Gastrointestinal malignancies are among the most common cancers globally,^1,2 with their prognosis often hinging on the occurrence of peritoneal metastasis. Once peritoneal metastasis occurs, there is a significant reduction in patient survival rates.^3-5 Accurately predicting the likelihood of peritoneal metastasis preoperatively is therefore crucial in developing personalized treatment plans. Circulating tumor cells (CTC) have gained increasing attention in cancer research in recent years.^6-9 They are tumor cells that have detached from the primary tumor or metastatic sites and entered the bloodstream. Considered as a form of “real-time liquid biopsy,” CTC are seen as a powerful tool for predicting cancer progression and response to treatment. However, the detection and identification of CTC remain challenging due to their extreme rarity in the blood and high heterogeneity. Folate receptor (FR), a cell surface high-affinity folate-binding protein, is overexpressed in many types of tumors, including ovarian, breast, and lung cancers.^10-12 The presence of FR-positive CTC (FR⁺-CTC) in gastrointestinal malignancies has been confirmed, with studies indicating that the quantity of FR⁺-CTC is closely associated with the clinical-pathological characteristics and prognosis of tumors.^13-15 Compared to other biomarkers, such as EpCAM+ CTC or traditional tumor markers like carcinoembryonic antigen, cancer antigen 19-9, and cancer antigen 125 (CA-125), FR⁺-CTC demonstrate unique advantages in gastrointestinal malignancies due to their higher specificity and relevance in tumor aggressiveness and metastasis.^16-18 For instance, while EpCAM-based detection is commonly employed, it has limitations in capturing mesenchymal-like CTC, which are critical in metastasis. FR⁺-CTC provide a broader detection spectrum that includes epithelial, mesenchymal, and hybrid phenotypes, offering a more comprehensive understanding of metastatic potential. To ensure the accuracy and sensitivity of detecting FR⁺-CTC, rigorous controls were implemented in our immunofluorescence and flow cytometry techniques. These controls included the use of established positive and negative cell lines to validate antibody specificity, as well as spike-in experiments to evaluate the efficiency of CTC recovery. Additionally, strict gating strategies were applied to distinguish FR⁺-CTC from other circulating cells, minimizing the risk of false positives. However, up to now, there is still a relative scarcity of research on the role of FR⁺-CTC in predicting peritoneal metastasis in gastrointestinal cancers. Most existing studies are preliminary observational studies based on small samples, lacking large-scale validation and in-depth mechanistic research. Moreover, despite their predictive value, how to accurately integrate FR⁺-CTC into clinical decision-making remains an unresolved issue. In this context, we have designed a new predictive model based on the detection of FR⁺-CTC, aimed at preoperatively predicting the risk of peritoneal metastasis in gastrointestinal malignancies, thereby guiding clinical practice.

1. Data sources

This retrospective study collected data from patients diagnosed with gastric and colorectal cancer through surgery at Henan Provincial People's Hospital from July 2021 to May 2023. Patients who underwent preoperative testing for FR⁺-CTC were selected. This study was approved by the Institutional Review Board of Henan Provincial People's Hospital (approval number: 202284), and it adhered to the principles of the 1975 Declaration of Helsinki and its subsequent amendments. Written informed consent was waived.

2. Inclusion and exclusion criteria

Inclusion criteria were: (1) patients confirmed to have gastric or colorectal cancer through preoperative gastroscopy or colonoscopy; (2) preoperative testing for FR⁺-CTC; (3) no other related antitumor treatment before surgery; and (4) patients underwent surgical treatment. Exclusion criteria included: (1) non-gastric or colorectal cancer patients; (2) patients who received related antitumor treatments preoperatively, such as radiotherapy, chemotherapy, neoadjuvant therapy, traditional Chinese medicine, etc.; (3) patients with a history of other malignancies; and (4) patients who did not undergo surgery.

3. Detection method for FR⁺-CTC

This method involves using the FR as a marker to detect and quantify CTC from blood samples. Initially, preoperative blood samples are collected from patients for detecting FR⁺-CTC. These cells are then separated from the blood samples using techniques such as immunomagnetic separation. Subsequently, folate (folic acid) is used as a probe to detect and quantify the expression of FR in these separated cells using techniques like immunofluorescence staining and flow cytometry. To ensure the accuracy and sensitivity of detecting FR⁺-CTC, rigorous and well-defined controls were employed. Positive control cell lines with known FR expression were utilized to validate antibody specificity, while negative control cell lines lacking FR expression minimized the risk of nonspecific binding. Additionally, spike-in experiments were conducted by introducing known quantities of FR⁺ tumor cells into blood samples from healthy donors to evaluate CTC recovery efficiency and sensitivity. For flow cytometry, strict gating strategies were applied, leveraging fluorescence intensity, cell size, and morphological parameters to differentiate FR⁺-CTC from other circulating cells, such as leukocytes and platelets. Furthermore, appropriate isotype controls and repeated measurements ensured reproducibility and minimized false positives. These measures collectively enhanced the reliability of FR⁺-CTC detection.

4. Data grouping

The “caret” package in R software (R Foundation for Statistical Computing, Vienna, Austria) is used for random grouping of patients. All patients are randomly divided into a training group and a validation group in a 1:1 ratio.

5. Construction and validation of the predictive model

The predictive model is constructed using a logistic regression model with the “glmnet” package in R software. The receiver operating characteristic (ROC) curve was plotted using the “pROC” package, and the area under the curve (AUC) is calculated to assess the predictive capability of the model. The model's validation is performed using a 10-fold cross-validation method, iterating 1,000 cycles to build predictive models and calculate the mean AUC values for validation. The accuracy of the predictive model is verified using the “riskRegression” and “prodlim” packages in R software to plot calibration curves.

6. Statistical methods

Correlation analysis is performed using the Spearman correlation method, with the “ggplot2,” “ggpubr,” and “ggExtra” packages in R software used to plot correlation curves. Non-parametric testing (Wilcox test) is used to analyze statistical differences in FR⁺-CTC values and clinical relevant indicators among different gastrointestinal cancer patients, visualized using the “ggplot2” and “reshape2” packages in R software to create difference box plots. The chi-square test is used to compare baseline data between the training and validation groups. All analyses in this study are conducted in R software (version 4.0.3), with an alpha level of 0.05, considering p<0.05 as statistically significant.

1. Basic clinical information of patients

A total of 396 patients meeting the criteria were included after screening. The baseline clinical characteristics, tumor pathological staging, levels of albumin, total protein, hemoglobin, FR⁺-CTC values, and tumor marker expressions of the included patients are summarized in Table 1.

Table 1 . Baseline Clinical Information of Patients Included in This Study.

Characteristic	Value (n=396)
Age, median (IQR), yr	62 (55–71)
Sex, No. (%)
Male	250 (63.13)
Female	146 (36.87)
Vessel invasion, No. (%)
No	176 (44.44)
Yes	220 (55.56)
Nerve invasion, No. (%)
No	221 (55.81)
Yes	175 (44.19)
T stage, No. (%)
T1	58 (14.66)
T2	81 (20.45)
T3	172 (43.43)
T4	85 (21.46)
N stage, No. (%)
N0	212 (53.53)
N1	50 (12.63)
N2	94 (23.74)
N3	40 (10.10)
M stage, No. (%)
M0	362 (91.41)
M1	34 (8.59)
Stage, No. (%)
I	109 (27.52)
II	118 (29.80)
III	135 (34.09)
IV	34 (8.59)
Tumor, No. (%)
Gastric cancer	123 (31.06)
Intestinal cancer	273 (68.94)
Peritoneal metastasis, No. (%)
No	363 (91.67)
Yes	33 (8.33)
Gastric cancer	19 (57.58)
Intestinal cancer	14 (42.42)
FR+-CTC, median (IQR), FU/3 mL	10.80 (8.88–13.73)
Albumin, median (IQR), g/L	39.35 (34.98–41.10)
Total protein, median (IQR), g/L	66.75 (60.48–71.83)
Hemoglobin, median (IQR), g/L	122.0 (109.0–136.0)
CEA, median (IQR), ng/mL	1.85 (1.11–3.43)
CA19-9, median (IQR), U/mL	9.94 (5.83–17.54)
CA-125, median (IQR), U/mL	9.92 (6.66–15.32)

IQR, interquartile range; FR⁺-CTC, folate receptor-positive circulating tumor cells; CEA, carcinoembryonic antigen; CA19-9, cancer antigen 19-9; CA-125, cancer antigen 125..

2. Logistic regression to identify risk factors for peritoneal metastasis in gastrointestinal malignancies

A logistic regression model was constructed using the “glmnet” package in R software, incorporating all risk factors into the model. As shown in Table 2, in this predictive model, FR⁺-CTC, albumin, total protein, and CA-125 were found to be statistically significant (p<0.05), while hemoglobin, cancer antigen 19-9, and carcinoembryonic antigen were not statistically significant (p>0.05). Therefore, the indicators ultimately included in the model are: FR⁺-CTC, albumin, total protein, and CA-125.

Table 2 . Logistic Regression Analysis of Peritoneal Metastasis Variable in Patients with Gastrointestinal Malignancies.

	SE	Z-value	OR	95% CI	p-value
Intercept	2.039	–2.569	0.01	0.00–0.29	0.010
FR+-CTC	0.026	4.553	1.13	1.07–1.19	<0.001
Albumin	0.060	–3.127	0.83	0.74–0.93	0.002
Total protein	0.036	2.729	1.10	1.03–1.18	0.006
Hemoglobin	0.012	0.874	1.01	0.99–1.03	0.382
CA-125	0.008	2.335	1.02	1.00–1.04	0.019
CA19-9	0.002	–0.128	1.00	0.98–1.01	0.899
CEA	0.003	–0.183	1.00	0.99–1.01	0.855

SE, standard error; OR, odds ratio; CI, confidence interval; FR⁺-CTC, folate receptor-positive circulating tumor cells; CA-125, cancer antigen 125; CA19-9, cancer antigen 19-9; CEA, carcinoembryonic antigen..

3. ROC curve analysis for diagnosing peritoneal metastasis in gastrointestinal malignancies with FR⁺-CTC, albumin, total protein, and CA-125

The ROC curve was used to evaluate the accuracy of FR⁺-CTC and conventional clinical indicators in diagnosing peritoneal metastasis in patients with gastrointestinal malignancies. As shown in Fig. 1, the AUC values for diagnosing peritoneal metastasis with FR⁺-CTC, albumin, total protein, and CA-125 were 0.774, 0.686, 0.531, and 0.750, respectively. The optimal cutoff values for FR⁺-CTC, albumin, total protein, and CA-125 were 15.05 FU/3 mL, 37.45 g/L, 58.65 g/L, and 17.91 U/mL, respectively. This indicates that FR⁺-CTC had the highest predictive accuracy for peritoneal metastasis in patients with gastrointestinal malignancies.

Figure 1. Receiver operating characteristic (ROC) curves for the diagnosis of peritoneal metastasis in patients with gastrointestinal malignancies based on FR⁺-CTC count, albumin level, total protein level, and CA-125 level. FR⁺-CTC, folate receptor-positive circulating tumor cells; AUC, area under the curve; CA-125, cancer antigen 125.

4. Correlation analysis among FR⁺-CTC, albumin, total protein, and CA-125

Spearman correlation analysis was conducted to study the relationships between FR⁺-CTC and albumin, total protein, and CA-125 levels. As shown in Fig. 2A, there was a significant negative correlation between FR⁺-CTC and albumin levels (correlation coefficient R=–0.21, p<0.001). Fig. 2B showed a significant negative correlation between FR⁺-CTC and total protein levels (correlation coefficient R=–0.10, p=0.047). Fig. 2C indicated a significant positive correlation between FR⁺-CTC and CA-125 levels (correlation coefficient R=0.15, p=0.004). Although the FR⁺-CTC value has a certain correlation with albumin level, total protein level, and CA-125 level, the correlation coefficients are all relatively low.

Figure 2. Correlation analysis between CTC count and albumin level, total protein level, and CA-125 level. (A) Correlation analysis between FR⁺-CTC count and albumin level. (B) Correlation analysis between FR⁺-CTC count and total protein level. (C) Correlation analysis between FR⁺-CTC count and CA-125 level. FR⁺-CTC, folate receptor-positive circulating tumor cells; CA-125, cancer antigen 125.

5. Statistical difference analysis of FR⁺-CTC, albumin, total protein, and CA-125 levels among different types of gastrointestinal malignancies

As shown in Fig. 3A, patients with peritoneal metastasis had significantly higher levels of FR⁺-CTC and CA-125, and lower levels of albumin compared to those without peritoneal metastasis. Fig. 3B illustrated that patients with lymph node metastasis had significantly higher levels of FR⁺-CTC and CA-125, and lower albumin levels than those without lymph node metastasis. In Fig. 3C, patients with vascular invasion showed significantly higher FR⁺-CTC count and lower albumin levels than those without vascular invasion. Fig. 3D showed that patients with neural invasion had significantly higher levels of FR⁺-CTC and CA-125, and lower albumin levels than those without neural invasion. As per Fig. 3E, patients with T3-4 stage tumors had significantly higher FR⁺-CTC count compared to those in T1-2 stage. Fig. 3F indicated that patients with stage III-IV pathology had significantly higher levels of FR⁺-CTC and CA-125, and lower albumin levels compared to those in stage I-II. The detailed statistical data are shown in Table 3.

Figure 3. Statistical analysis of the differences in FR⁺-CTC count, albumin level, total protein level, and CA-125 level among different types of gastrointestinal malignancies. (A) Analysis of differences in FR⁺-CTC count, albumin level, total protein level, and CA-125 level between patients with and without peritoneal metastasis (PM). (B) Analysis of differences in FR⁺-CTC count, albumin level, total protein level, and CA-125 level between patients with and without lymph node (N) metastasis. (C) Analysis of differences in FR⁺-CTC count, albumin level, total protein level, and CA-125 level between patients with and without vascular invasion. (D) Analysis of differences in FR⁺-CTC count, albumin level, total protein level, and CA-125 level between patients with and without neural invasion. (E) Analysis of differences in FR⁺-CTC count, albumin level, total protein level, and CA-125 level between patients with T3-4 stage tumors and those with T1-2 stage tumors. (F) Analysis of differences in FR⁺-CTC count, albumin level, total protein level, and CA-125 level between patients with stage III-IV tumors and those with stage I-II tumors. FR⁺-CTC, folate receptor-positive circulating tumor cells; CA-125, cancer antigen 125. ^*p<0.05, ^†p<0.01, ^‡p<0.001.

Table 3 . Statistical Analysis of FR⁺-CTC Count, Albumin Level, Total Protein Level, and CA-125 Level in Various Types of Gastrointestinal Malignancies.

Covariates	Type	FR+-CT, FU/3 mL			Albumin, g/L			Total protein, g/L			CA-125, U/mL
		median (IQR)	p-value		median (IQR)	p-value		median (IQR)	p-value		median (IQR)	p-value
Peritoneal metastasis	PM (–)	10.5 (8.8–13.3)	<0.001		39.9 (35.3–43.2)	<0.001		66.8 (60.6–71.9)	0.551		9.05 (6.59–14.16)	<0.001
	PM (+)	18.8 (12.2–25.8)			35.0 (31.1–38.4)			65.6 (57.4–70.8)			21.79 (9.67–78.92)
Lymph node metastasis	N (–)	10.4 (8.9–12.7)	0.028		40.5 (35.9–43.5)	0.003		66.3 (60.8–71.9)	0.726		8.76 (6.46–12.81)	<0.001
	N (+)	11.6 (8.8–15.5)			38.0 (34.7–42.3)			67.4 (60.0–71.8)			10.43 (6.82–19.41)
Vascular invasion	Vessel (–)	10.2 (8.8–12.5)	0.005		40.5 (36.0–43.4)	0.013		66.6 (61.4–71.5)	0.906		9.09 (6.59–13.58)	0.074
	Vessel (+)	11.6 (9.0–15.5)			38.2 (34.4–42.9)			67.0 (59.8–71.8)			9.46 (6.69–17.58)
Neural invasion	Nerve (–)	10.4 (8.8–12.7)	0.039		40.3 (35.2–43.5)	0.014		66.6 (60.6–72.0)	0.892		9.05 (6.58–13.77)	0.036
	Nerve (+)	11.2 (8.9–15.9)			38.1 (34.9–42.3)			67.1 (59.9–71.5)			9.80 (6.79–18.77)
T stage	T1–2	10.2 (8.2–12.0)	<0.001		40.2 (35.2–43.3)	0.536		66.2 (59.8–70.9)	0.331		9.18 (6.58–13.97)	0.290
	T3–4	11.4 (9.1–15.3)			39.2 (34.9–43.0)			67.1 (60.8–71.9)			9.26 (6.81–16.14)
Stage	Stage I–II	10.3 (8.9–12.4)	0.004		40.5 (35.6–43.6)	0.005		66.2 (60.0–71.9)	0.452		8.81 (6.55–13.65)	0.005
	Stage III–IV	12.0 (8.8–16.1)			38.0 (34.8–42.1)			67.4 (60.8–71.8)			10.0 (6.82–19.97)

FR⁺-CTC, folate receptor-positive circulating tumor cells; CA-125, cancer antigen 125; PM, peritoneal metastasis; lQR, interquartile range..

6. Construction of a predictive model for peritoneal metastasis in gastrointestinal malignancies

Patients were randomly divided into a trial group and a validation group in a 1:1 ratio. Detailed clinical information of patients in both groups is shown in Table 4. There were no significant differences between the training and validation groups in terms of age, sex, vascular invasion, neural invasion, T stage, N stage, M stage, stage, tumor location, and presence of peritoneal metastasis (p>0.05), indicating comparability between the two groups.

Table 4 . Clinical Data of Patients Included in the Training and Validation Groups.

Covariates	Type	Train set (n=198)	Validation set (n=198)	p-value
Age	≤62 yr	100 (50.51)	102 (51.52)	0.920
	>62 yr	98 (49.49)	96 (48.48)
Sex	Male	124 (62.63)	126 (63.64)	0.917
	Female	74 (37.37)	72 (36.36)
Vessel invasion	No	85 (42.93)	91 (45.96)	0.613
	Yes	113 (57.07)	107 (54.04)
Nerve invasion	No	111 (56.06)	110 (55.56)	0.998
	Yes	87 (43.94)	88 (44.44)
T stage	T1	28 (14.14)	30 (15.15)	0.854
	T2	39 (19.70)	42 (21.21)
	T3	85 (42.93)	87 (43.94)
	T4	46 (23.23)	39 (19.70)
N stage	N0	96 (48.48)	116 (58.58)	0.094
	N1	30 (15.15)	20 (10.10)
	N2	47 (23.74)	47 (23.74)
	N3	25 (12.63)	15 (7.58)
M stage	M0	178 (89.90)	184 (92.93)	0.369
	M1	20 (10.10)	14 (7.07)
Stage	I	47 (23.74)	62 (31.31)	0.293
	II	59 (29.80)	59 (29.80)
	III	72 (36.36)	63 (31.82)
	IV	20 (10.10)	14 (7.07)
Tumor	Gastric cancer	66 (33.33)	57 (28.79)	0.385
	Intestinal cancer	132 (66.67)	141 (71.21)
Peritoneal metastasis	No	178 (89.90)	185 (93.43)	0.275
	Yes	20 (10.10)	13 (6.57)

Data are presented as number (%)..

7. ROC curve and nomogram construction for the predictive model

As shown in Fig. 4A-C, the AUC values for the training group, validation group, and the entire patient cohort were 0.813, 0.874, and 0.837, respectively, indicating that the model has a strong predictive capability. Fig. 4D illustrates a nomogram, which visualizes the predictive model, enhancing the ability to predict the probability of peritoneal metastasis in patients with gastrointestinal malignancies, thus applying this predictive model in clinical settings. By determining the scores for each indicator and calculating the total score, the probability of a patient having peritoneal metastasis can be estimated. The specific parameters of this predictive model, as shown in Table 5, indicate that indicators such as FR⁺-CTC, albumin, total protein, and CA-125 are statistically significant (p<0.05).

Figure 4. Receiver operating characteristic (ROC) curve and nomogram for the predictive model. (A) ROC curve for the predictive model constructed with the training set. (B) ROC curve for the predictive model constructed with the validation set. (C) ROC curve for the predictive model constructed with all patient data. (D) Nomogram for the predictive model. AUC, area under the curve; FR⁺-CTC, folate receptor-positive circulating tumor cells; CA-125, cancer antigen 125.

Table 5 . Basic Parameters of the Predictive Model for Determining Peritoneal Metastasis in Patients with Gastrointestinal Malignancies.

	SE	Z-value	OR	95% CI	p-value
Intercept	1.895	–2.443	0.01	0.00–0.40	0.014
FR+-CTC	0.026	4.593	1.13	1.07–1.18	<0.001
Albumin	0.057	–3.036	0.84	0.75–0.94	0.002
Total protein	0.036	2.780	1.10	1.03–1.18	0.005
CA-125	0.005	3.468	1.02	1.01–1.03	0.001

SE, standard error; OR, odds ratio; CI, confidence interval; FR⁺-CTC, folate receptor-positive circulating tumor cells; CA-125, cancer antigen 125..

8. Validation of the predictive model through ten-fold cross-validation and calibration curve

The accuracy of the predictive model was validated using a 10-fold cross-validation method, iterating 1,000 cycles to build predictive models and calculate the mean AUC values for validation. As depicted in Fig. 5A-C, the cross-validation AUC values for the training group, validation group, and the entire patient cohort were 0.813, 0.886, and 0.844, respectively. These AUC values, all greater than 0.8, indicate a stable predictive capability of the model. The calibration curve, shown in Fig. 5D, demonstrates that the red curve closely aligns with the 45° diagonal line, suggesting that the probability of peritoneal metastasis predicted by the model closely matches the actual occurrence of peritoneal metastasis in patients, confirming the model's strong predictive accuracy.

Figure 5. Ten-fold cross-validation and calibration curve analysis for validating the accuracy of the predictive model. (A) Cross-validated receiver operating characteristic (ROC) curve of the predictive model constructed with the training set. (B) Cross-validated ROC curve of the predictive model constructed with the validation set. (C) Cross-validated ROC curve of the predictive model constructed with all patient data. (D) Calibration curve for the predictive model.

Gastrointestinal malignancies, such as gastric and colorectal cancers, are among the most common malignant tumors globally.^19-22 The risk associated with these cancers stems from their high invasiveness and tendency to metastasize, especially to the peritoneum. Peritoneal metastasis is a common late-stage phenomenon in gastrointestinal malignancies, significantly impacting patient quality of life and overall survival rates.^23,24 Traditional imaging methods have limitations in early detection of peritoneal metastasis, hence developing new, more sensitive, and specific methods for early prediction and diagnosis of peritoneal metastasis in gastrointestinal malignancies is of paramount importance.^25-27 FR⁺-CTC have received extensive attention in recent cancer research.^28-30 The FR is a glycoprotein associated with the cell membrane that efficiently mediates cellular uptake of folate, an indispensable element for cell growth and division. In certain types of tumor cells, the expression of the FR is significantly upregulated, making it a marker distinguishing tumor cells from normal cells. CTC are tumor cells that have shed from the primary tumor into the peripheral blood circulation. Due to their high specificity and ease of detection, FR⁺-CTC are recognized as powerful biomarkers for cancer diagnosis and prognosis. Studies have indicated that FR⁺-CTC are not only involved in tumor progression but also actively contribute to the metastatic cascade. Overexpression of FR has been linked to enhanced tumor cell proliferation, migration, invasion, and immune evasion, all of which are critical steps in metastatic development. Recent evidence suggests that FR⁺-CTC facilitates hematogenous dissemination by interacting with the microenvironment and promoting the establishment of pre-metastatic niches. For example, FR-mediated signaling pathways have been implicated in epithelial-mesenchymal transition, a process that endows tumor cells with increased motility and invasiveness, further enhancing their ability to metastasize.^31-33 These findings provide mechanistic insights into the role of FR⁺-CTC in peritoneal metastasis, highlighting their potential as both a predictive biomarker and a therapeutic target. Furthermore, the unique biology of FR⁺-CTC, including their ability to survive in circulation and evade anoikis, underscores their clinical relevance in predicting metastatic risk and guiding treatment strategies. Recent studies suggest that detecting FR⁺-CTC can predict the presence and metastasis of tumors at asymptomatic or early stages.

In our study, we have developed a predictive model based on FR⁺-CTC for predicting peritoneal metastasis in gastrointestinal malignancies. The establishment of this model not only provides a new perspective for understanding and predicting the clinical progression of gastrointestinal malignancies but also may offer substantial support for clinical decision-making. Firstly, we found that FR⁺-CTC, albumin, total protein, and CA-125 are significant factors in predicting peritoneal metastasis in gastrointestinal malignancies. FR⁺-CTC shows a negative correlation with albumin and total protein, and a positive correlation with CA-125. The observed negative correlation between FR⁺-CTC and albumin or total protein may reflect the systemic impact of advanced malignancies. Hypoalbuminemia and decreased total protein, often observed in cancer patients, are markers of malnutrition, systemic inflammation, and poor prognosis. Systemic inflammation, characterized by elevated levels of pro-inflammatory cytokines such as interleukin-6 and tumor necrosis factor-α, suppresses hepatic albumin synthesis and enhances protein catabolism, weakening the host’s immune defenses. These systemic changes create a permissive environment for the survival and dissemination of CTC by allowing them to evade immune surveillance and establish metastatic colonies. Moreover, hypoalbuminemia may reduce antioxidant capacity, further enhancing the resilience of tumor cells in circulation against oxidative stress. The decline in albumin and total protein also reflects the metabolic reprogramming associated with cancer cachexia, wherein resources are diverted to sustain tumor growth and progression. Conversely, the positive correlation between FR⁺-CTC and CA-125 underscores CA-125’s role as a biomarker of tumor burden and peritoneal involvement. CA-125, secreted by mesothelial cells lining the peritoneum, is frequently elevated in response to peritoneal irritation or tumor invasion. Elevated CA-125 levels may reflect a pro-inflammatory state in the peritoneal cavity, driven by tumor cell-secreted cytokines and growth factors. This inflammatory environment could facilitate FR⁺-CTC adhesion and implantation by promoting extracellular matrix remodeling via matrix metalloproteinases.^34-37 Additionally, CA-125 is involved in mediating cell adhesion and aggregation through its interactions with MUC16 and integrins, thereby supporting FR⁺-CTC driven metastatic cascades. These interactions between FR⁺-CTC and CA-125 highlight their complementary roles in local and systemic aspects of tumor progression and underscore their combined value as predictive biomarkers for peritoneal metastasis. These findings suggest that monitoring the dynamics of FR⁺-CTC in conjunction with albumin, total protein, and CA-125 could provide a more comprehensive assessment of tumor progression and metastatic potential. Previous studies have extensively researched the role of the FR in the growth and metastasis of tumor cells, with increased expression often correlating with poor prognosis in malignant tumor patients.^38,39 Our study not only quantitatively confirms this observation but also finds that the expression of FR⁺-CTC is significantly related to the risk of the tumor peritoneal metastasis. This finding further strengthens the importance of FR⁺-CTC as a potential biomarker, potentially offering a more accurate, biology-based predictive tool. Secondly, our predictive model demonstrates high predictive accuracy. While the AUC values reported for our model exceed 0.8 in the training group, validation group, and across all patient groups, additional metrics provide a fuller picture of the model’s performance. Specifically, our model demonstrated sensitivity, specificity, and overall accuracy exceeding 85%. This balance between sensitivity and specificity ensures the model’s capacity to accurately identify patients at risk while minimizing false positives, offering a robust tool for clinical application. These supplementary metrics, alongside the high AUC value, emphasize the model’s reliability and practicality. This indicates that our model can effectively differentiate patients at risk of peritoneal metastasis and provide accurate predictive results for clinical decision-making. More importantly, the model's advantage lies in using clinical indicators available from routine examinations, making it more feasible for practical application. This model aids understanding and prediction of the progression of gastrointestinal malignancies and potentially improving treatment strategies for patients.

However, despite the high predictive accuracy and clinical value of our predictive model, we must acknowledge some limitations in our study. Firstly, this study is based on a patient group from a single center, which may introduce sample selection bias. Future studies should be conducted across multiple centers to enhance the generalizability of the results. Secondly, although our predictive model performs well in the training and validation groups, its applicability to gastrointestinal malignancies in other countries needs further validation. One key limitation of these validation groups is the lack of subgroup analyses to assess the model’s performance across diverse clinical and demographic patient populations. Without this, the robustness and consistency of the model in varied settings remain unclear. Future studies should address these gaps by including larger, multi-center cohorts that better represent different populations and clinical contexts. Such analyses would help determine whether the predictive performance observed in the current study holds across different healthcare systems and patient demographics. Moreover, incorporating external validation datasets could further confirm the model’s reliability and minimize potential biases. These steps are crucial to enhance the clinical applicability of our findings. Future research should focus on elucidating the mechanisms linking FR⁺-CTC, albumin, total protein, and CA-125. Investigating whether the inflammatory state or specific cytokine profiles mediate these relationships could uncover new therapeutic targets. Exploring the dynamic interplay between nutritional status, systemic inflammation, and tumor biology will also be critical. Furthermore, incorporating additional biomarkers such as circulating tumor DNA or exosomal proteins could complement FR⁺-CTC analysis, enhancing the predictive model’s robustness. These efforts may not only refine the model but also provide a foundation for personalized therapeutic strategies. Current cancer research is focusing on the discovery of more precise biomarkers and personalized treatment strategies. In this context, the discovery of FR⁺-CTC and its application in predicting tumor metastasis is of significant importance. Future research could explore combining FR⁺-CTC with other emerging biomarkers to enhance the accuracy and clinical applicability of the predictive model.^40,41 Moreover, with the development of precision medicine and personalized treatment strategies, our model can be further refined to meet the specific needs of different patient groups.

In summary, this discovery provides new insights for the early diagnosis and treatment of gastrointestinal malignancies. Particularly in predicting peritoneal metastasis preoperatively, the detection of FR⁺-CTC may become an essential tool. The development of a preoperative predictive model based on FR⁺-CTC is a key step in translating these research findings into clinical practice. This model integrates patients' biomarker data and clinical characteristics, offering clinicians a quantitative assessment of the risk of peritoneal metastasis. This is crucial for formulating individualized treatment strategies, such as deciding on adjuvant chemotherapy and selecting surgical methods. We anticipate that future research will further expand and refine this model, enabling it to play a more significant role in the personalized treatment of cancer.

In conclusion, FR⁺-CTC can serve as a novel biomarker for gastrointestinal malignancies. A new model based on FR⁺-CTC demonstrates strong predictive capabilities for preoperative assessment of peritoneal metastasis in gastrointestinal cancers.

No potential conflict of interest relevant to this article was reported.

Study concept and design: R.H. Data acquisition: D.L. Data analysis and interpretation: D.L. Drafting of the manuscript: D.L. Critical revision of the manuscript for important intellectual content: C.L. Statistical analysis: C.L. dministrative, technical, or material support; study supervision: R.H. Approval of final manuscript: all authors.

The data of this study can be obtained from the corresponding author according to reasonable requirements.