This article is mentioned in:

Introduction

Gastric cancer is the fourth most common cancer type and the third leading cause of cancer-associated mortality worldwide (1). Currently, prediction of gastric cancer prognosis predominantly relies on the Tumor-Node-Metastasis (TNM) staging classification (2). Histological grading, which reflects tumor differentiation, is also widely employed for the categorization and prognostic prediction of gastric cancer (3). Poorly differentiated gastric cancers are often more aggressive, leading to earlier lymph node and distal metastasis.

However, via laboratory experiments and clinicopathological analyses, several genes that influence gastric cancer tumorigenesis have been identified to serve as potential biomarkers for the diagnosis, prognostic prediction and further clinical applications associated with the treatment of gastric cancer. For example, several studies have indicated that HER-2 is a negative prognostic factor for patients with gastric cancer (4,5). Moreover, Oh et al (6) reported that p53 status and HIF-1α expression in patients with gastric cancer may be markers of tumor invasion and lymph node involvement, and that high HIF-1α expression predicts a poor prognosis. Despite these findings, the mechanisms underlying the development and progression of gastric cancer remain unclear and more cancer-associated molecules are yet to be discovered.

Concurrent with the rapid advancement of sequencing technologies, computational bioinformatics approaches have become useful methods for systematically identifying the genes and mechanisms involved in tumorigenesis and progression. Public repositories such as the Gene Expression Omnibus (7) and cBioPortal (8) have provided access to functional genomic data, submitted by numerous research groups. Importantly, the datasets hosted by The Cancer Genome Atlas (TCGA; http://www.cancer.gov/) are widely used in the bioinformatics analysis of cancer. For instance, Wang et al (9) established a prognostic scoring system for gastric cancer, constructed using a 53-gene signature identified through the analysis of RNA-sequencing (RNA-seq) data from TCGA combined with a microarray dataset (GSE30727).

Nomograms are widely used as a prognostic method in oncology, providing a user-friendly digital interface of a statistical predictive model that generates the numerical probabilities of specific clinical events (10,11). For instance, in the field of gastric cancer, Lai et al (12) and Kim et al (13) constructed nomograms to predict the recurrence of gastric cancer following curative resection. In addition, several studies have focused on developing nomograms for predicting cancer-specific survival and disease-free survival rates (14,15). Moreover, Han et al (16) constructed a predictive nomogram by combining clinicopathological variables associated with the overall survival (OS) of patients with gastric cancer after gastric resection. Wang et al (17) and Liu et al (18) revealed that specific gene expression patterns could be integrated with clinical variables to provide a better prediction of prognosis using a nomogram.

In the present study, genes with prognostic value for predicting the OS of patients with gastric cancer were identified by analyzing transcriptomic data retrieved from TCGA database. Subsequently, multivariate Cox regression analyses were conducted to determine which genes and clinicopathological variables should be incorporated into a prognostic model. Finally, a nomogram was constructed, providing a visualized prognostic model for the prediction of OS of patients with gastric cancer.

Materials and methods

Patients

The present study was approved by the institutional review boards of the First Affiliated Hospital of Zhejiang University. Pathologically confirmed gastric cancer specimens of 59 patients (median age, 63 years; range, 30–81 years) were included in the present study. Written informed consent was obtained from each patient prior to sample collection and analysis. The specimens were frozen and stored in liquid nitrogen (−80°C) following curative or palliative surgical resection. Surgeries were performed in the First Affiliated Hospital of Zhejiang University (Zhejiang, China) between November 2011 and June 2015. Patient information, including age, sex, grade and TNM stage (determined according to the 7th edition of the American Joint Committee on Cancer staging manual), was documented. All patients enrolled in the study were of Han Chinese ethnicity. The primary outcome of interest was OS time, which was defined as the duration in months from the date of surgery to the date of death.

TCGA data retrieval and screening of survival-associated genes

RNA-seq gene expression data and the corresponding clinicopathological characteristics for patients with gastric cancer were downloaded from the TCGA data portal (https://tcga-data.nci.nih.gov) using TCGA-Assembler (search term, ‘STAD’) (19). The R package DESeq (20) was applied to screen for differentially expressed genes (DEGs). The screening was conducted by comparing early-stage tumors versus metastatic tumors and histologically well-to moderately differentiated tumors versus poorly differentiated tumors. Genes with a difference in expression level between groups at an adjusted P-value of <0.05 and an absolute fold-change of >2 were classified as DEGs. The common genes that were identified during both comparisons were selected as candidate survival-associated genes for further analysis.

Validation of the predictive value of the selected genes for estimating OS

Gene expression data and follow-up information were downloaded from TCGA. To validate the predictive value of the selected genes, the patients were classified into high and low expression groups for each gene. The cutoff points for expression ranges were determined using X-tile software (version 3.6.1) (21) and survival curves were generated using the Kaplan-Meier method. In addition, the online Kaplan-Meier plotter tool (http://kmplot.com/analysis/) (22) was used to validate the predictive value of the selected genes.

Cox regression analysis and risk stratification

All predictors of interest were added, separately and jointly, in the initial full model prior to selection, including tumor grade, sex, TNM stage, age and the selected survival-associated genes. A stepwise selection method was used during model selection to choose predictive variables. The risk score of each patient was then calculated by summing each stepwise-selected predictive variable multiplied by its corresponding coefficient. The risk score was then used to divide the patients into high-, medium- and low-risk signatures (cutoff points determined using X-tile software), in which a higher risk score indicated a poorer survival time for the patient. To reduce the discrepancy of gene expression values arising from the use of multiple detection platforms, the expression level of each gene was coded as 1, 2, 3, 4, 5 or 6 when it ranked in the ≤16.7th, >16.7th to ≤33.3rd, >33.3rd to ≤50.0th, >50.0th to ≤66.7th, >66.7th to ≤83.3rd or >83.3rd percentile of total gene expression, respectively.

Nomogram analysis

A nomogram was constructed, according to the results of the multivariate Cox regression analysis, using the rms (23) package in R version 3.3.0 (http://www.r-project.org/). The predictive performance of the nomogram was measured using the concordance index (C-index) and the calibration curves from internal and external validation. The C-index is correlated with the ability of a model to separate patients with different outcomes, whereas the calibration curves reflect the level of similarity between the outcomes predicted by a model and the actual outcomes.

Tissue RNA isolation and reverse transcription-quantitative (RT-q)PCR

For further validation of the RNA-seq data, the expression level of prickle planar cell polarity protein 1 (PRICKLE1) was assessed in gastric cancer specimens from patients using RT-qPCR. Gastric cancer tissue specimens were pulverized in liquid nitrogen and total RNA was extracted using TRIzol® reagent (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer's instructions. Reverse transcription was conducted using the GoScript™ reverse transcription system (Promega Corporation). Subsequently, qPCR was performed using SYBR® Premix Eq Taq™ reagents (Takara Bio, Inc.) and a StepOne™ Real-Time PCR System (Applied Biosystems; Thermo Fisher Scientific, Inc.) according to the manufacturer's protocol for 2-step PCR. The following thermocycling conditions were used: Initial denaturation at 95°C for 30 sec, followed by 95°C for 5 sec and 60°C for 30 sec (40 cycles). The reactions were performed in triplicate. The expression level of PRICKLE1 was normalized to that of GAPDH using the formula: PRICKLE1ΔCq=(Avg. PRICKLE1Cq-Avg. GAPDHCq) (24). Primers sequences were as follow: PRICKLE1 forward, 5′-TGCTGCCTTGAGTGTGAAAC-3′ and reverse, 5′-CACAAGAAAAGCAGGCTTCC-3′ (25); and GAPDH forward, 5′-GGAGCGAGATCCCTCCAAAAT-3′ and reverse, 5′-GGCTGTTGTCATACTTCTCATGG-3′.

Statistical analysis

Categorical variables were compared using the χ2 test or Fisher's test. The optimal cut-off value was determined using X-tile software (version 3.6.1) (21). Survival curves were depicted using the Kaplan-Meier method and compared using the log-rank test. Nomogram analysis was performed using the rms (23) package in R (version 3.3.0; http://www.r-project.org/). Statistical analysis was conducted using SAS 9.4 (SAS Institute, Inc.). For all of the analyses, P<0.05 in a two-tailed test was considered to be statistically significant.

Results

Strategy for the selection of survival-associated genes

The aim of the present study was to combine the two most commonly used classifications of gastric cancer, namely TNM stage and histological grade, to identify candidate genes associated with the survival of patients with gastric cancer.

RNAseqV2 transcriptomic data and corresponding clinicopathological data of 443 patients with gastric cancer were retrieved from TCGA database using TCGA-Assembler (19). As exhibited in Fig. 1, after standardizing the counts of the RNAseqV2 data using the DESeq R package, the DEGs (P<0.05 and absolute fold-change ≥2) were identified according to differences in the expression profiles of early-stage (TNM stage I; n=16) and metastatic (TNM stage IV; n=27) tumors. Subsequently, a similar analysis was performed comparing transcript levels between patients with histologically well-to-moderately differentiated (G1/G2; n=160) and poorly differentiated (G3; n=246) gastric cancer. The corresponding data regarding TNM stage and grade were missing in some samples; therefore, the total case number in the screening analysis (n=406) was slightly different from the total number of expression profiles collected (n=443).

Figure 1. Workflow for the selection of survival-associated genes in gastric cancer using data from TCGA database. TGCA, The Cancer Genome Atlas.

As a result, a total of 60 and 236 genes were classified as DEGs from the analysis of TNM staging (Table SI) and histological grading (Table SII), respectively. The following seven genes were common to the results of both analyses: SPON1, THBS4, SVEP1, PRICKLE1, ABCA8, SLIT2 and EFEMP1.

Validation of the prognostic value of seven candidate genes in gastric cancer

To validate the prognostic value of the seven candidate genes, Kaplan-Meier analyses based on TCGA data (Fig. 2) were conducted. The optimal cutoff value for classifying the expression of each gene as high or low was determined using the X-tile program (21). There was a trend of a higher mRNA level of each gene to be associated with a shorter OS time of patients with gastric cancer. The association reached statistical significance in EFEMP1 (P=0.006) and SVEP1 (P=0.011), and showed marginal significance in PRICKLE1 (P=0.053) and ABCA8 (P=0.055). Survival analyses based on the Kaplan-Meier (22) plotter database revealed that all seven candidate genes were significantly associated with poor OS (Fig. 3).

Figure 2. X-tile analysis of the association between the expression of each candidate gene and overall survival time of patients with gastric cancer. Survival curves were depicted using the Kaplan-Meier method and compared using the log-rank test. P<0.05 was considered to indicate a statistically significant difference. SPON1, spondin 1; THBS4, thrombospondin 4; SVEP1, Sushi, Von Willebrand factor type A, EGF and pentraxin domain containing 1; PRICKLE1, prickle planar cell polarity protein 1; ABCA8, ATP binding cassette subfamily A member 8; SLIT2, Slit guidance ligand 2; EFEMP1, EGF containing fibulin extracellular matrix protein 1.

Figure 3. Kaplan-Meier survival curves for gastric cancer patients according to tumor expression of each candidate gene. HR, hazard ratio.

Cox regression analysis

Univariate and multivariate Cox regression analyses were further conducted to determine the potential value of each of the seven genes as a predictor of OS. In the univariate Cox regression analysis, the expression levels of 5 out of 7 candidate genes (SVEP1, PRICKLE1, ABCA8, SLIT2 and EFEMP1), as well as age, TNM stage and histological grade were significantly associated with the OS of patients with gastric cancer. PRICKLE1 showed the highest hazard ratio for OS [1.159; 95% confidence interval (CI), 1.055–1.274] among the seven genes. The multivariate Cox regression analysis revealed that PRICKLE1 expression in gastric cancer remained a statistically significant predictor of OS (hazard ratio, 1.193; 95% CI, 1.017–1.400) when combined with TNM stage and age (Table I).

Table I. Univariate and multivariate cox regression analysis for overall survival of patients with gastric cancer.

Table I.

Univariate and multivariate cox regression analysis for overall survival of patients with gastric cancer.

	Univariate		Multivariate
Variable	Hazard ratio (95% CI)	P-value	Hazard ratio (95% CI)	P-value
Grade, (poorly vs. well to moderately)	1.477 (1.061–2.056)	0.021b	1.411 (0.991–2.010)	0.056
Sex, men vs. women	1.255 (0.879–1.792)	0.211	1.243 (0.864–1.787)	0.241
TNM stagea	1.607 (1.313–1.967)	<0.001c	1.681 (1.351–2.093)	<0.001c
Age, yearsa	1.018 (1.002–1.034)	0.025b	1.033 (1.015–1.051)	<0.001c
SPON1a	1.095 (0.997–1.202)	0.057	0.964 (0.798–1.164)	0.700
THBS4a	1.087 (0.990–1.192)	0.079	0.876 (0.719–1.067)	0.189
SVEP1a	1.145 (1.043–1.258)	0.005d	1.082 (0.886–1.321)	0.440
PRICKLE1a	1.159 (1.055–1.274)	0.002d	1.193 (1.017–1.400)	0.030b
ABCA8a	1.125 (1.022–1.238)	0.016b	1.028 (0.879–1.201)	0.731
SLIT2a	1.118 (1.018–1.228)	0.020b	0.976 (0.791–1.204)	0.821
EFEMP1a	1.125 (1.023–1.236)	0.015b	1.042 (0.854–1.270)	0.688

{ label (or @symbol) needed for fn[@id='tfn1-ol-0-0-11543'] } Expression of each gene was divided into 6 levels, level 1–6 according to their percentile of the total expression. The stepwise selection method was used for multivariate Cox regression model.

a Calculated as continuous variables.

b P<0.05

c P<0.001

d P<0.01. CI, confidence interval; TNM, Tumor-Node-Metastasis.

Risk stratification model using predictive variables

A risk stratification model was developed including the three statistically significant predictors identified using multivariate Cox regression analysis (age, TNM stage and PRICKLE1 expression). A prognostic index (PI) was introduced to the risk stratification, which was calculated as follows: 0.532 × TNM stage (stage 1–4) + 0.031 × age (years) + 0.160 × PRICKLE1 expression (graded on a 1–6 scale). X-tile software was used to calculate cutoff values of PI 3.8 and 4.3 to divide patients into low- (PI<3.8), medium- (4.3>PI≥3.8) and high-risk (PI≥4.3) signatures that could most effectively discriminate among differences in OS (Fig. 4A-C). Kaplan-Meier analysis according to the risk stratification based on the PI value exhibited a higher efficiency for discriminating OS than that based on TNM stage alone (stage IV vs. stage II + III vs. stage I) (Fig. 4D).

Figure 4. Risk stratification of gastric cancer patients using X-tile analysis. (A) According to prognostic index (PI), the optimal cut-points were calculated using X-tile. (B) Patients were accordingly divided into low-, medium- and high-risk groups (blue, gray and purple, respectively). (C) Kaplan-Meier survival curves were plotted of the aforementioned risk groups. (D) Survival curves based on TNM stage alone were also plotted.

Construction of a nomogram model including PRICKLE1 expression

To visualize the predictive model, a nomogram for predicting the OS of patients with gastric cancer was constructed (Fig. 5). The significant predictors identified during the multivariate analysis (age, TNM stage and PRICKLE1 expression) were used to build the model. Points were assigned for each patient based on each of the three predictors and were summed. The total points were then used to predict the probabilities of 1-, 3- and 5-year OS, which are displayed at the bottom of the nomogram.

Figure 5. Nomogram for predicting 1-, 3- and 5-year probabilities of overall survival in patients with gastric cancer. Total score of an individual patient is based on each variable. A line is drawn upward to determine the score received for each variable value. The sum of these scores is located on the Total Points axis, then a line is drawn downward to the survival axes to determine the likelihood of 1-, 3- or 5-year overall survival. PRICKLE1 expression was coded as 1, 2, 3, 4, 5 or 6 according to its percentile of the total gene expression. PRICKLE1, prickle planar cell polarity protein 1.

The nomogram was internally validated using the C-index and calibration curves. The C-index of the nomogram was 0.66 (95% CI, 0.61–0.71). The calibration curves of 1-, 3- and 5-year OS obtained from the nomogram are shown in the calibration plot in Fig. 6, indicating the relatively high consistency of this model.

Figure 6. Calibration curve for predicting patient survival at (A) 1 year, (B) 3 years and (C) 5 years. Nomogram-predicted probability of overall survival presented in the x-axis; actual overall survival presented in the y-axis. Error bars, standard deviation.

Validation of the prognostic value of PRICKLE1 in an independent cohort

The performance of the nomogram was externally validated using RT-qPCR to determine the expression level of PRICKLE1 in a cohort of 59 patients who underwent curative or palliative gastric cancer resection at the First Affiliated Hospital of Zhejiang University. There was a significant association between a high expression level of PRICKLE1 and lymph node metastasis (Table II). Tumors were grouped according to high or low expression of PRICKLE1 using the median value as the cut-off (median PRICKLE1ΔCq=−8.60; prickle/GAPDH ratio=0.00257). Kaplan-Meier analysis showed that a high expression level of PRICKLE1 was significantly associated with a poor prognosis (hazard ratio, 2.087; 95% CI, 1.016–4.544; P<0.05; Fig. 7). In this cohort, the C-index of the external validation was 0.63 (95% CI, 0.52–0.74), supporting the result of the internal validation.

Figure 7. Kaplan-Meier analysis of PRICKLE1 expression in an independent cohort of patients with gastric cancer. PRICKLE1, prickle planar cell polarity protein 1.

Table II. Association of PRICKLE1 expression with clinicopathological features.

Table II.

Association of PRICKLE1 expression with clinicopathological features.

	PRICKLE1 expression, n (%)
Clinicopathological features	Low (n=30)	High (n=29)	P-value
Sex			0.9367a
Male	22 (73.33)	21 (72.41)
Female	8 (26.67)	8 (27.59)
Age, years			0.0510a
≤65	11 (36.67)	18 (62.07)
>65	19 (63.33)	11 (37.93)
TNM stage			0.0776a
I–II	15 (50.00)	8 (27.59)
III–IV	15 (50.00)	21 (72.41)
T stage			0.2184a
T1-2	14 (46.67)	9 (31.03)
T3-4	16 (53.33)	20 (68.97)
N stage			0.0205b,c
N0	12 (40.00)	3 (10.34)
N1-3	18 (60.00)	26 (89.66)
M stage			0.2713b
M0	27 (90.00)	22 (75.86)
M1	3 (10.00)	7 (24.14)
Differentiation			0.3007b
Well to moderately	7 (23.33)	3 (10.71)d
Poorly	23 (76.67)	25 (89.29)d
Missing		1

a χ2

b χ2 with Yates' correction.

c P<0.05.

d Percentage from n=28. PRICKLE1, prickle planar cell polarity protein 1; TNM, Tumor-Node-Metastasis.

Discussion

In the present study, a bioinformatics analysis was conducted on transcriptomic data retrieved from TCGA database to screen for candidate genes associated with the tumor aggressiveness and prognosis of patients with gastric cancer. The screened candidate genes, SPON1, THBS4, SVEP1, PRICKLE1, ABCA8, SLIT2 and EFEMP1 all exhibited significant associations with a poorer prognosis of gastric cancer. Multivariate Cox regression analysis of the candidate genes revealed that PRICKLE1 was an independent predictor for the OS of patients with gastric cancer. Therefore, a nomogram was constructed including PRICKLE1 expression level as a factor to predict OS for patients with gastric cancer. As an external validation method, the nomogram was tested in an independent cohort of 59 patients with gastric cancer and exhibited promising results.

To improve the efficiency of the screening process, strict criteria were implemented in the present study. Genes with an adjusted P-value of <0.05 and an absolute fold-change of >2 were classified as DEGs. Significantly DEGs between certain TNM stages and histological grades were then identified, and the seven DEGs common to both categories were selected for further analysis. Notably, the two sets of DEGs identified by the two different strategies included several genes that are associated with cancer, including FAP, FGF2 and IGF1. FAP encodes fibroblast activation protein α, which is upregulated in cancer-associated fibroblasts, representing the predominant component of the stroma in various types of cancer (26), and regulates the invasion and migration of gastric cancer cells (27). Fibroblast growth factor 2 (encoded by FGF2) serves key roles in tumorigenesis and tumor progression and may also serve as a prognostic indicator in gastric cancer (28). Insulin-like growth factor 1 (encoded by IGF1) induces epithelial-mesenchymal transition in gastric cancer and its expression indicates a poor prognosis (29,30). The identification of these genes as DEGs lends support to the reliability of the screening method used.

The functions and clinical indications of the 7 selected candidate genes in the discovery stage were investigated in the literature. SPON1 encodes a secreted extracellular matrix (ECM) glycoprotein (F-spondin/VSGP) and has been reported to promote metastasis in human osteosarcoma (31). THBS4 encodes thrombospondin-4, which influences cellular migration, adhesion, attachment and proliferation (32). THBS4 is upregulated in cancer-associated fibroblasts in invasive breast cancer (33); moreover, it is upregulated in the ECM of diffuse-type gastric cancer (34). SVEP1 encodes a multidomain cell adhesion protein that has been demonstrated to mediate cell-cell adhesion in osteogenic cells (35) and serves a critical role in epidermal differentiation (36). However, its function in relation to carcinogenesis has not yet been investigated. PRICKLE1 regulates planar cell polarity in Drosophila as well as convergent extension in zebrafish and Xenopus (37,38). PRICKLE1 has been demonstrated to be required for tumor progression in a breast cancer xenograft mouse model (39). Moreover, a previous study revealed that its product, PRICKLE1, contributes to breast cancer cell dissemination via interaction with mTORC2. Furthermore, the upregulation of PRICKLE1 in basal breast cancer is associated with poor metastasis-free survival (25). ABCA8 encodes a member of the superfamily of ATP-binding cassette transporters. A high expression level of ABCA8 in primary tumors was associated with a reduced survival rate in patients with serous ovarian cancer (40). SLIT2 encodes a secreted protein that functions as a repulsive axon guidance cue by interacting with the roundabout receptor (ROBO1). It was previously reported that SLIT2 and ROBO1 are expressed in various malignant solid tumors and influence cell proliferation, migration, apoptosis and angiogenesis (41–43). EFEMP1 encodes a multifunctional ECM protein (fibulin-3) that is critical in maintaining the integrity of the basement membrane and the structural stability of the ECM, which was implicated to be correlated with the tumorigenicity of cervical and ovarian cancer, and pancreatic adenocarcinoma (44–46).

In the current study, Cox regression analysis revealed that PRICKLE1 expression was an independent prognostic factor for OS in patients with gastric cancer. Previous studies have revealed the functions of PRICKLE1 and its potential roles in cancer progression. PRICKLE1 is a regulator of the planar cell polarity signaling pathway, which serves multiple roles in epithelial tissue morphogenesis during embryonic development and also in abnormal tissue polarity, invasion and metastasis in breast cancer (39,47). Daulat et al (25) revealed that PRICKLE1 interacts with RICTOR, a member of the mTORC2 complex, to form a complex required for AKT activation, the regulation of focal adhesions and cancer cell dissemination. The study further indicated that PRICKLE1 is upregulated in basal breast cancer, and that its upregulation is correlated with a poor prognosis. Furthermore, the expression of PRICKLE1 mRNA positively correlated with AKT phosphorylation in basal breast cancer (25). Since the planar cell polarity signaling pathway is involved in gastric cancer (47) and aberrant activation of AKT is a key molecular signature in various human malignancies, including gastric cancer, PRICKLE1 may influence the AKT-mediated tumorigenesis of gastric cancer. However, further research is needed to reveal the biological function of PRICKLE1 and its underlying mechanistic association with gastric cancer progression.

The most commonly used prognostic prediction system for gastric cancer is the TNM staging (2). In addition, recent studies have explored new prognostic prediction systems from the perspective of omics analysis. For instance, Li et al (48) identified a seven-miRNA signature via miRNA expression profile analysis and validated the association of this signature with relapse-free survival and OS among patients with gastric cancer. Additionally, Wang et al (9) developed a novel gene expression-based prognostic scoring system to predict survival in gastric cancer. However, it is not sufficient to only include omics data as the prognostic variables of patients with gastric cancer. Crucial clinical information (e.g., age and TNM stage) should also be considered carefully. Several studies have combined genetic and clinical variables to determine the prognosis of patients (18,22,49), but the genes evaluated by the majority of these studies were derived from literature reviews. In the present study, identification of new survival-associated genes using RNA-seq expression data from TCGA was performed. In addition, a Cox regression model including age, TNM stage and PRICKLE1 expression revealed a good predictive power for OS in patients with gastric cancer. Liu et al (18) revealed that Jagged1 was a potential prognostic biomarker for OS and that it could be integrated with tumor depth (T stage), lymph node metastasis (N stage) and distant metastasis (M stage).

Recently, nomograms have been constructed and verified to be more accurate than the conventional staging systems in predicting the prognosis of patients with a variety of cancer types (11,50). In the present study, the factors integrated in the nomogram were independent predictors for OS, selected using multivariate Cox analysis. The C-index for the constructed nomogram was higher than that of TNM stage, with an 8.2% increase, indicating that the nomogram performed better in predicting OS in patients with gastric cancer. Moreover, the ability of this model to discriminate among patients with different prognoses was intuitively shown through risk stratification. Liu et al (18) constructed a nomogram integrating Jagged1 expression and TNM stage, which had a C-index of 0.718. However, few patients with gastric cancer at TNM stage IV were included in their study [6/302 (2.0%) vs. 40/406 (9.9%) in the current study], which imposes certain limitations for interpreting the clinical application of the results. Although the studies by Wang et al (17) and Liu et al (18) developed nomograms that included gene expression data, they did not conduct external validation, which is essential for ensuring the external applicability of a model. The nomogram produced in the present study was subjected to stringent external validation and exhibited high consistency between the predicted and observed results for OS in patients with gastric cancer.

To the best of our knowledge, this is the first study to screen survival-associated genes, along with their validation and combination with clinicopathological variables in order to construct a nomogram to predict the prognosis of patients with gastric cancer. In the study, seven genes were identified with prognostic value for predicting the OS of patients with gastric cancer. The PRICKLE1-based nomogram was developed and was capable of predicting the 1-, 3- and 5-year OS probability for patients with gastric cancer. The lack of validation of the other candidate genes in the present study is a limitation and should be further explored in future studies. In conclusion, the nomogram showed improved predictive accuracy when compared with the conventional TNM classification. The accuracy of the model was externally validated in a cohort from a single institute and indicated good applicability in this population.

Supplementary Material

Supporting Data

Acknowledgements

Not applicable.

Funding

The present study was supported by the Natural Science Foundation of Zhejiang Province (grant nos. LZ16H160001, LY15H160026 and LY15H160013).

Availability of data and materials

The databases used during the present study are available as follows: The Kaplan-Meier plotter database, http://kmplot.com/analysis/; TCGA, https://www.cancer.gov/. All data generated or analyzed during this study are included in this published article.

Authors' contributions

LT designed and supervised the study. YD, YC, MW and YH analysed the data. YD, YC, MW, LL and XY analysed the data and prepared the figures and tables. NX collected the clinical samples. YC, HYW, XY and HHW conducted the qPCR. YD, YC, MW and NX wrote the manuscript. YD, YC, XY, NX and LT revised the manuscript. All authors read and approved the final manuscript.

Ethics approval and consent to participate

This study was approved by the institutional review boards of the First Affiliated Hospital of Zhejiang University (approval no. 2017-860). Written informed consent was obtained from each patient for sample collection and analysis.

Patient consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

References