Systematic review and meta-analysis of studies assessing circulating tumor DNA kinetics in metastatic colorectal cancer

Abstract

Background

Circulating tumor DNA (ctDNA) kinetics have emerged as a promising biomarker for monitoring treatment response in metastatic colorectal cancer (mCRC). We carried out a systematic review and meta-analysis to comprehensively assess the prognostic value of ctDNA kinetics during palliative systemic therapy, aiming to consolidate the current evidence and clarify its potential role in clinical practice.

Materials and methods

PubMed, Embase, Cochrane Database of Systematic Reviews, and Cochrane Central Register of Controlled Trials were searched (last search 12 September 2025). Eligible studies assessed the association between ctDNA kinetics and outcomes in patients with mCRC receiving systemic palliative treatment. Meta-analyses were carried out to evaluate associations with survival outcomes.

Results

A total of 64 studies, including data from >2760 patients with mCRC receiving palliative systemic therapy, met the eligibility criteria. Across studies, unfavorable ctDNA kinetics were consistently associated with poorer outcomes, including shorter overall survival (OS) [pooled hazard ratio (HR) 2.6, 95% confidence interval (CI) 2.2-3.2, n = 1086] and progression-free survival (PFS) (pooled HR 2.7, 95% CI 2.4-3.1, n = 1093).

Conclusion

ctDNA kinetics during palliative systemic therapy have strong prognostic value in mCRC. However, clinical implementation is hampered by methodological heterogeneity, particularly the use of study-specific ctDNA markers and non-validated cut-offs. Standardized and externally validated approaches are needed to support clinical implementation.

Highlights

• •ctDNA kinetics consistently predict survival in mCRC.
• •This is the first focused meta-analysis on on-treatment ctDNA dynamics in mCRC.
• •Unfavorable ctDNA changes link to shorter OS and PFS across 64 studies.
• •Prognostic effect remains robust despite methodological heterogeneity.
• •Standardized ctDNA response criteria are essential for clinical translation.

Introduction

Colorectal cancer (CRC) is the third most common cancer worldwide, with 1.9 million new cases annually, and approximately half of patients develop metastatic disease [metastatic CRC (mCRC)].1, 2, 3 Most patients with mCRC present with disseminated disease lacking curative potential.⁴ Systemic anticancer therapy remains the cornerstone of care in disseminated CRC, aiming to prolong survival and preserve quality of life.

Currently, treatment efficacy is primarily evaluated through clinical assessment and radiological examinations using the Response Evaluation Criteria in Solid Tumors (RECIST).⁵ RECIST relies on measuring changes in target lesion diameters using computed tomography or magnetic resonance imaging. Although RECIST is widely applied in both clinical practice and research, it has several important limitations. Firstly, evaluations are typically carried out every 8-12 weeks, precluding real-time monitoring of treatment response or disease progression. Secondly, substantial changes in tumor volume are required before they are deemed significant on imaging, delaying recognition of treatment efficacy or disease activity not immediately reflected in measurable volume changes. Thirdly, novel treatment modalities have introduced response patterns, such as pseudoprogression during immunotherapy, that challenge conventional imaging-based interpretation. Consistent with these limitations, RECIST-defined response has been shown to be a suboptimal surrogate marker of overall survival (OS).⁶

Given these shortcomings, blood-based biomarkers have been explored as complementary or alternative monitoring strategies. In mCRC, carcinoembryonic antigen (CEA) remains the most commonly used marker, but its limited sensitivity, specificity, and inability to capture dynamic treatment responses in real time restrict its utility.7, 8, 9, 10

These combined limitations underscore an urgent need for improved tools for treatment monitoring. Such tools could enable earlier discontinuation of ineffective therapy and timely initiation of alternative strategies, thereby reducing unnecessary toxicity and ultimately improving survival.

Tumor cells shed small DNA fragments into the bloodstream. Thus, circulating tumor DNA (ctDNA) is accessible via a simple blood sample, and after the isolation of plasma, the amount of ctDNA can be measured by different analytical methods, e.g. digital droplet PCR (ddPCR) and next-generation sequencing. With an elimination half-life of only a few hours, ctDNA kinetics provides a near real-time snapshot of tumor biology.11, 12, 13 ctDNA kinetics seems to be a promising surrogate marker of OS and a valuable alternative for treatment monitoring.¹⁴

Several studies have suggested that ctDNA kinetics hold strong prognostic value during systemic palliative treatment of mCRC, although heterogeneity in study design and methodology has limited clinical implementation to date.¹⁵ Despite growing interest, ctDNA kinetics has not yet been incorporated into routine practice or tested in prospective randomized trials.

This systematic review and meta-analysis aims to synthesize the current evidence on the prognostic value of ctDNA kinetics in mCRC and to clarify its potential role in treatment monitoring and clinical decision making.

Materials and methods

Eligibility criteria

Eligibility criteria include the following: (i) human studies on patients with metastatic colorectal adenocarcinoma, (ii) studies reporting the value of circulating tumor-specific DNA in plasma or serum, and (iii) studies involving patients receiving palliative systemic anticancer treatment where ctDNA kinetics were associated with a survival endpoint and/or a response endpoint.

Information sources and systematic search strategy

The databases PubMed, Embase, and Cochrane Central Register of Controlled Trials were searched as of 12 September 2025 to identify relevant studies. We did not limit our search by language, year of publication, or type of publication. The specific search strategies and strings were previously published.¹⁵ Studies were managed using Covidence.¹⁶

Study evaluation and selection

Two researchers independently assessed all studies at each stage of the review process, including screening, assessment of eligibility, and final inclusion. Initial screening was based on titles and abstracts, followed by a full-text review of potentially relevant studies. Discrepancies were resolved through discussion until consensus was achieved. Key data elements were extracted and tabulated by two reviewers. A standardized data extraction form was used to ensure consistency in data collection. The extracted information included study characteristics (author, publication year, design), patient population size, treatment details (line and type), ctDNA marker and analytical method used, evaluated cut-off values, ctDNA detection rate at baseline, treatment response, and hazard ratios (HRs) for progression-free survival (PFS) and OS with corresponding 95% confidence intervals (CIs) and P values.

Summary and synthesis of results

The prognostic value of ctDNA kinetics was assessed across the included studies. ctDNA kinetics referred to changes observed from baseline to a time point during treatment as reported in the individual studies. Core data from the included studies were summarized in tables, including assessment of the association between ctDNA kinetics and clinical outcomes (Figure 1). For meta-analyses, univariate HRs for PFS and OS were extracted as reported in the individual studies and presented in forest plots. The HRs reflected comparisons based on study-specific cut-offs for dichotomization. Changes in ctDNA during treatment are referred to as favorable or unfavorable ctDNA kinetics as defined in individual studies. Both in tables and forest plots, studies were ordered chronologically by first publication date.

Figure 1.

Schematic overview of ctDNA kinetics. Schematic overview of ctDNA kinetics during systemic palliative treatment and outcome. ctDNA, circulating tumor DNA.

In studies reporting multiple results for the same clinical outcome, for example, using different ctDNA markers or cut-offs, all results were included if they were derived from distinct study cohorts. If multiple results originated from the same study cohort, the analysis with the most complete data (HR, 95% CI, P value, and sample size) was included. In cases where data completeness was equal, the result covering the largest patient population was prioritized. These criteria were also applied when selecting data for inclusion in the meta-analyses.

Statistical analyses

Studies providing HRs with corresponding 95% CIs from univariate survival analyses were eligible for inclusion in meta-analyses estimating the association between ctDNA and clinical outcomes, specifically PFS and OS. When multiple studies evaluated comparable ctDNA cut-off definitions, separate meta-analyses were conducted to enable comparison across cut-off categories.

Clinical and methodological heterogeneity was assessed based on differences in study design, patient populations, blood sampling time points, ctDNA analytical method, ctDNA markers, and cut-off values. Study-specific results were visualized in forest plots.

Statistical heterogeneity was evaluated using visual inspection of the forest plots and quantified using the inconsistency statistic (I²), with values above 50% and 70% considered indicative of moderate and high heterogeneity, respectively.¹⁷ In line with the aim of summarizing the existing literature, pooled estimates were reported unless the observed clinical or methodological heterogeneity exceeded expectations based on the predefined eligibility criteria. To account for the variability, a random-effects model was applied.¹⁸ In parallel, a structured descriptive synthesis summarized findings from all included studies.

All statistical analyses were conducted using Stata software version 19.5 (StataCorp, College Station, TX). A two-sided P value <0.05 was considered statistically significant.

Results

Eligibility assessment

The search identified 5629 unique studies, which were screened according to predefined eligibility criteria. Only studies available as full-text articles were considered, leaving 457 studies for full-text assessment. Reasons for exclusion at this stage included not meeting eligibility criteria, duplicate publications, availability of abstract only, reporting across multiple cancer types or disease stages, or inclusion of fewer than the required number of patients (Figure 2). Ultimately, 64 studies met the eligibility criteria and were included in the review.19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82

Figure 2.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses(PRISMA)diagram. Flow diagram of the study selection process.

In total, the included studies enrolled 8894 patients. Among these, the association between ctDNA kinetics and treatment or survival outcomes were reported for at least 2760 patients receiving palliative systemic therapy for mCRC. Of the eligible studies, 27 provided data suitable for inclusion in at least one meta-analysis (Table 1).

Table 1.

Study overview

Author names	Year	Line of treatment	Study design	Analytical method	ctDNA marker	Second sample	Groups compared	n	ctDNA marker associated with			Ref.
					Genetic mutation unless otherwise specified				Treatment response	PFSa	OSa
Tie et al.	2015	First-line	Prospective biomarker study	Targeted sequencing	Tumor-informed	After cycle 1	<10-fold reduction in ctDNA	52	Yes	No	No	19
							≥10-fold reduction in ctDNA
Wong et al.	2015	Later-line	Prospective open-label non-RCT	BEAMing	KRAS	Day 8 of treatment	—	14	NE	D	NE	20
Hong et al.	2016	Various/unknown	Prospective phase I 3 + 3 non-RCT	ddPCR + targeted sequencing	BRAF	After one dose of treatment	—	12	Yesb	NE	NE	21
Herbst et al.c	2017	First-line	Prospective phase III RCT	qPCR	Methylation of HPP1	After cycle 1	Detection at second sample	300	NE	NE	Yes	22
							Undetectable at BL and at second sample
Vidal et al.	2017	Various/unknown	Retrospective cohort	BEAMing	KRAS, NRAS	At 8-12 weeks of treatment	—	21	Yesd	NE	NE	23
Garlan et al.c	2017	Various/unknown	Prospective biomarker study	ddPCR	Tumor-informed (KRAS/BRAF/TP53) or hypermethylation of WIF1/NPY	Before cycle 2 or 3	Bad ctDNA responsee	108	Yes	Yes	Yes	24
							Good ctDNA responsee
Khan et al.	2018	Later-line	Prospective phase II non-RCT	ddPCR	RAS	At 2 months of treatment	No decrease in FA	21	NE	Yes	No	25
							Decrease in FA
Boeckx et al.	2018	First-line	Prospective biomarker study	ddPCR	Methylation of NPY and tumor-informed	Before cycle 2 and at first radiographic evaluation	—	23	D	NE	NE	26
Hsu et al.	2018	First-line	Retrospective cohort	Targeted sequencing	12 genes	Next available plasma sampling	ctDNA variant frequency reduction of ≤80%f	15	Yes	Yes	NE	27
							ctDNA variant frequency reduction of >80%f
Vandeputte et al.	2018	Later-line	Prospective biomarker study	ddPCR + targeted sequencing	47 genes	Day 14 of first cycle	Absolute increase in mutated copies/mlg	19	NE	Yes	Yes	28
							Absolute decrease in mutated copies/mlg
Corcoran et al.	2018	Various/unknown	Prospective phase I RCT	BEAMing	BRAF	At 4 weeks of treatment	—	71	Yesb	NE	NE	29
Barault et al.	2018	Later-line	Retrospective cohort	BEAMing	Methylation of EYA4, GRIA4, ITGA4, MAP3K14-AS, MSC	Lowest measured ASM during treatmenth	Best ASM change <0i	25	Yes	Yes	NE	30
							Best ASM change ≥0i
Liu et al.	2019	Various/unknown	Prospective observational study	ddPCR	HER2	Serial sampling	—	5	D	NE	NE	31
Lyskjær et al.	2019	First-line	Prospective biomarker study	ddPCR	Tumor-informed	Days 7, 14, 21, and 60 of treatment	Two temporary increases in MAFj	21	NE	Yes	No	32
							Less than two temporary increases in MAFj
Jia et al.	2019	First-line	Prospective biomarker study	Targeted sequencing	50 genes	After cycle 1	ctDNA log2 (C1/C0) >−0.126k	41	Yes	Yes	NE	33
							ctDNA log2 (C1/C0) ≤ −0.126k
Osumi et al.	2019	Second-line	Prospective biomarker study	Targeted sequencing	14 genes	At 8 weeks of treatment	>50% after/before ratio in ctDNA levelsl	26	Yes	Yes	Yes	34
							≤50% after/before ratio in ctDNA levelsl
Amatu et al.	2019	Later-line	Prospective biomarker study	BEAMing	Methylation of EYA4, GRIA4, ITGA4, MAP3K14-AS, MSC	Next available plasma sampling	Circulating methylated DNA increasem	52	NE	Yes	NE	35
							Circulating methylated DNA decreasem
Thomsen et al.	2020	First-line	Prospective biomarker study	ddPCR	Methylation of NPY	After cycle 1	Others	123	No	Yes	Yes	36
							Level = 0 at BL or 0 in 95% CI at second sample
Lueong et al.c	2020	First-line	Prospective phase III RCT	ddPCR	KRAS	15-22 days after start of treatment	Detectable at BL and at second sample	113	NE	Yes	Yes	37
							Detectable at BL and undetectable at second sample
							Undetectable at BL and detectable at second sample	38	NE	NE	Yes
							Undetectable at BL and at second sample
Moser et al.	2020	Various/unknown	Prospective biomarker study	Targeted sequencing	Tumor-informedn	52 h after treatment start	—	11	Yeso	NE	NE	38
Klein-Scory et al.	2020	First-line	Prospective biomarker study	ddPCR	RAS	Various (4-24 weeks of treatment)	—	12	D	NE	NE	39
Wang et al.	2020	Later-line	Retrospective cohort	Targeted sequencing	74 genes	At 4 weeks of treatment	Increase in ctDNA or emergence of new clones	13	D	NE	NE	40
							Decrease in ctDNA
Parikh et al.	2020	Various/unknown	Prospective biomarker study	ddPCR	Tumor-informed	At 4 weeks of treatment	<30% decrease in MAFp	55	Yes	Yes	NE	41
							≥30% decrease in MAFp
Kopetz et al.	2020	Various/unknown	Prospective phase II RCT	Targeted sequencing	BRAFV600E	Next available plasma sampling	—	34	D	NE	NE	42
Sefrioui et al.c	2021	Various/unknown	Prospective biomarker study	dPCR	Tumor-informed (RAS, BRAF)	At 6 weeks of treatment	ctDNA slope >−0.73	NR	Yes	Noq	Yesq	43
							ctDNA slope ≤−0.73
Jia et al.	2021	First-line	Prospective biomarker study	Targeted sequencing	1021 genes	After cycle 4	<0.8-fold reduction in mTBI	16	Yes	Yes	NE	44
							>0.8-fold reduction in mTBI
Stein et al.	2021	First-line	Prospective phase II non-RCT	Targeted sequencing	19 genes	Serial sampling	—	5	D	NE	NE	45
Lim et al.	2021	First-line	Prospective biomarker study	Targeted sequencing	11 genes/16 genes	At first response evaluation	Others	84	NE	Yes	NE	46
							Decrease in VAF to below 1%
Wang et al.	2021	First-line	Prospective cohort study	Targeted sequencing	RAS	Serial sampling	Remained mutated	61	NE	Yes	Yes	47
							Clearance
							Acquisition of mutation	84	NE	Yes	Yes
							Remained wild-type
Nakamura et al.	2021	Various/unknown	Prospective phase II non-RCT	Targeted sequencing	74 genes	At 3 weeks of treatment	Increased ctDNA fraction	28	D	Yes	Yes	48
							Decreased ctDNA fraction
Lin et al.	2021	Various/unknown	Prospective biomarker study	ddPCR	Methylation of MYO1-G	Serial sampling	—	NR	D	NE	NE	49
Hallermayr et al.c	2022 A	First-line	Prospective biomarker study	ddPCR	KRAS, BRAF	Serial sampling	—	5	D	NE	NE	50
Rachiglio et al.	2022	First-line	Prospective phase II RCT	Real-time PCR	KRAS, NRAS, BRAF	At 8 weeks of treatment	—	10	D	NE	NE	51
Kim et al.	2022	First-line	Prospective biomarker study	Targeted sequencing	106 genes	Serial sampling	No ctDNA clearancer	51	NE	Yes	NE	52
							ctDNA clearancer
Watanabe et al.	2022	Second-line	Prospective phase II non-RCT	PCR-rSSO	RAS, BRAF	First radiological evaluation	—	36	D	NE	NE	53
Sunakawa et al.	2022	First-line	Prospective biomarker study	Real-time PCR	KRAS, NRAS	At 8 weeks of treatment	Mutation positive at BL and at second sample	41	No	No	No	54
							Mutation positive at BL and negative at second sample
Palmer et al.	2022	Various/unknown	Phase I trial	Targeted sequencing	Tumor-informed	NR	No ctDNA decrease	7	D	NE	D	55
							ctDNA decrease
Hallermayr et al.c	2022 B	Various/unknown	Prospective biomarker study	WGS	LIFE-CNA	Serial sampling	—	7	D	NE	NE	56
Vidal et al.	2022	First-line	Prospective multicentric study	Targeted sequencing	14 genes	Before cycle 3	Increase in VAF trunk mutations	48	Yes	Yes	No	57
							Decrease in VAF trunk mutations
Raunkilde et al.	2022	First-line	Prospective biomarker study	ddPCR	Methylation of NPY	After cycle 1	Others	59	NE	Yes	No	58
							ctDNA detectable level at BL and undetectable or 95% CI crossing 0 at second sample
Ye et al.	2022	Various/unknown	Prospective biomarker study	Targeted sequencing	BRAF	At first scan	No ctDNA BRAF clearance	22	Yest	Yes	No	59
							ctDNA BRAF clearance
Lee et al.	2023	Later-line	Prospective biomarker study	Targeted sequencing	106 genes	After two cycles of treatment	<50% decrease in sum of/mean of/max VAF	98	Yes	Yes	Yes	60
							≥50% decrease in sum of/mean of/max VAF
Janssens et al.	2023	First-line	Prospective phase II RCT	ddPCR	Methylation of NPY	After cycle 2	Above median slope	36	NE	Yes	NE	61
							Below median slope
Tsai et al.	2023	First-line	Prospective biomarker study	PCR followed by MALDI-TOF mass spectrometry	KRAS, NRAS	Every 3 months	Acquired mutation	108	No	Yes	Yes	62
							No detectable mutation
Callesen et al.	2023	Various/unknown	Prospective biomarker study	ddPCR	KRAS, NRAS, BRAF	Before the third treatment cycle	Any decrease in ctDNA fraction without clearance	12	NE	Yes	No	63
							ctDNA clearance
Sacher et al.	2023	Various/unknown	Prospective phase I study	Targeted sequencing	KRASG12C	Day 15 of cycle 1 and day 1 of cycle 3	—	35	D	NE	NE	64
Urbini et al.	2023	First-line	Retrospective exploratory study	Targeted sequencing	KRAS, NRAS, BRAF, PIK3CA	At first clinical evaluation	≤98% decrease in VAF	34	D	Yes	Yes	65
							>98% decrease in VAF
Doleschal et al.	2023	Various/unknown	Prospective, observational study	ddPCR	Hypermethylation WIF1, NPY/tumor-informed	Before the second treatment cycle	Reduction of <57.45%/57.22% in ctDNA level	NR	D	D	Yes/no	66
							Reduction of >57.45%/57.22% in ctDNA level
Gambaro et al.	2023	Second-line	Phase II exploratory study	Targeted sequencing	14 genes	At radiological response	—	6	D	NE	NE	67
Lavacchi et al.	2023	First-line	Prospective biomarker study	Targeted sequencing	14 genes	Serial sampling	—	9	D	D	NE	68
Yang et al.	2024	Later-line	Prospective phase II study	Targeted sequencing	108 genes	Seventh week after first dose of SCT200	—	80	NE	D	D	69
Ahn et al.	2024	Various/unknown	Phase Ib clinical study	ddPCR	KRAS	Day 1 of cycle 2	Increase in MAF	15	D	NE	NE	70
							Decrease in MAF
Yaeger et al.	2024	Various/unknown	Phase I/II clinical study	ddPCR	KRASG12C	Day 1 of cycle 4	<90% decrease in VAF	15	D	NE	NE	71
							>90% decrease in VAF
Choi et al.c	2024	Various/unknown	Prospective phase I study	WES + targeted sequencing	Personal panel design based on WES	Day 15 of cycle 1	<Median ctDNA reduction	42	D	Yes	NE	72
							>Median ctDNA reduction
Xu et al.	2024	Later-line	Prospective phase II RCT	Targeted sequencing	769 genes	Day 15 of cycle 2	Reduction of <83.44% in ctDNA level	32	D	Yes	Yes	73
							Reduction of >83.44% in ctDNA level
Hamfjord et al.	2024	First-line	Prospective phase II study	ddPCR	NRAS, KRAS, BRAF	First radiological evaluation	Others	40	D	Yes	Yes	74
							ctDNA detectable level at BL and undetectable or 95% CI crossing 0 at second sample
Klein-Scory et al.	2024	First-line	Prospective phase II RCT	ddPCR	BRAF V600	The next available blood sample	Increase in MAF	39	Yes	Yes	Yes	75
							Decrease in MAF
Unseld et al.	2025	Later-line	Translational exploratory study	sWGS + targeted sequencing	77 genes/ichorCNA tumor fraction	Serial sampling	ctDNA trajectories	30	Yes	NE	D	76
Iwai et al.	2025	First-line	Prospective observational study	ddPCR	BRAFV600E	Consecutively	No decrease in VAF	14	NE	NE	Yes	77
							Decrease in VAF
Grancher et al.c	2025	First-line	Prospective biomarker study	ddPCR + targeted sequencing	6 genes	Day 1 of cycle 3 or 4	Reduction of <80% in ctDNA	152	Yes	Yes	Yes	78
							Reduction of ≥80% in ctDNA
Meric-Bernstam et al.	2025	Various/unknown	Phase IIa multiple-basket study	Targeted sequencing	74 genes	Day 1 of cycle 3	<Twofold reduction in mean VAF	25	Yes	Yes	Yes	79
							>Twofold reduction in mean VAF
Taïeb et al.	2025	Second-line	RCT	ddPCR	Hypermethylation WIF1, NPY	At 1 month of treatment	<Median change in ctDNA	74	NE	Yes	Yes	80
							>Median change in ctDNA
Shim et al.	2025	First-line	Prospective observational cohort study	ddPCR + targeted sequencing	46 genes	NR	—	9	D	NE	NE	81
Martelli et al.c	2025	First-line	Prospective multicentric study	Targeted sequencing	14 genes	8 weeks of treatment	Groups based on percentage change (10%) in aggregate VAF	66	NE	Yes	NE	82

n, number of patients with reported association between early dynamics of the ctDNA level during treatment and at least one outcome. For some outcomes, n may be lower.

ASM, average of selected markers; BL, baseline; CI, confidence interval; CNV, copy number variations; ctDNA, circulating tumor DNA; D, descriptively with no statistical test; dPCR, digital PCR; ddPCR, droplet digital PCR; FA, fractional abundance; LIFE-CNA, liquid biopsy fragmentation, epigenetic signature and copy number alteration analysis; MAF, mutant allele fraction; MALDI-TOF, matrix-assisted laser desorption/ionization—time of flight; mTBI, molecular tumor burden index; NE, not evaluated; NR, not reported; OS, overall survival; PCR, polymerase chain reaction; PCR-rSSO, PCR-reverse sequence-specific oligonucleotide; PD, progressive disease; PFS, progression-free survival; PR, partial response; qPCR, quantitative PCR; RCT, randomized controlled trial; Ref, reference; SD, stable disease; sWGS, shallow WGS; VAF, variant allele frequency; WGS, whole-genome sequencing.

^aUnivariate analysis.

^bPatients with partial response were more likely to demonstrate a deeper reduction in ctDNA fraction relative to those with stable disease or progressive disease at the time of first restaging (P < 0.01).

^cLueong et al. 2020 and Herbst et al. 2017: same cohort but different methods. Garlan et al. 2017, Sefrioui et al. 2021, and Grancher et al. 2025: overlapping cohorts but different methods. Hallermayr et al. 2022 A and Hallermayr et al. 2022 B: overlapping cohorts but different methods. Vidal et al. 2023 and Martelli et al. 2025: overlapping cohorts but different methods. Sacher et al. 2023 and Choi et al. 2024: same cohort but different methods.

^dMAF percentage of change was significantly lower in patients with progressive disease compared with patients with partial response or stable disease at the time of first restaging (P = 0.027).

^eA combined marker integrating baseline ctDNA level (below/above 0.1 ng/ml) and drop in ctDNA during the first 4 weeks (</>80%).

^fctDNA frequency reduction was defined as the difference between the frequency of the ctDNA variant with the highest variant frequency at treatment initiation and the frequency of the ctDNA variant with the highest variant frequency at follow-up.

^gOn average 3 mutations (range 1-4) were selected per patient based on the highest VAFs.

^hASM: calculated based exclusively on the loci which displayed positive methylation in the baseline.

ⁱASM change: ASM at a longitudinal time point was subtracted from the ASM at baseline, and the best change (lowest) over the course of treatment was selected for correlative assessment.

^jctDNA increase relative to previous sample. One patient-specific mutation is chosen for each patient.

^kFor each patient, the mutation of the maximal frequency in the pretreatment plasma ctDNA sample was selected as the candidate target for analysis.

^lctDNA level was defined as the highest allele frequency of the detected mutant alleles at each time point in each patient when two or more mutations were detected.

^mCirculating methylated DNA level is estimated as an average of the level of all positive markers.

ⁿIn the tumor of one patient, no mutation was identified. Instead, a mutation identified in plasma by targeted sequencing was tracked.

^oCompared with patients with progressive disease patients with stable disease/partial response/complete response ctDNA MAFs remained at decreased levels at the time of last blood collection, i.e. T9 (between T1 and T5: P = 0.03906; between T1 and T9: P = 0.01563).

^pFor patients with multiple assessable mutations the percent change in MAF of up to three mutations ctDNA was averaged.

^qOnly reported for multivariate analysis.

^rctDNA clearance was defined as the disappearance of all the mutations or CNVs above the cut-off value that were detectable at their baseline samples.

^sTrunk mutation is defined as the most frequently detected mutation within a specific plasma sample.

^tA higher proportion of patients with PR or SD showed a VAF decrease rate >50% compared with patients with PD (P = 0.046).

Study design and population

Across the 64 included studies, the most common design was a prospective biomarker study (n = 27). In the prospective studies, blood samples were collected prospectively but analyzed retrospectively (Table 1).

Most patients received life-prolonging systemic therapy comprising 5-fluorouracil, oxaliplatin, and/or irinotecan, with or without an epidermal growth factor receptor (EGFR) inhibitor or a vascular endothelial growth factor pathway inhibitor. A few patients received temozolomide monotherapy. Smaller subsets received targeted or experimental therapies, including anti-EGFR monotherapy; trifluridine/tipiracil or regorafenib; BRAF or MEK inhibitors; tyrosine kinase inhibitors; human epidermal growth factor receptor 2-directed therapy; KRAS G12C inhibitors; polo-like kinase 1 inhibitors; immune checkpoint inhibitors; and a neoantigen vaccine, either alone or in combination with chemotherapy.

Patients received treatment in first line (n = 29), second line (n = 4), or later line (n = 10). The treatment line was various or unknown in a substantial number of studies (n = 21) (Table 1).

Analytical methods and ctDNA markers

For the detection and quantification of ctDNA, most studies used either PCR-based analytical methods (n = 31), targeted sequencing (n = 25), or a combination of those (n = 4). Other approaches comprised whole-genome sequencing (WGS) (n = 1) and PCR followed by matrix-assisted laser desorption/ionization—time of flight spectrometry (n = 1) and a combination of targeted sequencing and shallow WGS or whole-exome sequencing (n = 2) (Table 1).

Across studies, the most frequently applied ctDNA marker was a tumor-agnostic genetic alteration (n = 47). These ranged from single-gene assays (e.g. RAS, BRAF, PIK3CA) to extended panels covering up to 1021 genes, with baseline detection rates varying widely from 0% to 100%. Eight studies employed a tumor-agnostic epigenetic marker, most commonly methylation of the NPY gene, with baseline detection rates of 72%-95%. A tumor-informed approach was less frequently used (n = 6), with baseline detection rates ranging from 85% to 100%. Finally, three studies combined a tumor-informed strategy with a tumor-agnostic epigenetic marker, reporting detection rates of 76%-93% (Table 1).

Sampling time points and cut-offs for ctDNA kinetics

According to the eligibility criteria, all included studies evaluated changes in ctDNA during systemic treatment by comparing a baseline blood sample with a subsequent sample collected during therapy (the ‘second sample’, Table 1). The timing of this second sample varied substantially, ranging from 52 h to 24 weeks after treatment initiation. Four studies compared baseline with ‘the next available plasma sample’ without specifying the exact time point. Ten studies carried out serial sampling. Barault et al.³⁰ compared baseline with the lowest measured average of selected markers and Gambaro et al.⁶⁷ with the sample obtained at radiological response, while two studies did not report the timing of the second sample (Table 1).

In the analysis of the association between ctDNA kinetics and clinical outcomes, the included studies applied different strategies to dichotomize the study populations (Table 1). Twenty studies classified patients according to a defined change in ctDNA levels (e.g. a 10-fold or 30% decrease) (Table 1). Cut-offs were determined by receiver operating characteristic analyses (n = 6),^{19,27,33,44,66,73} median value (n = 3),^61,72,80 previously reported cut-offs (n = 2),^71,82 recursive partitioning analysis (n = 1),⁶⁵ best separation of survival curves (n = 1),⁴³ log-rank test statistics (n = 1),²⁴ a 90% specificity threshold (n = 1),⁴¹ or by convention for biomarker response (n = 1)⁶⁰; in four studies, no justification for the chosen cut-off was reported.^34,46,78,79 For 13 studies, the defined change was data dependent.^{19,24,27,33,41,43,44,61,65,66,72,73,80}

Ten studies compared an increase versus a decrease in ctDNA levels, measured either as copies per milliliter plasma or as the fraction of total cell-free DNA (cfDNA). Seven studies dichotomized patients according to changes in ctDNA detectability. In three studies, dichotomization was based on quantitative changes in ctDNA incorporating 95% CIs obtained from ddPCR. Another three studies applied alternative approaches to dichotomization. The remaining 21 studies did not dichotomize patients but instead reported ctDNA kinetics descriptively (Table 1).

Overall survival

Among 28 studies with a statistical assessment of the association between favorable ctDNA kinetics (e.g. decline or clearance) and longer OS, or unfavorable kinetics (e.g. increase, persistence, or decline without reaching a defined threshold) and shorter OS, 19 studies found the association statistically significant, while 8 did not. In one study, analyses had diverging results depending on the ctDNA marker evaluated. Three studies presented an association descriptively with no statistical test (Table 1).

Nineteen studies had sufficient data for inclusion in a meta-analysis. Minimal heterogeneity was found among studies (I² = 0.0%). Unfavorable ctDNA kinetics were associated with shorter OS (pooled HR 2.6, 95% CI 2.2-3.2, n = 1086) (Figure 3A).

Figure 3.

Forestplots of the association between ctDNA kinetics and survival. Forest plots of the association between unfavorable ctDNA kinetics and OS (A) and PFS (B): all under the random-effects model. Studies are ordered according to publication date. CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio; n, number of patients included in the analysis; OS, overall survival; PFS, progression-free survival.

Progression-free survival

In 34 studies, ctDNA kinetics were statistically significantly associated with PFS, and 3 studies did not find a statistically significant association. Four studies presented an association descriptively with no statistical test (Table 1).

Twenty-six studies had sufficient data for inclusion in a meta-analysis. Minimal heterogeneity was found among studies (I² = 0.0%). Unfavorable ctDNA kinetics were associated with shorter PFS (pooled HR 2.7, 95% CI 2.4-3.1, n = 1093) (Figure 3B).

Treatment response

The relationship between ctDNA kinetics and treatment response was evaluated using statistical analysis in 23 studies, of which 20 found a statistically significant association, while 3 did not. There were no common effect estimates across studies, preventing comparative analyses. An additional 24 studies presented the relationship descriptively (Table 1).

Subgroup meta-analyses of comparable ctDNA cut-offs

Only two cut-off strategies were consistently applied across multiple cohorts and therefore eligible for separate pooling.

Six studies eligible for inclusion in a meta-analysis compared patients with an increase versus a decrease in ctDNA levels following treatment initiation (Table 1). An increase in ctDNA was associated with shorter OS (pooled HR 2.8, 95% CI 1.6-4.9, n = 155; I² = 0.0%) and PFS (pooled HR 3.0, 95% CI 1.9-4.7, n = 205; I² = 22.9%) (Figure 4A and B).

Figure 4.

Forest plots of the association between increase in ctDNA levels and survival. Forest plots of the association between increase in ctDNA levels and OS (A) and PFS (B): all under the random-effects model. Studies are ordered according to publication date. CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio; n, number of patients included in the analysis; OS, overall survival; PFS, progression-free survival.

Five studies eligible for inclusion in a meta-analysis evaluated ctDNA clearance (Table 1). Lack of clearance was likewise associated with worse OS (pooled HR 2.5, 95% CI 1.5-4.3, n = 115; I² = 0.0%) and PFS (pooled HR 2.4, 95% CI 1.6-3.7, n = 174; I² = 0.0%) (Figure 5A and B).

Figure 5.

Forest plots of the association between absence of ctDNA clearance and survival. Forest plots of the association between absence of ctDNA clearance and OS (A) and PFS (B): all under the random-effects model. Studies are ordered according to publication date. CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio; n, number of patients included in the analysis; OS, overall survival; PFS, progression-free survival.

Due to heterogeneity in definitions and cut-off selection, no additional meta-analyses of specific thresholds were feasible.

Discussion

This systematic review and meta-analysis, comprising 8894 patients across 64 studies, found a consistent association between on-treatment favorable ctDNA kinetics and improved OS and PFS. Despite the inter-study heterogeneity, the direction of the associations was remarkably consistent, supporting the potential of ctDNA kinetics as a prognostic biomarker in mCRC.

To the best of our knowledge, this systematic review represents the only comprehensive synthesis specifically addressing the association between ctDNA kinetics and treatment response and survival in patients with mCRC. While several reviews have broadly explored the clinical utility of ctDNA as a biomarker, none have focused exclusively on the dynamic changes in ctDNA levels during systemic therapy and their prognostic implications. Thus, the present review provides the most current and focused evidence base in this field, thereby reinforcing the unique contribution of this body of research to the ongoing evaluation of ctDNA as a tool for treatment monitoring and outcome prediction.

It should be acknowledged that a recent systematic review and meta-analysis by Holz et al. examined the association between cfDNA kinetics and outcomes during systemic palliative treatment of mCRC.⁸³ While this work provides valuable insights into the potential role of circulating nucleic acids as biomarkers, it did not distinguish between total cfDNA and tumor-derived ctDNA. This distinction is clinically relevant, as ctDNA more directly reflects tumor dynamics and therefore may offer greater specificity for treatment monitoring and prognostication. The present review specifically addresses this aspect, thereby complementing and extending the existing literature.

Pretreatment ctDNA detection rates varied substantially (0%-100%), influenced by patient selection, ctDNA marker choice, and analytical method. Notably, prior studies have demonstrated lower ctDNA levels in patients with metastases limited to the lungs, lymph nodes, or peritoneum compared with liver metastases,⁸⁴ emphasizing the biological variability of ctDNA shedding and its implications for both study interpretation and clinical application.

Methodological considerations are also important. The majority of studies employed PCR-based assays targeting RAS/BRAF mutations, whereas more recent approaches include targeted sequencing of broader gene panels and methylation-specific assays, reflecting ongoing technological advances in ctDNA analysis. With the highest detection rates found across different methods and targets, PCR-based tumor-agnostic approaches may be particularly well suited for longitudinal monitoring in the palliative context, owing to their feasibility, rapid turnaround times, and ability to capture ctDNA kinetics.

In this review, studies with low baseline detection rates still demonstrated associations with clinical outcomes, suggesting that ctDNA may provide prognostic insight even in settings with limited detectability. The considerable variability in detection rates underscores the need to explore the prognostic relevance of absolute ctDNA levels beyond dichotomized classifications. Furthermore, direct comparisons of ctDNA and CEA, both at baseline and during treatment, are warranted to clarify the relative value of these biomarkers. Such analyses remain hampered, however, by the lack of standardized thresholds for interpreting CEA dynamics in the treatment setting.

Across the included studies, definitions of ctDNA response varied considerably, with terms such as ‘clearance’, ‘decline’, and ‘undetectable’ used inconsistently. The timing of the ‘second sample’ likewise ranged from hours to several months after treatment initiation, contributing to methodological heterogeneity. Despite this variability, the consistency of reported associations across studies is noteworthy. Prospective trials employing predefined early or longitudinal sampling schedules and intention-to-treat analyses are needed to validate these observations.

Cut-offs for defining ctDNA kinetics varied considerably, were frequently derived from study-specific data, and generally lacked external validation. This heterogeneity in cut-off selection and underlying ctDNA definitions, together with a wide variation in sampling time points, raises important concerns regarding biological interpretability and clinical comparability across studies. To enable quantitative synthesis, these heterogeneous ctDNA definitions were necessarily collapsed into a binary classification of ‘favorable’ versus ‘unfavorable’ kinetics, a simplification that may oversimplify complex and biologically diverse ctDNA trajectories. None the less, forest plots demonstrated consistent effect sizes across studies, suggesting that the associations may be robust despite methodological heterogeneity. A major limitation, however, is the absence of standardized and validated criteria for ctDNA response, which continues to hinder clinical translation. Recently proposed frameworks, such as ctDNA-RECIST,⁸⁵ provide a promising foundation for harmonization and warrant prospective validation in future trials.

This review points toward an association between ctDNA kinetics and radiological response, but the paucity of available data limits firm conclusions. Elucidating the interplay between ctDNA changes and imaging could advance our understanding of metastatic disease biology. Future prospective studies integrating standardized ctDNA monitoring with imaging are essential to assess concordance and define the added value of ctDNA in treatment evaluation.

The studies included in this review showed considerable heterogeneity in design, methodology, and cohort size. Most were exploratory in nature, relying on retrospective analyses of plasma samples and conducted at single institutions. As a result, the generalizability of the findings to broader, unselected mCRC populations may be limited. As all meta-analyses were based on unadjusted HRs, and multivariable adjustment was not feasible due to inconsistent and heterogeneous reporting of adjusted analyses across studies, the magnitude of the observed associations may be overestimated.

While the majority of patients received standard first-line systemic therapy—facilitating cross-study comparisons—this limits the ability to assess the relevance of ctDNA kinetics in later treatment lines. In addition, ctDNA kinetics were frequently reported only for selected patient subsets, thereby reducing statistical power and likely contributing to variability in outcomes. Such selective reporting also raises the possibility of immortal time bias, which may influence estimates of the prognostic effect of ctDNA kinetics.

Despite the large number of studies evaluating the prognostic value of ctDNA kinetics, few applied comparable cut-offs, which further complicates cross-study comparisons and limits the ability to determine the most clinically relevant threshold. Among the few consistent strategies, both comparisons of ctDNA increase versus decrease and absence versus presence of ctDNA clearance demonstrated similar prognostic strength, with pooled effect sizes comparable to those observed in the overall meta-analyses. These findings suggest that, despite methodological variability, the prognostic signal of unfavorable ctDNA kinetics remains robust. Nevertheless, harmonization of ctDNA response definitions and external validation of specific cut-offs are essential to establish the true clinical utility of ctDNA kinetics.

Meta-analyses revealed no measurable statistical heterogeneity (I² = 0.0%-22.9%), suggesting robustness of the observed association across diverse clinical contexts and study designs. However, the exclusion of a proportion of studies from quantitative synthesis introduces a potential for selection bias, and the effect size should therefore be interpreted with care. Importantly, the narrative synthesis complements these findings by contextualizing clinical and methodological variability and linking it to survival outcomes.

However, the selection bias inherent in retrospective studies may systematically exclude patients with poor performance status and prognosis, thereby limiting the generalizability to this group. Thus, findings may not be representative of all clinical scenarios.¹⁷ Nevertheless, the observed associations within the included cohorts provide important insight into the direction and consistency of the effect.

As anticipated in a body of evidence largely composed of retrospective and exploratory studies, substantial heterogeneity was observed, resulting in a highly diverse dataset. Current guidelines recommend cautious interpretation of pooled estimates and emphasize the importance of complementing them with descriptive or narrative synthesis to maintain clarity and transparency. In this review, the primary objective was to provide a comprehensive overview of the existing literature rather than to directly inform clinical practice.⁸⁶ Accordingly, pooled estimates were reported with appropriate caution, as they still offer valuable insights to guide the direction of future research despite the heterogeneity of the underlying studies.

This review highlights the prognostic relevance of ctDNA kinetics in mCRC, supporting its potential as a minimally invasive, early-response biomarker. Several studies demonstrate a statistically significant association with OS, in some cases within two cycles of treatment, indicating that ctDNA may offer earlier insight into treatment benefit than imaging.

These results strengthen the evidence for the clinical validity of ctDNA kinetics as a prognostic biomarker; however, they do not establish the predictive value or clinical utility for treatment adaptation.⁸⁷ Well-designed prospective studies are essential to clarify whether treatment decisions based on ctDNA kinetics can offer superior outcomes relative to current practice. Accordingly, ctDNA kinetics should currently be regarded as a prognostic marker rather than a tool for guiding treatment decisions, and its use for treatment adaptation cannot yet be recommended outside the context of prospective, controlled clinical trials.

Future research should prioritize the prospective validation of predefined thresholds, such as those outlined in ctDNA-RECIST,⁸⁵ alongside the determination of optimal sampling time points and clinically meaningful cut-offs. These efforts will be essential to establish the clinical utility of ctDNA, including its potential to improve survival outcomes and quality of life through early treatment adaptation.

Taken together, the magnitude and consistency of the observed associations support ctDNA kinetics as a robust prognostic biomarker in metastatic colorectal cancer, demonstrating clinical validity but not clinical utility. Given the predominance of retrospective study designs, non-standardized methodologies, and heterogeneous definitions of ctDNA response, these findings should be interpreted with appropriate caution.

Importantly, prospective efforts to move ctDNA kinetics from prognostic relevance toward clinical utility are already under way. Ongoing studies within the ctDNA-RECIST program include prospective biomarker studies with predefined assays, standardized early sampling time points, and explicit ctDNA-based response criteria, as well as a randomized trial evaluating ctDNA-guided treatment decisions against standard RECIST-based management (ClinicalTrials.gov Identifier: NCT06562348). Together, these studies directly address key limitations identified in the current literature and represent critical next steps toward establishing standardized ctDNA response criteria suitable for clinical implementation.

Conclusions

This systematic review and meta-analysis demonstrates that ctDNA kinetics during palliative systemic therapy in mCRC have strong and consistent prognostic value. Across studies, unfavorable ctDNA changes were associated with markedly shorter OS and PFS, underscoring the potential of ctDNA kinetics as a dynamic biomarker of treatment response. However, methodological heterogeneity—particularly in sampling schedules, analytical approaches, and cut-off definitions—remains a major barrier to clinical implementation. Prospective studies employing standardized and validated ctDNA response criteria, such as ctDNA-RECIST, are warranted to confirm these findings and to define the clinical utility of ctDNA-guided treatment adaptation.