2.1. Patient's characteristics
For the entire cohort, the median age was 61 years (range 54–67) and 72.9% were male. 743 (44.6%) patients were diagnosed with metastatic GC. CEA, CA19-9 and CA72-4 in the metastatic group was significantly higher than that in the non-metastatic group (all P < 0.001). Increased absolute neutrophil count and decreased absolute lymphocyte count in the metastasis group resulted in a significant increase in NLR (P < 0.001). At the same time, the albumin level of the metastasis group decreased and the globulin level increased (both P < 0.001) (Table 1). Among all patients, 923 patients were diagnosed with primary tumors, 163 with peritoneal metastasis, 119 with osseous metastasis, 38 with pulmonary metastasis, 46 with lymph node metastasis, 192 with hepatic metastasis, 22 with ovarian metastasis, 122 with multisite metastasis, 8 with brain metastasis and 34 with metastasis of other organs.
Table 1
Demographic and baseline characteristics of patients
Characteristic | Overall (N = 1,667) | Non-metastasis (N = 924) | Metastasis (N = 743) | P value |
Gender, no. (%) | | | | < 0.001 |
Male | 1215(72.9) | 707(76.5) | 508(68.4) | |
Female | 452(27.1) | 217(23.5) | 235(31.6) | |
Age, year | 61(54–67) | 61(54–68) | 59(52–67) | 0.001 |
CEA, ng/ml | 2.82(1.58–7.07) | 2.345(1.37–4.19) | 3.88(1.94–17.25) | < 0.001 |
CA199, U/ml | 12.46(6.71–36.87) | 10.49(5.95–19.96) | 18.22(8.31–124.7) | < 0.001 |
CA724, U/ml | 2.98(1.35–9.43) | 2.11(1.17–5.56) | 5.32(1.85–21.31) | < 0.001 |
Erythrocyte,10^12/l | 4.2(3.7–4.6) | 4.28(3.8–4.67) | 4.07(3.6–4.47) | < 0.001 |
Hemoglobin, g/l | 124(104–140) | 129(107–143) | 119(102–133) | < 0.001 |
Leukocyte,10^9/l | 5.68(4.51–7.11) | 5.47(4.4–6.8) | 6.04(4.7–7.85) | < 0.001 |
Neutrophil,10^9/l | 3.56(2.67–4.82) | 3.29(2.48–4.29) | 3.97(2.92–5.75) | < 0.001 |
Lymphocyte,10^9/l | 1.42(1.08–1.83) | 1.53(1.18–1.92) | 1.29(0.95–1.67) | < 0.001 |
NLR | 2.45(1.67–3.83) | 2.1(1.52–3.02) | 3.11(2.08–4.86) | < 0.001 |
Platelet,10^9/l | 207(159–263) | 205(160-259.75) | 209(158–269) | 0.317 |
Monocyte,10^9/l | 0.4(0.29–0.52) | 0.39(0.28–0.5) | 0.4(0.3–0.53) | 0.018 |
Albumin,g/l | 38.2(34.8–41.4) | 38.6(35.4–41.7) | 37.7(34-40.9) | < 0.001 |
Globulin,g/l | 26.2(23.4–29.3) | 25.3(22.6–28.2) | 27.4(24.7–30.5) | < 0.001 |
Data are shown as number of cases (percentage) or median (interquartile range); NLR, neutrophil-to-lymphocyte ratio. |
2.2. NLR can more efficiently predict gastric cancer metastasis than tumor markers
ROC analysis was performed to evaluate the diagnostic efficacy of NLR in predicting GC metastasis. The optimal cutoffs for CEA, CA19-9, CA72-4 and NLR were determined to be 4.34 (a sensitivity of 48.5% and a specificity of 76.3%), 18.69 (a sensitivity of 49.7% and a specificity of 74.1), 6.96 (a sensitivity of 44.7% and a specificity of 80.7%) and 2.91 (a sensitivity of 55% and a specificity of 73.4%), respectively. The AUC of CEA was 0.647 [95% confidence interval (CI), 0.62–0.674, P < 0.001], of CA19-9 was 0.649 (95% CI, 0.622–0.675, P < 0.001), of CA72-4 was 0.662 (95% CI, 0.636–0.688, P < 0.001) and of NLR was 0.679 (95% CI, 0.653–0.705, P < 0.001) for predicting GC metastasis. The predicting efficacy of NLR was comparable to GC markers (Fig. 1a). The predicted probability of combining GC markers (P1) and the predicted probability of combining GC markers and NLR (P2) was obtained by binary logistic regression. The optimal cutoffs of P1 were 0.39 (a sensitivity of 53.8% and a specificity of 75%) and of P2 was 0.39 (a sensitivity of 64.9% and a specificity of 72.8%). The AUC of P1 was 0.716 (95% CI, 0.691–0.74, P < 0.001) and of P2 was 0.73 (95% CI, 0.705–0.754, P < 0.001). Although the AUC of P1 and P2 was comparable (P = 0.209), the combination of NLR slightly increased diagnostic performance (Table 2).
Table 2
AUC and the cut-point of diagnostic indicators at the maximum of Youden index for GC metastases (N = 1,667)
| AUC | 95% CI | Cut-off | Sen | Spe | Youden index | PPV | NPV | P value |
NLR | 0.679 | 0.653–0.705 | 2.91 | 0.55 | 0.734 | 0.284 | 0.549 | 0.734 | < 0.001 |
CEA | 0.647 | 0.62–0.674 | 4.34 | 0.485 | 0.763 | 0.248 | 0.546 | 0.719 | < 0.001 |
CA199 | 0.649 | 0.622–0.675 | 18.69 | 0.497 | 0.741 | 0.238 | 0.537 | 0.727 | < 0.001 |
CA724 | 0.662 | 0.636–0.688 | 6.96 | 0.447 | 0.807 | 0.254 | 0.447 | 0.799 | < 0.001 |
P1 | 0.716 | 0.691–0.74 | 0.39 | 0.538 | 0.75 | 0.333 | 0.568 | 0.758 | 0.209 |
P2 | 0.73 | 0.705–0.754 | 0.39 | 0.649 | 0.728 | 0.377 | 0.637 | 0.733 | Ref |
AUC, area under receiver operating characteristics; CI, confidence interval; Sen, sensitivity; Spe, specificity; PPV, positive predictive value; NPV, negative predictive value; P, P value for comparison of AUC of reference with other indicators using DeLong’s test; P1, prediction probability obtained by binary logistic regression of CEA, CA199, CA724; P2, prediction probability obtained by binary logistic regression of CEA, CA199, CA724 and NLR; Ref, reference. |
We also assessed diagnostic efficacy of NLR for single site metastasis or peritoneal metastasis, ROC analysis was performed using data stratified by different metastasis sites. For peritoneal metastasis (n = 163), AUC (95% CI) of NLR was 0.704 (0.662–0.747), CEA was 0.557 (0.507–0.608), CA19-9 was 0.626 (0.573–0.679) and CA72-4 was 0.694 (0.648–0.739), respectively (Supplementary Table 1). Diagnostic efficacy of prediction probability obtained by binary logistic regression of CEA, CA199, CA724 and NLR was significantly higher than NLR (P = 0.003), CEA (P < 0.001), CA19-9 (P < 0.001) and CA72-4 (P = 0.013) (Fig. 1b). For osseous metastasis (n = 119), AUC (95% CI) of NLR was 0.653 (0.599–0.707), CEA was 0.65 (0.594–0.707), CA19-9 was 0.555 (0.498–0.612) and CA72-4 was 0.586 (0.524–0.648), respectively (Supplementary Table 2). Similar to peritoneal metastasis, diagnostic efficacy of prediction probability was significantly higher than CA19-9 (P < 0.001) and CA72-4 (P = 0.017) but not NLR (P = 0.313) and CEA (P = 0.514) (Fig. 1c). As for hepatic metastasis (n = 192), AUC (95% CI) of NLR was 0.676 (0.633–0.718), CEA was 0.733 (0.693–0.774), CA19-9 was 0.713 (0.669–0.757) and CA72-4 was 0.626 (0.58–0.673) (Supplementary Table 3). Diagnostic efficacy of prediction probability was significantly higher than NLR (P < 0.001), CEA (P = 0.041), CA19-9 (P = 0.002) and CA72-4 (P < 0.001) (Fig. 1d). For multisite metastases, the results of the analysis were similar to the overall cohort (Supplementary Table 4, Fig. 1e). We did not analyze other site such as pulmonary metastasis (n = 38) and ovarian metastasis (n = 22) etc. on account of small sample size (Fig. 1f).
Correlation of NLR and other parameters was showed in the Table 3. In GC patients, NLR showed a positive correlation with CEA, CA19-9, CA72-4 (r = 0.157, 0.173 and 0.161, respectively, all P < 0.001) and no correlation of NLR with age (P = 0.205) was found.
Table 3
Correlation between NLR and clinicopathological characteristics
| Spearman correlation | P value |
Age | 0.031 | 0.205 |
CEA | 0.157 | < 0.001 |
CA19-9 | 0.173 | < 0.001 |
CA72-4 | 0.161 | < 0.001 |
Parameters are grouped by optimal cutoff value. Multiple logistic regression analysis showed that younger age (OR: 0.68, 95% CI: 0.547–0.844, P < 0.001), higher NLR (OR: 2.858, 95% CI: 2.295–3.559, P < 0.001), higher CEA (OR: 2.108, 95% CI: 1.671–2.661, P < 0.001), higher CA199 (OR: 1.844, 95% CI: 1.466–2.32, P < 0.001), higher CA724 (OR: 2.348, 95% CI: 1.851–2.978, P < 0.001) were independent risk factors for GC metastasis (Table 4).
Table 4
Univariate and multivariate analyses of variables for GC metastasis
| Univariate analysis | Multivariate analysis |
| Odds ratio | 95% CI | P value | Odds ratio | 95% CI | P value |
Age, years | | | | | | |
<60 vs. ≥60 | 0.739 | 0.609–0.898 | 0.002 | 0.68 | 0.547–0.844 | < 0.001 |
NLR | | | | | | |
Low vs. high | 3.375 | 2.748–4.145 | < 0.001 | 2.858 | 2.295–3.559 | < 0.001 |
CEA | | | | | | |
Low vs. high | 3.026 | 2.455–3.729 | < 0.001 | 2.108 | 1.671–2.661 | < 0.001 |
CA199 | | | | | | |
Low vs. high | 2.813 | 2.289–3.455 | < 0.001 | 1.844 | 1.466–2.32 | < 0.001 |
CA724 | | | | | | |
Low vs. high | 3.385 | 2.722–4.211 | < 0.001 | 2.348 | 1.851–2.978 | < 0.001 |