Shuaipeng Geng1 , Shiyong Li2 , Wei Wu2 , Yinyin Chang1 , and Mao Mao3,4

1.Clinical Laboratories, Shenyou Bio, Zhengzhou, China.

2.Research & Development, SeekIn Inc, Shenzhen, China.

3.Research & Development, SeekIn Inc, San Diego, California.

4.Yonsei Song-Dang Institute for Cancer Research, Yonsei University, Seoul, Republic of Korea.

Hongchao Du, Wenfeng Liu, and Yunfei Li contributed equally to this work.

*Corresponding author.

Address correspondence to: Mao Mao, MD, PhD, TwixBio. Address: 11F, Building 2, DBH Life Sciences & Health Industrial Park, 2028 Shenyan Road, Yantian, Shenzhen, Guangdong 518000, China. E-mail: maomao@yuhs.ac. Shiyong Li, TwixBio Inc. E-mail: lishiyonglee@gmail.com.

Abstract

In population-wide cancer screening, three key issues need to be focused on: the number of cancer cases identified, the number of false positives, and the cost. OncoSeek is a multi-cancer early detection (MCED) test using seven protein tumor markers and artificial intelligence. SeekInCare is an MCED test that integrates the seven protein tumor markers and four cancer genomic features from cell-free DNA by shallow whole-genome sequencing. In a two-step approach, the initial screening is conducted using OncoSeek, and SeekInCare is then used as the secondary test for individuals who tested positive by OncoSeek. We simulated a screening in five million adults ages ≥50 years with a cancer incidence rate of 1.9%. Whereas at 91.0% specificity OncoSeek had 441,450 false positives, using the two-step approach significantly reduced false positives to 34,335 (0.7%). Although SeekInCare and Galleri identified more cancer cases (32,015 and 27,455, respectively) than the two-step MCED (21,280), their total costs reached $3,750 million and $4,745 million, respectively. As the positive predictive value of two-step MCED (38.3%) was comparable with SeekInCare (27.7%) and Galleri (38.3%), it reduced the cost by 5.3-fold and 6.6-fold, respectively, amounting to a total cost of $713.6 million and a cost of $143 per individual screened. The cost of per cancer case detected was $117,133 for SeekInCare and $172,828 for Galleri, which were 3.5-fold and 5.2-fold higher, respectively, than the two-step MCED ($33,534). The two-step approach not only significantly reduces false positives but also cuts down the screening cost substantially, making it a cost-effective strategy for population-wide cancer screening.

Significance:

Large-scale screening inevitably leads to significant financial burdens on the healthcare system, which is a key factor constraining nationwide screenings. The two-step MCED approach not only maintains comparable performance but also substantially alleviates financial strains compared with the direct use of next-generation sequencing–based MCED tests for massive screenings.

Introduction

In population-wide cancer screening, three key aspects need to be focused on: the number of cancer cases identified, the number of false positives, and the cost of screening. A multi-cancer early detection (MCED) test is likely more effective than a single-cancer early detection test in a single screening session, given its capability to detect multiple cancer types at once (1). Multiple single-cancer early detection tests for screening multiple cancer types can have higher false positive rates as the false positives are accumulative from different tests, whereas an MCED test simultaneously targets multiple cancer types with a single, lower false positive rate (2, 3). A high false positive rate not only results in a significant waste of healthcare resources but also imposes unnecessary psychologic and financial burdens on individuals. Additionally, the high cost of screening can overburden the public healthcare system and impede the implementation of widespread screening programs. Large-scale screening inevitably leads to significant financial burdens on the healthcare system, which is a key factor constraining nationwide screenings, even in high-income countries. This challenge is exacerbated by the introduction of state-of-the-art MCED tests based on next-generation sequencing (NGS), even when the cost of sequencing will decrease in the coming years (1).

To date, there are well-established screenings for several cancer types, including breast, cervical, colon, lung, oral, prostate, skin, and stomach cancers. The screenings for prostate and colon cancers typically include two steps to avoid financial burdens on individuals and the public healthcare system. Prostate cancer screening starts with a PSA blood test and then uses an MRI to rule out false positives (4). Similarly, colon cancer screening starts with a stool-based fecal immunochemical test and then uses colonoscopy to verify positive results (5). This two-step approach is much more cost-effective than initiating with costly and resource-intensive MRI and colonoscopy. This approach has been implemented worldwide, especially in countries with universal health coverage such as the United Kingdom and the Netherlands.

In this study, we propose a two-step MCED screening approach using low-cost screening tests (such as OncoSeek) for the initial screening, followed by more superior screening tests (such as SeekInCare and Galleri) as the secondary test for individuals who tested positive in the initial screening. This two-step approach can significantly reduce the financial burden compared with using NGS-based MCED tests for population-level screening.

Materials and Methods

Study design

OncoSeek is a test for MCED using a panel of seven selected blood-based protein tumor markers (PTM) that was quantified by common clinical electrochemiluminescence immunoassay analyzers (6). The SeekInCare test, an advanced version of OncoSeek, further enhances MCED by integrating OncoSeek results with cancer-specific genomic features—copy-number aberration (CNA), fragment size (FS), end motifs, and oncogenic viruses—from circulating-free DNA (cfDNA) using shallow whole-genome sequencing (sWGS) at 3× coverage (7, 8). In our proposed two-step MCED approach, OncoSeek is initially used for primary screening, providing a probability of cancer (POC) index. Then SeekInCare is used for a second round of testing on the positive results with a POC index of 0.5 or higher, which generates a cancer risk score (CRS) to further evaluate the cancer risk. In the second-round test, the PTMs used by SeekInCare can carry over the results from OncoSeek in the primary screening, thereby avoiding the costs associated with retesting the PTMs. OncoSeek and SeekInCare received the conformité européene - in vitro diagnostic (CE-IVD) mark (i.e., the regulatory approval in the European Union) in 2022.

Study participants

Newly diagnosed patients with cancer (n = 617) who were admitted to five hospitals (Peking University Shenzhen Hospital, Sun Yat-sen Memorial Hospital, Sun Yat-sen University Cancer Center, The First Affiliated Hospital of Zhengzhou University, and Henan Cancer Hospital) were enrolled in this study. Noncancer individuals (n = 580) without cancer history or cancer-related symptoms were selected from the physical examination center of Sun Yat-sen Memorial Hospital as the control group. In addition, our employees and their parents were also included as noncancer controls. The study was approved by the independent ethics committee of the leading site, Peking University Shenzhen Hospital (No. 2019.052). Inclusion criteria for cancer patients included confirmed pathology diagnoses and no prior antitumor treatments. Noncancer individuals were excluded if they had a malignancy presence or history or if they had undergone organ transplantation or were pregnant. All participants provided written informed consent upon enrollment. A measure of 8 mL of peripheral blood from each participant was collected using a Cell-Free DNA BCT blood collection tube (Streck) after enrollment. All patients with cancer were treatment-naïve at the time of blood draw. For further details on the cohort, refer to previous publications (6–9).

OncoSeek algorithm for PTM analysis

Refer to our prior publication for the analysis of seven selected PTMs (AFP, CA125, CA15-3, CA19-9, CA72-4, CEA, and CYFRA 21-1; ref. 6) In brief, Roche cobas e411 (Roche Diagnostics GmbH) was utilized to measure the concentration of seven PTMs in 500 μL plasma samples following the manufacturer’s instructions. By integrating and leveraging the performance of all seven PTMs, an algorithm named OncoSeek was constructed based on the generalized linear model method.

Multi-omics SeekInCare analysis and weighted CRS algorithm

See the previous publications for plasma sample preparation, DNA extraction, sequencing library construction, and sWGS (7–9). In brief, cfDNA was extracted from plasma using QIAamp Circulating Nucleic Acid Kit (QIAGEN) following the manufacturer’s instructions. The isolated cfDNA was utilized for library construction using KAPA HyperPrep Kit (Kapa Biosystems), adhering to the provided protocol. These prepared libraries were then subjected to sWGS (∼3× coverage) on X Ten or the NovaSeq platform (Illumina).

Sequence alignment, cfDNA CNA detection, end motif analysis, and oncogenic viral DNA analyses were described in previous publications (7–9). We used the global FS short ratio P150, detailed in our previous studies (7, 9), along with the 5-Mb regional short-to-long ratio features, described in the publication (10), to calculate the FS score using the gradient-boosting machine machine-learning method.

Through the SeekInCare test, the genomic features (CNA, FS, end motif, and oncogenic virus) were analyzed by the sWGS data. Additionally, the PTM value from OncoSeek was incorporated as an additional feature. We used a linear regression approach to integrate the value of the five features to calculate the CRS with the following formula: CRS = a × CNA + b × FS + c × end motif + d × virus + e × PTMs, in which a, b, c, d, and e represent the weight values of CNA, FS, end motif, virus, and PTMs, respectively, which were determined by grid search to get the highest area under the ROC curve (AUC; ref. 11). Finally, the CRS cutoff at 98% specificity was set at 2.0; a score of CRS ≥2.0 classifies a case as cancer.

Statistical analysis

All statistical analyses were performed using R (v3.5.3, RRID: SCR_001905). pROC (v1.13.0, RRID: SCR_024286) was used to estimate the performance of ROC for classifying patients with cancer and noncancer subjects. Sensitivity, specificity, and predictive values were calculated with the epiR (v0.9-99, RRID: SCR_021673) package.

Data availability

All data generated or analyzed during this study are included in this article, its supplementary information files, and publicly available repositories. Sequencing data of the patients with cancer and noncancer individuals, as well as clinical information about these patients, are available with controlled access at https://ngdc.cncb.ac.cn with accession numbers PRJCA017828, PRJCA022396, and PRJCA022679.

Results

Participant demographics

A total of 1,197 participants were enrolled in this case–control study. The cancer group (n = 617) and noncancer group (n = 580) were nearly comparable in terms of gender (χ2 test, P = 0.05). The average age of patients with cancer was higher at 56.0 years compared with the noncancer group, which had an average age of 48.3 years (Table 1). Among the patients with cancer, majority (66.3%) exhibited localized disease (stage I: 26.3%; stage II: 18.6%; and stage III: 21.4%). There were 6.2% of patients with missing tumor stages and 5.3% of patients (e.g., leukemia) who were inapplicable for the American Joint Committee on Cancer Staging Manual (8th edition; ref. 12). A total of 27 cancer types were included in this case–control study (Supplementary Table S1).

Table 1

OncoSeek and SeekInCare performances in the case–control study

All 1,197 participants from this case–control study with both OncoSeek and SeekInCare results were included. OncoSeek and SeekInCare could distinguish patients with cancer from noncancer individuals with AUCs of 0.792 and 0.899, respectively (P < 0.001 by the DeLong test; Fig. 1). Among the genomic features in SeekInCare, the AUCs for CNA, FS, end motif, and oncogenic virus were 0.797, 0.865, 0.877, and 0.654, respectively (Supplementary Fig. S1). With a POC 0.5 as the cutoff for OncoSeek and a CRS 2.0 as the cutoff for SeekInCare, OncoSeek showed a sensitivity of 49.9% at 91.0% specificity, whereas SeekInCare demonstrated a sensitivity of 60.0% at 98.3% specificity (Fig. 1; Table 2). In GRAIL’s circulating cell-free genome atlas (CCGA) study (13), a similar case–control MCED study, involving 2,823 patients with cancer and 1,254 noncancer individuals, GRAIL’s Galleri test demonstrated a sensitivity of 51.5% at 99.5% specificity (Table 2).

Figure 1

Table 2

Sensitivity and specificity in case–control studies: OncoSeek, SeekInCare, two-step MCED, and Galleri

Furthermore, in terms of screening costs, although SeekInCare and Galleri identified more cancer cases (32,015 and 27,455, respectively) than the two-step MCED approach (21,280), their total costs were significantly high, reaching $3,750 million and $4,745 million respectively, equating to $750 for SeekInCare (based on the assumptions of reagent/consumable cost, instrument depreciation, labor cost, tax, profit margin, etc.) and $949 for Galleri per individual screened (Fig. 4B).

Whereas the positive predictive value (PPV) and the number needed to screen (NNS) for the detection of one cancer case of the two-step MCED approach (PPV, 38.3%; NNS, 235) were comparable with SeekInCare (PPV, 27.7%; NNS, 156) and Galleri (PPV, 38.3%; NNS, 182), it reduced the cost significantly (i.e., 5.3-fold and 6.6-fold reductions, respectively), amounting to a total cost of $713.6 million and cost per individual screened of $143 ($80 per test for OncoSeek and $670 additional cost per test for SeekInCare; Table 3). The cost of per cancer case detected using the two-step MCED approach was $33,534, which was substantially less than the costs of SeekInCare and Galleri, which were 3.5 times and 5.2 times higher, amounting to $117,133 and $172,828, respectively (Table 3). Hence, the two-step approach is more cost-effective than using superior screening tests like SeekInCare and Galleri alone.

Discussion

Screening for cancer types like breast, cervical, colon, lung, oral, prostate, skin, and stomach cancers is recommended for asymptomatic individuals based on test availability and cancer prevalence in various countries. In the United States, the United States Preventive Services Task Force recommends screening four cancer types, namely, breast, cervical, colon, and lung cancers, which account for approximately 25% of all cancers and may detect nearly 15% of all cancer cases (14, 15). Therefore, the upcoming blood-based MCED tests that detect cancer DNA and protein in the bloodstream have the potential to detect the remaining 75% of cancers not covered by the current screenings, as well as the cancer cases missed by the current screenings.

New screening tests inevitably place additional costs on the healthcare system. This is especially true with the introduction of the state-of-the-art MCED tests that are based on NGS. For example, Galleri is $949 per test, and SeekInCare is $750. SeekInCare leverages PTM results from OncoSeek as the secondary test, effectively eliminating the PTM retesting costs and reducing the overall cost to $670. Healthcare expenditure in both developed and developing countries has continued to grow over the recent decades. This is due to both significant medical advances as well as the spending related to aging. The challenge for all countries is to enhance health systems by improving healthcare quality and accessibility while controlling expenditure growth. According to a recent study, the United States spent $43 billion on screening in 2021 to detect five cancers (breast, cervical, colorectal, lung, and prostate; ref. 16). The largest driver of the expenditure was colonoscopy, which accounted for 55% of the total screening cost. The high cost of colon cancer screening is largely driven by colonoscopy being used as a primary screening for colon cancer in the United States instead of the two-step approach (fecal immunochemical test followed by colonoscopy) used in countries with universal health coverage, such as the United Kingdom and the Netherlands.

In this study, we started with assessing the performances of a low-cost MCED test (OncoSeek) and a high-cost MCED test (SeekInCare) in a case–control study and then demonstrated that the combined approach of the two tests (i.e., two-step MCED) would maintain comparable performance with NGS-based MCED tests (SeekInCare and Galleri). By simulating the two-step approach and high-cost/high-performance MCED tests (SeekInCare and Galleri) in a screening of five million people, the two-step approach demonstrated that the cost of per cancer case detected was 3.5 times and 5.2 times less than that of SeekInCare and Galleri, respectively, whereas the number of cancer cases identified was reduced by only 33.5% and 22.5%.

We would like to emphasize here that in resource-constrained settings, the actual priority is not a high specificity but a high sensitivity. If we miss detecting a cancer in such a population which does not usually seek medical advice until the cancer is at an advanced stage, it results in a loss of opportunity. The individual will be wrongly reassured, the cancer will progress, and the diagnosis will come too late. Therefore, in places such as low-income countries, OncoSeek would be the sole screening test. OncoSeek can also be effectively used as an auxiliary diagnostic tool for individuals with cancer-related symptoms in low-resource settings as it has sufficient sensitivity (52%) and accurate tissue of origin prediction (67%) across nine common cancer types (6).

As this study relies primarily on the simulations drawn from retrospective data and a simulated MCED screening cohort rather than real-world data, a prospective study in the real-world setting is needed to provide stronger evidence supporting the two-step approach for population-level screening. The healthcare expenditure over quality-adjusted life year gain should be simulated using the key parameters in this study and the cost of standard-of-care therapy for each common cancer type in the countries like the United States to address the cost-effectiveness of the two-step MCED approach.

In conclusion, the two-step MCED approach not only significantly reduces the false positive rate but also helps cut down a significant amount of screening cost, making it a cost-effective strategy for nationwide screening.

Authors’ Disclosures

S. Geng, W. Wu, and S. Li report a patent to Noninvasive, Computer-implemented Method, System, and Storage Medium for Two-step Cancer Risk Prediction pending. No other disclosures were reported by the other authors.

Authors’ Contributions

S. Geng: Formal analysis, writing–review and editing. S. Li: Formal analysis, writing–review and editing. W. Wu: Data curation, formal analysis. Y. Chang: Project administration, writing–review and editing. M. Mao: Formal analysis, methodology, writing–original draft, writing–review and editing.

Acknowledgments

We would like to express our gratitude to all the patients with cancer and noncancer individuals who participated in this study. We thank Dr. Catherine Sauvaget for helpful comments on the manuscript. We would like to acknowledge the English editing provided by Anthea Bull.

Note: Supplementary data for this article are available at Cancer Research Communications Online (https://aacrjournals.org/cancerrescommun/).