Nomogram for predicting pathological complete response to neoadjuvant chemotherapy in patients with advanced gastric cancer

Yong-He Chen, Jian Xiao, Xi-Jie Chen, Hua-She Wang, Dan Liu, Jun Xiang, Jun-Sheng Peng

Abstract

Key words: Advanced gastric cancer; Neoadjuvant chemotherapy; Nomogram;Pathological complete response

INTRODUCTION

Gastric cancer is the fourth most common malignant tumor in the world, and it is one of the most common causes of cancer-related death[1]. In China, the diagnosis rate of early gastric cancer is low as a large proportion of patients are diagnosed only at an advanced stage[2].

For potentially curable advanced gastric cancer (AGC) patients, surgical resection with D2 lymphadenectomy followed by adjuvant chemotherapy was traditionally the standard clinical practice in Asia[3,4]. However, the survival outcome of this treatment modality remained poor with a 5-year relative survival rate of 17.9%-54.2%[5,6]. In recent years, neoadjuvant chemotherapy (NAC) was introduced into the treatment modality of AGC, in hopes to improve the overall survival of those with potentially curative tumor site[7]. NAC was found to bring forth downstaging of tumor and improve the chance for curative complete resection of tumor[8,9].

Nevertheless, survival benefit of NAC compared to traditional approach remains a debatable issue[7,10]. Multiple studies had shown that the survival benefit of NAC is dependent on the pathological response to chemotherapy drugs[11], which is defined by the proportion of residual tumor cells in the resected specimen after NAC.Pathological complete response (pCR) is the most preferable pathological response because it means no residual tumor cell is present in the resected specimen. Those with pCR after NAC are more inclined to have ideal prolapsed-free survival and overall survival[12,13]. Meanwhile, those who showed limited response, in which large amount of residual tumor cells are found to be present in the resected tumor, are still associated with poor prognosis[12,13]. For these non-responders, NAC yielded no survival benefit, only toxicity and increased risk of tumor progression during chemotherapy, which may hinder surgical resection. To achieve personalization in precision medicine, we should clearly identify responders and non-responders prior to providing individualized treatment modality to patients from either group.

Much effort had been made to identify the predictor of response to NAC[14-16].However, these predictors were either not available in the routine clinical practice or their significance level was limited by the sample size. Furthermore, not much work has been done in this field.

Therefore, we sought to identify tumor biological characteristics and clinical parameter in this study that are associated with pCR after NAC. Secondly, we aimed to establish an individual nomogram for the prediction of pCR using only easily available pre-treatment clinical parameter, in an attempt to provide personalized treatment strategy to AGC patients.

MATERIALS AND METHODS

Study population and data collection

A total of 208 patients were identified from the gastric cancer database of The Sixth Affiliated Hospital, Sun Yat-sen University from March 2012 to July 2019. Inclusion criteria were as follows: (1) Patients with histologically confirmed adenocarcinoma of the stomach or esophagogastric junction in a clinical stage of T3N+ or T4N0/+ as evaluated by abdominal-pelvic computed tomography; and (2) Patients who underwent NAC followed by gastrectomy with standardized D2 lymphadenectomy.Patients with insufficient information were excluded from the study.

After initial screening, a total of 208 patients were included in this study. All available pre-treatment clinical information was retrieved from the database,including gender, age, body mass index, biopsy pathological differentiation, tumor staging information according to the staging system of American Joint Committee on Cancer (AJCC) 8thedition, routine hematological and biochemical tests result and tumor markers. Patient information is listed in Table 1.

This retrospective study was approved by the Institutional Review Board of The Sixth Affiliated Hospital, Sun Yat-sen University.

Chemotherapy

The treatment strategies for all patients were set by the multi-disciplinary team comprising of surgeons, medical oncologists and radiologists of The Sixth Affiliated Hospital, Sun Yat-sen University. The clinical staging was further confirmed by an experienced radiologist, and the eligibility for NAC was determined by both the medical oncologists and surgeons. Chemotherapy regimen adopted for NAC mainly consisted of mFLOT and FOLFOX (or its derived regimen such as SOX or XELOX). (1)mFLOT: Docetaxel 50-60 mg/m2+ Oxaliplatin 85 mg/m2+ Fluorouracil 2800 mg/m2intravenous injection over 48 h; every 2 wk; (2) SOX: Oxaliplatin 130 mg/m2intravenous injection + Tegafur Gimeracil Oteracil Potassium Capsule 40-60 mg bid D1-D14; every 3 wk; (3) XELOX: Oxaliplatin 130 mg/m2+ Capecitabine 1000 mg/m2bid D1-D14; every 3 wk; and (4) FOLFOX: Oxaliplatin 85 mg/m2+ Fluorouracil 2800 mg/m2continuous intravenous injection over 48 h; every 2 wk.

Surgery

After the completion of NAC, the resectability of the primary tumor site was confirmed again by the multi-disciplinary team. All patients who were enrolled received curative tumor resection (total or subtotal gastrectomy) with D2 lymphadenectomy. Open or laparoscopic surgery was chosen according to the preference of the surgeon. A throughout examination of the abdominal cavity was routinely performed to determine the status of peritoneum metastasis, while peritoneal washing cytology test was not routinely conducted.

Response assessment

All resection specimens were sent to the Department of Pathology and were examined by the attending pathologist according to the same standards. pCR was defined as absence of residual cancer cell in the primary tumor and the dissected regional lymph node. Pathological tumor staging was also conducted according to the 8thAJCC Tumor, Node, Metastasis staging system[17].

Data analysis

Univariate analysis: The normality of data was assessed using the Kolmogorov-Smirnov test and normal probability plots. Parameters that were not normally distributed were expressed in the form of median (upper quartile to lower quartile)and were analyzed using a non-parametric test: Mann-Whitney test or Kruskal-Wallis test, as appropriate. Normally distributed parameters were expressed in the form of mean ± standard deviation and were analyzed by Student’s t-test. Categorical variablewas analyzed by the chi-square test.

Table 1 Patient characteristics and P value of univariate analysis

WBC: White blood cell; RBC: Red blood cell; NEU: Neutrophil cell; NEUR: Neutrophil cell ratio; LYM:Lymphocyte; LYMR: Lymphocyte ratio; MONO: Monocyte; MONOR: Monocyte ratio; EOS: Eosinophil;EOSR: Eosinophil ratio; BASO: Basophil; BASOR: Basophil ratio; HGB: Hemoglobin; PLT: Platelets; UA: Uric acid; Cr: Creatinine; ALT: Alanine transaminase; AST: Aspartate transaminase; r-GT: Gamma-glutamyl transpeptidase; TP: Total protein; Alb: Albumin; AKP: Alkaline phosphatase; PA: Prealbumin; TBIL: Total bilirubin; DBIL: Direct bilirubin; LD: Lactate dehydrogenase; CH: Cholesterol; TG: Triglyceride; HDL: High density lipoprotein; LDL: Low density lipoprotein; CRP: C-reactive protein; CA125: Carbohydrate antigen 125; CEA: Carcinoembryonic antigen; CA199: Carbohydrate antigen 199; CA-153: Carbohydrate antigen 153;AFP: Alpha fetoprotein; BMI: Body mass index.

Multivariate analysis: Logistic regression was used to analyze variables related to the probability of pCR and to estimate regression coefficients and odds ratios with 95%confidence intervals for the parameters that had achieved a significance of P ＜ 0.05 in the univariate analysis.

Nomogram construction

Parameters that achieved a significance of P ＜ 0.05 in the logistic regression analysis were used to build a nomogram to predict the probability of pCR. Continuous variables enrolled in the model were transformed into categorical variables for clear comprehension. The concordance statistic was acquired for the nomogram, and internal validation using the bootstrap method was performed to determine the adjusted concordance statistic. Calibration curve of the nomogram was generated to show the relationship between the predicted and observed outcomes. All statistical analyses were performed using SPSS software ver. 22.0 (IBM, Armonk, NY, United States) and R version 3.6.1 software (The R Foundation for Statistical Computing,Vienna, Austria; www.r-project.org).

RESULTS

A total of 208 patients diagnosed with adenocarcinoma of the stomach or esophagogastric junction from March 2012 and July 2019 were enrolled in the study.The majority of patients were male (161/208, 77.4%), and the median age of the study cohort was 59-years-old (range: 50-64). Overall, 45.2% of the tumors were poorly differentiated adenocarcinoma (91/208) and radiologically suspicious lymph node metastasis was detected in almost all patients (204/208, 98.1%).

All patients received treatments as depicted in Table 1. Patients received a median of four cycles of NAC before surgery, 58.7% of patients received mFLOT regimen(122/208) while 26.1% of patients received FOLFOX or its analogue (75/208). Four point eight percent of patients received a modified 2-drug regimen of docetaxel with fluorouracil (10/208), and one patient received docetaxel monotherapy. Detailed toxicity profile is listed in Table 2. The most common grade 3/4 hematological toxicities were anemia (92/208, 44.2%) and neutropenia (86, 41.3%). The incidence rates of grade 3/4 thrombocytopenia and febrile-neutropenia were 12.0% (25/208)and 3.7% (7/208), respectively. Grade 3/4 hematological toxicities were more common in docetaxel contained regimens than oxaliplatin-based doublet regimens in terms of anemia (48.9% vs 36%) and febrile-neutropenia (4.5% vs 1.3%), but the differences were not statistically significant.

All patients received subsequent radical resection surgery. Peritoneal washing cytology test was conducted in three patients, and the results were all negative.

Postoperative complications were observed in 44 patients (21.2%). The incidence rate was not statistically different between pCR (41/181, 22.7%) and Non-pCR group(3/27, 11.1%). As listed in Table 3, abdominal abscess was the most frequent complication in both groups, and all were resolved by non-surgical management,such as percutaneous centesis drainage, enteral nutrition support and antibiotic therapy. Two patients underwent reoperation due to intestinal obstruction. One patient died of progressive pneumonia 6 wk after surgery in the intensive care unit.

Ninety-two percent (191/208) of patients received postoperative adjuvant chemotherapy. Most patients (126/208, 60.6%) received platin-based doublet regimen,such as FOLFOX or its analog (SOX, XELOX). Although mFLOT was the mainstream regimen (122/208, 58.7%) in NAC before surgery, only a minority of patients (15/208,7.2%) accepted mFLOT as adjuvant chemotherapy after surgery because of the rather intolerable toxicity. Other regimens include taxanes contained doublet regimen(27/208, 13.0%) and oral agents such as S-1 capsule or capecitabine (22/208, 10.6%), as listed in Table 4.

Based on the treatments above, only 13% (27/208) can be classified as achieving pCR.

Univariable associations between the clinical parameters and pCR are shown in Table 1. Statistically significant factors (P ＜ 0.05) include tumor differentiation,carcinoembryonic antigen (CEA) level, lymphocyte ratio (LYMR), monocyte (MONO)count, blood type and uric acid level. The multivariable analysis showed that higher CEA level and LYMR and lower MONO count and tumor differentiation grade are independent predictors of pCR with their respective odd ratios and corresponding 95% confidence intervals as shown in Table 5. The established logistic linear regression model was used to build a nomogram as shown in Figure 1, while the receiver operating characteristic curve of the nomogram is shown in Figure 2. Area under the curve is 0.823. The apparent concordance statistic is 0.767, indicating a strong discriminative ability in prediction. Calibration curves between predicted and actual observations were plotted for internal validation. The outcome demonstrated that this nomogram showed good statistical performance for predicting the probability of pCR, as shown in Figure 3.

DISCUSSION

NAC is emerging as a common perioperative treatment modality. However, there was lacking a prospective randomized controlled trial to prove that it can bring more survival benefit compared to adjuvant chemotherapy[7]. In a FLOT4 trial, researchers showed that NAC is able to elicit tumor downstaging, improve R0 resection rate and may eventually result in better survival in patients that are sensitive to chemotherapy drugs[8]. Nonetheless, NAC may result in tumor progression during chemotherapy for those who respond poorly to chemotherapy drugs, thus causing a probable hindrance to curative resection. This makes screening for sensitivity to NAC a crucial step before deciding the treatment modality for ACG patients. Henceforth, we set up an exploratory study to identify pre-treatment parameters that can predict sensitivity to NAC.

pCR was chosen as the marker for high sensitivity to NAC. Among all other potential markers, such as radiological response indicator (according to RECIST 1.1)[18]or Ryan’s classification system of tumor regression grading, pCR is more universal,objective and replicable. Many previous studies have proven that pCR is more closely related to prolonged survival after NAC and curative surgery[19-21]. pCR is also closely related to tumor downstaging (mostly to stage ypT0N0), and the 8thAJCC staging proved that downstaging to T0N0 showed favorable survival outcome with a 5-year relative survival rate of 89%. Therefore, we believe that predicting pCR alone is of great clinical significance.

Table 1 shows that 13.0% of patients achieved pCR after NAC, which is similar to previous reports (8.4%-17.4%)[13,22-24]. This again indicated that only a portion of patients can benefit from NAC, signifying the importance of individualized treatment strategy. Based on this, we used only pretreatment parameters to create a nomogram for pCR prediction. After internal validation by bootstrapping, tumor differentiation,CEA, LYMR and MONO count were found to be useful pretreatment parameters in predicting pCR (concordance statistic of 0.767 after correction for optimism). Wang et al[25]showed that tumor differentiation is an independent predictive marker for histopathologic response to cytotoxic therapy and that poorly differentiated tumor is clearly less sensitive to NAC. Previous reports indicated that lymphocytes possess potent antibody-dependent anti-cancer activities, and thus a high LYMR is associated with stronger lymphocyte-mediated immune response to the tumor, leading to a higher probability of pCR[26-28]. It is known that monocytes promote tumor genesis and angiogenesis while suppressing the host immune response to cancer cells[29]. Besides,monocytes in the blood circulation were proven to be an important source of soluble mediators that may support the evolution of malignant cells[30,31].

Table 2 Hematological toxicity of neoadjuvant chemotherapy, n (%)

Previous work showed that elevated monocyte count is associated with poor response to chemotherapy and confers a negative prognosis in cancer patients, which is in line with our findings[32,33]. Interestingly, as opposed to previous studies, our study indicated that higher CEA is associated with higher probability of pCR.Elevated CEA had always been associated with poor chemosensitivity[34-37], lower probability of pCR and a worse prognosis after perioperative chemotherapy[38,39]. A possible assumption to this phenomenon is that elevated CEA is associated with heavier tumor load and faster tumor growth rate[40,41], thus making it more susceptible to chemotherapy.

There were a few previous studies that aimed to find a predictor for response to NAC. It was reported that factors such as tumor blood supply, grade of differentiation and iodine uptake were associated with response to NAC[15,16]. Research also showed that certain radiomic features extracted from computed tomography image may be promising predictors for response to NAC[42-44]. Nevertheless, these studies require complicated additional examination, while some were limited by the small sample size. Besides, all these studies explored only one single factor or factors of a single dimension[15,16,42,43]. The major highlight of our study is the combination of four parameters to build a nomogram according to the weighting and importance of each parameter. Our nomogram is built upon CEA level, LYMR, MONO count and tumor differentiation grade, which are all easily assessable, making it practical for clinical use.

However, we acknowledge that there were some limitations to our study. Since most patients enrolled in the study underwent surgery in the recent 2 years, there were insufficient survival events to analyze the impact of the predictor and pCR on overall survival rate. Secondly, the chemotherapy regimen in our study was not unified, although we verified through chi square test that the different regimens were not associated with different responses (P = 0.217). Lastly, there was a lack of external validation in an independent cohort of patients.

CONCLUSION

A nomogram predicting pCR was built based on clinical parameters prior to the start of NAC. This nomogram showed satisfactory predicting power and can be useful in further development of an individualized treatment strategy for AGC patients.

Table 3 Postoperative complication and mortality, n (%)

Table 4 Postoperative adjuvant chemotherapy

Table 5 Result of multivariate analysis

Figure 1 Nomogram for predicting pathological complete response to neoadjuvant chemotherapy. The carcinoembryonic antigen axis, 1:0-5 ng/mL; 2:5-10 ng/mL; 3:10-15 ng/mL; 4:15-20 ng/mL; 5:20-25 ng/mL; 6: ＞ 25 ng/mL; The tumor differentiation axis: 1: Well-differentiated; 2: Moderately differentiated; 3: Moderatelypoorly differentiated; 4: Poorly differentiated; 5: Signet ring cell adenocarcinoma. CEA: Carcinoembryonic antigen; LYMR: Lymphocyte ratio; pCR: Pathological complete response.

Figure 2 Receiver operating characteristic curve for the nomogram model. AUC: Area under the curve.

Figure 3 Calibration curve for the nomogram model.

ARTICLE HIGHLIGHTS

are shown in Table 1. Statistically significant factors (P ＜ 0.05) include tumor differentiation,carcinoembryonic antigen level (CEA), lymphocyte ratio (LYMR), monocyte count (MONO),blood type and uric acid level. The multivariable analysis showed that higher CEA level and LYMR and lower MONO and tumor differentiation grade are independent predictors of pCR with their respective odd ratios and corresponding 95% confidence intervals, as shown in Table 5. The established logistic linear regression model was used to build a nomogram as shown in Figure 1, while the receiver operating characteristic curve of the nomogram is shown in Figure 2.Area under the curve was 0.823. The apparent concordance statistic was 0.767, indicating a strong discriminative ability in prediction. Calibration curves between predicted and actual observations were plotted for internal validation. The outcome demonstrated that this nomogram showed good statistical performance for predicting the probability of pCR, as shown in Figure 3.

Research conclusions

A nomogram predicting pCR is built based on clinical parameters prior to the start of NAC. In this model, higher CEA level and LYMR and lower MONO and tumor differentiation grade are correlated with higher probability of pCR. Interestingly, the correlation between CEA level and pCR is opposite to the previous report, in which higher CEA level is often the predictor of poor response to NAC. The model was internally validated using bootstrap method and showed satisfactory predictive power. However, it should also be acknowledged that there is a lack of external validation in an independent cohort, and the survival impact of pCR was not elucidated owning to insufficient data.

Research perspective

In the future, we plan to add an external cohort for validation to strengthen the reliability of the nomogram. In addition, we plan to analyze the impact of pCR on survival.

ACKNOWLEDGEMENTS

The authors would like to thank Miss Pheier Saw for her careful language assistance.