Clinical specimens and study design
A total of 112 patients with primary gastric cancer who underwent standard gastrectomy with lymph node dissection from 2008 to 2013 at Peking University First Hospital were enrolled. All patients underwent regular follow up for 7–10 years. Patients were observed until October 2019. Patients undergoing neoadjuvant treatment before surgery were excluded. The pathological diagnostic results were confirmed by two independent gastroenterology pathology doctors, and the gastric cancer classifications were made based on the 8th TNM staging classification for gastric cancer. The clinicopathological characteristics of the 112 patients are shown in Supplementary Table 1. The study was approved by the Peking University First Hospital Biomedical Research Ethics Committee (No. 2017-37). All patients enrolled in the study provided written informed consent.
Clinical information, expression of macrophage markers, and follow-up information from 2008 to 2011 (total of 67 patients) were divided into the training set, and stepwise regression was used to screen the most streamlined clinically meaningful indicators to build a prognostic survival forecasting model. At the same time, clinical information of patients from 2011 to 2013 (total of 45 patients) was used as the validation set. The value of postoperative survival prediction in patients with gastric cancer was evaluated in the training, validation, and overall datasets from the viewpoints of discrimination (concordance index), calibration, and social benefits (net benefit). A study design flow chart is shown as follows.
We performed immunohistochemical staining for macrophage markers on all tumor tissues obtained from patients. Murine anti-human CD68 (1:200; Santa Cruz Biotechnology, CA, USA), murine anti-human Stabilin-1 (1:250; Santa Cruz Biotechnology), and rabbit anti-human CD163 (1:1000; Abcam, Cambridge, England) antibodies were used to confirm the position of macrophages with respect to cancer tissues. Horseradish peroxidase-conjugated goat anti-rabbit and anti-murine IgG (Zsgb Bio, Beijing, China) were used as secondary antibodies. Macrophage morphology and spatial distribution were assessed by two independent pathologists blinded to patients’ information. A computerized imaging system, including an Olympus DP71 camera and Olympus BX51 microscope, was used to evaluate staining. To ensure homogeneity and representativeness, we scanned immunohistochemistry sections at a high magnification (×400) and captured five independent microscopic fields. The same settings were applied for each image capture. ImagePro Plus version 10.0 (Media Cybernetics, Bethesda, MD, USA) was used to measure the density of macrophage makers. The median relative density was used as the cut-off value for high and low expression. We measured the integrated optical density of all markers that stained positive in each image. We then calculated the ratio of integrated optical density to the total area of each image to obtain the relative density. Results were obtained by calculating the average density of five microscopic fields.
Stepwise regression algorithms were used to filter variables and build several Cox regression models using Stata for Windows (Stata Corp, TX, USA). We performed a univariate analysis to identify potential risk factors. A P value of ≤0.2 was used to select variables for the model. A multivariate analysis was performed to confirm the best-fit model after potential risk factor selection. We used a Cox regression model to select variables to construct a prediction model. We identified the Akaike information criterion (AIC) of these models, selected the variable with the lowest AIC value for the prediction model, and considered the practical clinical significance of the included variables. A nomogram was constructed based on a multivariate Cox regression analysis for further analysis. We used the likelihood ratio test to apply forward stepwise selection with AIC as the stopping rule.13 Discrimination was evaluated using the concordance index, and the calibration curve was evaluated using the unreliability U test.14 The rms package was used for the nomogram and calibration curve in RStudio version 3.5.1 (R Foundation for Statistical Computing, Oakland, New Zealand). To evaluate the clinical benefit of the diagnostic model, a decision curve analysis plot was developed. The net benefit of making the decision was measured using the following formula:
where n is the total number of patients and Pt is a given threshold probability. Decision Curve Analysis quantified the net benefits at different threshold probabilities in the validation set to determine the clinical usefulness of the nomogram.15, 16 All statistical analyses noted above were two-sided, and P ≤ 0.05 was considered statistically significant. Stata 15.0 for Windows was applied to carry out the calculation.
The correlation between macrophage marker expression and clinicopathology was evaluated using the χ2 test for categorical data, the Mann–Whitney U test for continuous and ordinally distributed data, and the Kruskal–Wallis test and Fisher’s exact test for ranked ordinal data. We performed a Kaplan–Meier analysis to calculate the survival duration and used the Breslow test (generalized Wilcoxon test) to analyze the significance between groups. The log-rank test was applied to compare survival between subgroups. A one-sample Kolmogorov-Smirnov test was used to test for normality. A multivariate analysis was performed using the stepwise Cox regression model. Variables with a P value of ≤0.2 in the univariate analysis were included in the multivariate regression models. Analysis of Cox regression was employed to compute multivariate hazard ratios and 95% confidence intervals. The results of these tests were analyzed using SPSS version 24.0 (IBM Corp, Armonk, NY, USA). All statistical tests were two-tailed.