An ascites grading system for predicting the prognosis of gastric cancer with peritoneum dissemination

Abstract Aim Gastric cancer with peritoneum dissemination is intractable with surgical resection. The evaluation of the degree of dissemination using computed tomography (CT) is difficult. We focused on the amount of ascites based on CT findings and established a scaling system to predict these patients’ prognoses. Methods We extracted individual data from a population‐based cohort. Patients diagnosed with histologically proven gastric adenocarcinoma with peritoneum dissemination were enrolled. Two raters evaluated the CT images and determined the grade of ascites in each patient: grade 0 indicated no ascites in all slices; grade 1 indicated ascites detected only in the upper or lower abdominal cavity; grade 2 indicated ascites detected in both the upper and lower abdominal cavities; and grade 3 indicated ascites extending continuously from the pelvic cavity to the upper abdominal cavity. We evaluated the relationship between the ascites grade and survival time. After adjusting for other clinical factors, we calculated hazard ratios of each ascites grade. Results A total of 718 patients were enrolled. The number of patients with grades 0, 1, 2, and 3 were 303, 223, 94, and 98, respectively. The median overall survival times were 16.0, 8.7, 5.4, and 3.0 months for ascites on CT grades 0, 1, 2, and 3, respectively (P < .001). The adjusted hazard ratios for the survival time were 1.74 (1.33‐2.26, P < .001), 3.20 (2.25‐4.57, P < .001), and 4.76 (3.16‐7.17, P < .001) for grades 1, 2, and 3, respectively. Conclusion We established a new grading system of pretreatment ascites to better predict the prognosis of gastric cancer.


| INTRODUC TI ON
Peritoneal dissemination (PD) is one of most common metastatic patterns of gastric cancer (GC). 1,2 The prognosis of GC patients with PD remains poor, despite recent advances in anticancer agent development. 3 The progression of PD is often accompanied by various symptoms, including abdominal pain, fullness, and vomiting caused by intestinal obstruction or massive ascites.
Surgeons often detect PD during surgery for GC patients scheduled for radical resection. 4 As imaging examinations cannot detect the small nodules characteristic of PD, 5 laparotomy or staging laparoscopy is sometimes required to observe the intra-abdominal cavity. 6 Serious problems hampering the treatment of GC patients with PD include difficulty assessing the degree of PD and predicting the prognosis. Jacquet and Sugarbaker previously proposed a scaling method of PD for all kinds of malignant disease, 7 and Fujimura et al reported a modified version of this scaling system for GC patients with PD to predict patients' prognoses. 8 These methods are based on macroscopic findings; however, a scaling system without invasiveness is also required from a physician's perspective, as many patients with PD do not receive surgery.
Previous reports have described several prognostic factors in patients with PD, such as the nutritional condition, 3,9,10 performance status, 3 tumor markers, 11 and the presence of ascites. 3,4,12,13 However, which of these is most useful for estimating the degree of ascites as a surrogate marker of PD in patients with GC is unclear.
In the present study, we focused on the amount of ascites at the initial diagnosis and the scaling system based on computed tomography (CT) findings. We investigated the relationship between the amount of ascites on CT and the prognosis of GC with PD using our established large-scale cohort in an effort to establish a physician-friendly prediction tool involving the assessment of ascites to predict the prognosis of GC with PD.

| Study design and cohort development
The present study was a population-based historical cohort study.
All nine designated hospitals for cancer treatment in Fukushima Prefecture participated in this study. First, we listed the patients with stage IV GC using hospital-based cancer registries. We then obtained these patients' individual data along with additional information, including the Charlson's comorbidity index, 14 symptoms, cTNM stage, 15 CT findings, laboratory data, and treatment. Finally, we merged the datasets from each participating institute after anonymizing the information.
We enrolled patients in this study according to the criteria described below. Patients were diagnosed with GC (ICD-10, C16.0-16.9) with synchronous PD and histologically proven adenocarcinoma from a primary lesion from 2008 to 2015.
Patients who did not undergo abdominopelvic CT before any treatment, who were lost to follow-up, who had remnant GC after previous gastrectomy, who had other synchronous malignancies, or who had a medical history of liver cirrhosis, congestive heart failure, or use of diuretics were excluded. Investigators confirmed that the helical CT scans included the area from the diaphragm to the symphysis pubis in 5-to 10-mm-thick transverse sections at the same intervals.
This study was conducted in accordance with the Declaration of Helsinki and all of the applicable local laws and regulations. Approval for the protocol was obtained from the institutional review boards of all of the participating hospitals.

| The diagnosis of PD and grading of ascites
We reviewed medical records of all patients in this cohort and diagnosed the PD according to the findings as follows: having visible nodules near the primary lesion on CT images, positive cytology from aspirated ascites or lavage cytology, or histologically proven disseminated lesions during/after surgery, including staging laparoscopy.
In addition, two gastrointestinal surgeons (MH and HK), who were blinded to the survival outcome, reviewed abdominal CT before initial treatment and classified patients into four categories depending on the amount of ascites detected at the initial diagnosis.
We established the ascites on CT (AC) grading system by referencing the methods used in published clinical trials 16

| Outcomes and statistical analyses
The primary outcome was the adjusted hazard ratio (HR) of each AC grade for the overall survival (OS). To compare the OS by AC grade, we evaluated the descriptive statistics and isolated potential confounding factors. After adjusting for the age, sex, nutritional condition, comorbidities, tumor markers, and treatment with or without chemotherapy as confounding factors, we calculated the HR and 95% confidence interval (95% CI) of each AC grade using the Cox proportional hazard model. We also calculated the adjusted HR in patients with no metastatic lesions other than PD as subclass analysis. In addition, we evaluated the survival curve for these patients using the Kaplan-Meier method for each AC grade and performed the Wilcoxon test.
Secondary outcomes were nutritional indicators, tumor markers, proportion of chemotherapy or gastrectomy, and incidence of synchronous liver metastasis in each AC grade. The descriptive statistics were evaluated, and as necessary, continuous variables were compared using Student's t-test and categorical variables by Fisher's exact test. All statistical tests were two-sided, and P values of .05 or less were considered to indicate statistical significance. Figure 1 shows the patient enrollment flow. A total of 1366 patients with stage IV GC were identified from the cohort databases, including 819 patients diagnosed with PD; ultimately, 718 patients were enrolled in this study. Table 1 shows the patients' characteristics.

| D ISCUSS I ON
Our study proposed the AC grading system, which is a new scale of ascites using abdominopelvic CT as a prognostic tool in GC patients with PD. Patients with a larger amount of ascites generally have a poorer nutritional condition and higher proportion of other metastatic organs than those with less ascites. We confirmed in the present study that the AC grade indicated a linear increase in the HR for the OS in each grade after adjusting for these factors. Abbreviations: AC Grade, ascites on computed tomography grade; BMI, body mass index; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen.

TA B L E 3 Adjusted hazard ratio in patients with no metastatic lesions other than PD
IV. 8 Several other studies have further identified specific proteins or mRNA in ascites or lavage fluid that may be candidate biomarkers for predicting the prognosis. [21][22][23] However, many PD patients receive treatment without ever undergoing surgery, as the effectiveness of debulking surgery has been denied by the previous reported phase III clinical trial. 24 The development of a simple and low-invasive prediction tool would therefore be quite useful for both patients and physicians.
We established our AC grading system by referencing the method of assessing ascites in a previous phase III clinical trial evaluating combination chemotherapy of 5-fluorouracil, l-leucovorin, and paclitaxel. 16  Other strong points of this study are its high external validity for the targeted population and the accuracy of the measurements. We established a large, population-based cohort in a medical field, and fixed investigators reviewed CT images of all cases and decided the AC grades based on predetermined criteria.
One of the limitations in this study is that we did not evaluate the indication of chemotherapy and therapeutic response.
Although most of the physicians prescribed chemotherapy according to Japanese guidelines in participating institutes, the decisions of regimen or dosage were not completely same. In addition, the trends in ascites, clinical symptoms, and serum tumor marker levels may be useful for deciding to change a regimen to a later line. We are planning a further evaluation using the trend in the ascites volume during chemotherapy. Another limitation was the accuracy of the measurement modality. To minimize information bias, two fixed raters evaluated AC grade in this study; however, the CT parameters, such as the slice range setting or use of contrast medium, were not unified due to the multicenter retrospective nature.
In conclusion, we established the clinical meaning of the new grading system for pretreatment ascites that can predict the prognosis of GC patients with PD.

ACK N OWLED G EM ENTS
We would like to express our gratitude to Seria Sato, Koji Uehara, Nobuko Kanno, Mika Yusa, Kazuhira Saito, Tomoko Oya, Yosinobu Yamazaki, Yoko Endo, Chieko, Tairako, and Yumi Inaba for their contribution to data collection.