Preoperative Body Mass Index, Blood Albumin and Triglycerides Predict Survival for Patients with Gastric Cancer

Abstract

Background: Gastric cancer (GC) is common and its prognosis is often poor due to difficulties in early diagnosis and optimal treatment strategies. TNM staging system is useful in predicting prognosis but only possible after surgery. Therefore, it is desirable to investigate prognostic factors/markers that may predict prognosis before surgery by which helps appropriate management decisions preoperatively.

Methods: A total of 320 GC patients were consecutively recruited from 2004 to 2013 and followed up for 127 months (10.6 years) after surgery. These patients' were examined for body mass index (BMI) and blood levels of albumin, triglyceride, total cholesterol, low density lipoprotein cholesterol (LDL-C), and high density lipoprotein cholesterol (HDL-C). Kaplan-Meier method and log rank test were used to analyze long-term survival using the above potential risk markers. We first employed medians of these variables to reveal maximal potentials of the above prognostic predictors.

Results: Three major findings were obtained: (1) Preoperative BMI was positively correlated with albumin (r = 0.144, P<0.05) and triglyceride (r = 0.365, P<0.01), but negatively correlated with TNM staging (r = -0.265, P<0.05). Preoperative albumin levels were positively correlated with triglyceride (r = 0.173, P<0.05) but again, negatively correlated with TNM staging (r = -0.137, P<0.05); (2) Poor survival was observed in GC patients with lower levels of BMI (P = 0.028), albumin (P = 0.004), and triglyceride (P = 0.043), respectively. Receiver operating characteristic (ROC) curve analyses suggested BMI, albumin and triglyceride to have survival-predictor powers similar to TNM system; and (3) Cox multi-factorial analyses demonstrated that age (P = 0.049), BMI (P = 0.016), cell differentiation (P = 0.001), and TNM staging (P = 0.011) were independent overall survival-predictors for GC patients.

Conclusions: Preoperative BMI, albumin, and triglyceride levels are capable of predicting survival for GC patients superior to postoperative TNM system in terms of timing for management. As potential survival-predictors, preoperative tests of BMI, albumin and triglyceride, combined with clinical imaging, may help personalized management for GC patients including planning surgical strategy, optimal radio-chemotherapy and appropriate follow-up intervals after surgery.

Fig 1. Survival analyses against the established risk factors affecting survival validate the GC patient cohort for analyzing potential new factors.

The cohort included 320 GC patients, of whom 151 were alive and 169 were dead. To validate the cohort for survival analysis, general survival is analyzed against several established factors affecting survival in GC patients using Kaplan-Meier method. Panel T shows the impact of tumor infiltration depth (T1, T2, T3 and T4) on the survival with the T3 and T4 patients showing the poorest survival. Panel N demonstrates that the more metastatic lymph nodes (from N0 to N3) in a patient, the poorer the survival. Panel M indicates that patients with distant metastasis (M1) have worse survival than patients without metastasis (M0). Panel TNM staging depicts a typical survival hierarchy that patients with the earliest stage I have the best survival while patients with the latest stage IV show the poorest survival. T, N, M, and TNM staging were determined according to the AJCC TNM staging system [5].

Fig 2. ROC curves reveal performance abilities for 8 factors affecting patient survival.

As shown, the diagonal black line is the reference line and, TNM staging has the largest AUC (purple line, AUC = 0.685) followed by BMI (blue line, AUC = 0.636), Alb (light blue line, AUC = 0.633), and TG (dark green line, AUC = 0.629) (see Table 4 for statistical comparisons). It is very interesting to note that comparisons indicate the 3 new potential survival predictors TG, Alb, and BMI to have AUCs similar to, or even better than, AUCs of the conventional survival predictors T, N, M, and TNM staging (see Table 4, list of individual AUCs). In keeping with Figs 1, 3, 4 and 5, these AUC observations again demonstrate that these newer predictors have very similar, if not the same, power in predicting survival prognosis as do the conventional survival predictors.

Fig 3. Low levels of BMI correlate with poor overall survival for GC patients by both reference range grouping and median range grouping.

Kaplan-Meier survival graphs are shown using either clinical reference ranges (left panel) or median ranges (right panel). It is clear that poorer survival is seen in patients with lower BMI by either reference range or median range. It is also interesting to note that grouping by median range appears to have more statistical power in terms of probability or P value to predict potential differences in survival rates among GC patients than grouping by reference range. The censored patients in Kaplan-Meier graphs are represented by a dot in the line.

Fig 4. Low levels of albumin (Alb) correlate with poor overall survival for GC patients by both reference range grouping and median range grouping.

Similar to BMI analyses as shown in Fig 3, GC patients with lower levels of Alb have worse survival according to either clinical reference range (left) or median range (right) than those with higher levels of Alb. Again, grouping by median range appears to have more statistical power in terms of probability (P value) to predict potential differences in survival rates among GC patients. The censored patients in Kaplan-Meier graphs are represented by a dot in the line.

Fig 5. Low levels of TG correlate with poor overall survival for GC patients by both reference range grouping and median range grouping.

Triglyceride (TG) shows a trend that, although not statistically significant (P = 0.112), patients with lower TG levels have inferior survival to patients with higher TG levels when grouped by reference range (left panel); however, this trend becomes significant (P = 0.043) when patients are grouped by median range (right panel). Again, it is evident that grouping by median range shows more statistical power in terms of probability or P value to predict potential differences in survival rates among GC patients. The censored patients in Kaplan-Meier graphs are represented by a dot in the line.

Fig 6. BMI, Alb and TG are positively correlated with overall survival time among patients with gastric cancer.

As shown, the positive correlations of these 3 prognostic markers with overall survival time of gastric cancer patients are revealed using scatter diagram method, in keeping with the survival curves analyzed using Kaplan-Meier method (see Figs 3, 4 and 5).