Malnutrition significantly affects the prognosis of patients with cancers. [28–30] Our study suggests that more than 50% of patients have a high prognostic risk based on a predictive model constructed by nutritional indicators. Nutritional indicators, including albumin and prealbumin, were independent prognostic factors for OS. Our study also suggests that sex, age, baseline weight, food intake reduction grade, emerging disease, ECOG performance status, hospitalization frequency, clinical stage, hemoglobin suppression grade, platelet suppression grade, and liver function classification are significantly associated with OS. If screening was based solely on the NRS-2002, only 37.5% of the patients in our study would have required nutritional intervention, but 50.7% (682/1346) of the patients in the training cohort and 53.5% (299/559) of the patients in the validation cohort had a high prognostic risk according to our model.
In our study, the established nomogram contained 13 clinically easy-to-obtain variables. The C-indexes at 1, 3, 5 and 10 years in the training cohort were 0.848, 0.826, 0.814 and 0.799, respectively. The C-indexes at 1, 3, 5 and 10 years in the validation cohort were 0.851, 0.819, 0.814, and 0.801, respectively. The model had great calibration in both the training and validation cohorts. Thus, the nutritionally high-risk status according to the NRS-2002 method prevents a large proportion of patients from receiving early nutritional intervention and does not provide prognostic information. Our study makes up for this deficiency. Our model included additional prognostic risk factors beyond those used in traditional nutritional screening. Early nutritional treatment can be carried out for high-risk patients, and personalized condition assessments and nutritional treatment decisions can be made.
In our model, we combined hunger, complications of chemoradiotherapy, and inflammatory indicators to derive the actual nutritional status of patients and correlated them with clinical prognostics. For hunger, there are acute and chronic forms. We considered a recent reduction in oral intake as a candidate variable for acute hunger and baseline weight and recent weight loss as candidate variables for chronic hunger. Multivariable analysis showed that baseline weight and food intake reduction grade were independent prognostic factors. Reduced food intake in patients leads to muscle mass reduction and cachexia. The current weight loss grading methods do not take the potential benefit of a higher baseline weight into account in the risk assessment of patients with cancer or treatment-related weight loss. [31] Patients with a higher baseline body weight may have more energy reserves, thus conferring a survival advantage. [32]
The patients enrolled in the analysis were admitted to the hospital for antitumor therapy. The processes of radiotherapy and chemotherapy inevitably lead to a reduction in trilineage hemocytes. Patients with malnutrition lack raw materials for cell division, [33] which limits the recovery of trilineage hemocytes. Thus, malnutrition would lead to a delay in antitumor therapy, extended treatment cycles and ultimately a shortened survival time in patients. [1] In addition, due to the side effects of antitumor therapy such as chemoradiotherapy, [1] patients often have impaired liver function, which leads to reduced protein synthesis or hypoproteinemia. [34] Those with complications such as hydrothorax or ascites have a worse prognosis. Due to the routine preventive application of leukopenia drugs, fewer patients delayed treatment due to leukopenia, [35–37] so we did not include white blood cell suppression in the model. In summary, hemoglobin suppression grade, platelet suppression grade, and liver function classification were included in the final model.
Inflammation is recognized as an increasingly important potential factor that increases the risk of malnutrition and may lead to an unsatisfactory effect of nutritional therapy and an increased risk of death. [14] Various forms of acute and chronic inflammation (complications caused by the primary disease, hospital-acquired infection and chemoradiotherapy) lead to increased catabolism and further malnutrition. [38, 39] These studies suggest that inflammation increases nutritional risk and that malnutrition further leads to a poorer prognosis, which is similar to our conclusion. In our study, CRP showed statistical significance in the univariable analysis but was not included in the model because it was not statistically significant in the multivariable analysis. PCT, which had too many missing values, was also not included in the model. Prealbumin and albumin are both nutritional and inflammatory indicators. In our enrolled patients, 53.9% had a prealbumin level that was lower than normal, and 59.6% had an albumin level that was lower than normal. Approximately half of the patients’ prealbumin and albumin levels decreased below normal. Therefore, we considered the mean score of the model as the cutoff point. This may be in part why the model can stratify patient prognosis so well.
Studies have shown that acute or chronic inflammation of different degrees is closely related to the occurrence of malnutrition, and in diseases or injuries, metabolic and dietary intake changes occur. [15, 40, 41] There is a close interaction between inflammation and malnutrition. The existence of inflammation contributes to the development of malnutrition and limits the effectiveness of nutritional interventions. [39, 42] Similarly, malnutrition may reduce the effectiveness of drug therapy. Patients can benefit from early nutritional intervention, which may be due to its regulatory effect on systemic inflammation. [20, 21] Our study only identified inflammatory indicators as independent prognostic factors but could not determine the optimal intervention time for drug regulation. Inflammatory indicators are not limited to leukocytes, CRP, PCT, albumin and prealbumin. Cytokines, especially IL-1, IL-6 and TNF-alpha, [21, 43] should also be collected and analyzed as candidate variables. Therefore, it is necessary to further design prospective interventional clinical trials to confirm our findings.
This study has several limitations. First, we constructed a prediction model based on pan-cancer to explore the relationship between nutritional indicators and pan-cancer prognosis. However, different types of cancer have different prognoses. In our study, no significant prognostic difference was found between different cancer types based on the site of the primary tumor. The main reason for this phenomenon is that we tended to enroll patients at a later stage. A prognostic model based on nutritional indicators according to cancer type may be a better choice, but a sufficient sample size is needed for each cancer type. Second, this study did not consider possible differences in survival due to different treatment regimens. Third, despite all our efforts to improve the quality and quantity of the data, missing data were inevitable. However, the proportion of missing data was low, so we did not impute the missing data.