Diagnostic performance of multimodal biomarkers in colorectal cancer

Abstract

To evaluate the diagnostic performance of combined biomarkers for colorectal cancer (CRC), this prospective observational study enrolled 188 CRC patients and 693 non-CRC controls from Hunan Provincial People’s Hospital. Binary logistic regression was used to establish a predictive model for CRC risk factors, and receiver operating characteristic (ROC) curve analysis was performed to assess diagnostic efficacy. Results showed that the positive rates of plasma methylated SEPT9 (mSEPT9) and fecal occult blood test (FOBT) in the CRC group were significantly higher than those in the non-CRC group (both P < 0.001). Additionally, levels of carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), red blood cell distribution width-coefficient of variation (RDW-CV), red blood cell distribution width-standard deviation (RDW-SD), neutrophil-to-lymphocyte ratio (NLR), and platelet-to-lymphocyte ratio (PLR) were markedly elevated in CRC patients (all P < 0.001). Multivariate regression identified mSEPT9, CEA, CA19-9, FOBT, RDW-CV, and PLR as independent CRC-related factors. Notably, their combined model achieved an area under the curve (AUC) of 0.939, with a sensitivity of 0.920 and specificity of 0.839, offering a valuable non-invasive strategy to enhance CRC diagnosis.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37280-w.

Introduction

Colorectal cancer (CRC) is a leading gastrointestinal malignancy globally, with increasing younger onset and high morbidity/mortality in China^1,2. Early diagnosis significantly improves treatment outcomes, but most patients are asymptomatic in the early stage and diagnosed at an advanced stage with poor prognosis^3,4. Thus, early screening is crucial for enhancing efficacy and reducing mortality. Current non- invasive screening methods include fecal occult blood test (FOBT) and serum tumor markers have suboptimal sensitivity/specificity, while invasive colonoscopy suffers from low compliance due to complex preparation and potential risks⁵. Therefore, developing non-invasive, highly sensitive/specific combined screening strategies is clinically urgent, especially for colonoscopy-reluctant individuals⁶.​.

DNA methylation is a key epigenetic event in CRC carcinogenesis, and studies have demonstrated that such methylation-based blood tests even outperform traditional tumor markers in CRC screening^7,8. Among validated methylation biomarkers (SEPT9, NDRG4, BMP3, etc.)⁹, methylated SEPT9(mSEPT9) is aberrantly methylated in CRC. mSEPT9 is detectable in CRC tissues and patient blood, and its level correlates with CRC progression, making it a promising non-invasive diagnostic indicator^10–12.

Beyond methylation markers, inflammation and hematological indices, such as the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), and lymphocyte-to-monocyte ratio (LMR), and hematological indices red cell distribution width (RDW) have been demonstrated closely associated with tumorigenesis/ progression and prognosis for gastrointestinal tumors^13–16. Meanwhile, classic serum tumor markers (CEA, CA19-9) combined with FOBT are commonly used in clinical practice but have limitations in CRC diagnosis¹⁷. Notably, these markers have complementary diagnostic mechanisms: mSEPT9 reflects epigenetic aberrations, CEA/CA19-9 are tumor-specific antigens, FOBT indicates intestinal lesions, and RDW/PLR reflect the inflammatory microenvironment. This complementarity supports combined detection to overcome the shortcomings of single/partial markers, as evidenced by studies showing improved diagnostic accuracy and reduced missed/misdiagnosis with mSEPT9 combined with CEA, CA19-9, NLR/PLR, or FOBT^18,19.

However, no study has comprehensively analyzed the predictive and combined diagnostic performance of mSEPT9 with CEA, CA19-9, FOBT, RDW, and inflammation-related indices for CRC. Herein, we constructed a binary logistic regression model to screen CRC-related predictive risk factors and used receiver operating characteristic (ROC) curves to evaluate the diagnostic value of these indices alone and in combination. This study aims to provide a new strategy for CRC screening and diagnosis.

Materials and methods

Study subjects

A prospective observational study was conducted by enrolling 188 patients pathologically diagnosed with CRC at Hunan Provincial People’s Hospital from January 2024 to December 2024 as the CRC group, and 693 patients with digestive symptoms (such as abdominal pain, diarrhea, changes in bowel habits, or hematochezia) who underwent relevant treatment and were diagnosed as non-CRC at the same hospital during the same period as the control group.Inclusion criteria: CRC patients were pathologically confirmed as CRC via colonoscopy prior to enrollment according to the Chinese Colorectal Cancer Diagnosis and Treatment Guidelines (2023 Edition), and complete clinical data were ensured. Exclusion criteria: (1)Female patients who were lactating or pregnant; (2)Patients with other malignant tumors; (3)Patients who had received previous radiotherapy, chemotherapy, immunotherapy, surgical treatment, or systemic therapy.This study has been granted a waiver of informed consent and was approved by the Medical Ethics Committee of the Hunan Provincial People’s Hospital and The first-affiliated hospital of Hunan normal university (Approval No.2024297) and all trails were carried out in according to the relevant guidelines and protocols, as well as the Declaration of Helsinki.

Research methods

mSEPT9 detection

The detection process was performed in strict accordance with the instructions of the mSEPT9 detection kit (Beijing BioChain Co., Ltd., China) instructions as shown: (1)Sample collection and storage: Venous blood (10 mL) was collected using an EDTA-K2 anticoagulant vacuum blood collection tube, centrifuged at 1350×g for 12 min at room temperature (25 °C), and at least 3.5 mL of plasma was separated and stored at −25 to −15 °C for detection within 1 week. (2)DNA extraction and bisulfite conversion from plasma: 3.5 mL of plasma was lysed with lysis buffer, bound to magnetic beads, washed, and eluted to obtain free DNA. Free DNA was subjected to bisulfite conversion, bound to magnetic beads again, washed, dried, and eluted to obtain bisulfite-converted DNA samples (BisDNA). (3)Preparation of PCR reaction system: Each PCR reaction required 32 µL of PCR reaction solution and 1.6 µL of polymerase. The PCR reaction solution and polymerase were mixed in the specified ratio in a 2.0 mL centrifuge tube, vortexed thoroughly to ensure uniform distribution, and centrifuged to collect the liquid. 30 µL of the pre-mixed PCR solution was added to a 96-well plate, followed by 30 µL of BisDNA into the corresponding wells. The plate was sealed with a film and centrifuged at 110 g for 1 min. (4)PCR reaction: A real-time fluorescent quantitative PCR (RT-qPCR) instrument (Hongshi Medical Technology Co., Ltd., Shanghai, China) was used to set up the PCR reaction program. The reaction system was activated at 94 °C for 20 min, followed by 45 thermal cycles (62 °C for 5 s, 55.5 °C for 35 s, 93 °C for 30 s), and then cooled down at 40 °C for 5 s. The human housekeeping gene β-actin (ACTB) was selected as the internal reference for the PCR reaction. (5)Result interpretation: Valid positive control: Septin9 Ct ≤ 41.1 and ACTB Ct ≤ 29.8; Valid negative control: No Septin9 Ct value and ACTB Ct ≤ 37.2; Valid internal control (ACTB): The Ct value meets the corresponding criteria of positive/negative controls, ensuring that the reaction system does not fail. A positive result was defined as Septin9 Ct ≤ 41.0 and ACTB Ct ≤ 32.1; a negative result was defined as no detectable Septin9 Ct or Septin9 Ct > 41.0 with ACTB Ct ≤ 32.1; an invalid result was defined when ACTB Ct > 32.1 regardless of Septin9 Ct value.

FOBT detection

FOBT was performed within 24 h after stool specimen collection using the colloidal immunochromatographic assay reagents (W.H.P.M. Bioresearch & Technology Co., Ltd., China) and strictly following the manufacturer’s instructions.

CEA and CA19-9 detection

Venous blood (3–5 mL) was collected using a vacuum blood collection tube with a separating gel and coagulant accelerator. The sample was centrifuged at 1500×g for 12 min. The upper-layer serum was transferred to a 2 mL centrifuge tube and stored at −80 °C until testing. Detection was performed by ‌iFlash 3000 chemiluminescence immunoassay analyzer‌ and supporting reagents (Shenzhen YHLO Biotech Co., Ltd., Shenzhen, China). In this study, the normal reference ranges for CEA and CA19-9 were 0–5 (ng/mL) and 0–37 (U/mL), respectively. An abnormal result was defined as CEA ≥ 5 (ng/mL) or CA19-9 ≥ 37 (U/mL).

CBC detection

Fasting venous blood (2 mL) was collected from patients in the morning. Using a SYSMEX-XN9000 hematology analyzer and supporting reagents (Sysmex Corporation, Kobe, Japan), parameters such as RDW-CV, RDW-SD, neutrophil count, platelet count, lymphocyte count, and monocyte count were measured. NLR, PLR, and LMR were calculated based on the obtained data.

Statistical analysis

Statistical analyses were performed using SPSS 26.0 software. Measurement data conforming to normal distribution were expressed as mean ± standard deviation (±s). The t-test was used for comparing measurement data between two groups, while the chi-square test was applied to count data. Non-normally distributed measurement data were presented as median (first quartile, third quartile) [M (Q1, Q3)], and the Mann-Whitney U test was used for intergroup comparisons. Binary logistic regression was employed to analyze factors associated with CRC occurrence. The AUC of the ROC curve was used to evaluate the diagnostic efficacy of individual and combined indicators for CRC, and sensitivity, specificity, odds ratios (OR), and 95% confidence intervals (CI) were calculated. A P < 0.05 was considered statistically significant.

Results

Baseline clinical information of the two groups

The baseline clinical information of the CRC group and non-CRC control group is shown in Fig. 1A total of 881 participants (564 males and 317 females) were enrolled in this study, including 188 cases in the CRC group (111 males and 77 females, median age 63 years) and 693 cases in the non-CRC control group with similar digestive symptoms (453 males and 240 females, median age 52 years). There was no significant difference in gender distribution between the two groups (P > 0.05) as shown in Fig. 1A, while the age difference was statistically significant (P < 0.001). The CRC group had a higher proportion of patients aged ≥ 60 yearsas shown in Fig. 1B.

Fig. 1.

Clinical characteristics of CRC and non-CRC groups. (A) Gender distribution in CRC and non-CRC groups; (B) Age distribution between CRC and non-CRC groups; *P < 0.05; **P < 0.01;***P < 0.001; ns, no statistically significant.

Tumor distribution and clinicopathological characteristics of the CRC group

In the CRC group (n = 188), the tumor stage distribution showed that stage II accounted for the highest proportion (35.6%, 67/188), followed by stage III (28.7%, 54/188), stage IV (25.0%, 47/188), and stage I (10.6%, 20/188). For the N stage, N0 (48.4%, 91/188) was the most common, followed by N1 (38.3%, 72/188) and N2 (13.3%, 25/188). In terms of M stage, 75.0% (141/188) of patients were M0, and 25.0% (47/188) were M1. Regarding tumor location, colon cancer (54.3%, 102/188) was slightly more prevalent than rectal cancer (45.7%, 86/188). For the degree of cancer differentiation, moderate differentiation was the dominant type (62.2%, 117/188), followed by moderate-low differentiation (22.9%, 43/188), high-moderate differentiation (6.9%, 13/188), high differentiation (5.3%, 10/188), and low differentiation (2.7%, 5/188). In terms of tumor size, tumors smaller than 5 cm accounted for 63.8% (120/188), while those 5 cm and above accounted for 36.2% (68/188). For vascular infiltration, 83 cases (44.1%) were absent, 63 cases (33.5%) were present, and 42 cases (22.4%) were unknown. Regarding neural infiltration, 82 cases (43.6%) were absent, 65 cases (34.6%) were present, and 41 cases (21.8%) were unknown, as shown in Table 1.

Table 1.

Tumor distribution and clinicopathological characteristics of CRC patients.

	Number(N = 188)
Cancer stage
I	20(10.6)
II	67(35.6)
III	54(28.7)
IV	47(25.0)
N stage
N0	91(48.4)
N1	72(38.3)
N2	25(13.3)
M stage
M0	141(75.0)
M1	47(25)
Tumor location
Colon	102(54.3)
Rectum	86(45.7)
Cancer differentiation degree
Low	5(2.7)
Moderate-low	43(22.9)
Moderate	117(62.2)
High-moderate	13(6.9)
High	10(5.3)
Tumor size(cm)
< 5	120(63.8)
≥ 5	68(36.2)
Vascular infiltration
Absent	83
Present	63
Unknow	42
Neural infiltration
Absent	82
Present	65
Unknow	41

Detection results of mSEPT9 and FOBT in the two groups

The positive rates of mSEPT9 and FOBT in the CRC group were 61.2% (115/188) and 85.4% (135/158), respectively, while those in the control group were 5.5% (38/693) and 20.3% (70/344), respectively. The positive rates of both mSEPT9 and FOBT in the CRC group were significantly higher than those in the control group (P < 0.001, χ² test), as shown in Table 2.

Table 2.

Comparison of mSEPT9 and FOBT between CRC and non-CRC groups.

Parameters	Overall (n = 881)	CRC (n = 188)	Non -CRC (n = 693)	χ²	P
mSEPT9				319.557	< 0.001
Positive	153	115	38
Negative	728	73	655
FOBT				189.886	< 0.001
Positive	205	135	70
Negative	297	23	274

Detection results of CEA, CA19-9, RDW-CV, RDW-SD, NLR, PLR, and LMR in the two groups

None of the seven indicators of CEA, CA19-9, RDW-CV, RDW-SD, NLR, PLR, and LMR in the CRC group and the control group followed a normal distribution, so M (Q1, Q3) was used for statistical description. Mann-Whitney U test was used for analysis. The levels of CEA, CA19-9, RDW-CV, RDW-SD, NLR and PLR in the CRC group were all higher than those in the control group (P < 0.001), and the LMR level was lower than that in the control group (P < 0.001), as shown in Table 3.

Table 3.

Comparison of CBC and tumor markers between CRC and non-CRC groups.

Parameters	CRC M(Q1,Q3)	Non -CRC M(Q1,Q3)	Z	P
CEA(ng/ml)	5.63(2.50,15.10)	1.46(0.88,2.22)	−13.819	< 0.001
CA19-9(U/ml)	13.70(5.06,51.35)	5.93(3.39,11.03)	−7.495	< 0.001
RDW-CV	13.40(12.70,14.60)	12.70(12.20,13.40)	−7.683	< 0.001
RDW-SD	43.80(40.93,46.28)	42.60(40.30,45.00)	−3.597	< 0.001
NLR	3.10(2.17,4.95)	1.99(1.45,3.27)	−7.294	< 0.001
PLR	176.52(126.80,245.30)	129.02(98.17,179.36)	−7.175	< 0.001
LMR	3.26(2.06,4.63)	4.18(2.75,5.52)	−4.987	< 0.001

Binary logistic regression analysis for predicting risk factors of CRC occurrence

Indicators with statistical significance (P < 0.05) between the CRC group and control group (mSEPT9, FOBT, CEA, CA19-9, RDW-CV, RDW-SD, NLR, PLR, LMR) were selected as independent variables, and CRC status (with or without CRC) was used as the dependent variable to construct a binary logistic regression prediction model. The results showed that in the univariate binary logistic regression analysis, all 9 above-mentioned indicators were independent predictive factors for CRC occurrence. Among them, 8 indicators including mSEPT9, FOBT, CEA, CA19-9, RDW-CV, RDW-SD, NLR, and PLR were risk factors for CRC occurrence (OR > 1), while LMR was a protective factor for CRC occurrence (OR < 1). The above 8 independent risk factors were included in the multivariate binary logistic regression analysis, and the results showed that a total of 6 indicators, including mSEPT9, FOBT, CEA, CA19-9, RDW-CV, and PLR, could be used to distinguish whether clinical individuals were more likely to suffer from CRC, as shown in Table 4.

Table 4.

Univariate and multivariate binary logistic regression analysis of CRC risk factors.

	Variable	β	SE	Waldχ²	P	OR	95%CI
Univariate binary logistic regression	Age	0.053	0.006	75.206	< 0.001	1.055	1.042 ~ 1.067
	mSEPT9	3.302	0.224	216.973	< 0.001	27.154	17.500 ~ 42.133
	FOBT	3.134	0.262	142.755	< 0.001	22.975	13.739 ~ 38.420
	CEA	0.370	0.044	71.764	< 0.001	1.447	1.329 ~ 1.576
	CA19−9	0.030	0.006	26.996	< 0.001	1.031	1.019 ~ 1.042
	RDW-CV	0.249	0.040	38.406	< 0.001	1.283	1.185 ~ 1.388
	RDW-SD	0.059	0.015	15.901	< 0.001	1.061	1.030 ~ 1.092
	NLR	0.062	0.018	11.863	0.001	1.064	1.027 ~ 1.103
	PLR	0.004	0.001	28.899	< 0.001	1.004	1.003 ~ 1.006
	LMR	−0.204	0.045	20.696	< 0.001	0.815	0.747 ~ 0.890
Multivariate binary logistic regression	mSEPT9	1.782	0.468	14.487	< 0.001	5.943	2.374 ~ 14.879
	FOBT	3.099	0.382	65.790	< 0.001	22.186	10.491 ~ 46.920
	CEA	0.240	0.073	10.733	0.001	1.272	1.101 ~ 1.468
	CA19−9	0.020	0.008	5.693	0.017	1.020	1.004 ~ 1.037
	RDW-CV	0.197	0.094	4.369	0.037	1.218	1.012 ~ 1.468
	PLR	0.005	0.002	4.081	0.043	1.005	1.000 ~ 1.009

Diagnostic efficacy analysis of mSEPT9, FOBT, CEA, CA19-9, RDW-CV and PLR for CRC

A diagnostic model was constructed to distinguish between CRC and non-CRC individuals using the predictive risk factors (mSEPT9, FOBT, CEA, CA19-9, RDW-CV, and PLR) identified by multivariate binary logistic regression analysis. The clinical efficacy of each indicator for independent CRC diagnosis is shown in Table 5. The sensitivities of mSEPT9, FOBT, CEA, CA19-9, RDW-CV, and PLR were 0.612 (95% CI 0.541–0.678), 0.854 (95% CI 0.791–0.901), 0.710 (95% CI 0.636–0.772), 0.526 (95% CI 0.452–0.598), 0.596 (95% CI 0.524–0.663), and 0.649 (95% CI 0.578–0.714), respectively. The specificities were 0.945 (95% CI 0.926–0.960), 0.797 (95% CI 0.751–0.836), 0.880 (95% CI 0.850–0.904), 0.819 (95% CI 0.781–0.851), 0.674 (95% CI 0.638–0.708), and 0.610 (95% CI 0.574–0.646), respectively. The AUCs (95% CI) were 0.778 (0.732–0.823), 0.826 (0.785–0.866), 0.847 (0.808–0.885), 0.692 (0.641–0.741), 0.683 (0.640–0.725), and 0.670 (0.628–0.713), respectively. The ROC curves for independent CRC diagnosis by each indicator are shown in Fig. 2A.

Table 5.

Diagnostic value of mSEPT9 and other biomarkers for CRC detection.

Parameters	AUC(95%CI)	P	SD	Sensitivity(95%CI)	Specificity(95%CI)	Cutoff value	Youden index
mSEPT9	0.778(0.732ཞ0.823)	< 0.0001	0.022	0.612(0.541ཞ0.678)	0.945(0.926ཞ0.960)	0.500	0.557
FOBT	0.826(0.785ཞ0.866)	< 0.0001	0.020	0.854(0.791ཞ0.901)	0.797(0.751ཞ0.836)	0.500	0.651
CEA	0.847(0.808ཞ0.885)	< 0.0001	0.019	0.710(0.636ཞ0.772)	0.880(0.850ཞ0.904)	3.000	0.590
CA19-9	0.692(0.641ཞ0.741)	< 0.0001	0.026	0.526(0.452ཞ0.598)	0.819(0.781ཞ0.851)	13.405	0.345
RDW-CV	0.683(0.640ཞ0.725)	< 0.0001	0.022	0.596(0.524ཞ0.663)	0.674(0.638ཞ0.708)	13.150	0.270
PLR	0.670(0.628ཞ0.713)	< 0.0001	0.022	0.649(0.578ཞ0.714)	0.610(0.574ཞ0.646)	148.613	0.259
mSEPT9 + FOBT	0.897(0.865ཞ0.929)	< 0.0001	0.016	0.918(0.864ཞ0.951)	0.765(0.717ཞ0.806)	0.246	0.683
mSEPT9 + CEA + CA19-9	0.889(0.856ཞ0.921)	< 0.0001	0.017	0.757(0.688ཞ0.815)	0.900(0.870ཞ0.924)	0.166	0.657
mSEPT9 + RDW-CV	0.830(0.792ཞ0.867)	< 0.0001	0.019	0.686(0.617ཞ0.748)	0.887(0.861ཞ0.909)	0.154	0.573
mSEPT9 + PLR	0.832(0.795ཞ0.869)	< 0.0001	0.019	0.692(0.622ཞ0.753)	0.876(0.849ཞ0.898)	0.135	0.567
mSEPT9 + FOBT + CEA + CA19-9 + RDW-CV + PLR	0.939(0.912ཞ0.966)	< 0.0001	0.014	0.920(0.865ཞ0.953)	0.839(0.789ཞ0.880)	0.377	0.758

Fig. 2.

ROC curve analysis of mSEPT9, FOBT, CEA、CA19-9、RDW-CV、PLR for CRC diagnosis.(A) The diagnostic values of mSEPT9,FOBT, CEA, CA19-9、RDW-CV、PLR for detecting CRC; (B) The diagnostic values of the combination of mSEPT9 ,FOBT, CEA, CA19-9、RDW-CV、PLR for detecting CRC.

The diagnostic performance of mSEPT9 combined with other indicators is shown in Table 5. When mSEPT9 was combined with FOBT, the sensitivity increased to 0.918 (95% CI 0.864–0.951), but the specificity decreased to 0.765 (95% CI 0.717–0.806) compared with individual detections. The combination of mSEPT9 with tumor markers CEA and CA19-9 improved the sensitivity for CRC diagnosis to 0.757 (95% CI 0.688–0.815), with a slightly lower specificity of 0.900 (0.870–0.924) than that of mSEPT9 alone. The combinations of mSEPT9 with RDW-CV and PLR showed similar and relatively low sensitivity and specificity compared with other combinations. When mSEPT9 was combined with FOBT, CEA, CA19-9, RDW-CV, and PLR, the diagnostic efficacy for CRC was the highest, with the AUC (95% CI) increasing to 0.939 (0.912–0.966), and the sensitivity (95% CI) and specificity (95% CI) reaching excellent levels of 0.920 (0.865–0.953) and 0.839 (0.789–0.880), respectively. The ROC curves for combined CRC diagnosis by each indicator are shown in Fig. 2B.

Discussion

Common symptoms of colorectal cancer (CRC) include changes in bowel habits, rectal bleeding, abdominal pain, abdominal distension, anemia, and unexplained weight loss, with most patients diagnosed at intermediate or advanced stages when symptoms manifest. Early diagnosis and treatment are critical for improving the five-year survival rate, and the CAN-SCREEN model highlights that non-invasive screening strategies with high compliance yield favorable health outcomes (e.g., averted CRC deaths, improved quality-adjusted life years [QALYs])²⁰, emphasizing the urgency of selecting compliant and efficient diagnostic methods. The SEPT9 gene (chromosome 17q25.3) encodes a cell cycle-related protein; its methylation-induced silencing disrupts cytokinesis, promoting malignant proliferation²¹. Methylated SEPT9 (mSEPT9) has demonstrated diagnostic value in multiple malignancies, particularly CRC²². In our study, the positive rate of mSEPT9 was significantly higher in the CRC group (61.2%) than in the control group (5.5%), with an AUC (95% CI) of 0.778 (0.732–0.823) for independent CRC diagnosis—consistent with multicenter study results^19,23, supporting its potential as a non-invasive screening biomarker, especially for colonoscopy-reluctant high-risk individuals⁶˒²³. Fecal occult blood test (FOBT) based on fecal immunochemical analysis is a cost-effective screening tool for average-risk individuals^24–26. Our data showed a higher positive rate of FOBT in the CRC group (85.4% vs. 20.3% in controls) and an AUC (95% CI) of 0.826 (0.785–0.866), indicating moderate diagnostic value similar to mSEPT9. Carcinoembryonic antigen (CEA) (AUC = 0.847, 95% CI 0.808–0.885) exhibited superior diagnostic performance among single tumor markers, while carbohydrate antigen 19 − 9 (CA19-9) showed relatively low efficacy (AUC = 0.692, 95% CI 0.641–0.741)^27–29.

Inflammation contributes to CRC development via DNA damage and epithelial barrier disruption³⁰, and composite inflammatory indices (NLR, PLR, LMR) better reflect systemic inflammatory status than single cell counts^31,32. Consistent with previous findings¹⁶, our study found higher NLR/PLR and lower LMR in the CRC group (all P < 0.001). While univariate logistic regression identified NLR, PLR, and LMR as predictive factors, only PLR remained significant in multivariate analysis (AUC = 0.670, 95% CI 0.628–0.713), indicating limited independent diagnostic value. Red blood cell distribution width (RDW), including RDW-CV and RDW-SD, reflects tumor-associated inflammation or iron metabolic disorders and correlates with CRC prognosis^33,34. We observed significantly higher RDW-CV and RDW-SD in CRC patients (P < 0.05), with RDW-CV identified as an independent risk factor (AUC = 0.683, 95% CI 0.640–0.725)—comparable to PLR, suggesting modest diagnostic utility^35–37.

This study compared the diagnostic efficacy of plasma mSEPT9 detection with several predictive risk factors for CRC occurrence, evaluating the model’s potential to effectively distinguish between CRC patients and control subjects. It was found that individual plasma mSEPT9 detection exhibited the highest specificity, reaching 0.945 (95% CI 0.926–0.960), while its sensitivity was higher than that of CA19-9 and RDW-CV, indicating that plasma mSEPT9 holds significant diagnostic value for CRC and may emerge as a promising screening biomarker. Individual FOBT screening showed the highest sensitivity for CRC (0.854) but relatively lower specificity (0.797)—lower than that of mSEPT9, CEA, and CA19-9—while CEA demonstrated the highest diagnostic value among single indicators with an AUC (95% CI) of 0.847 (0.808–0.885), confirming its continued role as an important CRC screening tool. In contrast, individual CA19-9, RDW-CV, and PLR showed unsatisfactory diagnostic performance (AUC < 0.700), suggesting they are not suitable for independent CRC screening^16,37. Furthermore, combining plasma mSEPT9 with CEA, CA19-9, FOBT, RDW-CV, or PLR individually improved both AUC and sensitivity compared to single-indicator detection. To the best of our knowledge, no previous studies have reported the combined efficacy of mSEPT9, CEA, CA19-9, FOBT, RDW-CV, and PLR in CRC diagnosis; our results showed that this six-indicator combination achieved an exceptionally high AUC (95% CI) of 0.939 (0.912–0.966), with sensitivity (0.920) and specificity (0.839) that significantly exceeded those of all single indicators and pairwise combinations. By integrating measurements of multiple biomarkers to calculate disease probability, such models comprehensively capture the complex multi-layered characteristics of cancer development^18,19, outperforming single-analyte detection methods and offering excellent value for non-invasive CRC screening.

Notably, when contextualized against current CRC diagnostic standards, these specific biomarkers demonstrate distinct clinical relevance. Colonoscopy, as the gold standard for definitive diagnosis, is constrained by invasiveness and low population compliance, limiting its use in large-scale screening^5,6. In contrast, mSEPT9’s high specificity (0.945) outperforms conventional serum markers (CEA: 0.880; CA19-9: 0.819) and reduces unnecessary colonoscopy referrals, addressing a key limitation of FOBT (specificity = 0.797)^23,27. CEA, with the highest single-indicator AUC (0.847), complements mSEPT9 by capturing tumor-specific antigen expression, while FOBT’s superior sensitivity (0.854) effectively identifies bleeding-related lesions^17,26. RDW-CV and PLR, though individually modest (AUC < 0.700), add critical insights into tumor-associated inflammation and metabolic disorders, enhancing the model’s multi-dimensional coverage1^16,34. This complementarity underscores their collective value beyond standalone non-invasive tools, bridging the gap between clinical utility and patient compliance.

The superior performance of the combined model also highlights a critical consideration in CRC screening: balancing sensitivity and specificity. The core significance of tumor screening lies in early identification of potential patients to maximize treatment opportunities, making sensitivity a key priority—missed diagnoses of early CRC can lead to irreversible progression and poor prognosis. In our study, individual FOBT (AUC = 0.826) and CEA (AUC = 0.847) exhibited higher sensitivity than mSEPT9 (sensitivity = 0.612, AUC = 0.778), aligning with the clinical goal of “catching all potential cases.” However, the inherent trade-off in diagnostic medicine means higher sensitivity often increases false-positive risks, as seen with FOBT’s relatively low specificity which may misclassify nearly one-fifth of non-CRC individuals as high-risk, leading to unnecessary follow-ups, increased healthcare burdens, and psychological distress. Striking a context-specific balance requires tailoring strategies to screening objectives and populations: our combined model achieves this by integrating high-sensitivity indicators (FOBT, CEA) with high-specificity markers (mSEPT9), ensuring most CRC cases are identified while controlling false positives to clinically acceptable levels. For high-risk groups (e.g., family history of CRC, chronic inflammatory bowel disease), sensitivity should be prioritized over slight specificity losses, whereas average-risk populations benefit from a more balanced profile; optimizing indicator cutoff values can further refine this equilibrium, with the ideal screening tool achieving tailored alignment with clinical needs, population characteristics, and healthcare resource availability rather than pursuing absolute sensitivity or specificity in isolation.

Beyond diagnostic performance, the operability and economy of the multimodal screening model are critical for its translation into clinical practice and population-based screening, which merit further discussion. All biomarkers included in our model are clinically accessible with favorable operability: plasma mSEPT9 (detected via commercialized quantitative methylation-specific PCR with routine peripheral blood samples), CEA, CA19-9, and CBC parameters (RDW-CV, NLR, PLR) (detected by universally equipped automated analyzers), and FOBT (non-invasive with at-home fecal sampling) can be tested using mature, standardized techniques available in both tertiary hospitals and primary healthcare settings. A single venipuncture for blood-based indicators and independent at-home fecal sampling for FOBT minimize patient discomfort and improve compliance without increasing operational complexity. In terms of economy, the combined model balances diagnostic performance with cost-effectiveness: routine serological indicators and FOBT are low-cost, while the relatively higher cost of mSEPT9 is offset by its high specificity (reducing unnecessary colonoscopies and associated costs). A stepwise screening strategy (first low-cost indicators, then mSEPT9 for high-risk individuals) can further optimize cost-effectiveness for resource-limited settings, consistent with real-world evidence of reduced colonoscopy demand and long-term healthcare costs^27,30. Although the cost of mSEPT9 may be a barrier in some regions, future optimization of detection technology and simplified model exploration will enhance its economic sustainability. Collectively, our multimodal model exhibits high operability and favorable cost-effectiveness, supporting its potential for clinical application and population-based CRC screening.

Although this study has made some clinically significant discoveries, there are still several limitations. First, all participants were recruited from a single tertiary hospital in a specific geographic region, leading to a lack of ethnic and demographic diversity. This restricts the generalizability of our findings to broader populations, especially those with distinct genetic backgrounds or lifestyle factors that may influence the biomarkers we investigated. We have also proposed multi-center studies with diverse populations as a key direction for future research to address this limitation. Second, this study did not evaluate the application value of plasma mSEPT9 combined with other tests in monitoring CRC recurrence and treatment response, nor did it explore the model’s diagnostic performance across different CRC stages (particularly early-stage CRC [stage I–II]). Third, the current biomarker panel may be further optimized by incorporating additional promising indicators (e.g., other methylation markers or circulating tumor DNA). Corresponding future research will focus on validating the combined model in larger multi-center cohorts, investigating its efficacy in early CRC screening and recurrence monitoring, and refining the indicator panel to enhance diagnostic accuracy and clinical utility.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Investigation, Project administration and Writing—original draft, Shuang Yang; Data curation, software and Methodology, Yuqing Wang; Project administration, Data curation and Supervision, Jiang Li; Funding acquisition, Supervision and Writing—review and editing, Kaiwu Xu. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Scientific Research Foundation of Hunan Provincial Education Department, Grant number 24C0009; and by the Natural Science Foundation of Hunan Province, China, Grant number 2025JJ80731; and by the Natural Science Foundation of Hunan Province, China, Grant number 2024JJ9138.

Data availability

The raw data are included in the supplementary information files.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

This study has been granted a waiver of informed consent and was approved by the Medical Ethics Committee of the Hunan Provincial People’s Hospital and The first-affiliated hospital of Hunan normal university (Approval No.2024297).