Prognostic significance of Fusobacterium nucleatum in colorectal cancer using reconstructed survival data

Abstract

BACKGROUND

Fusobacterium nucleatum has been implicated in colorectal cancer (CRC) progression, recurrence, immune evasion, and treatment resistance. However, clinical studies evaluating its prognostic significance have reported inconsistent findings.

AIM

To determine the prognostic significance of tissue F. nucleatum abundance in CRC.

METHODS

We searched PubMed, Scopus, Web of Science, and the Cochrane Library through August 10, 2025. Eligible observational studies evaluated tissue F. nucleatum abundance in CRC and reported survival outcomes. Individual patient data were reconstructed from Kaplan–Meier curves. Cox proportional hazards models, random-effects meta-analysis, flexible parametric survival models, and restricted mean survival time analyses were performed.

RESULTS

Twelve cohort studies involving 4358 patients were included. Using the high/positive F. nucleatum expression group as the reference, the low/negative expression group had a significantly lower risk of disease-free survival (DFS) events, corresponding to worse DFS among patients with high/positive expression (hazard ratio = 0.45, 95% confidence interval: 0.37–0.56). A similar association was observed for cancer-specific survival (hazard ratio = 0.34, 95%CI: 0.26–0.44). No significant association was observed for overall survival (hazard ratio = 0.99, 95% confidence interval: 0.82–1.20). Restricted mean survival time differences were 21.8 months for DFS and 38.4 months for cancer-specific survival.

CONCLUSION

High tissue F. nucleatum abundance is associated with poorer DFS and cancer-specific survival, but not overall survival, in CRC.

Key Words: Fusobacterium nucleatum; Colorectal cancer; Prognosis; Survival; Meta-analysis; Individual patient data; Kaplan–Meier curve; Restricted mean survival time

Core Tip: This meta-analysis evaluated the prognostic significance of tissue Fusobacterium nucleatum abundance in colorectal cancer using reconstructed individual patient data from Kaplan–Meier curves. High/positive F. nucleatum expression was associated with poorer disease-free survival and cancer-specific survival, whereas no significant association was observed for overall survival. These findings suggest that F. nucleatum may be more closely associated with recurrence and cancer-specific mortality than with all-cause mortality, although further validation is required.

INTRODUCTION

Colorectal cancer (CRC) is the second leading cause of cancer-related death worldwide, with nearly 2 million new cases reported annually[1]. Unfortunately, evidence shows that frequent follow-up does not necessarily lead to earlier detection of recurrence[2]. Even after intended curative resection, CRC still recurs in about 45% of patients[2]. Several factors are known to play important roles in prognosis[3,4]. Therefore, incorporating prognostic biomarkers into daily clinical practice for CRC may significantly improve treatment decisions and follow-up strategies.

In recent years, the gut microbiota has emerged as an important factor in human health and disease[5-8], and it now appears to be a key factor to consider in many disorders, including cancer. Specifically, dysbiosis of the gut microbiome is strongly associated with CRC pathogenesis and progression[5,9,10]. In fact, there is a close association between CRC and the gut microbiota, which constitutes a major component of the tumor microenvironment[9]. Evidence has linked CRC tumorigenesis to gut dysbiosis, suggesting that the microbiota may serve not only as a causative factor but also as a potential biomarker for metastasis[9,11,12]. The ability of the gut microbiota to interact with colonic epithelial and immune cells through a wide range of metabolites and proteins highlights its role in modulating the tumor environment, as demonstrated in mechanistic studies[13,14]. Moreover, it has been shown to influence patient responses to chemotherapy and immunotherapy, which are significant factors in CRC prognosis[15]. Together, these findings position the gut microbiota as an important factor to consider when investigating CRC prognosis.

Among the microbial species studied, Fusobacterium nucleatum has attracted particular attention in CRC pathogenesis and prognosis. Fusobacterium nucleatum promotes cancer development through different mechanisms, including activation of oncogenic signals, suppression of host immunity via the Fap2 protein, microRNA-21–mediated induction of CRC, and enhancement of chemoresistance[16-20]. Moreover, F. nucleatum has been detected in CRC liver metastases, which suggests its involvement in metastatic spread and further supports its importance as a prognostic factor[21].

However, studies examining the prognostic role of F. nucleatum in CRC have yielded conflicting results, with some demonstrating associations with worse survival outcomes and others finding no significant link[22-33]. An earlier meta-analysis in 2022 suggested that F. nucleatum was associated with worse CRC prognosis[34]. Yet, this analysis was largely based on published hazard ratios (HRs) (summary-level data) without reconstructing survival outcomes at the individual level. To address these gaps, we conducted an updated meta-analysis to clarify the prognostic significance of tissue F. nucleatum abundance in CRC. This study extends previous evidence by reconstructing individual patient data from published Kaplan–Meier curves and applying time-to-event survival analyses, restricted mean survival time estimates, and flexible parametric survival models. This approach allows a more detailed evaluation of the association between F. nucleatum abundance and CRC survival outcomes while preserving the observational nature of the available evidence.

MATERIALS AND METHODS

In this meta-analysis, we followed the PRISMA statement guidelines[35]. The study was also conducted according to the Cochrane Handbook for Systematic Reviews of Interventions[36]. This meta-analysis was registered in the Open Science Framework under registration No. 10.17605/OSF.IO/8432A.

Literature search and screening

A comprehensive electronic search of PubMed, Scopus, Web of Science, and the Cochrane CENTRAL Library was conducted from inception to August 10, 2025. The search strategy combined terms related to CRC, F. nucleatum, tissue-based detection, and prognostic outcomes. The search included the following terms: [“Colorectal Neoplasms” (MeSH) OR “colorectal cancer” OR “colon cancer” OR “rectal cancer”] AND [“Fusobacterium nucleatum” (MeSH) OR “Fusobacterium nucleatum” OR “F. nucleatum”] AND (“quantitative PCR” OR “qPCR” OR “digital droplet PCR” OR “ddPCR” OR “molecular detection” OR “bacterial load” OR “bacterial abundance” OR “bacterial density”) AND (“survival” OR “prognosis” OR “outcome” OR “mortality” OR “recurrence” OR “disease-free survival” OR “cancer-specific survival” OR “overall survival” OR “high levels” OR “low levels”). The search strategy was intentionally focused on tissue-based studies using molecular detection methods because these methods allow classification of CRC samples according to F. nucleatum abundance, which is necessary for survival comparison and Kaplan–Meier reconstruction. The detailed search strategy for each database is presented in Supplementary Table 1. Duplicates were detected and removed using EndNote X9 and manually verified[37]. Articles were screened using Rayyan software[38]. Backward citation screening was also conducted to identify additional relevant studies. Screening was performed in 2 steps: Title and abstract screening followed by full-text assessment of potentially eligible studies.

Eligibility criteria and endpoints

We included full-length articles published in English that detected F. nucleatum in human CRC tissue samples, categorized CRC samples into high and low (or high and low/negative) F. nucleatum levels, and reported clinicopathological, molecular, or survival-associated data correlated with F. nucleatum levels. We excluded studies that were abstracts or posters only, could not be dichotomized into high and low F. nucleatum levels, had insufficient data, used multiple types of biospecimens to detect F. nucleatum, or assessed F. nucleatum levels in stool samples. The survival outcomes evaluated were overall survival (OS), cancer-specific survival (CSS), and disease-free survival (DFS).

Potential cohort overlap was assessed by comparing study location, recruitment period, patient population, sample size, and reported outcomes. When potential overlap was suspected, the study with the most complete survival data was prioritized to avoid double counting.

Statistical analysis

A standardized extraction sheet was used to collect study characteristics, baseline population data, survival outcomes, and risk-of-bias domains. Individual patient data for OS, DFS, and CSS were reconstructed from published Kaplan–Meier curves using the method described by Liu et al[39]. First, WebPlotDigitizer 4.5 was used to extract survival probabilities at each time point from eligible Kaplan–Meier curves[40]. When available, numbers at risk were incorporated into the reconstruction process. Individual patient data were reconstructed using the “IPDfromKM” package in R software.

For quality control, each reconstructed Kaplan–Meier curve was visually compared with the corresponding published curve. When number-at-risk tables were available, reconstructed risk tables were also compared with the published values. Reconstruction accuracy was assessed using root mean square error, mean absolute error, and maximum absolute error. Reconstruction was considered acceptable when visual agreement was satisfactory and the predefined accuracy criteria were met, including root mean square error ≤ 0.05, mean absolute error ≤ 0.02, and maximum absolute error ≤ 0.05. Curves showing marked deviation from the original published curves or reported risk tables were re-digitized and reconstructed.

Cox proportional hazards models were used to calculate HRs and 95% confidence intervals (CIs) from reconstructed individual patient data. A 2-stage meta-analysis was then performed using random-effects models to pool study-level HRs. When HRs were not reported in the original study, they were recalculated from the reconstructed individual patient data. Flexible parametric survival models using B-spline functions were applied to assess time-varying HRs. Restricted mean survival time was calculated using the “survRM2” package in R. Unless otherwise stated, HRs compared the low/negative F. nucleatum expression group with the high/positive expression group as the reference.

Quality assessment

Two reviewers independently collected data from the eligible articles using a standardized form. The extracted data included the first author’s name, year of publication, total number of patients, number of F. nucleatum-positive patients, type of biospecimen (formalin-fixed paraffin-embedded tissue or fresh-frozen tissue), method of F. nucleatum detection, survival data, and clinicopathological variables associated with F. nucleatum levels. The methodological quality of the included studies was independently assessed by the 2 reviewers using the Newcastle–Ottawa Scale[41]. This tool evaluates studies across 3 domains: Selection, comparability, and outcome. Based on this assessment, studies were categorized as “good quality,” “fair quality,” or “poor quality.” A score exceeding 6 out of a possible 9 points was predefined as the threshold for a study to be considered of acceptable quality. Any discrepancies between the 2 reviewers were resolved through discussion until consensus was reached.

RESULTS

Literature search and characteristics

The initial search yielded 572 articles, of which 400 were eligible for screening after removing duplicates. After title and abstract screening, 26 articles were considered for full-text screening. Twelve cohort studies[22-33] were included, comprising 4358 patients (2432 [55.8%] males; mean age, 63.9 years; follow-up duration, 1–17 years). The PRISMA flow diagram is shown in Figure 1. In total, 3331 (76.4%) patients were allocated to the low/negative expression group, and 1027 (23.6%) were allocated to the high/positive expression group for F. nucleatum detection. Further details on the included studies and baseline characteristics of the population are presented in Table 1.

Figure 1 PRISMA flow diagram of study selection. ¹The number of records identified from each database is reported separately.

Table 1 Characteristics and baseline data of included studies on Fusobacterium nucleatum in colorectal cancer.

Study	Study interval	Country	Study design	Total cases, n	Positive cases, n	Biospecimen	Method	Group	Patients, n	Age, years	Male sex, %	Stage I, n	Stage II, n	Stage III, n	Stage IV, n
Ito et al[25], 2015	2001–2013	Japan, Asia	Cohort study	511	143	FFPE	qPCR	Negative	225	66.1 ± 11.9	57	23	63	112	27
								Low expression	143	69.1 ± 10.9	57	20	44	58	21
								High expression	143	66.5 ± 12.2	54	13	53	65	12
Kunzmann et al[26], 2019	2008–2018	Czech Republic, Europe	Cohort study	190	61	FF	qPCR	Low expression	129	65.6 ± 18.27	69	25	36	44	22 + (2 unknown stage)
								High expression	61	67 ± 17.6	67.2	9	23	14	14 + (1 unknown stage)
Oh et al[28], 2019	2005–2013	South Korea, Asia	Cohort study	593	204	FFPE	qPCR	Negative/low expression	389	56.6 ± 9	61.2	362		27
								High expression	204	66.3 ± 12	60.8	188		16
Chen et al[24], 2019	2017–2019	China, Asia	Cohort study	91	25	FFPE	qPCR	Negative/low expression	66	58.52 ± 10.42	57.6	-	36	30	-
								High expression	25	58.72 ± 12.64	68	-	15	10	-
Mima et al[27], 2016	Up to 2008	Japan, Asia	Cohort study	1069	67	FFPE	qPCR	Negative	935	69.2 ± 8.8	43	225	274	239	115
								Low	67	71.1 ± 9.0	39	9	23	25	5
								High	67	68.8 ± 8.3	34	7	28	14	13
Wei et al[31], 2016	2010–2012	China, Asia	Cohort study	180	90	FF	qPCR	Low expression	90	63.5 ± 10.2	63.3	53		37
								High expression	90	69.8 ± 12.5	56.7	42		48
Sun et al[30], 2016	2009–2010	China, Asia	Cohort study	152	59	FF	qPCR	Negative	34	66.5 ± 11.2	36.4	2	16	11	5
								Positive	118	69.2 ± 10.9	63.6	4	41	53	20
Yamaoka et al[32], 2018	1999–2008	Japan, Asia	Cohort study	100	50	FF	ddPCR	Low	50	66.7 ± 12.8	50	11	19	13	7
								High	50	67.0 ± 10.0	52	3	17	17	13
Yan et al[33], 2017	2007–2015	China, Asia	Cohort study	280	187	FFPE	qPCR	Low	93	65.3 ± 11.1	54.8	-	-
								High	187	68.4 ± 11.5	57.2	-	-
Sakamoto et al[22], 2021	2001–2018	Japan, Asia	Cohort study	181	8	FFPE	qPCR	Negative	173	63.5 ± 11.1	65.3	-	-
								Positive	8	58.5 ± 12.0	50	-	-
Phipps et al[29], 2025	1998–2007	United States, North America	Cohort study	859	86	FFPE	ddPCR	Negative	683	57 ± 12	48	212	375	80	16
								Low	90	58 ± 12	33.0	27	48	13	2
								High	86	58 ± 13	37	21	56	9	0
de Carvalho et al[23], 2019	2008–2015	Brazil, South America	Cohort study	152	18	FFPE	qPCR	Negative	117	60.6 ± 13.7	54.7	39	74	4	0
								Low	17	-	58.8	2	13	2	0
								High	18	-	38.9	2	15	0	0

Age is presented as mean ± SD. Expression categories reflect the definitions used in the original studies; when studies reported negative, low, and high categories, negative and low categories were combined as appropriate for the low/negative comparison group in the survival analysis. ddPCR: Digital droplet polymerase chain reaction; FF: Fresh-frozen tissue; FFPE: Formalin-fixed paraffin-embedded tissue; qPCR: Quantitative polymerase chain reaction; -: Data not reported or not available.

Quality assessment

The included cohort studies were assessed using the Newcastle–Ottawa Scale. The studies scored between 6 and 9 points, indicating fair to good quality (Supplementary Table 2).

Outcomes

Analysis of OS: According to the reconstructed individual patient data, 8 studies reported Kaplan–Meier curves for OS, including 1091 patients: 682 in the low/negative F. nucleatum expression group and 409 in the high/positive expression group (Figure 2A). Using the high/positive expression group as the reference, no significant difference in OS was observed between the low/negative and high/positive groups (HR = 0.99, 95%CI: 0.82–1.20, P = 0.93). The overall restricted mean survival time was 92.5 months (95%CI: 87.18–96.90) in the low/negative expression group and 91.4 months (95%CI: 85.81–97.20) in the high/positive expression group, with a nonsignificant difference of 1.0 month (P = 0.462).

Figure 2 Kaplan–Meier survival curves stratified by tissue Fusobacterium nucleatum expression. A: Overall survival; B: Disease-free survival; C: Cancer-specific survival. CI: Confidence interval; HR: Hazard ratio.

DFS: Three studies reported Kaplan–Meier curves for DFS, including 1053 patients: 572 in the low/negative F. nucleatum expression group and 481 in the high/positive expression group. Using the high/positive expression group as the reference, the low/negative expression group had a significantly lower risk of DFS events (HR = 0.45, 95%CI: 0.37–0.56, P < 0.001). This finding indicates poorer DFS among patients with high/positive F. nucleatum expression. The restricted mean survival time was 102.4 months (95%CI: 99.0–105.9) in the low/negative expression group and 80.5 months (95%CI: 75.8–85.3) in the high/positive expression group, corresponding to a significant difference of 21.8 months (P = 0.002) (Figure 2B).

CSS: Four studies reported Kaplan–Meier curves for CSS, including 630 patients: 185 in the low/negative F. nucleatum expression group and 445 in the high/positive expression group. Using the high/positive expression group as the reference, the low/negative expression group had a significantly lower risk of cancer-specific death (HR = 0.34, 95%CI: 0.26–0.44, P < 0.001) (Figure 2C). This finding supports an association between high/positive F. nucleatum expression and poorer CSS. The restricted mean survival time was 83.4 months (95%CI: 77.0–90.0) in the low/negative expression group and 45.0 months (95%CI: 40.4–49.5) in the high/positive expression group, corresponding to a significant difference of 38.4 months (P = 0.002). Restricted mean survival time and restricted mean time lost over the entire follow-up are shown in Figure 3A and B, respectively. Figure 4 presents the analysis of time-varying HRs for mortality based on flexible parametric survival models with B-splines.

Figure 3 Restricted mean survival time and restricted mean time lost stratified by tissue Fusobacterium nucleatum expression. A: Restricted mean survival time; B: Restricted mean time lost. RMST: Restricted mean survival time; RMTL: Restricted mean time lost.

Figure 4 Time-varying hazard ratios for overall survival based on flexible parametric survival models with B-splines. HR: Hazard ratio.

DISCUSSION

Contextual interpretation of our findings

Previous clinical and experimental evidence has linked F. nucleatum to CRC carcinogenesis and progression, with increasing attention to its potential prognostic relevance[42-45]. Mechanistic studies suggest that F. nucleatum may promote tumor growth and metastasis through E-cadherin/β-catenin signaling, immune evasion, induction of chemoresistance, and persistence within metastatic lesions. These biological properties make F. nucleatum an organism of interest for understanding CRC prognosis and tumor behavior[45].

In this meta-analysis, high/positive tissue F. nucleatum expression was associated with poorer DFS and CSS, whereas no significant difference was observed for OS. This pattern suggests that F. nucleatum may be more closely related to cancer progression, recurrence, and CRC-specific mortality than to all-cause mortality. However, this interpretation should be considered cautiously because the number of studies contributing to DFS and CSS was limited.

OS reflects all-cause mortality and is therefore influenced by deaths unrelated to CRC. The studies included in our analysis often had long follow-up durations and older patient populations (Tables 1 and 2), which likely introduced competing risks such as cardiovascular disease or other malignancies. These competing causes may have diluted the cancer-specific effect of F. nucleatum on OS. By contrast, the significant associations with DFS and CSS indicate that high F. nucleatum levels are strongly linked to increased recurrence and CRC-specific mortality, but this effect becomes less apparent when non-cancer deaths are included in the OS analysis.

Table 2 Survival outcomes based on Fusobacterium nucleatum expression in colorectal cancer.

Outcome	Studies, n	Patients (low vs high), n	HR	95%CI	P value	RMST low (95%CI)	RMST high (95%CI)	RMST difference	RMST P value
Overall survival	8	1091 (682 vs 409)	0.99	0.82–1.20	0.93	92.5 (87.18–96.9)	91.4 (85.81–97.2)	1.0	0.462
Disease-free survival	3	1053 (572 vs 481)	0.45	0.37–0.56	< 0.001	102.4 (99.0–105.9)	80.5 (75.8–85.3)	21.8	0.002
Cancer-specific survival	4	630 (185 vs 445)	0.34	0.26–0.44	< 0.001	83.4 (77.0–90.0)	45.0 (40.4–49.5)	38.4	0.002

Hazard ratios compare the low/negative Fusobacterium nucleatum expression group with the high/positive expression group as the reference. A hazard ratio < 1 indicates a lower event risk in the low/negative expression group. CI: Confidence interval; HR: Hazard ratio; RMST: Restricted mean survival time.

In addition, treatment-related factors may have contributed. Patients with recurrent disease or cancer-specific events may still benefit from salvage therapies, leading to extended OS despite worse DFS or CSS. In such cases, OS becomes less sensitive to differences directly attributable to F. nucleatum.

However, it should be noted that the number of studies contributing to each analysis differed substantially. Only 3 and 4 studies were included for DFS and CSS, respectively, compared with 8 studies for OS. This imbalance adds a degree of uncertainty to the pooled results for DFS and CSS compared with OS.

Overall, our findings suggest that F. nucleatum primarily influences tumor biology and progression rather than overall mortality. Thus, DFS and CSS may serve as more reliable and sensitive endpoints than OS when assessing the prognostic value of F. nucleatum in CRC. However, future studies should place greater emphasis on reporting DFS and CSS to strengthen the evidence base.

Compared with the previous meta-analysis by Kim et al[34], which reported that high F. nucleatum expression was associated with poor OS, DFS, and CSS in patients with CRC, our findings are consistent for DFS and CSS but differ for OS. This discrepancy may be partly related to differences in study inclusion and analytical approach. Their OS analysis included 6 studies, whereas our review included 8 studies. In addition, the previous analysis relied mainly on published HRs, whereas our study reconstructed individual patient data from Kaplan–Meier curves, allowing time-to-event analysis and restricted mean survival time estimation. Therefore, our results should be viewed as an updated and methodologically distinct synthesis rather than a direct contradiction of previous evidence.

Implications

The findings of this meta-analysis have several implications. First, the observed associations with DFS and CSS support the prognostic relevance of tissue F. nucleatum abundance in CRC. Patients with high/positive F. nucleatum expression may represent a subgroup with greater risk of recurrence and CRC-specific death. Second, the lack of a significant association with OS emphasizes the importance of endpoint selection in prognostic biomarker research. OS may be influenced by competing causes of death, comorbidities, and treatment after recurrence, whereas DFS and CSS may more directly reflect tumor-related prognosis.

Nevertheless, these findings should not be interpreted as sufficient evidence for routine clinical screening at present. The available studies are observational, and the number of studies contributing to DFS and CSS remains limited. Further prospective studies using standardized assays, harmonized cutoff definitions, and clinically relevant endpoints are needed before F. nucleatum testing can be recommended for routine risk stratification.

Strengths and limitations

This meta-analysis has several strengths. First, the pooled sample size was relatively large, which increased the statistical power of the analysis. Second, instead of relying only on reported HRs, we reconstructed individual patient data from Kaplan–Meier curves. This enabled direct time-to-event analysis and calculation of restricted mean survival time, which provides clinically interpretable estimates of survival differences. Third, the use of flexible parametric survival models allowed assessment of time-varying effects. Finally, the inclusion of OS, DFS, and CSS provided a more nuanced evaluation of the prognostic significance of F. nucleatum in CRC.

Several limitations should also be acknowledged. The number of studies available for DFS and CSS was limited compared with OS. Therefore, although the associations with DFS and CSS were statistically significant, these findings should be interpreted with caution because the pooled estimates may be more sensitive to the influence of individual studies. In addition, the included studies differed in geographic region, biospecimen type, assay method, tumor stage distribution, treatment context, and cutoff definition for high F. nucleatum expression. Although these variables may modify the prognostic association, formal subgroup analyses were not feasible because of the limited number of studies per endpoint and inconsistent reporting of relevant covariates. Many included studies were conducted in East Asia, which may limit generalizability to other populations. Finally, all included studies were observational, and residual confounding cannot be excluded.

CONCLUSION

This meta-analysis suggests that high/positive tissue F. nucleatum abundance is associated with poorer DFS and CSS in CRC, whereas no significant association was observed for OS. These findings support the prognostic relevance of F. nucleatum, particularly for recurrence and CRC-specific mortality. However, the results should be interpreted cautiously because the DFS and CSS analyses were based on a limited number of studies. Further prospective studies using standardized detection methods, harmonized cutoff values, and diverse populations are needed to validate the prognostic role of F. nucleatum before routine clinical implementation can be recommended.