11].

Diagnostic inefficiencies have negative consequences for patients without cancer as well. The time and testing it takes to evaluate for cancer extends identification of the true disease and exposes the patient to unnecessary and potentially invasive tests and harms. Whether or not a patient presenting with nonspecific symptoms ultimately is diagnosed with cancer, a prolonged diagnostic journey could introduce an increased financial burden for patients, influenced by variable insurance coverage, out-of-pocket costs, and productivity and income loss [12]. Financial stress, along with the fear and uncertainty of eventual diagnosis, has negative impacts on mental wellbeing. Multiple studies have also linked extended diagnostic intervals with worse health-related quality of life and poorer survival [6].

Taken together, there is a clear unmet clinical need for innovations to support triage and identify the most suitable diagnostic pathways for patients who present with nonspecific signs and symptoms.

2. Novel tools to direct & improve efficiency of diagnosis

A promising new approach to single and multi-cancer detection is liquid biopsy, in which samples of biofluids are analyzed for cancer biomarkers. Blood is a common liquid biopsy biofluid because it is easy and non-invasive to collect peripherally and can contain biomarkers from throughout the body. Blood-based tests utilize isolated tumor-derived analytes from peripheral blood. Rapidly dividing tumor cells are prone to replication errors that lead to increased rates of necrosis and apoptosis. The large quantity of this tumor cell debris overwhelms normal clearance mechanisms and is released into the bloodstream. These circulating analytes, including circulating tumor cells, exosomes, proteins, metabolites and nucleic acids, have been examined by multiple studies of early cancer detection [13].

Of these circulating analytes, tumor-derived cell-free DNA (cfDNA), either alone or in combination with other analytes, has successfully been translated to clinical use for cancer screening, diagnosis, disease profiling and disease monitoring. For example, multiple companion diagnostic tests are commercially available to evaluate which patients may benefit from specific cancer treatments (e.g., Guardant360® CDx, Epi proColon®, therascreen® PIK3CA RGQ PCR kit, FoundationOne Liquid® CDx) and to detect molecular residual disease and recurrence (Signatera™). These technologies are limited in focus, with most reliant on cfDNA mutation analysis and tailored to a single or a few cancer types, meaning these tests would not be used as diagnostic aids for patients presenting with nonspecific symptoms that do not point to a particular cancer.

Pairing next-generation sequencing and machine learning algorithms has enabled the development of MCED tests as cancer screening tools. Current MCED assays include somatic mutation, fragmentation profile or methylation profile analysis to distinguish cancer-derived cfDNA from noncancer cfDNA. Several blood-based MCED tests are in development (reviewed in [14]) with the CancerSEEK test (now being developed as Cancerguard™, Exact Sciences Corporation) and Galleri® test (GRAIL, Inc.) demonstrating success in prospective, interventional clinical trials of cancer detection in asymptomatic screening populations (DETECT-A [15] and PATHFINDER [16], respectively). DETECT-A included 10,006 female participants and ultimately diagnosed 96 cancers [15]. The version of the CancerSEEK test evaluated in DETECT-A included a four-step process: baseline blood test, confirmatory blood test, multi-disciplinary review committee case review and diagnostic PET-CT (for localization of potential cancers). The blood test evaluated 16 gene mutations and 9 protein biomarkers. Specificity was 99.6% and the positive predictive value (PPV) was 28.3% for the blood test plus diagnostic PET-CT. The version of the test evaluated in DETECT-A did not include a machine learning component; machine learning methods were added to a later version of the test to improve sensitivity and specificity and provide a localization algorithm to predict the two most likely sites of origin out of seven potential sites (colorectum, ovary, pancreas, breast, upper gastrointestinal, lung or liver) [17].

This test continues to be refined and the subsequent case-control study, ASCEND 2, which included 6354 participants (1438 cancers and 4916 noncancers), reported a specificity and sensitivity of 98.5 and 50.9%, respectively, for all cancers but did not report on a cancer localization prediction feature [18]. The version of the test trained and tested in this study used methylation and protein biomarkers. After finalizing biomarker selection, the test will be assessed in a prospective, interventional, longitudinal trial [19].

The Galleri test evaluated in PATHFINDER (NCT04241796) utilized targeted methylation sequencing of cfDNA and machine learning classifiers to identify a shared cancer signal. When a cancer signal was detected, the test predicted the cancer signal origin (CSO; i.e., the organ or tissue type where the tumor arose) from 21 predetermined categories corresponding to major tumor types [16]. PATHFINDER included 6621 participants and diagnosed 122 cancers in 121 participants. The test demonstrated 99.1% specificity and 38% PPV, with CSO prediction accuracy of 97%. Refinement of this test is ongoing to improve test performance, and it is being evaluated in several clinical trials, including a randomized controlled trial (ISRCTN91431511). The Galleri test received Breakthrough Device designation from the US Food and Drug Administration (FDA) in 2019 and is currently the only MCED test that is commercially available.

In addition to screening, a possible extension of these MCED technologies is application in the diagnostic testing space. However, to the best of our knowledge, there are no publications reporting MCED tests that have been optimized for use as diagnostic (not screening) tools in symptomatic patients; rather, studies thus far have evaluated the potential diagnostic performance of an MCED test optimized for screening asymptomatic populations [20,21]. Diagnostic MCED tests would ideally have acceptable sensitivity and negative predictive value (NPV) for multiple cancers to avoid numerous exploratory tests and risk of false reassurance, capture cancers for which there are no recommended screening tests, point to the likely organ of origin and have acceptable specificity to limit unnecessary diagnostic workup in those without cancer.

3. Beyond screening: initial studies show potential for the diagnostic application of MCED technology in symptomatic populations

Blood-based multi-cancer detection is an area of active research, and several studies have reported development, validation and performance of cfDNA tests for cancer detection over the past several years. Yet only two recently published studies—both with the GRAIL MCED test—have evaluated performance in a symptomatic population [20,21]. Because the methylation profiling of this MCED test allows for accurate CSO predictions, the test is well suited for translation as a diagnostic tool. Despite being optimized for screening [22], initial studies of the MCED test in symptomatic individuals have demonstrated its promise as a potential diagnostic aid to stratify cancer risk and facilitate efficient workups.

To begin to investigate if the GRAIL MCED test could function as a diagnostic tool, performance of the MCED test was evaluated in a subgroup of participants from the case-control Circulating Cell-free Genome Atlas (CCGA) substudy 3 who had symptoms that raised suspicion of cancer prior to diagnosis [20]. Participants with clinically presenting cancers either had a cancer diagnosis following symptoms before study enrollment or had symptoms at enrollment and were later confirmed to have cancer within the study enrollment window. Ultimately, 2036 participants with clinically presenting cancer were analyzed. Noncancer participants with underlying medical conditions were also included (n = 548) to represent the population with symptoms that could be attributable to cancer or another disease. Overall specificity was 99.5% (95% CI: 98.4–99.8), suggesting that specificity was not affected by underlying biological aberrations related to symptoms. Overall sensitivity was 64.3% (95% CI: 62.2–66.4), and among patients with a confirmed cancer diagnosis, 80% (1115/1438) of cancers stage II to IV were detected. For 10 cancers with generally poor prognosis (lung, ovary, anus, pancreas, esophagus, head and neck, colon/rectum, urothelial tract, liver/bile duct and cervix), sensitivity exceeded 75%. Interestingly, sensitivity for GI cancers (colon/rectum, esophagus, gallbladder, liver/bile duct, pancreas, stomach) was particularly high at 84.1% (95% CI: 80.6–87.1) across all stages. Importantly, the test identified signals associated with multiple cancer types, including those that lack robust screening tests. CSO was correctly predicted for 90.3% (95% CI: 88.6–91.9) of participants with a positive test and a reported CSO, indicating that this test could provide valuable direction to diagnostic workups.

The SYMPLIFY study (ISRCTN10226380) was the first large-scale trial designed to assess the performance of the GRAIL MCED test in symptomatic patients referred from primary care [21]. The study was conducted in the UK, and enrolled participants who were referred for urgent investigation to diagnostic centers for suspected gynecological, lung, or upper or lower gastrointestinal cancers or to a rapid diagnostic center for nonspecific symptoms suspicious for cancer. Ultimately, 5461 participants were analyzed (n = 368 with diagnosed cancer and n = 5093 without diagnosed cancer). Performance largely aligned with the CCGA substudy 3 subpopulation analysis [20]. Overall specificity was 98.4% (95% CI: 98.1–98.8) and sensitivity was 66.3% (95% CI: 61.2–71.1) [21]. For stage II to IV cancers, 81.7% (210/257) were detected. Across all cancer types, PPV was 75.5% (95% CI: 70.5–80.1) and NPV was 97.6% (95% CI: 97.1–98.0). CSO prediction accuracy was 85.2% (95% CI: 79.8–89.3), although further investigation suggests that predictions classified as erroneous were largely explained by shared tumor biology (e.g., cancers caused by human papillomavirus, such as head and neck, anus and cervix). Again, sensitivity for cancer signal detection was highest for upper GI cancers (esophagus, pancreas, stomach, gallbladder, liver), with 80.4% (95% CI: 66.1–90.6) sensitivity and 99.1% (95% CI: 98.2–99.6) NPV for participants with upper GI cancer symptoms.

Results from the SYMPLIFY study also supported the predictive capability of the MCED test for both positive (cancer signal detected) and negative (no cancer signal detected) results [21]. For individuals with a positive result, the overall probability of cancer increased approximately tenfold. Conversely, for individuals with a negative result, the overall probability of cancer decreased by approximately two-thirds. Probabilities ranged by cancer type and symptoms. This post-test cancer risk information, combined with the high specificity and PPV of the MCED test, suggests that a positive test could be used to indicate a need for a more aggressive cancer evaluation. Combining the post-test risk information with the test's CSO prediction could guide primary care providers (PCPs) regarding urgency, extent and focus of a workup. Moreover, the CSO prediction could identify cancer sites that may not have been suspected during the original referral, saving valuable resources and time. Indeed, in the SYMPLIFY study, 47 and 25% of cancers diagnosed in the upper GI and gynecological pathways, respectively, were incongruent with the original referral pathway. The performance characteristics of the SYMPLIFY study could be further improved (e.g., higher NPV to improve utility in ruling out further investigations for cancer) by optimizing the machine learning algorithms of the MCED test for a symptomatic population with higher background cancer incidence than a general asymptomatic population.

Overall, these results support the feasibility of further developing this test as a diagnostic tool for PCPs and specialists to stratify patients by their likelihood of cancer and thus help resolve cancer suspicion. Implementation of and experience with MCED tests for diagnostic applications will continue to provide guidance on the clinical application of such tests.

4. Next steps & considerations for clinical application

When considering the clinical application of MCED tests for diagnostic purposes, it is important to assess their impact on health equity and their cost. On one hand, MCED tests have the potential to improve health equity. An important advantage to blood-based MCED tests is that they are noninvasive and thus easy to implement in the primary care setting. This reduces logistical barriers to access, such as geography or socioeconomics. Furthermore, increasing the efficiency of diagnostic intervals may reduce stress and financial costs [12], which is critical for scaling efforts toward more efficient cancer detection. However, the potential for a positive impact on health equity will likely depend on test coverage by private and public insurance plans. Insurance coverage is a challenge, and clinical utility data will be needed to attain meaningful coverage that facilitates access. Further, studies of the cost–effectiveness of an MCED test for diagnostic use in populations with nonspecific symptoms will be needed to support implementation into clinical practice.

Continued improvement of MCED tests and an understanding of how to implement their use in a clinical setting will benefit from accrual of additional evidence. An advantage of studies within large practice networks would be access to a broad patient population, allowing for evaluation of critical metrics beyond test accuracy, such as the potential impacts on imaging orders, referrals, time to diagnosis, misdiagnoses, patient anxiety and costs. Research to engage with and understand the priorities of patients, caregivers and practitioners will also be essential. MCED tests therefore provide a promising opportunity for test developers to partner with academia and community-facing organizations to evaluate how this type of test could be used for diagnostic applications and could affect PCP and patient consultation patterns as well as their empowerment and control of diagnostic pathways.

Funding Statement

GRAIL, Inc.

Author contributions

Article content was developed by and is the responsibility of the authors. Conceptualization, AG Singal, KN Kurtzman and MJ Thompson; writing-original draft preparation, reviewing and editing, AG Singal, KN Kurtzman and MJ Thompson. All authors have read and agreed to the published version of the manuscript.

Financial disclosure

AG Singal has served as a consultant or on advisory boards for FujiFilm Medical Sciences, Exact Sciences, Roche, Glycotest, GRAIL, Freenome, Abbott and Universal Dx and has research support from the National Cancer Institute (U01 CA271887 and U01 CA283935). KN Kurtzman was an employee of GRAIL, Inc. and holds equity in Illumina, Inc. MJ Thompson is an employee of Google, LLC and holds shares in Alphabet, but all work contained in this manuscript predates his employment at Google.

Competing interests disclosure

AG Singal has served as a consultant or on advisory boards for FujiFilm Medical Sciences, Exact Sciences, Roche, Glycotest, GRAIL, Freenome, Abbott and Universal Dx and has research support from the National Cancer Institute (U01 CA271887 and U01 CA283935). KN Kurtzman was an employee of GRAIL, Inc. and holds equity in Illumina, Inc. MJ Thompson is an employee of Google, LLC and holds shares in Alphabet, but all work contained in this manuscript predates his employment at Google. The authors have no other competing interests or relevant affiliations with any organization or entity with the subject matter or materials discussed in the manuscript apart from those disclosed.

Writing disclosure

Medical writing and editorial support were provided by AL Thomas and J Hepker of Prescott Medical Communications Group (Chicago, IL, USA) and were funded by GRAIL, Inc. (CA, USA; no grant number).