Integrating C-reactive protein flare and early MRI dynamics for enhanced prediction of immunotherapy response

Abstract

Background

The C-reactive protein (CRP) flare response, an indicator of early immune activation, has emerged as a promising and cost-effective biomarker for predicting response to immune checkpoint inhibitors (ICIs) across various tumor types. This study evaluates the utility of CRP dynamics as a tumor-agnostic biomarker and integrates systemic inflammatory markers with advanced multiparametric MRI metrics to uncover the biological mechanisms underlying the CRP flare phenomenon and its relationship with treatment response.

Methods

Patients were stratified into three groups based on CRP kinetics: (1) flare-responders, characterized by an initial doubling of baseline CRP followed by a decrease below baseline; (2) CRP responders, defined as patients with no flare increase but a CRP reduction of at least 30% below baseline and (3) CRP non-responders. Multiparametric MRI was performed at baseline, early (1–3 weeks), and intermediate (6–8 weeks) time points to assess tumor size and microstructural features, including cell density and vascularization. Clinical benefit and survival outcomes, including progression-free survival (PFS) and overall survival (OS), were analyzed using Kaplan-Meier curves and log-rank tests. Cox regression analyses were performed to identify independent predictors of clinical outcomes, while intergroup differences in MRI metrics were assessed using Wilcoxon rank-sum and Kruskal-Wallis tests.

Results

Among the 121 evaluable patients with solid tumors enrolled in the PREDICT trial, CRP flare-responders demonstrated significantly longer PFS (5.6 months) and OS (12.1 months) compared with responders (PFS: 3.4 months, OS: 8.0 months) and non-responders (PFS: 3.2 months, OS: 6.7 months; p=0.01 and p<0.01, respectively). Additionally, clinical benefit was achieved in 50% of flare-responders, compared with 13% of responders (p=0.05) and 23% of non-responders (p<0.01). Tumor growth was interrupted early after treatment initiation in CRP flare-responders, whereas non-responders exhibited marked increases in tumor size. In the pilot subset of 33 patients with MRI data, diffusion MRI revealed stable or increased apparent diffusion coefficient values in CRP flare-responders, indicative of reduced tumor cellularity just after 1–3 weeks of treatment.

Conclusions

This study highlights the potential of combining early CRP dynamics with non-invasive imaging metrics to identify ICI responders as early as 2 weeks after treatment initiation. By integrating systemic inflammatory biomarkers with MRI-derived insights into tumor size and microstructural changes, these findings optimize therapeutic strategies and advance understanding of immunotherapy-driven tumor dynamics.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Early C-reactive protein (CRP) dynamics identify immunotherapy responders as early as 2 weeks after treatment initiation, independent of tumor type.

WHAT THIS STUDY ADDS

• Integrating systemic inflammatory markers with advanced MRI metrics provides a non-invasive window into tumor biology, revealing early tumor growth arrest and reduced cellularity in CRP flare-responders.

• CRP flare-responders achieve superior clinical outcomes, including longer progression-free and overall survival, compared with other CRP response groups.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This proof-of-concept study supports the combined use of blood-based biomarkers and functional imaging for refined patient stratification and early response assessment.

Introduction

The advent of immune checkpoint inhibitors (ICIs) has revolutionized cancer treatment, marking a significant advancement in the therapeutic landscape.1,3 However, overall response rates to ICIs remain low, with only a subset of patients deriving long-term benefit. This variability underscores the pressing need for the development of novel predictive biomarkers to identify early therapy failure and avoid unnecessary side effects and costs.^{4 5}

A widely used biomarker in immunotherapy is PD-L1 score, which has been Food and Drug Administration-approved for certain solid tumors.⁶ Other tumor-agnostic biomarkers such as microsatellite instability and tumor mutational burden (TMB) have also been linked to an increased likelihood of response to ICIs. However, the predictive power of all these biomarkers remains limited, and there is an urgent need for new and more reliable biomarkers of ICI response.7,9

In the search for new biomarkers, several studies have shown that serum C-reactive protein (CRP) levels, an indicator of systemic inflammation, correlate with ICI response.¹⁰ Specifically, early CRP kinetics after ICI treatment start can identify a subgroup of patients known as ‘flare-responders,’ characterized by a sudden early spike in CRP levels (at least doubling of CRP within the first month) followed by a subsequent decrease below baseline levels and excellent ICI outcomes. This CRP flare kinetics might reflect the induction of an antitumor immune response, as this group has been linked to better ICI response in several solid tumor types: urothelial carcinoma,^{11 12} renal cell carcinoma,13,15 non-small cell lung cancer,^{16 17} hepatocellular carcinoma,¹⁸ head and neck cancer,¹⁹ and melanoma.²⁰ Additionally, a pan-cancer analysis has supported the tumor-agnostic potential of CRP as a predictive biomarker in patients treated with ICIs.²¹ Notably, in the phase III OAK trial of non-small cell lung cancer,^{17 22} CRP flare response was linked to favorable outcomes only in the atezolizumab arm and not in the chemotherapy arm, reinforcing that this phenomenon is specific to immunotherapy.

This CRP flare response underscores the potential of early tumor-associated inflammatory changes as predictors of long-term ICI treatment outcomes. However, a clearer understanding is needed of how the CRP flare response correlates to the tumor and its microenvironment at both macroscopic and microscopic levels.

Multiparametric MRI (mpMRI) provides comprehensive insights into both the macro and microenvironments of tumors. Structural high-resolution MRI scans enable the assessment of macroscopic features of tumors, as their volume and its change on treatment.23,25 Moreover, the quantitative scans of an mpMRI protocol, for example, those based on diffusion and perfusion imaging,²⁶ enable the estimation of biological properties in each pixel of a tumorous mass, as for example blood perfusion and permeability²⁶ or cell density and cellularity.27,29 While at present, the use of quantitative MRI in immunotherapy remains experimental and confined to the research settings, advanced mpMRI metrics offer promise as powerful non-invasive assays of the tumor microenvironment.^{23 29} Specifically, by assessing early changes in biologically specific biomarkers, quantitative mpMRI may help identify new characteristics of the CRP flare-responders group that might contribute to their favorable response to treatment. Furthermore, these insights could lead to novel ways of classifying patients, enhancing personalized treatment strategies.

In this study, we aimed to deepen the understanding of immunotherapy response by integrating early CRP dynamics with advanced MRI-derived metrics. First, we validated the CRP flare response as a tumor-agnostic biomarker in a diverse cohort of patients with solid tumors. Next, we used mpMRI to characterize how CRP-based response groups differ in both macroscopic tumor evolution and microenvironmental changes during treatment. By linking CRP kinetics with imaging-based tumor dynamics, this study seeks to refine response biomarkers, enabling earlier and more precise identification of immunotherapy responders.

Methods

Study design

All participants in this study were included in the prospective PREDICT study, which involved patients with advanced solid tumors treated with ICIs. The PREDICT study aimed to evaluate the potential of imaging and circulating biomarkers in predicting treatment response in patients with advanced solid tumors undergoing cancer immunotherapies. These therapies included ICIs, such as anti-PD-1, anti-PD-L1, anti-CTLA-4, and LAG3, as well as emerging immunotherapy combinations involving bispecific antibodies and T-cell engagers, administered either as monotherapies or in combination with other immunotherapies.

The PREDICT trial was approved by the Vall d’Hebron University Hospital Ethics Committee (PR (AG)29/2020), and all patients provided informed written consent for the acquisition and analysis of MRI scans, blood samples, and tumor biopsies.

Blood sample analysis

Blood samples were collected and analyzed to determine baseline and longitudinal concentrations of CRP. A CRP concentration exceeding 5 mg/L was defined as elevated, which is an established standard for CRP studies.¹⁶ Additional biomarkers were assessed, including lactate dehydrogenase (LDH), albumin, and the neutrophil-to-lymphocyte ratio (NLR). These biomarkers were also considered as elevated when above thresholds extracted from literature. For LDH, a cut-off value of 250 U/L was used.³⁰ For albumin, the threshold was set at 3.5 g/dL.³¹ For NLR, although there is no standard threshold, a value of four was used following recent literature.³²

Radiological variables

All patients included in the analysis underwent standard CT imaging at the start of treatment and subsequent time points. Of the 121 evaluable patients, baseline CT scans (thorax-abdomen-pelvis) were available for 100 individuals and were manually segmented by a radiologist to extract key radiological variables at baseline. The variables included in the analysis are total tumor volume (TTV) and presence of liver metastasis. TTV was defined as the sum of the volumes of all RECIST 1.1-measurable lesions (primary and metastatic) identified on CT scans; non-measurable disease was excluded. Lesion volumes were computed as voxel volume × voxel count for each manually delineated lesion. The presence of liver metastasis was recorded as a binary variable. Given the pan-cancer nature of the cohort and the absence of a validated tumor-agnostic cut-off (see online supplemental table 3 for mean TTV values in our cohort per tumor type), we used the cohort median TTV (61.3 mL) as a study-internal threshold to define high burden, ensuring balanced groups across heterogeneous tumor types; TTV was also summarized as a continuous variable.

Treatment response assessment

Treatment response was evaluated using two primary metrics: overall survival (OS) and progression-free survival (PFS). OS was defined as the duration from the start of immunotherapy to the patient’s death from any cause or the last recorded follow-up. PFS was defined as the time from the initiation of treatment to either disease progression or death, whichever occurred first. Disease progression was determined based on RECIST 1.1,³³ assessed through standard CT imaging. In cases of clinical deterioration where imaging findings were inconclusive, progression was further evaluated by the principal investigator. The best treatment response was evaluated according to RECIST: progressive disease (PD), stable disease (SD), partial response (PR), and complete response (CR). Objective response rate (ORR) was defined as the proportion of patients who achieved PR or CR by RECIST 1.1. Clinical benefit was defined as achieving CR, PR, or SD for 4 months or more, whereas patients with PD or SD and a PFS of less than 4 months were considered to have no benefit.

Definition of early CRP kinetics

Patients were categorized into three groups based on their early CRP kinetics, following previously established criteria.^{13 16} The group of CRP flare-responders included patients whose CRP levels exhibited at least a two-fold increase within the first 30 days of treatment, followed by a decrease below baseline levels by week 12. CRP responders comprised patients who achieved a reduction in CRP levels of 30% or more below baseline within 12 weeks, without experiencing the initial two-fold increase characteristic of flare-responders. The remaining patients, who did not meet the criteria for either of these two groups, were classified as CRP non-responders. This categorization was used to explore the potential relationship between the CRP flare phenomenon and treatment responses, as well as its implications for clinical outcomes.

mpMRI analysis

mpMRI was performed at multiple time points to assess patients and extract quantitative metrics from various MRI sequences. Scans were obtained at baseline, within 28 days prior to the initiation of ICI treatment, and at and early time point, 1–3 weeks after treatment began. Additionally, MRI scans were conducted alongside standard CT scans for response evaluation 6–8 weeks after the start of treatment. The MRI scans were acquired using two different scanners (a 1.5T Siemens Avanto and a 3T GE SIGNA Pioneer systems) and encompassed several imaging modalities:

• Anatomical MRI: tumors were manually delineated on the T2-weighted anatomical series (5 mm slice thickness) using 3D Slicer by a board-certified radiologist with >15 years’ experience in oncologic imaging (RP-L), with cross-reference to all MRI contrasts to ensure precise boundary definition. Tumor volume was computed as the volume (voxel volume × number of voxels) of such manually outlined tumors. All RECIST 1.1-measurable cancerous lesions visible within the MRI field of view were included for the tumor volume calculation, including both primary and metastatic tumors.

• Diffusion MRI (dMRI): metrics such as apparent diffusion coefficient (ADC), apparent diffusion kurtosis (K), vascular ADC (ADCv), and vascular fraction (Fv) were derived from diffusion-weighted imaging, offering insights into tissue microstructure, water diffusion, and tumor heterogeneity. ADC and K are, respectively, negatively and positively correlated to cellularity, while K also captures diffusion heterogeneity.³⁴ The metrics have been used with success in various cancer applications, for example, to monitor response in metastatic melanoma immunotherapy²⁶ or in neoadjuvant chemotherapy of breast cancer.³⁵ Conversely, Fv and ADCv from segmented intra-voxel incoherent motion imaging³⁶ are surrogate markers of vascularization and capillary flow.³⁴ The segmented IVIM fitting used here has been recently demonstrated for tumor microperfusion quantification in vivo.³⁷

• Relaxation MRI: T1, T2, and T2* relaxation times were acquired to provide quantitative insights into tissue microenvironment characteristics, including water content, cellular and iron density, and potential alterations induced by treatment. Relaxation metrics have been previously used in cancer imaging, for example, for diagnosis and grading of prostate cancer.^{38 39}

• Perfusion MRI: permeability metrics from dynamic contrast-enhanced (DCE) imaging, including Ktrans and extracellular volume fraction, were calculated to assess tumor vascularization, tissue perfusion, and vessel permeability. DCE metrics have already shown promise in several cancer treatments, such as targeted melanoma therapy or immunotherapy.²⁶

Quantitative MRI-derived maps were generated through mathematical fitting of various signal models following standard image postprocessing, producing voxel-level measurements for all patients. The obtained metrics were harmonized with ComBat to bring them to a common scale accounting for the differences in the scanners. These harmonized voxel-based values were then aggregated at the patient level by averaging all voxels within each tumor lesion, yielding a comprehensive metric representative of each patient’s tumor characteristics. Tumors, segmented on a high-resolution T2-weighted scan, were registered to all other MRI contrasts (diffusion/perfusion) through non-linear registration. The outlines of the registered masks were manually edited by the radiologist to correct for misregistration, ensuring the highest accuracy and precision of quantitative metric computation. Acquisition parameters, including slice thickness, were kept constant across time points to limit through-plane variability.

Details on the MRI acquisition and image processing are included in the online supplemental methods.

Statistical analysis

Statistical analyses were conducted using R (V.4.4.1). Intergroup comparisons of variables were performed using the χ² test for categorical variables and the Kruskal-Wallis test for continuous variables. Kaplan-Meier survival curves were constructed for both PFS and OS, stratified by CRP response groups, and differences were tested using log-rank tests.

To evaluate the prognostic value of the CRP-defined groups compared with baseline characteristics, univariable and multivariable Cox regression analyses were conducted for both PFS and OS. Wald tests were used to assess the significance of individual variables within the regression models. Univariate and multivariate Cox proportional hazards regression analyses were conducted to identify independent predictors of PFS and OS across the study cohort. Variables included in the analysis were selected based on their clinical relevance and statistical significance in univariate analyses (p<0.05). HRs with 95% CIs were calculated.

For MRI metrics, pairwise comparisons between CRP-defined groups were conducted at the patient level using the Wilcoxon rank-sum test. All statistical tests were two-sided, with a p value of <0.05 considered statistically significant. Besides, Kruskal-Wallis Rank-Sum tests on absolute value changes were also performed for general intergroup comparison.

Results

Study population

A total of 139 patients were enrolled in the PREDICT trial (figure 1a), with blood samples collected at baseline and multiple time points throughout treatment for CRP analysis in 121 patients. Applying the classification criteria for CRP kinetics by Fukuda et al (figure 1b), 25% (n=30) of patients met the criteria for CRP flare-responders. Additionally, 26% (n=31) of patients were classified as CRP responders, and the remaining 49% (n=60) were classified as CRP non-responders (figure 1a,c).

Figure 1. Study population of C-reactive protein (CRP) kinetics and group definition. (a) CONSORT diagram depicting the selection process of the study population for the defined CRP kinetics groups, shown for both the complete cohort and the subset with multiparametric MRI (mp-MRI) scans. (b) Analytical kinetics of the CRP groups. (c) Kinetics of the CRP groups defined in our pan-cancer cohort. BL, baseline; CONSORT, Consolidated Standards of Reporting Trials.

Patient characteristics of the complete cohort are summarized in table 1, along with those of the three different CRP groups. All patients had an Eastern Cooperative Oncology Group performance status of 0 or 1.⁴⁰ There were no significant differences in age (p=0.6) or sex distribution (p=0.9) among the groups, indicating a comparable demographic composition. Primary tumor location distributions were also not significantly different (p=0.3), suggesting a homogeneous tumor type representation across the groups. TTV was higher than the median for 67% of responders and 52% of non-responders, while only for 34% of flare-responders, indicating that the flare-responder group has lower TTV (p=0.007). A more visual representation of these distributions is available in online supplemental figure 1, where the histograms show that CRP flare-responders have the distribution centered below the median value. Baseline levels of LDH and albumin did not show significant differences between groups. However, some baseline biomarker levels revealed significant differences in CRP concentrations across the groups. Most CRP flare-responders (83%) and non-responders (92%) had baseline CRP levels below 5 mg/L, whereas the majority of CRP responders (55%) had baseline CRP levels above 5 mg/L (p<0.001). NLR was also enriched in flare-responders, being high for 70% of them compared with 52% and 38% of responders and non-responders, respectively (p=0.02).

Table 1. Comparison of baseline clinical parameters between CRP flare-responders, CRP responders, and CRP non-responders in the complete study cohort.

Characteristic	Overall N=121	Flare-responder N=30	Responder N=31	Non-responder N=60	P value
Age (years)					0.6
Mean, IQR	62 (55–71)	65 (59–68)	62 (55–71)	60 (53–68)
Range	18–86	18–86	18–80	18–83
Sex					0.9
Male	70 (58%)	17 (57%)	17 (60%)	36 (60%)
Female	51 (42%)	13 (43%)	14 (40%)	24 (40%)
Primary tumor location					0.3
Skin	17 (14%)	1 (3%)	4 (13%)	12 (20%)
Colon	17 (14%)	4 (13%)	4 (13%)	9 (15%)
Gastric	14 (12%)	2 (7%)	2 (7%)	10 (17%)
Head and neck	8 (7%)	2 (7%)	2 (7%)	4 (7%)
Lung	7 (6%)	3 (10%)	3 (10%)	1 (2%)
Pancreatic	8 (7%)	2 (7%)	0 (0%)	6 (9%)
Endometrial	6 (5%)	2 (7%)	2 (7%)	2 (3%)
Rectal	6 (5%)	2 (7%)	2 (7%)	2 (3%)
Ureteral	6 (5%)	3 (10%)	0 (0%)	3 (5%)
Renal	5 (4%)	1 (3%)	1 (3%)	3 (5%)
Breast	4 (3%)	1 (3%)	2 (7%)	1 (2%)
Prostate	3 (2%)	1 (3%)	2 (7%)	0 (0%)
Hepatocarcinoma	4 (3%)	0 (0%)	2 (7%)	2 (3%)
Ovarian	6 (5%)	1 (3%)	3 (10%)	2 (3%)
Cholangiocarcinoma	2 (2%)	1 (3%)	0 (0%)	1 (2%)
Mesothelioma	2 (2%)	2 (7%)	0 (0%)	0 (0%)
Thyroid	2 (2%)	1 (3%)	0 (0%)	1 (2%)
Sacral chordoma	1 (1%)	0 (0%)	1 (3%)	0 (0%)
Cervix	1 (1%)	0 (0%)	0 (0%)	1 (2%)
Ampuloma	1 (1%)	0 (0%)	1 (3%)	0 (0%)
Epithelial sarcoma	1 (1%)	1 (3%)	0 (0%)	0 (0%)
Total tumor volumes (N=100)					0.007
>61.3 mL	50 (50%)	10 (34%)	14 (67%)	26 (52%)
≤61.3 mL	50 (50%)	19 (66%)	7 (33%)	24 (48%)
Presence of liver lesions (N=100)					0.1
Yes	60 (60%)	13 (45%)	12 (57%)	35 (70%)
No	40 (40%)	16 (55%)	9 (43%)	15 (30%)
Baseline CRP					<0.001
>5 mg/L	27 (22%)	5 (17%)	17 (55%)	5 (8%)
≤5 mg/L	94 (78%)	25 (83%)	14 (45%)	55 (92%)
Baseline LDH					0.2
>250 UI/L	58 (48%)	11 (37%)	19 (61%)	28 (47%)
≤250 UI/L	63 (52%)	19 (63%)	12 (39%)	32 (53%)
Baseline NLR					0.08
>4	40 (33%)	13 (43%)	13 (42%)	14 (23%)
≤4	81 (67%)	17 (57%)	18 (58%)	46 (77%)
Baseline albumin					0.4
>3.5 g/dL	111 (92%)	27 (90%)	27 (87%)	57 (95%)
≤3.5 g/dL	10 (8%)	3 (10%)	4 (13%)	3 (5%)
Overall survival (OS, months)					0.002
Mean (IQR)	8.4 (3.7–10.4)	12.1 (5.6–15.7)	8.0 (3.8–8.0)	6.7 (3.4–8.3)
Range	1.4–34.5	2.3–34.5	2.1–32.2	1.4–23.2
Progression-free survival (PFS, months)					0.01
Mean (IQR)	3.9 (1.4–4.3)	5.6 (1.8–7.5)	3.4 (1.6–3.9)	3.2 (1.2–2.8)
Range	0.2–22.1	0.7–21.0	0.2–16.6	0.4–22.1

Intergroup comparisons of variables were performed using χ2 test for categorical variables and Kruskal-Wallis rank-sum test for continuous variables and p values < 0.05 were considered as significant.

CRP, C-reactive protein; LDH, lactate dehydrogenase; NLR, neutrophil-to-lymphocyte ratio.

Early CRP kinetics predict ICI response and survival in patients with solid tumors

CRP dynamics serve as a promising indicator of response to ICIs and of patient clinical outcomes. Notably, CRP flare-responders demonstrated significantly improved PFS and OS compared with responders and non-responders (p=0.028 and p=0.047, respectively) (figure 2a,b). The median OS for flare-responders was 14.4 months, markedly longer than the 6.5 months for responders and 6.7 months for non-responders. Similarly, flare-responders had a median PFS of 3.8 months, substantially exceeding the 1.7 months observed for both responders and non-responders.

Figure 2. C-reactive protein (CRP) dynamics: correlation with survival and treatment response. (a, b) Kaplan-Meier survival curves showing overall survival (OS) and progression-free survival (PFS) stratified by CRP groups. P values are calculated using the log-rank test. Dotted lines indicate the median OS and PFS for each CRP group and the colored areas show the 95% CIs of the curve of each CRP group. (c, d, e) Distributions of RECIST responses, ORR percentage, and binarized response outcomes across CRP groups, with p values calculated using the χ² test. CR, complete response; FR, flare-responder; NR, non-responder; PD, progression disease; PR, partial response; R, responder; SD, stable disease.

These findings were further supported by the analysis of the best treatment response based on RECIST 1.1, including ORR and clinical benefit in each group. While the majority of CRP non-responders (62%) and responders (52%) experienced PD as their best response by RECIST, only 37% of flare-responders fell into this category. In contrast, most CRP flare-responders achieved SD (47%), PR (13%), or CR (3%) (figure 2c). When grouping patients by ORR, the proportion of responders was modest in all groups: 17% in CRP flare-responders, 10% in CRP responders, and 8% in CRP non-responders. Although flare-responders showed a numerically higher ORR, the differences were not statistically significant (figure 2d). By comparison, while only 18% of CRP non-responders and 23% of responders achieved clinical benefit, up to 50% of CRP flare-responders experienced clinical benefit (p=0.004 and p=0.049, respectively) (figure 2e).

To further investigate the potential of the CRP flare phenomenon, we examined how early this dynamic response could be detected. On average, the CRP flare response became identifiable at approximately 4.8 weeks, which corresponds to the time required for CRP levels in flare-responders to decrease below baseline. However, individual variation was observed, with an IQR of 2.0–5.6 weeks.

These results highlight the utility of CRP kinetics as a promising biomarker for the early prediction of patient outcomes and a potential tool for optimizing immunotherapy strategies.

Predictive value of baseline hematological markers for ICI response

Univariate analysis using Cox regression of variables, including hematological baseline characteristics, identified several with prognostic significance (table 2). CRP flare response emerged as a protective prognostic factor for both PFS (HR=0.61, p=0.03) and OS (HR=0.46, p=0.01). Among other hematological markers, LDH levels at baseline demonstrated strong prognostic value. High baseline LDH levels were identified as a detrimental factor for both survival and response (HR=1.85, p<0.005 for PFS; HR=1.67, p=0.04 for OS). Interestingly, high TTV at baseline was also associated with worse outcomes (HR=1.61, p=0.02 for PFS; HR=1.86, p=0.01 for OS). The remaining variables, including CRP, NLR, and baseline albumin levels, were not significantly associated with survival outcomes.

Table 2. Univariable and multivariable Cox regression analysis regarding overall survival (OS) and progression-free survival (PFS) with baseline characteristics and the C-reactive protein (CRP) defined groups.

Characteristic	PFS				OS
	N	HR	95% CI	P value	N	HR	95% CI	P value
Univariate
CRP flare-responder	121	0.61	0.39 to 0.94	0.03	121	0.46	0.25 to 0.83	0.01
CRP responder	121	1.07	0.69 to 1.67	0.75	121	1.19	0.70 to 2.02	0.52
Age >64 years	121	0.87	0.59 to 1.27	0.47	121	1.10	0.68 to 1.77	0.70
Sex (male)	121	0.94	0.64 to 1.38	0.76	121	1.15	0.71 to 1.86	0.58
Tumor volume BL >61.3 mL	100	1.61	1.09 to 2.39	0.02	100	1.86	1.13 to 3.07	0.01
Presence of liver metastatic lesions	100	1.46	1.01 to 2.15	0.06	100	1.42	0.88 to 2.29	0.15
CRP BL >5 mg/L	121	1.22	0.77 to 1.92	0.40	121	1.44	0.84 to 2.47	0.18
LDH BL >250 UI/L	121	1.65	1.12 to 2.43	0.01	121	1.52	0.94 to 2.45	0.09
NLR BL >4	121	1.14	0.76 to 1.71	0.53	121	1.37	0.84 to 2.24	0.21
Albumin BL >3.5 g/dL	121	1.06	0.54 to 2.11	0.86	121	0.71	0.34 to 1.49	0.36
Multivariate
CRP flare responder	121	0.75	0.47 to 1.22	0.25	121	0.54	0.29 to 1.01	0.05
Tumor volume BL >61.3 mL	100*	1.38	0.91 to 2.09	0.13	100*	1.67	1.00 to 2.77	0.05
LDH BL >250 UI/L	1.38	1.38	0.90 to 2.10	0.14	121	1.16	0.70 to 1.93	0.56

Wald tests are use to test statistical significance and p values < 0.05 were considered as significant.

^*For the multivariate Cox model featuring the significant features of the univariate models, missing values in the baseline (BL) tumor volume variable were imputed using an iterative imputer available form the python package sklearn.

BL, baseline; LDH, lactate dehydrogenase; NLR, neutrophil-to-lymphocyte ratio.

In the multivariate Cox regression analysis, which included all significant variables from the univariate analysis, no variable was significant for PFS while being a CRP flare-responder (HR=0.54, p=0.05) and having a TTV above the median (HR=1.67, p=0.05) showed to be protective and detrimental, respectively, for OS. In contrast, LDH lost its statistical significance in the multivariate analysis.

While baseline CRP values did not show significance when considering high and low levels according to the established threshold of 5 mg/L, we show in online supplemental figure 2 that when selecting as threshold the median value of our population (1.6 mg/L), CRP baseline concentrations stratify high and low risk groups with statistical significance for both OS (p=0.0007) and PFS (p=0.0017). Besides, it shows a strong protective effect for PFS according to univariate Cox regression considering the variable as continuous (PFS: HR=1.07, p=0.028, and OS: HR=1.03, p=0.1).

MRI characterization of CRP groups

To further characterize the CRP flare response phenomenon, we performed an analysis of different mpMRI metrics and their relations to the defined CRP groups. A total of 55 patients underwent mpMRI at baseline, within 28 days before starting therapy. From these, 41 of them underwent early follow-up performed between 1 week and 3 weeks after starting treatment (referred to as MRI week 1). After matching the patients with CRP measurements available, the final MRI cohort consisted of 33 patients (figure 1a). The resulting mpMRI data provided comprehensive insights into tumor characteristics, integrating metrics from anatomical, diffusion, and perfusion imaging modalities.

Patient characteristics of the MRI cohort are provided in online supplemental table 1. Similarly to the complete cohort, age, sex, and primary tumor location showed no differences between groups. Baseline CRP levels followed the same trend, being higher in responders compared with flare and non-responders (p=0.002). In this cohort, NLR showed no significant differences, while LDH was significantly lower in flare-responders than in the other groups (p=0.006).

MRI-derived metrics enabled us to non-invasively better characterize the CRP flare phenomenon by providing detailed insights into tumor size, microstructure (such as cell density), and vascularization. The correlations between these imaging biomarkers and CRP-defined groups elucidate the biological underpinnings of CRP kinetics and their relationship to treatment outcomes (figure 3a).

Figure 3. MRI metrics and changes from baseline to week 1 by C-reactive protein (CRP) response. (a) Schematic of the multiparametric MRI acquisition pipeline, including a summary of the evaluated metrics. (b, c) Percentage changes in tumor volumes (b) and ADC values (c) from baseline to week 1 for each CRP group. P values were calculated using the Wilcoxon rank-sum test. ADC, apparent diffusion coefficient; FR, CRP flare-responders; F_vascular, vascular fraction; Ktrans, volume transfer constant; NR, CRP non-responders; R, CRP responders; V_{extracellular}, extracellular volume fraction.

CRP flare-responders showed a slower tumor growth rate, with a smaller increase in tumor volume from baseline to week 1 (mean increase: 2.9 mL, +2%), compared with CRP responders and non-responders (mean increases: 12.08 mL and 22.847 mL, +26% and +18%, respectively). Although the results did not reach statistical significance for the pairwise comparisons (CRP flare vs responders: p=0.09, CRP flare vs non-responders: p=0.09, CRP responders vs non-responders: p=0.84), there is a trend toward a halt in tumor growth in flare-responders (figure 3b, table 3, online supplemental table 2).

Table 3. Changes in MRI-derived parameters between C-reactive protein (CRP) flare-responders, CRP responders, and CRP non-responders in the study cohort, as obtained comparing the baseline and first follow-up (follow-up value - baseline value).

MRI metric	Overall (n=33)	Flare-responder (n=8)	Responder (n=8)	Non-responder (n=17)	P value
ΔTumor volume (mL)	15.401 (−0.130 to 20.16)	2.9 (−0.864 to 4.420)	12.08 (1.857 to 38.008)	22.847 (1.625 to 31.467)	0.18
ΔADC (um2/ms)	0.050 (−0.076 to 0.171)	0.166 (−0.003 to 0.211)	0.122 (0.050 to 0.182)	−0.039 (−0.108 to –0.029)	0.01
ΔADCvascular (um2/ms)	−1.891 (−12.490 to 14.276)	1.612 (−9.078 to 15.639)	−12.594 (−28,890 to 11.797)	1.497 (−8.091 to 13.253)	0.43
ΔKurtosis	−0.076 (−0.219 to 0.017)	−0.071 (−0.187, 0.019)	−0.177 (−0.285 to –0.062)	−0.031 (−0.145 to 0.019)	0.18
ΔT1 (ms)	15.849 (−109.646 to 105.094)	98.428 (−79.974 to 145.858)	−74.830 (−180.545 to 61.415)	19.660 (−65.618 to 105.396)	0.36
ΔT2 (ms)	5.670 (−44.223 to 32.622)	−32.078 (-61.242 to –22.372)	31.795 (10.624 to 41.083)	11.139 (−86.569 to 32.622)	0.09
ΔT2* (ms)	−0.424 (−5.596 to 4.470)	1.345 (−4.819 to 5.213)	−1.750 (−7.951 to 7.380)	−0.632 (−3.591 to 3.490)	0.88
ΔKtrans (1 /min)	−0.128 (−0.209 to 0.015)	−0.116 (−0.208 to 0.099)	−0.205 (−0.307 to 0.015)	−0.095 (−0.167 to 0.010)	0.60
ΔFvascular	−0.006 (−0.026 to 0.036)	−0.004 (−0.022 to 0.031)	0.022 (−0.011 to 0.056)	−0.021 (−0.026 to 0.024)	0.57
ΔVextracellular	−0.005 (−0.080 to 0.076)	0.012 (−0.085 to 0.143)	−0.028 (−0.017 to 0.071)	−0.002 (−0.029 to 0.064)	0.86

Values correspond to the mean of the distributions and the IQR of the change. Statistical significance is tested with the Kruskal Wallis rank-sum test and p values < 0.05 were considered as significant.

ADC, apparent diffusion coefficient; F_vascular, vascular fraction; Ktrans, volume transfer constant; V_{extracellular}, extracellular volume fraction.

Interestingly, dMRI provided further insights into the tumor microstructure. ADC revealed a notable decrease in ADC in non-responders after 1 week of treatment (−2.67%), potentially suggestive of increased tumor cellularity and disease progression. In contrast, CRP flare-responders (13.4%, p=0.01 with respect to non-responders) and responders (9.0%, p=0.01 with respect to non-responders) showed increased ADC values, possibly related to a reduction in cellularity and a favorable response to therapy (figure 3c).

Perfusion MRI metrics assessing vascular properties, including vascular permeability and vascular fraction, showed no significant changes at 1 week or 6 weeks across the CRP-defined groups (online supplemental figure 3). This indicates that vascular alterations are not a dominant feature of the early response associated with CRP dynamics (figure 4).

Figure 4. Representative MRI maps illustrating different metrics. One patient from each C-Reactive Protreactive protein group is shown, with anatomical images and apparent diffusion coefficient (ADC) and apparent excess kurtosis (AKC) maps zoomed in on the lesions. A table summarizing the values and percentage changes for each metric is displayed for each patient.

Figure 4 shows examples of tumors and mpMRI maps in a representative flare responder, responder, and non-responder patient. The differences in tumor outline are due to actual changes in tumor volume/shape during treatment. Trends in mpMRI metrics are also apparent, in line with the quantitative changes seen above (eg, ADC increase in the flare-responder and ADC decrease in the non-responder).

Discussion

The CRP flare response has emerged as a promising biomarker in the context of immunotherapy, reflecting early immune activation mechanisms. This phenomenon has been associated with favorable outcomes across multiple tumor types and provides an accessible, cost-effective tool for predicting immunotherapy response.11,20 This study validates the utility of CRP dynamics as a tumor-agnostic biomarker for categorizing response to ICIs and predicting clinical outcomes in a pan-cancer cohort. Moreover, by integrating early CRP changes with advanced mpMRI metrics, we provide novel insights into the biological underpinnings of the CRP flare phenomenon and its correlation with tumor response.

Based on previously defined criteria, patients were stratified into three groups according to CRP kinetics: flare-responders, responders, and non-responders. Flare-responders, characterized by an initial spike followed by a decline in CRP levels, demonstrated significantly longer PFS (3.8 months) and OS (14.4 months) compared with responders and non-responders (PFS: 1.7 months, OS: 6.6 months). These findings were further supported by improved ORR based on RECIST 1.1 and clinical benefit, with 50% of flare-responders achieving clinical benefit compared with 18% and 23% for non-responders and responders, respectively.

MRI was included in this study as a non-invasive tool capable of assessing not only the macroscopic characteristics of the tumor, such as volume, but also relevant aspects of the tumor microenvironment. Advanced MRI metrics, such as the ADC, provide insights into tumor cell density, reflecting cellularity changes during therapy.41,43 Perfusion MRI, on the other hand, can evaluate vascular architecture and permeability, shedding light on the vascular components of the tumor microenvironment.^{23 44} These imaging approaches enabled a comprehensive assessment of tumor dynamics.

Notably, in our cohort, flare-responders exhibited tumor growth interruption as early as 1–2 weeks after treatment initiation, in contrast to non-responders who showed markedly increased tumor volume. Given known measurement variability, small fluctuations in tumor burden were interpreted with caution; in line with Quantitative Imaging Biomarkers Alliance repeatability guidelines, changes within typical repeatability limits were not considered biologically significant.⁴⁵ These findings are consistent with prior studies indicating that early changes in CRP levels and tumor volume are predictive of treatment outcomes in solid tumors.

Additionally, dMRI metrics, such as the ADC, revealed that progressors experienced decreases in ADC values, potentially indicative of increased cellularity. In contrast, flare-responders maintained stable or increased ADC values, a finding consistent with reduced cell density. These cellular changes underscore the biological basis of the CRP flare response and its association with effective tumor control. The observed trends are consistent with prior studies demonstrating that early treatment response to immune checkpoint blockade can be characterized by reductions in tumor cellularity, as quantified by increases in ADC values. In particular, mpMRI studies in metastatic melanoma have shown that an increase in ADC as early as 3 weeks after treatment initiation is indicative of tumor cell death or reduced cellular density, distinguishing responders from non-responders.²⁴ Computer simulations based on computational models of the dMRI signal,^{46 47} included as supplementary material (online supplemental figure 4), corroborate this interpretation. The figure shows that lymphocyte infiltration can potentially lead to some reductions of ADC, due to the considerably smaller size of lymphocytes compared with cancer cells.2748,50 The simulation also shows that such small reductions in ADC are likely to go undetected in practice, owing to the cytotoxic action of the lymphocytes themselves. Their activity would inevitably lead to cancer cell density reduction, which would become apparent as a clear increase in ADC, thus blurring out the reduction in ADC associated with the lymphocyte presence. This further supports the role of dMRI metrics in complementing CRP kinetics as a non-invasive approach for assessing immunotherapy response dynamics. Interestingly, perfusion MRI metrics showed no significant changes in vascular density or permeability across CRP-defined groups, suggesting that vascular alterations may not serve as a primary driver of the CRP flare response in this context.

Despite its strengths, including the novel integration of CRP dynamics with MRI metrics for non-invasive assessment of macroscopic and microscopic tumor characteristics, this study has some limitations. The relatively small sample size, particularly within the MRI subgroup, limits the generalizability of the findings. Moreover, as a single-center study, the results may not fully reflect broader patient populations or the diversity of tumor types. Nevertheless, the inclusion of early non-invasive imaging data offers valuable proof-of-concept evidence. This work advances our understanding of CRP dynamics and highlights the potential of integrating systemic inflammatory markers with imaging biomarkers to improve patient stratification and enable early identification of immunotherapy responses. We acknowledge the modest size of the MRI subgroup, which limits generalizability. These findings should be viewed as proof of concept and will benefit from confirmation in larger, multicenter cohorts, including patients with earlier-stage disease and scans acquired across multiple vendors and scanners. Ongoing work is planned to validate these results in substantially bigger samples.

Finally, although this study does not address CRP dynamics at the time of disease progression, largely due to the limited number of MRI scans available at later stages, its main contribution lies in delineating the early biological patterns that emerge during immunotherapy and linking these with their radiological correlates. Capturing these early signals is critical for guiding clinical decisions and improving early response assessment. At the same time, we acknowledge that extending these observations to longer treatment courses will be highly informative, offering new insights into sustained responses and mechanisms of resistance. Future research should therefore aim to integrate both early and longitudinal perspectives to provide a more complete picture of immunotherapy dynamics in solid tumors.

In conclusion, this study demonstrates the potential of CRP dynamics as a robust biomarker for predicting immunotherapy response and highlights the value of MRI-derived metrics in elucidating the biological mechanisms underlying the CRP flare phenomenon. This integrative approach indicates new opportunities for improved patient stratification, personalized treatment strategies, and enhanced monitoring of immunotherapy outcomes.

Data availability statement

No data are available.