Longitudinal investigation of albumin-to-globulin ratio for human cancers demonstrates benefit in postoperative serial remeasurement

Abstract

Previous studies on serum albumin-to-globulin ratio (AGR) in human cancers were limited to its preoperative level, with postoperative serial AGR remeasurements ignored. In this study, 2844 CRC patients, 2267 NSCLC patients and 507 HCC patients who underwent curative resection were included. Postoperative AGR was a prognostic factor independent to preoperative AGR, performing a L shaped relation with OS. The 5-year OS rates for the persistently normal, normalized, lowered and persistently low perioperative AGR groups were 84.0%, 80.7%, 78.5% and 70.2%. Three longitudinal AGR trajectory groups were identified within 12 months after surgery. Compared with the normal-stable group, the adjusted HRs on OS for the rising-decreasing and decreasing-rising groups were 1.38 (95% CI: 1.13–1.68, P = 0.001) and 2.78 (95% CI: 2.30–3.36, P < 0.001). Similar results were observed for RFS. In conclusion, a routine follow-up of AGR in the postoperative surveillance will improve prognosis risk stratification of cancer patients.

Introduction

Serum albumin-to-globulin ratio (AGR), a widely available nutritional and inflammatory marker¹, has been recently recognized to be a prognostic factor in several malignancies², such as colorectal cancer (CRC)³, non-small cell lung cancer (NSCLC)⁴ and hepatocellular carcinoma (HCC)⁵. The efficacy of AGR for prognosis prediction of cancer patients has been systematically reviewed in previous studies, and low pretreatment AGR was reported to be associated with decreased long-term survival in many types of solid tumors^6,7. However, for non-metastatic cancer patients undergoing radical resection, previous studies only focused on preoperative AGR^3–5, neglecting the dynamic changes of AGR after surgery. So far, perioperative and longitudinal AGR dynamic changes have not been characterized, and the prognostic significance of following-up AGR in postoperative surveillance of human cancers is still uncertain.

Postoperative recurrence or metastasis accounts for the poor prognosis of cancer patients, and follow-up is crucial to identify the survival heterogeneity among cancer patients⁸. Previously, the postoperative change profiles of some inflammation and nutrition markers have been demonstrated to be independent prognostic indicators of various cancers^9,10. For example, perioperative C-reactive protein to albumin ratio profile was reported to be associated with stratified risks of death¹¹, and the remeasurement of systemic inflammatory response levels appeared to be beneficial for prognosis prediction¹². Postoperative decrease of serum albumin was indicative of serious complications and poor prognosis of patients after tumor resection¹³, and the combination of preoperative and postoperative prognostic nutritional index enabled a more precise prognosis prediction¹⁴. Similarly, AGR dynamic changes during the period of postoperative follow-up, reflecting the transformation of patients’ nutritional and immune status after interventions including surgery, drugs and diets, may also provide independent prognostic information.

Based on a pan-cancer cohort with CRC, NSCLC and HCC patients, this study aims to identify the perioperative and longitudinal dynamic changes of AGR, and elucidate the prognostic value of following up postoperative AGR in cancer prognosis surveillance.

Methods

Patients

This study collected pan-cancer data from retrospective cohorts of CRC, NSCLC, and HCC patients who underwent radical resection without neoadjuvant treatment at Yunnan Cancer Hospital. The study design is shown in Fig. 1. The retrospective cohort study was approved by the Yunnan Cancer Hospital Ethics Committee (KY2019141). The requirement for informed consent was waived by the board, owing to the study’s retrospective nature. This study was carried out according to the research protocol in compliance with the Declaration of Helsinki. All the patient data in the survey were anonymized.

Fig. 1.

Workflow.

Serum marker determination

Serum AGR was evaluated as albumin divided by total protein minus albumin value in serum. Serum total protein and albumin were measured by colorimetric analysis using Roche cobas8000 c702 analyzer (Roche Diagnostics, Germany). Preoperative AGR was defined as the measurement closest to the time of surgery within 1 month before surgery. Postoperative AGR was defined as the last measurement within three months after surgery and before starting postoperative adjuvant chemotherapy. Due to individual differences in health status and adherence to follow-up, AGR was repeatedly measured with different intervals and times for participants during 12 months after surgery.

Surveillance protocol and outcome

Patients were followed up regularly according to guidelines^15–18. In this study, the follow-up of the CRC, NSCLC and HCC cohort ended on Sep 30, 2022, Oct 31, 2022, and May 31, 2023, respectively. The primary endpoint was overall survival (OS), defined as the time from surgery to death or the last follow-up. The secondary endpoint was recurrence-free survival (RFS), defined as the time from surgery to recurrence, death, and the last follow-up. The recurrence included the local recurrence and distant metastases confirmed via histology of biopsy samples or positive imaging. Patients who were alive without recurrence at the last follow-up were defined to be censored.

Covariates

Covariates included age, sex, histology (CRC vs. NSCLC, and HCC), T stage, N stage, and adjuvant chemotherapy. For CRC, histologic subtype, primary site, tumor differentiation, microsatellite instability status, lymphovascular invasion, and perineural invasion were further collected. For NSCLC, histologic subtype, epidermal growth factor receptor (EGFR) gene status, anaplastic lymphoma kinase (ALK) gene status, surgical procedure, and tumor differentiation were further collected. For HCC, Child-Pugh score, Eastern Cooperative Oncology Group (ECOG) score, liver resection percentage, Edmondson-Steiner grade, and microvascular invasion were further collected.

Statistical analysis

Nonlinear relationships of continuous preoperative and postoperative AGR with the survival outcomes were investigated using restricted cubic spline (RCS) curves based on Cox models¹⁹. The number of knots of RCS determines the shape of the fitted curve. To select the optimal model, multiple RCS models with 3–10 knots were established, and evaluated according to the Akaike information criterion (AIC). The reference value (HR = 1) was set at the cut-off value of AGR (1.5), which is the lower limit of the normal AGR range.

Based on AGR levels, patients were grouped into the normal (≥1.5) and low (<1.5) preoperative or postoperative AGR groups. Considering preoperative and postoperative AGR levels jointly, patients were assigned into four perioperative groups: patients with normal preoperative and postoperative AGR (persistently normal group), patients with low preoperative but normal postoperative AGR (normalized group), patients with normal preoperative but low postoperative AGR (lowered group), and patients with low preoperative and postoperative AGR (persistently low group).

A joint latent class mixed model (JLCMM)²⁰ was used to identify different longitudinal AGR trajectory patterns from preoperative to 12 months after surgery. In JLCMM, each latent class of subjects is characterized by a class-specific linear mixed model estimating the trajectory of AGR over time and a class-specific proportional hazard model for overall survival²¹. To select the optimal shapes and number of classes, we tested linear, quadratic and cubic trajectory parameters with the number of classes changing from 1 to 5, and chose the one that gave the best combination of these indicators: (1) Bayesian information criterion decreased at least 20; (2) high mean posterior probabilities (>0.7); (3) high mean posterior class membership probabilities (>65%).

Baseline characteristics across AGR groups were compared using Kruskal–Wallis tests for continuous variables shown as median [inter-quartile range (IQR)], and χ² test for categorical variables shown as number (percentage). OS and RFS of AGR groups were estimated using the Kaplan-Meier method, and the overall difference of survival curves was determined using the log-rank test. Associations between the AGR groups and cancer outcomes were evaluated in three Cox proportional hazard models: model 1 adjusted for no covariate; model 2 adjusted for age and sex; model 3 further adjusted for histology, pathological T stage, pathological N stage, and adjuvant chemotherapy. Additionally, preoperative AGR was adjusted in model 2 and model 3 of postoperative and longitudinal AGR group. The type of recurrence (local vs. systemic vs. both) in relation to AGR status was analyzed, and multinomial logistic regression models were performed to evaluate associations between AGR groups and recurrence type.

Subgroup analysis

To study potential heterogeneity, repeated Cox analyses were performed in subgroups stratified by histology, age, sex, pathological T stage, pathological N stage, and adjuvant chemotherapy, with test for the existence of interaction. Subgroup analyses were performed in each tumor using a multivariable model with adjustment for tumor-specific features.

Results

Cohort characteristics

2844 CRC patients undergoing surgery between February 2010 and February 2019, 2267 NSCLC patients undergoing surgery between November 2013 and December 2018, and 507 HCC patients undergoing surgery between October 2013 and September 2020 were enrolled in the CRC, NSCLC and HCC cohort, respectively. A total of 5618 cancer patients (3310 (58.9%) male; median [interquartile range (IQR)] age, 59 [51, 66] years) were included in this pooled study, with a median [IQR] follow-up period of 61.5 [47.2, 82.1] months. During follow-up, 1258 (22.4%) patients occurred death and 1141 (20.3%) patients occurred recurrence. Baseline characteristics of the three cohorts of CRC, NSCLC and HCC patients are listed in Supplementary Table 1. Tumor-specific prognostic clinicopathological features of each cancer were shown in Supplementary Tables 2–4. The 5-year OS rates of the CRC, NSCLC, and HCC cohorts were 82.9% (95% confidence interval (CI): 81.5–84.4%), 79.5% (95% CI: 77.8–81.3%), and 62.9% (95% CI: 58.5–67.6%), and the corresponding 3-year RFS rates were 77.5% (95% CI: 75.8–78.9%), 77.6% (95% CI: 75.8–79.3%), and 51.5% (95% CI: 46.7–56.3%), respectively (Supplementary Fig. 1).

The study flow chart is shown in Supplementary Fig. 2. After the exclusion of 317 patients without preoperative AGR, the remaining 5301 patients were incorporated in the analysis of preoperative AGR, of which 115 patients without postoperative AGR were further excluded for the analysis of postoperative AGR and perioperative AGR changing patterns. In the analysis of longitudinal AGR, 4726 patients with four or more AGR measurements during preoperative to 12 months after surgery were included.

The associations of preoperative and postoperative AGR with prognosis

L-shaped relation between the preoperative AGR and OS was observed, which reduced substantially within the lower range of preoperative AGR and remained relatively flat thereafter (Fig. 2a). Preoperative AGR was normal in 3793 patients and low in 1508 patients, with median [IQR] preoperative AGR levels of 1.8 [1.6, 2.0] and 1.4 [1.2, 1.4]. Patients with low preoperative AGR had a greater age and pathological stage than those with normal preoperative AGR (Supplementary Table 5). The 5-year OS rates were 82.4% (95% CI: 81.2–83.7%) and 72.9% (95% CI: 70.5–75.3%) in patients with normal and low preoperative AGR, with a log-rank P < 0.001. (Fig. 2b).

Fig. 2. Association of preoperative and postoperative AGR with overall survival in pan-cancer patients.

a Association between preoperative AGR and overall survival on a continuous scale; (b) Kaplan-Meier curves of Normal (5-year OS rate: 82.4%) and Low (5-year OS rate: 72.9%) preoperative AGR groups for overall survival; (c) Association between postoperative AGR and overall survival on a continuous scale; (d) Kaplan–Meier curves of Normal (5-year OS rate: 83.6%) and Low (5-year OS rate: 74.5%) postoperative AGR groups for overall survival. In (a) and (c), Solid yellow lines are unadjusted hazard ratios, with dashed yellow lines showing 95% confidence intervals derived from restricted cubic spline regressions. Reference lines for no association are indicated by the solid red lines at a hazard ratio (HR) of 1.0. Blue histograms show the fraction of the population with different levels of AGR. Arrows indicate the A/G level with HR of 1.0. AGR albumin globulin ratio, CI confidence interval, E number of events, HR hazard ratio, N number of patients, CRC colorectal cancer, NSCLC non-small cell lung cancer, HCC hepatocellular carcinoma, OS overall survival.

The association of postoperative AGR with death risk was also L-shaped, showing a stronger trend of reduction than preoperative AGR (Fig. 2c). Descriptive statistics for the 3055 patients with normal postoperative AGR and 2131 patients with low postoperative AGR are summarized in Supplementary Table 6. The median [IQR] postoperative AGR levels of the normal and low postoperative AGR groups were 1.8 [1.6, 1.9] and 1.3 [1.2, 1.4], respectively. The 5-year OS rate for patients with low postoperative AGR was 74.5% (95% CI: 72.6–76.4%), significantly lower than the 83.6% (95% CI: 82.2–85.0%) in patients with normal postoperative AGR (Fig. 2d).

Compared with the normal preoperative AGR group, the low preoperative AGR group had a higher risk of death (HR: 1.59, 95% CI: 1.41–1.79, P < 0.001) in the unadjusted model, and the estimated risk attenuated slightly but remained significant after adjusting demographic and clinicopathological covariates in model 3 (HR: 1.32, 95% CI: 1.17–1.50, P < 0.001). Similar to preoperative AGR, the unadjusted HRs on OS in patients with low vs. normal postoperative AGR was 1.69 (95% CI 1.51–1.90, P < 0.001). The higher risk of the low postoperative AGR group was still evident in multivariable adjustment including preoperative AGR (HR: 1.33, 95% CI: 1.17–1.51, P < 0.001) (Table 1).

Table 1.

Cox proportional hazard regression analysis of the AGR groups on overall survival in pan-cancer patients

	N	Death (%)	Model 1		Model 2		Model 3
			Hazard ratio (95% CI)	P value	Hazard ratio (95% CI)	P value	Hazard ratio (95% CI)	P value
Preoperative AGR groups
Normal	3793	732 (19.3)	Reference		Reference		Reference
Low	1508	444 (29.4)	1.59 (1.41–1.79)	<0.001	1.62 (1.43–1.82)	<0.001	1.32 (1.17–1.50)	<0.001
Postoperative AGR groups
Normal	3055	534 (17.5)	Reference		Reference		Reference
Low	2131	613 (28.8)	1.69 (1.51–1.90)	<0.001	1.53 (1.35–1.74)	<0.001	1.33 (1.17–1.51)	<0.001
Perioperative AGR groups
Persistently Normal	2663	452 (17.0)	Reference		Reference		Reference
Normalized	392	82 (20.9)	1.21 (0.96–1.53)	0.109	1.26 (0.99–1.59)	0.057	1.16 (0.91–1.47)	0.223
Lowered	1096	275 (25.1)	1.49 (1.28–1.73)	<0.001	1.50 (1.29–1.74)	<0.001	1.32 (1.14–1.54)	<0.001
Persistently Low	1035	338 (32.7)	2.01 (1.75–2.32)	<0.001	2.05 (1.78–2.36)	<0.001	1.56 (1.34–1.81)	<0.001
Longitudinal AGR groups
Normal-stable	4016	739 (18.4)	Reference		Reference		Reference
Rising-decreasing	449	149 (33.2)	2.00 (1.67–2.38)	<0.001	1.80 (1.50–2.17)	<0.001	1.38 (1.13–1.68)	0.001
Decreasing-rising	261	147 (56.3)	4.18 (3.50–4.99)	<0.001	3.66 (3.05–4.40)	<0.001	2.78 (2.30–3.36)	<0.001
Longitudinal AGR groups in patients with persistently normal perioperative AGR
Normal-stable	2216	358 (16.2)	Reference		Reference		Reference
Rising-decreasing	162	43 (26.5)	1.78 (1.30–2.45)	<0.001	1.76 (1.28–2.42)	<0.001	1.33 (0.95–1.86)	0.099
Decreasing-rising	39	16 (41.0)	3.56 (2.16–5.88)	<0.001	3.41 (2.06–5.62)	<0.001	2.36 (1.42–3.93)	0.001

Model 1 was unadjusted.

Model 2 was adjusted for age, sex (female vs. male).

Model 3 was adjusted for age, sex (female vs. male), histology (NSCLC & HCC vs. CRC), T stage (T4 → T1), N stage (N3 → N0), and adjuvant chemotherapy (yes vs. no).

Preoperative AGR was further adjusted in Model 2 and Model 3 of the postoperative and longitudinal AGR group.

The associations of perioperative AGR changing patterns with prognosis

The risk of death increased with the decrease of postoperative AGR levels in patients with both normal and low preoperative AGR, and the slope of reduction was steeper in patients with low preoperative AGR than those with normal preoperative AGR (Fig. 3a, b). Of 5186 patients with accessible preoperative and postoperative AGR, four perioperative AGR groups were identified: the persistently normal group (n = 2663), the normalized group (n = 392), and the lowered group (n = 1096), and the persistently low group (n = 1035). Compared with patients with persistently normal perioperative AGR, patients in the other three groups had more pathological factors associated with poor prognosis (Supplementary Table 7).

Fig. 3. Association of perioperative AGR with overall survival in pan-cancer patients.

a Association between postoperative AGR and overall survival on a continuous scale in patients with normal preoperative AGR. b Association between postoperative AGR and overall survival on a continuous scale in patients with low preoperative AGR. c Kaplan-Meier curves of Persistently Normal (5-year OS rate: 84.0%), Normalized (5-year OS rate: 80.7%), Lowered (5-year OS rate: 78.5%), and Persistently Low (5-year OS rate: 70.2%) perioperative AGR groups for overall survival. In (a), solid lines are unadjusted hazard ratios, with dashed lines showing 95% confidence intervals derived from restricted cubic spline regressions. Reference lines for no association are indicated by the solid grey lines at a hazard ratio (HR) of 1.0. AGR albumin globulin ratio, CRC colorectal cancer, NSCLC non-small cell lung cancer, HCC hepatocellular carcinoma, OS overall survival.

The 5-year OS rates for the persistently normal, normalized, lowered and persistently low perioperative AGR groups were 84.0% (95% CI: 82.5–85.5%), 80.7% (95% CI: 76.7–84.9%), 78.5% (95% CI: 76.0–81.1%), and 70.2% (95% CI: 67.4–73.2%), with the overall log-rank P < 0.001 (Fig. 3c). Cox analyses revealed that lowered (adjusted HR: 1.32, 95% CI, 1.14–1.54, P < 0.001) and persistently low (adjusted HR: 1.56, 95% CI, 1.34–1.81, P < 0.001) perioperative AGR, but not normalized perioperative AGR (adjusted HR: 1.16, 95% CI, 0.91–1.47, P = 0.223), were associated with higher risk of death (Table 1).

The associations of longitudinal AGR trajectories with prognosis

The 3-class quadratic JLCMM model was optimal in the trajectory analysis of AGR (Supplementary Table 8), with parameters shown in Supplementary Table 9. Three longitudinal AGR trajectory groups were identified within 12 months after surgery, labeled as normal-stable (n = 4016, 84.98%), rising-decreasing (n = 449, 9.50%), and decreasing-rising (n = 261, 5.52%) (Fig. 4a). In the normal-stable group, AGR was maintained at a low normal level. In the rising-decreasing group, AGR increased from preoperative to 6 months after surgery and then decreased, while an opposite trend was observed in the decreasing-rising group. Decreasing-rising AGR trajectory was found to be associated with poor prognostic features including higher age and pathologic stage (Supplementary Table 10).

Fig. 4. Association of longitudinal AGR with overall survival in pan-cancer patients.

a AGR trajectory groups during preoperative to 12 months after surgery; (b) Kaplan-Meier curves of Normal-stable (5-year OS rate: 83.6%), Rising-decreasing (5-year OS rate: 67.5%), and Decreasing-rising (5-year OS rate: 48.0%) longitudinal AGR trajectory groups for overall survival; (c) Kaplan–Meier curves of Normal-stable (5-year OS rate: 84.7%), Rising-decreasing (5-year OS rate: 71.4%), and Decreasing-rising (5-year OS rate: 58.7%) longitudinal AGR trajectory groups for overall survival in patients with persistently normal perioperative AGR. AGR albumin globulin ratio, CRC colorectal cancer, NSCLC non-small cell lung cancer, HCC hepatocellular carcinoma, OS overall survival.

The 5-year OS rate in the normal-stable group was 83.6% (95% CI: 82.4–84.8%), which was significantly higher than that of the rising-decreasing group (67.5%, 95% CI: 63.2–72.2%) and the decreasing-rising group (48.0%, 95% CI: 42.0–54.8%) (Fig. 4b). Compared with the normal-stable group, the adjusted HRs on OS for the rising-decreasing and decreasing-rising groups were 1.38 (95% CI: 1.13–1.68, P = 0.001) and 2.78 (95% CI: 2.30–3.36, P < 0.001) (Table 1). In patients with persistently normal perioperative AGR, the 5-year OS rates for the three AGR trajectory groups were 84.7% (95% CI: 83.1–86.3%), 71.4% (95% CI: 64.4–79.2%) and 58.7% (95% CI: 44.6–77.4%), respectively (Overall log-rank P < 0.001) (Fig. 4c). And the adjusted HR of the decreasing-rising group was still significant (HR: 2.36, 95% CI: 1.42–3.93, P = 0.001) (Table 1).

The associations of AGR status with the type of recurrence

Similar results to OS were observed when analyses were repeated using RFS (Supplementary Fig. 3 and Supplementary Table 11). Compared with the normal AGR group, patients with low AGR had higher tendency to develop local recurrences rather than systematic recurrences (Supplementary Tables 12–15). In univariate model, the low preoperative AGR group, low postoperative group, persistently low perioperative AGR group, and decreasing-rising longitudinal AGR group had higher risks of local or both recurrence than systemic recurrence. However, the risks of these groups became insignificant after multivariate adjustment (Supplementary Table 16). Therefore, the capability of AGR to differentiate recurrence type may depend on other prognostic factors.

Subgroup analyses

Separate analyses using data from each cancer cohort found significant associations of preoperative, perioperative and longitudinal AGR groups with OS. Similarly, a lower overall survival curve of the low postoperative AGR group was observed in each cancer, albeit not statistically significant in HCC patients (Supplementary Figs. 4–6). After multivariable adjustment, persistently low perioperative AGR and decreasing-rising AGR trajectory were still significantly associated with both OS and RFS in CRC, NSCLC and HCC subgroup, suggesting that the prognostic value of AGR was independent to tumor-specific features in each cancer (Supplementary Table 17–19). Subgroup analysis by demographic and clinicopathological characteristics yielded similar results to the overall population (Supplementary Figs. 7–9).

Discussion

In this pan-cancer study in depth to the existing literature, we analyzed postoperative AGR measurements, identified perioperative and longitudinal dynamic changes of AGR, and assessed their associations with cancer prognosis. Postoperative AGR was an independent prognostic factor in pan-cancer patients, and its prognostic value was significant in patients with both normal and low preoperative AGR. The survival of patients with low preoperative but normal postoperative AGR was indistinguishable from those with normal preoperative and postoperative AGR, while lowered and persistently low perioperative AGR suggested a poorer prognosis. We identified three longitudinal AGR trajectories during preoperative to 12 months after surgery, characterized by normal-stable, rising-decreasing and decreasing-rising. The rising-decreasing and decreasing-rising AGR trajectory groups were found to be associated with a significantly increased risk of death even in patients with normal preoperative and postoperative AGR, demonstrating the importance of following-up AGR in cancer prognosis surveillance. The findings were robust in subgroup analysis, further accentuating the consistency of these associations.

Clinical and fundamental researches have shown that inflammation and nutrition played crucial roles in progression of malignancies and survival of patients^22,23. Our findings on the association between preoperative AGR and cancer survival are consistent with previous studies^6,7. Besides, we found that postoperative AGR can further stratify the prognosis of high-risk and low-risk patients identified by preoperative AGR, indicating that measuring postoperative AGR in addition to preoperative AGR may contribute to the prognosis management of cancer patients. Different from the preoperative level, postoperative AGR, affected by surgical resection, may indicate the occurrence of postoperative complications and thus predict the prognosis^13,24. Although the prognostic value of postoperative AGR has not been reported, previous studies have revealed that postoperative levels of albumin and inflammatory markers such as CRP are associated with cancer prognosis^25,26, which to some extent supports the conclusions of our study.

Monitoring dynamic inflammatory process and nutritional status may improve immune and nutrition precision medicine in cancer^27,28. Limited by short follow-up durations, previous studies only considered AGR measured at a single time point, leaving the longitudinal dynamic trend of AGR blank^3–6. In this study, we collected repeated AGR measurements during follow-up of our longitudinal cohort, giving an overall profile of AGR change process within 12 months after surgery. A previous study divided patients into decreased and stable albumin change groups based on preoperative level and 1-year postoperative level of serum albumin, and confirmed that postoperative albumin reduction had a negative impact on the prognosis of cancer patients²⁹. However, the rising-decreasing and decreasing-rising AGR trajectory suggested that the delta between two measurements may be unreliable, and it is necessary to consider dynamic serial changes. Although there was overlap between the longitudinal and perioperative AGR changing groups, the longitudinal AGR trajectory could perform risk stratification for each perioperative AGR subgroup (Supplementary Fig. 10), emphasizing the independent prognostic value of longitudinal AGR trajectory.

In subgroup analysis, the risk of low preoperative AGR on death was heterogeneous across histology, although a significant correlation existed in each cancer. Based on a stratified meta-analysis by cancer systems, Lv et.al also found the associations between low pretreatment AGR and poor OS were distinguishing in respiratory cancers and digestive cancers⁶. Notably, the relationship between AGR and HCC is complex by the fact that liver is the albumin-producing organ^30,31. Although the majority of HCC patients included in this study belonged to Child-Pugh A (91.7%), with liver resection percentage less than 50% (97.0%), we risk evaluating HCC patients not based on an inflammation marker, but on an index of reduced liver function. The prognostic significance of AGR in HCC should be interpreted with caution. In addition, significant interactions of age, pathological N stage and adjuvant chemotherapy with AGR were observed. HRs of risk AGR groups were found to be higher in young patients with early-stage cancer and without postoperative adjuvant chemotherapy. For patients with elder age and/or higher disease stage and/or patients who needed adjuvant treatment, the risk of poor prognosis may be linked to their own more severe disease, whereas for those mild patients, AGR might help prognosis risk stratification as an inflammatory marker. The presence of risk AGR status in mild patients may indicate a worse prognosis, and these patients should be closely monitored.

Based on a combined cohort of pan-cancer patients with a large sample size, our results may represent the real-world situation. Using the repeated AGR measurements obtained during postoperative follow-up of CRC, NSCLC and HCC patients, we elucidated the prognostic value of AGR dynamic changes in terms of perioperative changing patterns and longitudinal trajectories, filling the gaps in current studies and providing new insights into the clinical application of AGR. The comprehensive analysis and stable results ensured the validity and reliability of this study. However, our study also had several limitations. Due to the lack of relevant records, some factors affecting serum albumin and globulin levels, such as inflammation³², were not considered in the inclusion of patients. In addition, older adults were more likely to have malnutrition and hypoalbuminemia, which may affect the results³³. We performed subgroup analyses stratified by age and observed no significant interactions of age with preoperative and postoperative AGR. Compared with NSCLC, results of cancers in colorectal and liver, respectively serving as a nutrient absorption site and an albumin production site, should be interpreted carefully. Although we have made gradual adjustments for potential confounders, the possibility of bias caused by residual confounders cannot be ruled out. Finally, the patients included in our study were all from the Yunnan Cancer Hospital, and whether the conclusions can be extrapolated to a wider population needs further study.

In conclusion, postoperative AGR dynamic change, including perioperative changing patterns and longitudinal trajectories of AGR, were independent prognostic factors of pan-cancer. We recommend a routine follow-up of AGR in the postoperative surveillance of cancer patients, and the improved risk stratification based on dynamic AGR change can help patients and clinical researchers adjust their surveillance strategy.

Author contributions

L.Z., L.C., Y.D., L.W., and Z.T. did the concept and study design. L.Z., Y.R., L.Y., L.L., P.H., L.M., L.W., and Y.D. participated in the collection and assembly of data. L.C. and Z.T. did the statistical analysis and interpreted the results. L.Z. and L.C. drafted the manuscript. L.W., Z.T., and Y.D. revised the manuscript. All authors reviewed and commented on the manuscript, and approved its final submission. L.Z. and L.C. denote equal contributions.

Data availability

The data underlying this article cannot be shared publicly due to individuals’ privacy that participated in the study. The data will be shared at a reasonable request to the corresponding author.

Code availability

All statistical analyses were performed using R software (version 4.1.3), with two-sided statistical significance set at a P value < 0.05. JLCMM was implemented with package “lcmm” (version 1.9.2). Multinomial logistic regression models were implemented with package “nnet” (7.3–17).

Competing interests

All authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41698-025-00809-9.