Nomogram for predicting the survival of gastric adenocarcinoma patients who receive surgery and chemotherapy.

BACKGROUND
Surgery is the only way to cure gastric adenocarcinoma (GAC), and chemotherapy is the basic adjuvant management for GAC. A significant prognostic nomogram for predicting the respective disease-specific survival (DSS) rates of GAC patients who receive surgery and chemotherapy has not been established.


OBJECTIVE
We were planning to establish a survival nomogram model for GAC patients who receive surgery and chemotherapy.


METHODS
We identified 5764 GAC patients who had received surgery and chemotherapy from the record of Surveillance, Epidemiology, and End Results (SEER) database. About 70% (n = 4034) of the chosen GAC patients were randomly assigned to the training set, and the rest of the included ones (n = 1729) were assigned to the external validation set. A prognostic nomogram was constructed by the training set and the predictive accuracy of it was validated by the validation set.


RESULTS
Based on the outcome of a multivariate analysis of candidate factors, a nomogram was developed that encompassed age at diagnosis, number of regional lymph nodes examined after surgery, number of positive regional lymph nodes, sex, race, grade, derived AJCC stage, summary stage, and radiotherapy status. The C-index (Harrell's concordance index) of the nomogram model was some larger than that of the traditional seventh AJCC staging system (0.707 vs 0.661). Calibration plots of the constructed nomogram displayed that the probability of DSS commendably accord with the survival rate. Integrated discrimination improvement (IDI) revealed obvious increase and categorical net reclassification improvement (NRI) showed visible enhancement. IDI for 3-, 5- and 10- year DSS were 0.058, 0.059 and 0.058, respectively (P > 0.05), and NRI for 3-, 5- and 10- year DSS were 0.380 (95% CI = 0.316-0.470), 0.407 (95% CI = 0.350-0.505), and 0.413 (95% CI = 0.336-0.519), respectively. Decision curve analysis (DCA) proved that the constructed nomogram was preferable to the AJCC staging system.


CONCLUSION
The constructed nomogram supplies more credible DSS predictions for GAC patients who receive surgery and chemotherapy in the general population. According to validation, the new nomogram will be beneficial in facilitating individualized survival predictions and useful when performing clinical decision-making for GAC patients who receive surgery and chemotherapy.


Background
Gastric cancer (GC) is a common type of cancer worldwide, with more than 1 million new cases in 2018, and it causes nearly 800,000 deaths. The GC causing deaths take part in one twelfth global deaths. Thus, GC is the fifth most usually diagnosed cancer and makes the third predominant cause of cancer-related deaths [1]. GC has a routine appearance of adenocarcinoma in 90% of cases, with gastric adenocarcinoma (GAC) being the mostcommon subtype of GC [2]. The incidence of GC varies between regions, with approximate 70% of cases taking place in developing countries [3]. The incidence rate in men is two fold higher than that in women. Among men, GC is the most important diagnosed type of cancer and the predominant cause of cancer-related deaths in some countries in western Asian, including Iran, Turkmenistan, and Kyrgyzstan. The incidence rates of GC are also obviously elevated in countries in Eastern Asia, such as Mongolia, Japan, and the Republic of Korea [1].
The main treatment modality for locally advanced GC is stomach resection by surgery, and complete resection is the essential treatment for curing locally advanced GC. However, while complete resection by surgery can eliminate the cancer that is visible in the surgical field [4], cancer recurrence remains possible since complete resection cannot extinguish any micrometastatic cancer cells that exist outside of the surgical field. Such unseen cancer cells inevitably reproduce to become a lump that can be diagnosed on imageological examinations or physical examinations, corresponding to recurrence [5]. The objective of adjuvant therapy is to extinguish micrometastatic tumor cells before and/or after surgery in order to increase the probability of a good survival outcome for the cancer patient. The timing to perform chemotherapy is different in the world. In the European Union and the USA, preoperative chemotherapy is advocated while postoperative adjuvant chemotherapy is encouraged in Asia [6].
It is popular to use nomograms for cancer prognoses because nomograms simplify complex statistical predictive models containing large quantity of factors to a single brief numerical estimate model to predict the probability of an event. Such as death or recurrence of cancer. A nomogram is specific to an individual patient [7,8]. Nomogram is easily mastered graphical interfaces to inform clinical decision-making. Nomogram have in general been generally used as graphical representations of complex mathematical formulas. Nomograms combine some independent factors to get a statistical prognostic model. And the model was estimated the prognosis in multiple malignancies [9].
Some nomograms for the survival prognosis of GC or GAC have been reported [10][11][12][13][14][15]. However, no nomograms are available for the 10-year survival prognosis of GAC patients who have received surgery and chemotherapy. In the current study we planed to construct a survival nomogram to predicting the survival of GAC patients who receive surgery and chemotherapy.

Patients
The Surveillance, Epidemiology, and End Results (SEER) database contains nearly 30% of the total US population. It is composed of 18 registries form different cities containing important clinical information on patients in the US who suffered from tumors. We obtained clinical information on GAC patients from the SEER database that could be detailed analyses of survival in GAC. This study performed a retrospective review of all GAC patients in the SEER database who had received surgery and chemotherapy between 2004 and 2015. In order to assess the effect of lymph node status, patients with enough information about the number of regional lymph nodes examined (RNE) were selected in the current study. Finally, accord to the inclusion criteria, a total of 5764 GAC patients were selected as the primary cohort. Approximately 70% (n = 4034) of these patients were randomly assigned to the training set, and the rest 1729 patients were defined as the external validation set.
The inclusion criteria for GAC patients in the current study were described as follows: identifying information. Since the data are available to all the researchers when they signed to obey the data use agreement, no ethical approval was required for this study.

Data collection
The potential factors associated with the survival of GAC patients who receive surgery and chemotherapy were identified by obtaining information on the clinicopathological characteristics of these patients such as the age at diagnosis, RNE, number of positive regional lymph nodes (RNP), sex, race, grade, derived AJCC stage, summary stage, and radiotherapy status. The end point of this study was the disease-specific survival (DSS) rate. DSS was defined as the time period from surgery to cancer-caused death or the last follow-up. DSS was calculated and survival curves were produced using the Kaplan-Meier method and the outcome of the curves were validated by the log-rank test.

Statistical analysis
Age, RNE, and RNP were continuous variables in this study. Age and RNE are in accordance with normal distribution and they are expressed as the form of mean ± SD values, while RNP is expressed as median and interquartile-range values since it dose not follow the law of the normal distribution. The rest variables are categorical and are presented as percentages. Independent factors predicting the survival time were determined using the Kaplan-Meier and Cox proportional-hazards models [16]. Variables that were significant were further identified by a multivariate Cox proportional-hazards model via backward stepwise analysis.  Abbreviations; RNE Number of regional nodes examined, DSW divorced & separated &widowed, RNP Number of regional nodes positive, AJCC American Joint Committee on Cancer A nomogram to predict the 3-, 5-, and 10-year DSS rates was constructed using the results of the multivariate analyses. The predictive accuracy of the constructed nomogram was estimated using Harrell's concordance index (C-index) and the area below the time-dependent receiver operating characteristic curve (AUC). Calibration was assessed graphically by plotting the relationship between the predicted probability and the actual outcome using the Hosmer goodness-of-fit test [17]. The integrated discrimination improvement (IDI) and the net reclassification improvement (NRI) were calculated to evaluate the improved advantage in the predictive accuracy of the new prediction model [18].. Finally, decision-curve analysis (DCA) was employed to evaluate the clinical applicability of the constructed nomogram by quantifying the net improved benefits at various threshold probabilities [19].
All statistical analyses were carried out using SPSS software (version 24.0, SPSS, Chicago, IL, USA) and R software. A two-sided P value of ≤0.05 was regarded as existing statistical significance.
Patients in the validation set showed similar characteristics to those in the training set. The comprehensive clinicopathological characteristics of the GAC patients involved in the training and validation sets are displayed in Table 1.

Nomogram construction
After the multivariable Cox analysis, the outcomes revealed that the age at diagnosis, RNE, RNP, sex, race, grade, summary stage, and radiotherapy status can independently predict the DSS of GAC patients who receive surgery and chemotherapy ( Table 2). All of the statistically potential independent risk factors that were related Abbreviations; RNE number of regional nodes examined, DSW divorced & separated &widowed, RNP number of regional nodes positive, AJCC American Joint Committee on Cancer with DSS were incorporated in the prognostic nomogram developed in this study (Fig. 1).

Validation of the nomogram for DSS of GAC patients who receive surgery and chemotherapy
The constructed nomogram was externally validated using the validation set. The predictive capacity of the constructed nomogram was directly contrast the seventh edition of AJCC staging system for GC. The C-indexes for the training and validation sets were larger for the nomogram (0.694 and 0.707, respectively) than for the seventh AJCC staging system (0.651 and 0.661). The 3-, 5-, and 10-year AUCs for the nomogram were 0.744, 0.746, and 0.743, respectively, for the training set, and 0.744, 0.747, and 0.75 for the validation set, indicating a good model discrimination ability that was better than that of the seventh AJCC staging system (Fig. 2). Calibration plots for the proposal nomogram displayed that the predicted 3-, 5-, and 10-year DSS probabilities for the training and validation sets of patients from SEER database were almost identical to the actual observations (Fig. 3).

Decision curve analysis
DCA plots for the 3-, 5-, and 10-year DSS discrimination ability are depicted in Fig. 4. The proposed nomogram was found to consistently perform better than the seventh AJCC staging system.

Discussion
This study developed a nomogram to predict the 3-, 5-, and 10-year DSS of GAC patients who receive surgery and chemotherapy based on multi-center and multi-population and multi-ethnic data in the SEER database. The AJCC staging system has been the most commonly used and most-effective program for predicting the prognosis of GAC patients [20]. However, receiving both surgery and chemotherapy will result in many more important risk factors influencing the DSS, such as age, race, sex, marital status, primary cancer site, grade, and summary stage. We therefore implemented a more-comprehensive prognosis model in the form of a nomogram. This nomogram not only includes the AJCC staging system, but also system demographics and other important clinical parameters. The overall survival of GAC patients is prolonged after they receive surgery and chemotherapy [21][22][23][24]. The increasing number of GAC survivors-especially long-term survivors-makes the use of a nomogram for predicting their long-term survival prognosis highly desirable.
To our knowledge, the present nomogram is the first for predicting the 10-year DSS for GAC patients who receive surgery and chemotherapy. Zhong and colleagues constructed a similar nomogram for the 10-year survival prognosis of GC patients after they receive curative surgery, but that nomogram did not include the chemotherapy status [25]. In addition, an external cohort of GAC patients from the same database was used to validate the present nomogram. The obtained results suggest that we have successfully constructed a reliable nomogram for predicting the 3-, 5-, and 10-year DSS of GAC patients who receive surgery and chemotherapy, since the nomogram validation demonstrated favorable discrimination and calibration.
The constructed nomogram includes several independent prognostic factors. Many studies have indicated that the age of cancer patients is an important prognostic factor for DSS [13,[26][27][28]. Multivariate analyses indicated that the RNP and older age were statistical independent risk factors for the DSS of GAC patients who receive surgery and chemotherapy, with survival being worse in older patients. The current study found that the DSS of GAC patients who receive surgery and chemotherapy was negatively correlated with age. Moreover, the race that not white or black appeared to be a protective factor compared with being white (HR = 0.765, 95% CI = 0.668-0.877, P < 0.001), while cancer with a primary site of the pylorus seemed to be protective compared with the cardia(HR = 0.729, 95% CI = 0.649-0.818, P < 0.001). Undifferentiated classification was a risk factor compared with well-differentiated (HR = 1.581, 95% CI = 1.026-2.437, P < 0.001), a higher AJCC stage was associated with a worse DSS, and compared with a localized summary stage, a distant summary stage was a risk factor for DSS (HR = 1.726, 95% CI = 1.263-2.659, P < 0.001).
The proposed predicting nomogram model includes risk factors that are easily gotten and collected through clinical historical records. To further estimate whether the prognostic nomogram model expressed better than the traditional AJCC staging system, we assessed its performance based on calibration, discrimination, IDI, NRI, and DCA. The proposed nomogram displayed a good discrimination ability by producing a C-index of 0.694 for the training set and 0.707 for the validation set. Moreover, the C-indexes of the AJCC staging system were weaker than those of the proposed nomogram, as were the AUC values. The discriminative efficiency of the nomogram was obviously preferably comparing with that of the AJCC staging system. The plots for both the training set and validation set resembled a 45-degree line, showing that the predictions of the proposed nomogram were well calibrated (Fig. 2). The NRI and IDI are more sensitive indicators than the C-index, and the NRI indicated that the proposed model reclassified the risk probabilities better than did the AJCC staging system, while the IDI demonstrated the superior ability of the constructed nomogram to distinguish cases compared with the AJCC staging system. Numerous previous studies have found benefits of applying DCA [29][30][31][32][33], and the results of the current study showed that the 3-, 5-, and 10-year DCA curves displayed net benefits larger than those of the AJCC staging system, both in the training set and validation set (Fig. 4).
While the present nomogram model demonstrated high accuracy in predicting DSS, several limitations of this study must be considered. Firstly, data of the patients were collected from the SEER database, in which the chemotherapy status is only reported as either "yes" and "no/unknown." Although all of the cases included in the current study were GAC patients who had received chemotherapy (by excluding the "no/unknown" ones in the SEER database), the lack of detailed chemotherapy information might have influenced the obtained results. Secondly, there are potentially other factors that could influence the prognosis of GAC cancer patients, and so further clinical research should be carried out to improve the nomogram. Thirdly, since the current study had a retrospective research type design, some important data of including patients might had been missing inevitably, decreasing the number of eligible cases. Fourthly, the results of this study would be more meaningful if the nomogram model was well externally validated by another real world, independent, large-quantity, high-quality cohort, which would prove whether our findings are more-widely acceptable. Although there are so many limitations, The results show that our prognostic nomogram is an instructive and efficient model to predict the accurate individual survival outcomes of GAC patients who receive surgery and chemotherapy.

Conclusion
We have constructed and validated a prognosis nomogram for GAC that has a high accuracy. The prognostic effect of the proposed nomogram was better than that of the traditional seventh AJCC staging system alone. The validation process indicated that the current nomogram provides more efficient DSS predictions for GAC patients who receive surgery and chemotherapy in the general population. The nomogram will be beneficial for personalized survival prediction and helpful in clinical decision-making for GAC patients who receive surgery and chemotherapy.