A novel nutrition-related nomogram for the survival prediction of colorectal cancer-results from a multicenter study

Clinicopathological characteristics

We collected patients’ demographic and clinicopathological information from the two cohorts. The primary cohort comprised 1,373 patients, of whom 58.5% were men (n = 803). The median follow-up time for patients with CRC in the primary cohort was 34.1 months. A total of 329 deaths occurred during the follow-up period. Among the tumor stages, 99 patients (7.2%) were in stage I, 344 (25.1%) in stage II, 441 (32.1%) in stage III, and 489 (35.6%) in stage IV. Based on the PGSGA criteria, 924 (67.3%) patients were diagnosed with malnutrition, and 449 (32.7%) patients were well nourished (Table 1).

Table 1 Demographic and clinicopathological characteristics

The internal validation cohort was a total of 409 patients with 60.4% of them being male (n = 247). The median follow-up time for patients with CRC in the internal validation cohort was 34.9 months. A total of 98 deaths occurred during the follow-up period. Among the tumor stages, 33 patients (8.1%) were in stage I, 116 (28.4%) in stage II, 116 (28.4%) in stage III, and 144 (35.2) in stage IV. Based on the PGSGA criteria, 260 (63.6%) patients were diagnosed with malnutrition, and 149 (36.4%) patients were well nourished (Table 1).

Survival curves and sensitivity analyses of tumor-node-metastasis stage and Patient-Generated Subjective Global Assessment

Our survival curve showed that TNM stage and PGSGA had a statistically significant difference in the prognosis of CRC (all P < 0.001). However, we also found that the prognostic prediction ability between TNM stage I and II was poor in CRC 3 years after the patients were diagnosed with cancer. In contrast, the prognostic prediction ability of PGSGA was relatively better than that of the TNM stage (Fig. 2A, B). Similarly, our sensitivity analysis also found consistent results in patients with cancer for 3 years. In addition, the 1-year sensitivity analysis showed that the prediction ability was poor among TNM stages I, II, and III (Fig. 2C–F).

Fig. 2figure 2

The Kaplan–Meier survival curves and sensitivity analyses of CRC OS. A Survival curves of TNM stage; B Survival curves of PGSGA; C 3-year sensitivity analysis of TNM stage; D 3-year sensitivity analysis of PGSGA; E 1-year sensitivity analysis of TNM stage; F 1-year sensitivity analysis of PGSGA. Notes: CRC, colorectal cancer; OS, overall survival; TNM stage, tumor-node-metastasis stage; PGSGA, patient generated subjective global assessment

Development of nomogram

The LASSO showed that the TNM stage, radical resection, reduced food intake, activities and function declined, and low albumin level were selected as optimal indicators. Additionally, multivariable Cox regression analyses also suggested that these indicators were associated with CRC OS and were identified as prognostic indicators for the new nomogram model (Fig. 1C). Finally, we took advantage of these prognostic indicators to build a prognostic nomogram model of CRC (Fig. 3A), and the total points were obtained by adding the scores of each indicator. The greater the number of points obtained, the greater the risk and the lower the probability of survival. This nomogram gives a more accurate prediction of survival. The calibration plots showed that the nomogram had an excellent survival prediction consistency for patients with CRC (Fig. 3B, D).

Fig. 3figure 3

Development of the nomogram model in CRC. A Prognostic nomogram predicting 1-, 2-, 3-, 4-, and 5-year OS probability using the five prognostic indicators. BD Calibration curve of the nomogram predicting the 1-, 2-, and 3-year probability of OS in CRC. Notes: CRC, colorectal cancer; OS, overall survival

Analysis and comparison of the prognostic value among different models

We found that TNM staging was a weak predictor of survival in the first three years of patients with CRC, so we compared the predictive value of our constructed model, TNM staging, and PGSGA for survival in the first three years. The C-indexes of the different models were as follows: 0.74 in nomogram model (95% CI, 0.72–0.77), 0.70 in TNM stage model (95% CI, 0.67–0.72, comparable P < 0.001), 0.58 in PGSGA model (95% CI, 0.55–0.61, comparable P < 0.001), and 0.73 in TNM stage + PGSGA model (95% CI, 0.67–0.76, comparable P = 0.004) (Additional file 3: Table S1). The 1-, 2-, and 3-year time-dependent ROC curves showed that the area under the curve (AUC) of the nomogram model (1-year, AUC = 78.9; 2-year, AUC = 79.1; 3-year, AUC = 74.1) was higher than that of TNM stage(1-year, AUC = 73.4; 2-year, AUC = 73.5; 3-year, AUC = 70.2), PGSGA(1-year, AUC = 62.0; 2-year, AUC = 60.5; 3-year, AUC = 56.3), and TNM stage + PGSGA models (1-year, AUC = 77.4; 2-year, AUC = 77.0; 3-year, AUC = 72.2) (Fig. 4A–C). Finally, the nomogram model’s reliability and benefit were evaluated by DCA, which indicated that the nomogram model had a better effect in predicting the 3-year survival of patients with CRC than the other models (Fig. 4D).

Fig. 4figure 4

Analysis and comparison of the prognostic value among different models. A The time-dependent ROC curves for 12-, 24-, and 36-month CRC OS in the primary cohort; B DCA of the nomogram model. Nomogram model (red line), TNM staging system model (blue line), PGSGA model (cyan line), and TNM staging system combined with PGSGA model (green line). Notes: CRC, colorectal cancer; OS, overall survival; ROC, receiver operating characteristic curve; TNM Stage, tumor-node-metastasis stage; PGSGA, patient generated subjective global assessment

Validation of nomogram based on internal validation cohort

The calibration plots of nomogram showed a good survival prediction consistency of patients with CRC (Fig. 5A–C). We also validated the prognostic value of the nomogram in the internal validation cohort. The C-indexes of the different models were as follows: 0.75 in nomogram model (95% CI, 0.72–0.77), 0.68 in TNM stage model (95% CI, 0.63–0.73, comparable P < 0.001), 0.59 in PGSGA model (95% CI, 0.53–0.64, comparable P < 0.001), and 0.71 in TNM stage + PGSGA model (95% CI, 0.66–0.77, comparable P = 0.026) (Additional file 3: Table S1). The 1-, 2-, and 3-year time-dependent ROC curves showed that the AUC of the nomogram model (1-year, AUC = 83.2; 2-year, AUC = 79.8; 3-year, AUC = 73.9) was higher than that of TNM stage (1-year, AUC = 72.1; 2-year, AUC = 70.1; 3-year, AUC = 69.5), PGSGA (1-year, AUC = 65.9; 2-year, AUC = 62.4; 3-year, AUC = 67.2), and TNM stage + PGSGA models (1-year, AUC = 77.5; 2-year, AUC = 75.1; 3-year, AUC = 69.5) (Fig. 5D–F). Finally, the nomogram model’s reliability and benefit were evaluated by DCA, which indicated that the nomogram model had a better effect in predicting the 3-year survival of patients with CRC than the other models (Fig. 5G).

Fig. 5figure 5

Validation of the prognostic value of nomogram in internal validation cohort. AC The 1-, 2-, and 3- year calibration plots of nomogram; DF The time-dependent ROC curves for 12-, 24-, and 36-month CRC OS; G DCA of the nomogram model. Nomogram model (red line), TNM staging system model (blue line), PGSGA model (cyan line), and TNM staging system combined with PGSGA model (green line). Notes: CRC, colorectal cancer; OS, overall survival; ROC, receiver operating characteristic curve; TNM Stage, tumor-node-metastasis stage; PGSGA, patient generated subjective global assessment

Risk group stratification of nomogram in primary cohort

After successfully constructing the nomogram model, we incorporated it into the nutrition-related prognostic prediction of patients with CRC. We performed Kaplan–Meier curves for each TNM stage and the nomogram score could well stratify the survival outcome of those patients with stages I-IV, stage I-III, stage I-II, stage III-IV, stage I (all with low nomogram score), stage II, stage III, and stage IV (all with high nomogram score) (Fig. 6). Meanwhile, we also performed the Kaplan–Meier curves for each TNM stage in the internal validation cohort and found that the nomogram score showed consistent results with the primary cohort (Additional file 4).

Fig. 6figure 6

Risk group stratification within each TNM stage of CRC OS in primary cohort. A All patients; B TNM stage I, II and III; C TNM stage I and II; D TNM stage III and IV; E TNM stage I; F TNM stage II; G TNM stage III; H TNM stage IV. Notes: CRC, colorectal cancer; OS, overall survival; TNM stage, tumor-node-metastasis stage

留言 (0)

沒有登入
gif