Defining nomograms for predicting prognosis of early and late recurrence in gastric cancer patients after radical gastrectomy

Abstract

There are few studies on the predictive factors of early recurrence (ER) and late recurrence (LR) of advanced gastric cancer (GC) after curative surgery. Our study aims to explore the independent predictors influencing the prognosis between ER and LR in patients with advanced GC after curative intent surgery respectively. And we will further develop nomograms for prediction of post recurrence survival (PRS). Data of patients with GC who received radical gastrectomy was retrospectively collected. Recurrence was classified into ER and LR according to the 2 years after surgery as the cutoff value. Multivariate Cox regression analyses were used to explore significant predictors in our analysis. Then these significant predictors were integrated to construct nomograms. The 1-, 2- and 3-year probabilities of PRS in patients with ER were 30.00%, 16.36% and 11.82%, respectively. In contrast, the late group were 44.68%, 23.40%, and 23.30%, respectively. Low body mass index (hazard ratio [HR] = 0.86, P = .001), elevated monocytes count (HR = 4.54, P = .003) and neutrophil–lymphocyte ratio (HR = 1.03, P = .037) at the time of recurrence were risk factors of PRS after ER. Decreased hemoglobin (HR = 0.97, P = .008) and elevated neutrophil–lymphocyte ratio (HR = 1.06, P = .045) at the time of recurrence were risk factors of PRS after LR. The calibration curves for probability of 1-, 2-, and 3-year PRS showed excellent predictive effect. Internal validation concordance indexes of PRS were 0.722 and 0.671 for ER and LR respectively. In view of the different predictive factors of ER and LR of GC, the practical predictive model may help clinicians make reasonable decisions.

1. Introduction

Gastric cancer (GC) is the 6th most common cancer and the third leading cause of cancer-related death worldwide.^[1] Despite evident advances in medical technology, there has been no substantial breakthrough in reducing the mortality associated with GC. At the present, radical gastrectomy combined with adequate lymphadenectomy remains the only curative intent treatment strategy. However, the prognosis of GC remains extremely poor because more than half may recur after curative surgery.^[2] Recurrence is the most important factor affecting the prognosis of patients with GC.^[3]

After the systematic screening program for early GC was implemented successively in Japan and South Korea, the GC-related mortality was greatly reduced.^[3–5] Therefore, early prevention and detection of recurrence combined with effective intervention are of great significance to improve the prognosis of patients with GC after surgery.

Currently, most studies have focused on the analysis of predictors related to survival and recurrence of GC after radical resection.^[6–10] Several studies have investigated the independent risk factors for early recurrence (ER) of GC. However, there are few studies focused on the prediction of late recurrence (LR) in patients with advanced GC after radical gastrectomy. Moreover, the majority of these studies usually focused solely on early GC or considered early and advanced GC together.^[7,8,11] The aim of our study is to explore the independent predictors influencing the prognosis of ER and LR in patients with advanced GC respectively. And we will further develop nomograms for prediction. This may be helpful to make sensible decisions for clinicians to improve postoperative survival.

2. Methods

2.1. Study patients

Patients with GC who received radical gastrectomy in our institution between June 2016 and December 2019 were retrospectively collected. The study was approved by the ethics committee at our hospital (Registration number: 2022080-Y-01). All enrolled patients received informed consent. Inclusion criteria were as follows: patients who did not undergo neoadjuvant therapy; patients with histologically proven primary gastric adenocarcinoma (stage II/III) undergoing radical gastrectomy (R0 resection with surgical margin); the number of D1/D2 lymph node dissections was not less than fifteen; postoperative histopathological or radiological diagnosis confirmed recurrence. Exclusion criteria were as follows: experiences with other major abdominal surgeries; patients who have suffered from other primary tumors or serious diseases; the American Society of Anesthesiologists classification^[12] exceeded 3; data was incomplete. The surgical approach was performed in accordance with the latest Japanese Gastric Cancer Treatment Guidelines.^[13] The tumor, node, metastasis (TNM) staging was based on the 8th edition of American Joint Committee on Cancer classification.^[14] Postoperative complications were evaluated based on the Clavien-Dindo classification system.^[15] Based on patients practical physical condition, fluorouracil- and platinum-based regimens (generally 3-week cycles of capecitabine/S-1-and oxaliplatin) as adjuvant chemotherapy was recommended.^[16,17]

2.2. Demographic and clinicopathologic variables

Detailed information on demographic and clinicopathologic variables were shown. Demographic variables included sex, age and body mass index (BMI). Clinicopathologic variables included nutritional-, inflammatory-, tumor- and operative-related variables. Nutritional-related variables included hemoglobin (Hb), red blood cells (RBC), total protein, albumin, globulin (GLB) and albumin-globulin ratio (A/G). Inflammatory-related variables included neutrophils count, monocytes count, platelets count, C-reactive protein, neutrophil-lymphocyte ratio (NLR), lymphocyte-C-reactive ratio (LCR), platelet-lymphocyte ratio (PLR), lymphocyte-monocyte ratio. Tumor-related variables included tumor’s size, differentiation, location, Borrmann classification and TNM stage. Operaitve-related variables included resection range, surgical method, surgical procedure, operative time, positive lymph node rate, time to first flatus and recurrence patterns.

2.3. Definitions

The cutoff value discriminating ER and LR was defined as 2 years according to the widely accepted standard of other relevant studies.^[18–20] Post recurrence survival (PRS) was defined as the time interval from the initial diagnosis of tumor recurrence to death. Recurrence included local recurrence and metastasis (including abdominal, distant and multiple metastasis).^[21] Local recurrence was defined as tumor recurrence in situ or local lymph node recurrence. Abdominal cavity metastasis consisted of peritoneal and implantation metastasis. Distant metastasis included metastasis of the tumor to other parenchymal organs such as the liver, lung, bone and distant lymph nodes such as the pleura, neck, axilla and subclavian lymph nodes. Multiple metastasis was defined as 2 or more sites of metastasis.

2.4. Follow-up

Patients were followed every 2 months for the first 2 years after discharge and then every 3 months until recurrence or the last follow-up period. Follow-up items included blood tests, chest radiography, abdominal computed tomography or magnetic resonance imaging, and endoscopic tissue biopsy if necessary. Survival data was obtained through Center for Disease Control and Prevention. Our follow-up endpoint was the 3-year survival time of patients after recurrence.

2.5. Statistical analyses

Statistical analyses were performed with SPSS software (version 25.0, IBM Corp., Armonk, NY) and R software (version 4.1.2; R Foundation for Statistical Computing). For continuous variables distributed normally in the whole cohort, the mean and standard deviation were calculated, and a t-test was used to assess differences between groups. Otherwise, the median and the interquartile range (IQR) were calculated, and compared using a Wilcoxon test. Categorical variables were described as frequency (%) and analyzed using a chi-square test or Fisher exact test. Survival analyses were conducted using the Kaplan–Meier method with log-rank tests. Univariable and multivariable Cox regression analyses were used to identify independent risk factors of influencing PRS. The receiver operating curves (ROC) curve was used to evaluate the predictive power of the nomogram by calculating the area under the curve (AUC). The “rms” package was used to draw the nomogram, the bootstrap method (frequency = 1000) was used for internal validation, the concordance index (C-index) was calculated and the calibration curve was drawn. All P values were 2-sided, and those < 0.05 were considered statistically significant.

3. Results

3.1. Characteristics of the patients during the perioperative period and at the time of recurrence

Clinical data of 157 patients with recurrent GC after radical gastrectomy (including 110 patients with ER and 47 patients with LR) were retrospectively collected. The flowchart was displayed in Figure 1. Preoperative baseline characteristics of the patients were shown in Table S1, Supplemental Digital Content, http://links.lww.com/MD/K422. There were significantly statistical differences between the 2 groups in some indicators of blood routine including RBC, albumin, A/G, neutrophils, monocytes and NLR (all P < .05). Intraoperative and postoperative characteristics were presented at Table S2, Supplemental Digital Content, http://links.lww.com/MD/K423. We found obvious differences in some postoperative features between the 2 groups, including positive lymph node rate, time to first flatus and TNM stage (all P < .05). However, patients with ER and LR had similar recurrence patterns (P = .385). In terms of the characteristics of the 2 groups at the time of recurrence, patients with ER had lower RBC counts (P = .016) and LCR (P = .037), and higher monocytes counts (P = .032) and C-reactive protein (P = .025). (Shown in Table S3, Supplemental Digital Content, http://links.lww.com/MD/K424).

Figure 1.

Flowchart of screening for gastric patients with recurrence.

3.2. Kaplan–Meier survival curves for recurrence

The median PRS for all patients was 7 months (IQR 2-18 months). And the 1-year, 2-year and 3-year survival rates after recurrence were 34.39%, 18.47% and 15.28%, respectively. (Shown in Figure S1, Supplemental Digital Content, http://links.lww.com/MD/K425) The median PRS was 4 months (IQR 2–16 months) for early and 11 months (IQR 3–22 months) for LR. The 1-year, 2-year and 3-year PRS rates of patients with ER were 30.00%, 16.36% and 11.82%, respectively. In contrast, the LR group were 44.68%, 23.40%, and 23.30%, respectively. The median PRS for ER was worse than LR (4 months vs 11 months, P = .038) (Shown in Fig. 2).

Figure 2.

PRS for patients with early and late recurrence. PRS = post recurrence survival.

3.3. Cox regression analysis of influencing factors for PRS

In this study, Cox regression analysis was used to investigate the factors affecting PRS of GC after recurrence. For ER, univariate analysis showed intraoperative positive lymph node ratio and resection range, and BMI, albumin, globulin, A/G, neutrophils, monocytes, NLR, LCR, PLR, and lymphocyte-monocyte ratio at the time of recurrence were significantly correlated with ER (all P < .05). However, in multivariate analysis, only BMI (hazard ratio [HR] = 0.86, P = .001), monocytes count (HR = 4.54, P = .003), and NLR (HR = 1.03, P = .037) at the time of recurrence were independent predictors of PRS after ER. (Shown in Table 1.) Multivariate analysis showed only Hb (HR = 0.97, P = .008) and NLR (HR = 1.06, P = .045) were independent predictors of PRS after LR. (Shown in Table 2).

Table 1.

Univariate and multivariate cox regression analysis of post recurrence survival in gastric cancer with early recurrence.

Variables	Univariate analysis		Multivariate analysis
	HR (95 CI%)	P value	HR (95 CI%)	P value
Sex (female vs male)	1.06 (0.67-1.69)	0.785
Age (yr)	1.02 (1.00–1.04)	0.045*
FOBT (positive vs negative)†	0.96 (0.63-1.47)	0.859
BMI† (kg/m2)	0.97 (0.93-1.02)	0.213
HB† (g/L)	1.00 (0.99–1.01)	.982
RBC† (×10^12/L)	0.89 (0.66–1.20)	.440
TP† (g/L)	0.99 (0.96-1.02)	.552
ALB† (g/L)	0.97 (0.93–1.01)	0.154
GLB† (g/L)	1.01 (0.96-1.06)	0.658
A/G†	0.46 (0.18–1.15)	.095
NEUT† (×10^9/L)	1.02 (0.96–1.10)	.508
LY† (×10^9/L)	1.03 (0.68–1.55)	0.895
MONO† (×10^9/L)	3.38 (0.98-11.70)	0.054
CRP† (mg/L)	1.00 (0.99-1.01)	0.936
PLT† (×10^9/L)	1.00 (1.00-1.00)	0.420
NLR†	1.01 (0.96-1.06)	0.732
LCR†	1.00 (0.96-1.05)	.942
PLR†	1.00 (1.00–1.00)	.622
LMR†	0.92 (0.80–1.06)	0.236
FOBT (Positive vs Negative)‡	1.22 (0.75–1.98)	.426
BMI‡ (kg/m2)	0.86 (0.78-0.93)	.001*	0.86 (0.79–0.94)	.001*
HB‡ (g/L)	1.00 (0.98-1.01)	.255
RBC‡ (×10^12/L)	0.73 (0.51–1.04)	0.082
TP‡ (g/L)	1.00 (0.97-1.02)	.512
ALB‡ (g/L)	0.96 (0.92–0.99)	.010
GLB‡ (g/L)	1.04 (1.00–1.08)	.037
A/G‡	0.32 (0.16-0.64)	.001*
NEUT‡ (×10^9/L)	1.10 (1.04–1.17)	.001
LY‡ (×10^9/L)	0.75 (0.48–1.18)	.216
MONO‡ (×10^9/L)	2.62 (1.45-4.71)	.001*	4.54 (1.69–12.21)	.003*
CRP‡ (mg/L)	1.00 (1.00–1.01)	.177
PLT‡ (×10^9/L)	1.00 (1.00–1.00)	.570
NLR‡	1.03 (1.01–1.06)	0.007*	1.03 (1.00–1.06)	.037*
LCR‡	0.82 (0.72–0.94)	.003*
PLR‡	1.00 (1.00–1.00)	.039*
LMR‡	0.82 (0.71–0.95)	.009*
Recurrence pattern
Local	Reference
Abdominal	1.27 (0.60–2.73)	.532
Distant	0.71 (0.33–1.55)	.39‡
Multiple	1.91 (0.81–4.49)	.137
Positive lymph node ratio	2.47 (1.20-5.05)	.014*
Differentiation (Poor vs Moderate)	0.90 (0.54–1.53)	.705
Tumor size (cm)	1.03 (0.96–1.10)	.396
Resection range (subtotal vs total)	0.66 (0.44–0.99)	.044*
Surgical method
Billroth I	Reference
Billroth II	1.46 (0.66–3.22)	.351
Roux-en-Y	2.11 (0.94-4.73)	.070
Operative type (LS vs OS)	0.94 (0.63–1.40)	.744
Tumor location
High	Reference
Middle	1.36 (0.69–2.67)	.377
Low	0.98 (0.50–1.88)	.939
Borrmann classification
I	Reference
II	0.41 (0.11–1.54)	.186
III	0.70 (0.28–1.74)	.440
IV	1.02 (0.39–2.65)	.970
pTNM (III vs II)	1.27 (0.76–2.12)	.365
Postoperative chemotherapy (yes vs no)	0.69 (0.43–1.11)	.125
Clavein Dindo classification
I	Reference
II	1.56 (1.03–2.36)	.035*
III	3.65 (0.86–15.44)	.078
IV	6.10 (0.81–45.93)	.079

A/G = albumin-globulin ratio, ALB = albumin, BMI = body mass index, CI = confidential intervals, CRP = C-reactive protein, FOBT = fecal occult blood test, GLB = globulin, HB = hemoglobin, HR = hazard ratio, LCR = lymphocyte-C-reactive protein ratio, LMR = lymphocyte-monocyte ratio, LS = laparoscopic surgery, LY = Lymphocyte, MONO = monocyte, NEUT = neutrophil, NLR = neutrophil–lymphocyte ratio, OS = open surgery, PLR = platelet-lymphocyte ratio, PLT = platelet, RBC = red blood cell, TNM = tumor, node, metastasis, TP = total protein.

^*P < .05.

^†Preoperative variables.

^‡Variables at the time of recurrence.

Table 2.

Univariate and multivariate cox regression analysis of post recurrence survival in gastric cancer with late recurrence.

Variables	Univariate analysis		Multivariate analysis
	HR (95% CI)	P value	HR (95% CI)	P value
Sex (female vs male)	0.90 (0.45–1.80)	0.760
Age (yr)	1.01 (0.97–1.05)	.709
FOBT (positive vs negative)†	0.99 (0.50–1.99)	0.992
BMI† (kg/m2)	0.94 (0.88–1.02)	0.126
HB† (g/L)	1.00 (0.99–1.01)	.815
RBC† (×10^12/L)	0.99 (0.61–1.62)	.964
TP† (g/L)	0.98 (0.93–1.03)	.405
ALB† (g/L)	0.96 (0.89–1.04)	.340
GLB† (g/L)	0.97 (0.88–1.07)	.612
A/G†	1.01 (0.15–6.89)	.992
NEUT† (×10^9/L)	0.89 (0.73–1.08)	.241
LY† (×10^9/L)	1.13 (0.70–1.84)	.615
MONO† (×10^9/L)	0.41 (0.02–7.46)	0.548
CRP† (mg/L)	0.96 (0.91–1.02)	.201
PLT† (×10^9/L)	1.00 (1.00–1.01)	.702
NLR†	0.88 (0.72–1.10)	.254
LCR†	0.97 (0.90–1.05)	.406
PLR†	1.00 (1.00–1.00)	.894
LMR†	1.16 (0.91–1.49)	.231
FOBT (Positive vs Negative)‡	2.37 (1.15–4.90)	.020*
BMI‡ (kg/m2)	0.88 (0.75–1.04)	0.127
HB‡ (g/L)	0.98 (0.96–0.99)	.038*	0.97 (0.96–0.99)	.008*
RBC‡ (×10^12/L)	0.72 (0.38–1.37)	0.317
TP‡ (g/L)	0.92 (0.87–0.97)	.001*
ALB‡ (g/L)	0.91 (0.85–0.97)	.002*
GLB‡ (g/L)	0.94 (0.86–1.04)	0.226
A/G‡	0.38 (0.11–1.39)	.145
NEUT‡ (×10^9/L)	1.20 (1.04–1.38)	0.012*
LY‡ (×10^9/L)	0.67 (0.31––1.46)	.317
MONO‡ (×10^9/L)	0.81 (0.11–5.79)	.837
CRP‡ (mg/L)	1.01 (0.99–1.02)	.199
PLT‡ (×10^9/L)	1.00 (1.00–1.00)	0.964
NLR‡	1.07 (1.02–1.13)	0.009*	1.06 (1.00–1.12)	.045*
LCR‡	0.83 (0.68–1.00)	0.057
PLR‡	1.00 (1.00–1.00)	0.059
LMR‡	1.04 (0.83–1.30)	.757
Reference pattern
Local	Reference
Abdominal	1.17 (0.34–4.10)	0.801
Distant	0.92 (0.27–3.14)	.892
Multiple	0.90 (0.15–5.39)	.907
Positive lymph node ratio	3.97 (0.93–16.93)	.062
Differentiation
Well	Reference
Moderate Poor	0.80 (0.08–7.74) 3.33 (0.45–24.52)	.850 .238
Tumor size (cm)	1.08 (0.95–1.23)	.232
Resection range (subtotal vs total)	0.62 (0.32–1.20)	0.150
Surgical method
Billroth I	Reference
Billroth II	0.84 (0.30–2.32)	.736
Roux-en-Y	1.28 (0.47–3.51)	.627
Surgical procedure (LS vs OS)	0.81 (0.42-1.57)	.536
Tumor location
High	Reference
Middle	0.74 (0.26–2.09)	0.573
Low	0.93 (0.34-2.54)	0.881
Borrmann classification
I	Reference
II	3.48 (0.39–31.35)	.266
III	3.07 (0.41–22.78)	0.272
IV	4.17 (0.48–36.04)	0.194
pTNM (III vs II)	1.83 (0.88–3.77)	.104
Postoperative chemotherapy (yes vs no)	1.67 (0.51–5.46)	0.400
Clavein Dindo classification
I	Reference
II	0.76 (0.37–1.55)	0.450
III	0.53 (0.07–3.92)	.533

A/G = albumin-globulin ratio, ALB = albumin, BMI = body mass index, CI = confidential intervals, CRP = C-reactive protein, FOBT = fecal occult blood test, GLB = globulin, HB = hemoglobin, HR = hazard ratio, LCR = lymphocyte-C-reactive protein ratio, LMR = lymphocyte-monocyte ratio, LS = laparoscopic surgery, LY = Lymphocyte, MONO = monocyte, NEUT = neutrophil, NLR = neutrophil–lymphocyte ratio, OS = open surgery, PLR = platelet-lymphocyte ratio, PLT = platelet, RBC = red blood cell, TNM = tumor, node, metastasis, TP = total protein.

^*P < .05.

^†Preoperative variables.

^‡Variables at the time of recurrence.

3.4. Subgroup analysis for PRS

We stratified patients with ER into low BMI group (<18.5 kg/m²), normal-BMI group (18.5–23.9 kg/m²) and high-BMI group (≥24 kg/m²) using Chinese-specific criteria.^[22] The prognosis of recurrent patients with low BMI was worse than those with normal and high-BMI (P = .015). (Figure S2, Supplemental Digital Content, http://links.lww.com/MD/K426) Median PRS, 1-year, 2-year and 3-year PRS rate in different BMI cohorts were listed in Table S4, Supplemental Digital Content, http://links.lww.com/MD/K435.

And patients with LR were classified into anemia group (Hb < 120 g/l for men and < 110g/l for women) and non-anemia group. The prognosis of recurrent patients with anemia was worse than those with non-anemia (P = .004). (Figure S3, Supplemental Digital Content, http://links.lww.com/MD/K427) Median PRS, 1-year, 2-year, and 3-year PRS rate were listed in Table S5, Supplemental Digital Content, http://links.lww.com/MD/K428.

The cutoff values of NLR were calculated with ROC curve, and 3.11 value for ER and 2.06 for LR. We found the prognosis of higher NLR was worse than lower NLR significantly in the ER group (P < .001). However, for patients with LR, this effect disappeared. (Figure S4, Supplemental Digital Content, http://links.lww.com/MD/K429 Figure S5, Supplemental Digital Content, http://links.lww.com/MD/K430) Median PRS, 1-year, 2-year and 3-year PRS rate were listed in Table S6, Supplemental Digital Content, http://links.lww.com/MD/K431 Table S7, Supplemental Digital Content, http://links.lww.com/MD/K432.

For monocytes count, the cutoff value was 0.41 × 10^{^9} g/l. according to ROC curve. As the same with NLR, the prognosis of higher monocytes count was worse than lower ones in the ER group (P = .001). (Figure S6, Supplemental Digital Content, http://links.lww.com/MD/K433) Median PRS, 1-year, 2-year, and 3-year PRS rate were listed in Table S8, Supplemental Digital Content, http://links.lww.com/MD/K434.

3.5. Construction and evaluation of clinical predictive models

The nomogram for predicting 1-, 2-, and 3-year PRS were built based upon the results of multivariate models in the training cohort (Fig. 3). The C-index index of the calculated prediction model for ER and LR was 0.722 (95% CI, 0.660–0.790) and 0.671 (95% CI, 0.573–0.769), respectively. The predictive power of the nomogram model was evaluated and quantified by measuring the degree of fit between the C-index and the baseline time predicted by the nomogram in the standard curve. As can be seen from the 1-, 2-, 3-year survival probability calibration curve shown in Figure 4. The nomogram model has a better predictive effect on recurrence.

Figure 3.

Nomograms for predicting 1-, 2-, and 3-year PRS in the early (A) and late (B) recurrence group. PRS = post recurrence survival. 2 variables at the time of recurrence.

Figure 4.

Internal calibration plots. Calibration curves of nomograms for predicting 1-, 2-, 3-year PRS in the early (A–C) and late (D–F) recurrence group. PRS = post recurrence survival.

At the same time, based on the predictors above, the ROC curve was plotted (Fig. 5). The results showed that ER the AUC were 0.791 and 0.844 for 1- and 2-year PRS, respectively. For LR group, the AUC were 0.707 and 0. 710, respectively. The larger the area under the curve, the better the prediction effect.

Figure 5.

ROC curves for 1-year and 2- year PRS in the early (A) and late (B) recurrence group. PRS = post recurrence survival, ROC = receiver operating characteristic.

4. Discussion

Although radical surgery provides a glimmer of hope for patients with GC, the high recurrence rate after surgery is still a major problem plaguing general surgeons. We constructed predictive models for the early and late group respectively, and found that the predictors of the 2 groups were similar but not identical. The PRS of ER was much worse than that of LR. Low BMI, elevated monocytes count and NLR were independent risk factors for ER. In contrast, Hb and elevated NLR independently influenced LR. The our established predictive models based on nomograms showed excellent effect. Therefore, our study may provide a basis for personalized surveillance and treatment of patients with recurrent GC.

The survival analysis of all patients with recurrence showed that the 1-year, 2-year and 3-year survival rates after recurrence were 34.39%, 18.47% and 15.28%, respectively. Overall median survival after recurrence was 7 months. The median survival time after recurrence was 4 months for ER and 11 months for LR. Slightly different with our survival data, Kang et al,^[8] who followed up 194 GC patients with recurrence, found that the median survival time of patients with ER and LR was 6 months (95% CI, 6.01–7.02) and 7 months (95% CI, 4.21–5.80), respectively (P = .045). But they also included patients with early GC. The 1-year, 2-year and 3-year survival rates of patients with ER were 31.17%, 16.88% and 12.99%, respectively. In contrast, the LR were 37.50%, 20.00%, and 17.50%. There was statistically significant difference in overall survival after recurrence between the 2 groups (P = .038). In addition, we found that recurrence patterns were similar for ER and LR (P = .385), with distant metastasis being the most common pattern (40.26% vs 41.25%). This was consistent with the findings of Sun et al^[20] in the follow-up of patients with recurrent GC with negative lymph nodes. In terms of all patients with recurrence, the top 2 recurrence patterns were distant metastasis and peritoneal metastasis (40.76% and 38.85%), which was basically in accordance with the retrospective study conducted by Spolverato et al^[9] in a large population in the United States.

Inflammation is considered one of the hallmarks of cancer. More and more evidence showed that host inflammatory response might play an important role in the occurrence and development of cancer.^[23,24] Interestingly, markers of inflammation at the time of recurrence (e.g., NLR, LCR, PLR), but not preoperative inflammation, were significantly associated with survival of ER and LR, more so in patients with ER. Yasui et al^[25] reached the same conclusion in patients with stage III colon cancer. We speculated that the inflammatory state of patients could be altered by tumor resection, because the systemic inflammation in the preoperative cancerous state may be different from that in the postoperative temporary noncancerous state. The result of our subgroup analysis for patients with ER demonstrated the prognosis of NLR ≥ 3.1 was worse than NLR < 3.1 evidently. Therefore, for patients with recurrent GC after radical gastrectomy (especially those with ER), it is recommended to monitor blood inflammation indicators regularly. Studies have shown that timely anti-inflammatory can significantly reduce the chance of tumor recurrence in patients with elevated NLR indices.^[26] In the future, randomized controlled trials such as these are needed in large, multi-center patients with GC.

Furthermore, We confirmed that the physical nutritional status was a significant predictor of prognosis of GC after radical surgery. Our results supported the reported data on patients with low BMI at the time of recurrence tend to have adverse outcomes. Consistent with our study, in a retrospective study of 518 GC patients with peritoneal metastases, low BMI (BMI < 18.5kg/m2) was demonstrated as an independent predictor of poor prognosis (HR = 0.812, P = .016).^[27] Chen et al^[28] reported that low BMI was associated with more severe postoperative complications and poor prognosis, compared to patients with a normal-BMI. More importantly, in a large study conducted in Chinese population to dynamically monitor the association between postoperative BMI and prognosis in GC patients, BMI loss of more than 10% within the first year after surgery was found to be associated with poor prognosis.^[29] Overall survival was significantly lower in the anemia group than normal group for patients with stage I GC.^[30] These studies were consistent with our results, which showed an adverse effect of low BMI and anemia on postoperative survival.

Ma et al^[31] first developed a nomogram for predicting ER of stage II/III GC after surgery based on tumor’s location, pTNM stage, lymphocyte count, postoperative complications, adjuvant chemotherapy and positive lymph node ratio. Their C-index value was 0.780 with excellent practicability. Similarly, Liu et al^[32] accurately predicted the role of age, Lauren classification, preoperative CA 19 to 9 levels, pathological stage, major pathological response, and postoperative complications in early postoperative recurrence of GC. However, unlike these studies, we focused on the predictive role of inflammatory and nutritional markers. And our models showed satisfactory outcomes.

Our study has the following limitations: Firstly, due to the design type of retrospective study, selection bias was inevitable. Secondly, the cutoff value that divided ER and LR was based on the consensus view of most scholars. So, evidence-based research is needed in the future. Third, certain novel pathological features that were important for recurrence were not included into our nomograms, such as the result of immunohistochemistry, tumor specific genes, etc. Finally, our study was single-center with a limited sample size. So missing external validation is one of the key limitations of this study. In the future, the estimation of more prognostic factors based on a larger sample possible should be performed to verify the usefulness and generalizability of our predictive model. Notwithstanding these limitations, to the best of our knowledge, our study is the first to develop nomograms to predict ER and LR of GC following curative resection respectively.

5. Conclusion

BMI, NLR and monocytes count at the time of recurrence were independent predictors of survival after ER. Slightly differently, Hb and NLR at the time of recurrence influenced survival after LR independently. Therefore, we suggest that on the basis of regular anti-inflammatory, it may be helpful to improve BMI and correct anemia for prognosis. It is suggested to distinguish the different focus of the treatment of ER and LR. Our findings may provide a basis for future personalized treatment strategies for patients with ER and LR of GC.

Acknowledgments

The study was supported by Shaoxing People’s Hospital Youth Fund Project (2022YA05). The authors declared that they had no competing of interest. We were thankful for the colleagues from the general surgery due to their generous assistance.

Author contributions

Conceptualization: Chenming Liu, Liang An.

Data curation: Liang An.

Formal analysis: Feng Tao, Liang An.

Funding acquisition: Liang An.

Investigation: Liang An.

Methodology: Jialiang Lu, Liang An.

Project administration: Liang An.

Resources: Liang An.

Software: Liang An.

Supervision: Sungsoo Park, Liang An.

Validation: Liang An.

Visualization: Liang An.

Writing – original draft: Chenming Liu.

Writing – review & editing: Liang An.