DOI: https://doi.org/10.21203/rs.3.rs-1716282/v1
Background: This study analyzed the influencing factors of malnutrition in patients with gastric cancer three months after operation, and established a multi-dimensional risk model to predict postoperative malnutrition.
Methods: The clinical data of gastric cancer patients hospitalized for the first time and receiving laparoscopic surgery in the general surgery department of our hospital were retrospectively analyzed through the hospital information system, and divided into training set and validation set in the ratio of 7:3. Nutritional status was assessed using the Patient Generated Subjective Global Assessment scale and follow-up records of patients three months after surgery. They were divided into: non-malnutrition group and malnutrition group. A risk prediction model was established and displayed in the form of a nomogram.
Results: In this study, 344 patients were included, with 242 in the training and 102 in the validation set. The incidence of malnutrition three months after operation in the training and validation set was 43.4% and 38.2%, respectively. tumor node metastasis stage, cardiac function class as per the New York Heart Association, prealbumin, neutrophil-to-lymphocyte ratio, and classification of anxiety were independent risk factors. The prediction model was constructed with the above five variables. The area under the curve (AUC) values of the model for predicting malnutrition three months after surgery in gastric cancer patients was better than the AUC value of the Nutrition Risk Screening 2002 in predicting the effect, implying that the model possessed good discrimination. The model’s calibration and clinical applicability were also confirmed using the calibration curve and the Hosmer-Lemeshow goodness-of-fit test, and the clinical decision curve, respectively.
Conclusions: A clinical prediction model including multi-dimensional variables was established based on the independent risk factors of postoperative malnutrition in patients with gastric cancer. The model yields greater prediction accuracy of the risk of postoperative malnutrition in patients with gastric cancer, helps screen high-risk patients, formulates targeted nutritional prescriptions early, improves the overall prognosis of patients with gastric cancer, and increases the survival rate.
Gastric cancer, as a more common clinical gastrointestinal tumor, is in the top five tumor-related diseases in terms of the number of incidents; moreover, according to epidemiological surveys, it is the third most common cause of tumor-related death [1]. Although surgery as an essential treatment for gastric cancer has reduced the mortality rate, surgical treatment cannot eliminate the emergence of gastric cancer-related complications, which are crucial factors influencing the prognosis of gastric cancer patients. Gastric cancer has a considerable influence on the patient's appetite, digestion, and absorption functions and can induce a decrease in the patient's intake. Additionally, the resting energy expenditure of patients with gastric cancer increases, and both are mutually causal. This increase promotes patients' demand for energy. However, the feedback mechanism between the resting energy expenditure and food intake is destroyed owing to the decrease in the appetite of patients with gastric cancer; consequently, their food intake does not increase, resulting in a significant negative energy balance [2]. Additionally, factors such as continuous stress, chronic inflammation, and consumptive metabolic disorder can impact the nutritional status of patients with gastric cancer [3, 4]. Finally, surgery and chemotherapy are the mainstream treatment methods. Surgical resection of a part or all of the stomach causes not only trauma to the body but also further decline or loss of gastric function. Nearly 50% of patients with gastric cancer are in a state of malnutrition after the operation, and their average weight is reduced by 10–20% compared with that before the operation [5]. Chemotherapy leads to severe gastrointestinal reactions and immunosuppression, affecting the nutritional status of patients [6]. From the perspective of pathophysiology, the nutritional status of patients is closely associated with material metabolism, functional operation of important organs, immune response, and cell membrane stability. Clinically, postoperative malnutrition induces an increase in the incidence of complications such as infection, difficult wound healing, and anastomotic fistula, which aggravate malnutrition and form a vicious circle. Furthermore, about 40% of patients with gastric cancer die from postoperative malnutrition rather than their treatment factors. Postoperative malnutrition is positively correlated with recurrence rate, disease-free survival rate, and overall survival rate [7]. It is currently believed that malnutrition after gastric cancer surgery is one of the main factors influencing the prognosis of gastric cancer patients in addition to surgery.
The current methods of evaluating postoperative nutritional status mostly take 2‒3 months after the operation as the evaluation time point, presenting a time lag. The opportunity for intervention is severely delayed, though positive judgment is made. Moreover, postoperative malnutrition of gastric cancer involves many mechanisms, and a single index cannot fully explain it. Therefore, a multi-dimensional risk prediction model of postoperative malnutrition was established using the relevant perioperative indicators based on the occurrence mechanism of postoperative malnutrition in gastric cancer to achieve early warning for high-risk patients and provide theoretical support for early intervention. This can improve the postoperative nutritional status of gastric cancer patients, reduce complications, increase survival rate, and result in more clinical benefits.
A total of 345 patients who received gastric cancer surgery from January 2019‒December 2021 in the General Surgery Department of the Second Hospital of Anhui Medical University were selected as research participants. Inclusion criteria were as follows: 1) laparoscopic surgery was successfully performed and postoperative pathology was confirmed as gastric cancer; and 2) the patient's hospitalization data and follow-up data of three months after the operation were complete and available for review under the hospital information system (HIS). Exclusion criteria were as follows: 1) patients younger than 30 years old or older than 80 years old; 2) presence of severe liver and kidney dysfunction; 3) presence of diseases of the blood system or rheumatic immune system; 4) patients with distant tumor metastasis; 5) patients who died within three months after surgery (Fig. 1).
Patient data were extracted from HIS database. This study was approved by the Ethics Commission of the hospital (The Second Hospital of Anhui Medical University).
The effective sample size in predictive studies (modeling and validation) is determined by the number of outcome events. There must be at least 10 positive outcome events per variable to guarantee accuracy and feasibility [8]. According to the incidence of malnutrition after gastric cancer surgery reported in previous studies[9], clinical data of at least 200 patients were required to construct the model; thus, seven or fewer predictors could enter the model in the final multivariate logistic regression model. The sample size of the model established in this study was 242, and the number of positive results was 105, all of which significantly exceeded the events per variable method and could provide a reliable assessment.
≪Insert Fig. 1 here≫
This was a retrospective case-control study. Target data included general data, perioperative clinical data, and follow-up data recorded by the HIS, pertaining to gastric cancer patients who were hospitalized for the first time and who completed laparoscopic surgery in the general department of our hospital from January 2019‒December 2021.
The following general and perioperative data were collected: general data comprising gender, age, body mass index (BMI), and mid-arm muscle circumference (MAMC) using four items; history information comprising history of smoking, alcohol consumption, hypertension, and diabetes mellitus using four items; clinical data comprising pain grading on admission, anxiety score on admission, self-care grading on admission, cardiac function grading (NYHA), pulmonary function grading, surgical approach, TNM stage of the tumor, degree of differentiation, postoperative chemotherapy, postoperative complications, and nasal feeding within 48 hours after surgery using 12 items; and laboratory results comprising hemoglobin, neutrophil-to-lymphocyte ratio (NLR), albumin, prealbumin (PAB), Aspartate aminotransferase (AST), Alanine aminotransferase (ALT), creatinine, urea nitrogen, triglycerides, total cholesterol, serum sodium, serum potassium, C-reactive protein, carcinoembryonic antigen and glycosylated hemoglobin using 15 items. The above indicators were used as independent variables, and a prediction model was constructed after being screened using statistical indicators and professional theories relating to the pathophysiological mechanism of postoperative malnutrition in patients with gastric cancer.
Follow-up data of three months after operation were divided into two groups (non-malnutrition group and malnutrition group) based on whether postoperative malnutrition occurred. To evaluate postoperative malnutrition, nutritional status evaluation was performed using the Patient Generated Subjective Global Assessment scale, recommended by the American Dietetic Association for the assessment of nutritional status in oncology patients. This scale divides nutritional status into three levels: good nutritional status (0–3 points), moderate malnutrition (4–8 points), and severe malnutrition (> 8 points). In this scale, a score greater than or equal to 4 is classified as malnutrition, and less than 4 is classified as non-malnutrition. Nutritional status of patients was assessed after surgery based on the peak period of malnutrition in patients at three months after gastric cancer. Moreover, patients needed to be routinely reviewed and were easily evaluated at this time [10]. This index was adopted as the dependent variable (outcome variable) for the construction of the prediction model.
The scores of the Adult Nutrition Risk Screening Scale (NRS2002) on admission were collected. NRS2002 was published by the European Society of Parenteral and Enteral Nutrition in 2002 [11]. It can prospectively and dynamically judge changes in the nutritional status of patients, presenting appropriate validity and reliability.
The scale comprises three parts: the score of the impact of disease on nutritional status (the highest score was 3), score of impaired nutritional status (the highest score was 3), and a score for age (1 point for greater than or equal to 70 years old). The total score is the sum of three items. A score greater than or equal to 3 indicates nutritional risk; a score less than 3 indicates no nutritional risk. This scale, as a classic tool for judging nutritional risk, is frequently used in cancer patients. It was employed in this study to compare the prediction efficiency with the constructed prediction model, and to test and identify the discrimination of the model.
First, history of hypertension was determined according to the diagnosis certificate of the second-level or higher hospital or on meeting the relevant standards of the “ 2014 Evidence-Based Guideline for the Management of High Blood Pressure in Adults (Eighth Joint National Committee).” Second, history of diabetes was determined following the diagnosis certificate of secondary or higher hospital or meeting the relevant standards of “Standards of Medical Care in Diabetes-2020” issued by the American Diabetes Association. Third, history of smoking was determined based on the average daily smoking being greater than or equal to one cigarette for at least one year. Fourth, history of drinking was determined based on the average ethanol intake being more than 40g/d for at least five years. Fifth, surgical method was divided into total gastrectomy, subtotal gastrectomy, and distal gastrectomy based on the disease. Sixth, TNM staging of gastric cancer was determined based on the TNM staging standard of The Union for International Cancer Control / American Joint Committee on Cancer Staging in 2010. Seventh, postoperative complications were composed of abdominal infection, anastomotic leakage, anastomotic bleeding, aspiration pneumonia, and deep vein thrombosis.
First, BMI was divided into the following four grades: <18kg/m2, 18–24kg/m2, 24–27.5kg/m2, > 27.5kg/m2. Second, regarding MAMC, for men, it was divided into < 25.3cm and ≥ 25.3cm, whereas for women, it was divided into < 23.2cm and ≥ 23.2cm. Third, anxiety score on admission was determined based on the Self-Rating Anxiety Scale (SAS) performed during admission; this scale was prepared by William W.K. Zung and divides respondents into four grades: normal (< 50 points), mild anxiety (50–59 points), moderate anxiety (60–69 points), and severe anxiety (> 70 points). Fourth, pain classification on admission was divided into four grades (no pain, mild pain, moderate pain, and severe pain) according to the World Health Organization standards. Fifth, grade of self-care ability of admission was categorized into five grades following the Barthel index: independent (100 points), mild dependence (91–99 points), moderate dependence (61–90 points), severe dependence (21–60 points), and complete dependence (0–20 points). Sixth, cardiac function classification was divided into four grades (I, II, III, and IV) based on NYHA's heart failure classification. Seventh, pulmonary function classification was divided into four grades considering the percentage of residual gas volume and total lung volume; these grades were more than 80% (normal), 65–79% (slight reduction), 50–64% (moderate reduction), and 35–49% (severe reduction). Eighth, degree of differentiation was divided into three grades based on the postoperative pathological results; these grades were low differentiation, moderate differentiation, and high differentiation. Finally, the situation of postoperative chemotherapy was divided into three grades: no chemotherapy, incomplete chemotherapy, and complete chemotherapy.
The original data were processed using SPSS 25.0 statistical software. If the measurement data exhibited a normal distribution, they were expressed as x¯± s , and the comparison between groups was performed by t-test; if the measurement data presented a non-normal distribution, the median [M (P25, P75)] was adopted, and the Mann-Whitney U test was conducted for comparison between groups. Count or categorical data were expressed as n (%), and comparisons between groups were performed with the chi-square test or Fisher's exact test.
The training set data were employed to establish the model. The nutritional status of patients three months after the operation was considered the dependent variable (non-malnutrition, malnutrition, dichotomous variables). The variables with a statistical difference by simple correlation analysis (P < 0.05) combined with the variables considered significant for the outcome by clinical professionals were taken as the independent variables. With the SPSS 25.0 statistical software, multivariate logistic regression analysis was included, and the forward stepwise method was adopted for analysis. Following the variables with statistical significance (P < 0.05) in multivariate analysis, the influence of each factor on the outcome was determined by the odds ratio (OR), and the independent variables for constructing the risk prediction model were finally determined in combination with clinical practice. A nomogram for the display and application of the predictive model was drawn with the “rms” package in the R language (R 3.6.1) software.
The data of the training set and verification set were employed for internal and external verification of the model. The receiver operating characteristic (ROC) curve of the model was depicted using the “pROC” package in the R language (R 3.6.1) software. The area under curve (AUC) of the training and validation sets was calculated and compared with the prediction performance of the NRS2002 scale to test the discrimination of the model. The “rms” package was adopted to draw a calibration curve, and the model calibration was evaluated using the calibration curve and the Hosmer-Lemeshow goodness-of-fit test. The clinical suitability was evaluated by plotting the clinical decision curve using the “rmda” package model. The internal verification was realized by bootstrap self-sampling 1000 times. P < 0.05 was taken to indicate statistically significant differences.
In the training set, 242 patients were evaluated according to the Patient Generated Subjective Global Assessment scale, and three months after surgery was selected as the evaluation time. Among them, 105 patients developed malnutrition after surgery, and the incidence rate was 43.4%. The patients were further divided into the malnutrition group (105 cases) and non-malnutrition group (137 cases) following their nutritional status three months after the operation. Additionally, the differences in general and clinical data between the two groups were tested. The results demonstrated statistical differences (P < 0.05) in age, TNM stage of the tumor, cardiac function classification (NYHA), carcinoembryonic antigen, albumin, PAB, hemoglobin, NLR, triglycerides, nasal feeding within 48 hours after the operation, and anxiety scores, as detailed in Table 1.
|
Total |
Non-malnutrition group N = 137 |
Malnutrition group N = 105 |
Z / t / χ² |
p |
|
|---|---|---|---|---|---|
|
Gender (n/%) |
0.242 |
0.623 |
|||
|
Male |
179 (73.97) |
103 (75.18) |
76 (72.38) |
||
|
Female |
63 (26.03) |
34 (24.82) |
29 (27.62) |
||
|
Age (n/%) |
32.618 |
<0.001 |
|||
|
༜60 |
88 (36.36) |
71 (51.82) |
17 (16.19) |
||
|
≥ 60 |
154 (63.64) |
66 (48.18) |
88 (83.81) |
||
|
BMI (n/%) |
4.053 |
0.303 |
|||
|
༜18 |
3 (1.24) |
0 (0.00) |
3 (2.86) |
||
|
18–24 |
142 (58.68) |
82 (59.85) |
60 (57.14) |
||
|
24–27.5 |
66 (27.27) |
38 (27.74) |
28 (26.67) |
||
|
༞27.5 |
31 (12.81) |
17 (12.41) |
14 (13.33) |
||
|
MAMC (n/%) |
1.096 |
0.295 |
|||
|
Abnormal |
92 (38.02%) |
56 (40.88%) |
36 (34.29%) |
||
|
Normal |
150 (61.98%) |
81 (59.12%) |
69 (65.71%) |
||
|
Smoking history (n/%) |
3.057 |
0.080 |
|||
|
No |
133 (54.96) |
82 (59.85) |
51 (48.57) |
||
|
Yes |
109 (45.04) |
55 (40.15) |
54 (51.43) |
||
|
History of diabetes (n/%) |
1.066 |
0.302 |
|||
|
No |
194 (80.17) |
113 (82.48) |
81 (77.14) |
||
|
Yes |
48 (19.83) |
24 (17.52) |
24 (22.86) |
||
|
History of hypertension (n/%) |
0.000 |
0.986 |
|||
|
No |
173 (71.49) |
98 (71.53) |
75 (71.43) |
||
|
Yes |
69 (28.51) |
39 (28.47) |
30 (28.57) |
||
|
Drinking history (n/%) |
1.088 |
0.297 |
|||
|
No |
143 (59.09) |
77 (56.20) |
66 (62.86) |
||
|
Yes |
99 (40.91) |
60 (43.80) |
39 (37.14) |
||
|
Pulmonary function classification (n/%) |
1.723 |
0.632 |
|||
|
1 |
96 (39.67) |
55 (40.15) |
41 (39.05) |
||
|
2 |
104 (42.98) |
60 (43.80) |
44 (41.90) |
||
|
3 |
38 (15.70) |
21 (15.33) |
17 (16.19) |
||
|
4 |
4 (1.65%) |
1 (0.73) |
3 (2.86) |
||
|
Cardiac function classification (n/%) |
8.032 |
0.045 |
|||
|
I |
86 (35.54) |
58 (42.34) |
28 (26.67) |
||
|
II |
107 (44.21) |
57 (41.61) |
50 (47.62) |
||
|
III |
45 (18.60) |
21 (15.33) |
24 (22.86) |
||
|
IV |
4 (1.65) |
1 (0.73) |
3 (2.86) |
||
|
Pain grading on admission (n/%) |
7.029 |
0.071 |
|||
|
0 |
149 (61.57) |
92 (67.15) |
57 (54.29) |
||
|
1 |
52 (21.49) |
28 (20.44) |
24 (22.86) |
||
|
2 |
27 (11.16) |
13 (9.49) |
14 (13.33) |
||
|
3 |
14 (5.79) |
4 (2.92) |
10 (9.52) |
||
|
Anxiety classification for admission (n/%) |
14.536 |
0.001 |
|||
|
0 |
95 (39.26) |
68 (49.64) |
27 (25.71) |
||
|
1 |
99 (40.91) |
45 (32.85) |
54 (51.43) |
||
|
2 |
48 (19.83) |
24 (17.52) |
24 (22.86) |
||
|
Grading of self-care ability on admission (n/%) |
1.084 |
0.781 |
|||
|
0 |
92(38.02) |
52(37.96) |
42(40.00) |
||
|
1 |
103(42.56) |
59(43.07) |
44(41.90) |
||
|
2 |
32(13.22) |
17(12.41) |
15(14.29) |
||
|
3 |
13(5.37) |
9(6.57) |
4(3.81) |
||
|
Surgical approach (n/%) |
2.064 |
0.356 |
|||
|
Whole stomach |
146 (60.33) |
78 (56.93) |
68 (64.76) |
||
|
Large stomach |
37 (15.29) |
21 (15.33) |
16 (15.24) |
||
|
Distal stomach |
59 (24.38) |
38 (27.74) |
21 (20.00) |
||
|
TNM staging (n/%) |
19.578 |
<0.001 |
|||
|
I |
60 (24.79) |
45 (32.85) |
15 (14.29) |
||
|
II |
86 (35.54) |
45 (32.85) |
41 (39.05) |
||
|
III |
57 (23.55) |
21 (15.33) |
36 (34.29) |
||
|
IV |
39 (16.12) |
26 (18.98) |
13 (12.38) |
||
|
Degree of differentiation (n/%) |
2.777 |
0.249 |
|||
|
Low differentiation |
98 (40.50) |
50 (36.50) |
48 (45.71) |
||
|
Medium differentiation |
111 (45.87) |
65 (47.45) |
46 (43.81) |
||
|
High differentiation |
33 (13.64) |
22 (16.06) |
11 (10.48) |
||
|
C-reactive protein |
5.75 [3.30;11.97] |
5.50 [3.30;10.30] |
5.90 [3.50;17.60] |
0.056 |
0.056 |
|
Glycosylated hemoglobin |
5.50 [4.90;6.10] |
5.60 [4.90;6.10] |
5.50 [4.90;5.80] |
-0.853 |
0.394 |
|
Carcinoembryonic antigen |
3.29 [1.61;6.72] |
2.91 [1.53;5.66] |
3.77 [1.89;13.26] |
-2.775 |
0.006 |
|
Alb |
37.15 [33.00;39.90] |
38.30 [34.70;40.70] |
35.40 [31.10;38.10] |
-4.579 |
<0.001 |
|
Prealbumin |
220.00 [176.50;263.00] |
245.00 [205.00;279.00] |
186.00 [157.00;214.00] |
-7.657 |
<0.001 |
|
ALT |
18.00 [12.00;28.00] |
19.00 [13.00;28.00] |
17.00 [11.00;27.00] |
-1.268 |
0.205 |
|
AST |
21.00 [17.00;31.00] |
20.00 [16.00;30.00] |
24.00 [18.00;32.00] |
-1.934 |
0.053 |
|
Creatinine |
65.00 [53.00;78.00] |
64.00 [55.00;76.00] |
65.00 [51.00;81.00] |
-0.200 |
0.841 |
|
Urea nitrogen |
5.54 [4.52;6.91] |
5.57 [4.70;6.82] |
5.42 [4.12;7.45] |
-0.809 |
0.419 |
|
Triglyceride |
1.07 [0.77;1.51] |
1.30 [0.98;1.69] |
0.90 [0.69;1.18] |
-6.220 |
<0.001 |
|
Serum sodium |
140.90 [139.40;143.07] |
141.30 [139.70;143.00] |
140.50 [138.90;143.20] |
-1.825 |
0.068 |
|
Serum potassium |
4.02 [3.72;4.32] |
4.01 [3.73;4.30] |
4.05 [3.70;4.34] |
-0.525 |
0.599 |
|
Hemoglobin |
108.00 [82.25;130.00] |
113.00 [87.00;132.00] |
103.00 [73.00;127.00] |
-2.400 |
0.016 |
|
Neutrophils/lymphoc |
2.54 [1.77;4.21] |
2.48 [1.58;3.58] |
2.92 [1.92;5.39] |
-2.680 |
0.007 |
|
Postoperative chemotherapy (n/%) |
2.271 |
0.132 |
|||
|
No |
75 (31.12) |
48 (35.04) |
27 (25.96) |
||
|
Yes |
166 (68.88) |
89 (64.96) |
77 (74.04) |
||
|
Postoperative complications (n/%) |
1.220 |
0.269 |
|||
|
No |
210 (86.78) |
116 (84.67) |
94 (89.52) |
||
|
Yes |
32 (13.22) |
21 (15.33%) |
11 (10.48) |
||
|
Nasal feeding within 48 hours after the operation |
3.957 |
0.047 |
|||
|
No |
65 (26.86) |
30 (21.90) |
35 (33.33) |
||
|
Yes |
177 (73.14) |
107 (78.10) |
70 (66.67) |
≪Insert Table 1 here≫
The indicators with statistical differences in the difference test (see the above) and the indicators clinically exerting an impact on the outcome (classification of self-care ability, postoperative complications, and postoperative chemotherapy) were used as independent variables. The nutritional status (non-malnutrition and malnutrition) of patients with gastric cancer three months after surgery was used as the dependent variable for multivariate logistic regression analysis. The results suggested that TNM stage of tumor, cardiac function classification (NYHA), NLR, and anxiety classification were independent risk factors for the development of postoperative malnutrition in gastric cancer patients (P < 0.05, OR > 1), whereas PAB was an independent protective factor (P < 0.05, 0 < OR < 1). There was no statistical collinearity between the variables (TOL < 1, 1 < VIF < 10). The TNM stage of the tumor, cardiac function classification, prealbumin, NLR, and anxiety of admission were adopted for classification and the construction of a risk prediction model for malnutrition after gastric cancer surgery. Details are provided in Table 2 and Fig. 2.
The multivariate logistic regression equation is as follows:
P = Y/1 + Y and P > 0.5 implied the occurrence of postoperative malnutrition.
|
β |
SD |
Wald |
P |
OR |
95% CI |
TOL |
VIF |
||
|---|---|---|---|---|---|---|---|---|---|
|
Lower Upper |
|||||||||
|
TNM Stage |
0.330 |
0.161 |
4.205 |
0.040 |
1.391 |
1.015 |
1.906 |
0.928 |
1.077 |
|
NYHA |
0.470 |
0.205 |
5.244 |
0.022 |
1.601 |
1.070 |
2.394 |
0.990 |
1.010 |
|
PAB |
-0.020 |
0.003 |
36.376 |
༜0.001 |
0.980 |
0.974 |
0.987 |
0.933 |
1.072 |
|
NLR |
0.155 |
0.053 |
8.687 |
0.003 |
1.168 |
1.053 |
1.295 |
0.969 |
1.032 |
|
Anxiety Rating |
0.562 |
0.224 |
6.314 |
0.012 |
1.754 |
1.132 |
2.719 |
0.918 |
1.089 |
|
Constants |
2.117 |
0.824 |
6.605 |
0.010 |
8.305 |
||||
≪Insert Fig. 2 here≫
The five independent variables in the above prediction model (TNM stage of tumor, cardiac function classification (NYHA), prealbumin, NLR, and anxiety grade) were employed to construct a nomogram of the prediction model, as illustrated in Fig. 3.
To interpret the nomogram, a vertical line was drawn on the horizontal axis where each independent variable index of a patient was located, and the value on the horizontal axis of the corresponding “Point” was a specific score. The scores corresponding to the five independent variables were added to obtain the total score, and then a vertical line was drawn downward. The predicted value on the horizontal axis corresponding to “Risk” was the predicted value of the risk of the patient.
≪Insert Fig. 3 here≫
The difference test revealed no statistical difference in the general data and perioperative data between the training and the validation sets (P > 0.05), confirming that the two data sets were comparable (Table 3).
Table 3
Comparison of general and clinical data between the training set and validation set
|
|
Training set N=242 |
Validation set N=102 |
Z / t / χ² |
p |
|
Gender (n/%) |
|
|
1.425 |
0.233 |
|
Male |
179 (73.97) |
69(67.65) |
|
|
|
Female |
63 (26.03) |
33(32.35) |
|
|
|
Age (n/%) |
|
|
0.787 |
0.375 |
|
＜60 |
88 (36.36) |
32(31.37) |
|
|
|
≥60 |
154 (63.64) |
70(68.63) |
|
|
|
BMI (n/%) |
|
|
4.179 |
0.243 |
|
＜18 |
3(1.24) |
3(2.94) |
|
|
|
18–24 |
142(58.68) |
50(49.02) |
|
|
|
24–27.5 |
66(27.27) |
30(29.41) |
|
|
|
＞27.5 |
31(12.81) |
19(18.63) |
|
|
|
MAMC (n/%) |
|
|
3.584 |
0.058 |
|
Abnormal |
92(38.02) |
50(49.02) |
|
|
|
Normal |
150(61.98) |
52(50.98) |
|
|
|
Smoking history (n/%) |
|
|
0.709 |
0.400 |
|
No |
133(54.96) |
51(50) |
|
|
|
Yes |
109(45.04) |
51(50) |
|
|
|
History of diabetes (n/%) |
|
|
1.853 |
0.173 |
|
No |
194(80.17) |
75(73.53) |
|
|
|
Yes |
48(19.83) |
27(26.47) |
|
|
|
History of hypertension (n/%) |
|
|
1.778 |
0.182 |
|
No |
173(71.49) |
80(78.43) |
|
|
|
Yes |
69(28.51) |
22(21.57) |
|
|
|
Drinking history (n/%) |
|
|
0.949 |
0.330 |
|
No |
143(59.09) |
66(64.71) |
|
|
|
Yes |
99(40.91) |
36(35.29) |
|
|
|
Pulmonary function classification (n/%) |
|
|
6.683 |
0.083 |
|
1 |
96(39.67) |
45(44.12) |
|
|
|
2 |
104(42.98) |
41(40.2) |
|
|
|
3 |
38(15.7) |
10(9.8) |
|
|
|
4 |
4(1.65) |
6(5.88) |
|
|
|
Cardiac function classification (n/%) |
|
|
1.747 |
0.627 |
|
I |
86(35.54) |
30(29.41) |
|
|
|
II |
107(44.21) |
50(49.02) |
|
|
|
III |
45(18.6) |
19(18.63) |
|
|
|
IV |
4(1.65) |
3(2.94) |
|
|
|
Anxiety classification for admission (n/%) |
|
|
1.927 |
0.179 |
|
0 |
95(39.26) |
36(35.29) |
|
|
|
1 |
99(40.91) |
52(50.98) |
|
|
|
2 |
48(19.83) |
14(13.73) |
|
|
|
Pain grading on admission (n/%) |
|
|
1.198 |
0.754 |
|
0 |
149(61.57) |
61(59.8) |
|
|
|
1 |
52(21.49) |
19(18.63) |
|
|
|
2 |
27(11.16) |
15(14.71) |
|
|
|
3 |
14(5.79) |
7(6.86) |
|
|
|
Grading of NRS on admission (n/%) |
|
|
2.037 |
0.153 |
|
＜3 |
184(76.03) |
70(68.63) |
|
|
|
≥3 |
58(23.97) |
32(31.37) |
|
|
|
Grading of self-care ability on admission (n/%) |
|
|
2.750 |
0.432 |
|
0 |
92(38.02) |
45(44.12) |
|
|
|
1 |
103(42.56) |
42(41.18) |
|
|
|
2 |
32(13.22) |
8(7.84) |
|
|
|
3 |
13(5.37) |
7(6.86) |
|
|
|
Alb |
37.15 [33.00;39.90] |
37.20[34.65;39.50] |
0.490 |
0.624 |
|
Prealbumin |
220.00 [176.50;263.00] |
225.50[183.00;266.00] |
0.747 |
0.455 |
|
Neutrophils/lymphoc |
2.55[1.76;4.23] |
2.63 [1.67;4.35] |
0.177 |
0.860 |
|
Postoperative complications (n/%) |
|
|
3.776 |
0.052 |
|
No |
210(86.78) |
80(78.43) |
|
|
|
Yes |
32(13.22) |
22(21.57) |
|
|
|
Nasal feeding within 48 hours after the operation |
|
|
0.445 |
0.505 |
|
No |
65(26.86) |
31(30.39) |
|
|
|
Yes |
177(73.14) |
71(69.61) |
|
|
≪Insert Table 3 here≫
The predictive performance (discrimination) of the model was evaluated by plotting the ROC curves of the training and validation sets. In predicting the risk of postoperative malnutrition in gastric cancer patients, the AUC of the model in the training and validation sets was 0.842 (95%CI: 0.790–0.886) and 0.815 (95%CI: 0.790–0.886), respectively. Both values were better than those of the NRS2002 (AUC: 0.794, 95%CI: 0.738–0.844) (Fig. 4).
≪Insert Fig. 4 here≫
The accuracy of the prediction model was evaluated by drawing a calibration curve. The model was visually observed in the training and validation sets. The curves between the predicted outcome and actual outcome were highly consistent. Moreover, the Brier values of curves of the training and validation sets were 0.161 and 0.195, respectively, verifying that the model had high prediction accuracy. Concurrently, goodness-of-fit tests revealed no statistically significant deviation between the predicted and actual values of risk for the training (χ² = 14.070, P = 0.08) and validation sets (χ² = 1.989, P = 0.98).
≪Insert Fig. 5 here≫
The clinical applicability of the model was evaluated by drawing a clinical decision curve analysis (DCA). The occurrence of malnutrition after gastric cancer surgery was taken as a state variable, and the predicted value of the risk obtained from the nomogram was used as a test variable. Then, a DCA curve was drawn. The DCA results suggested that the net benefit of the model with a threshold probability in the interval of 0.1–0.7 was better when all participants were either positive or negative, as indicated in Fig. 6.
≪Insert Fig. 6 here≫
Malnutrition indicates chronic nutritional deficiencies in the body caused by insufficient food intake, impaired digestion and absorption, and excessive wastage during the process of ingesting nutrients. It is generally divided into four categories: primary malnutrition (chronic starvation, inadequate intake), secondary malnutrition (significant stress, disease, and excessive nutrient depletion), age-related malnutrition (frailty, sarcopenia), and mixed malnutrition. The risk of malnutrition elucidates the impact of undernutrition on adverse clinical events or outcomes in patients. Huang et al. retrospectively analyzed 597 older adult patients (> 60 years old) who underwent radical gastrectomy for gastric cancer. Approximately, 34.5% of the elderly patients presented postoperative malnutrition [12]. Skeie et al. retrospectively analyzed 6110 patients in Norway's National Gastric Cancer Surgery Registry using Global Leadership Initiative on Malnutrition (GLIM) criteria. They discovered that 35.4% of the patients had postoperative malnutrition, of whom 15.6% were severely malnourished [13]. Thus, the nutritional risk of patients after gastric cancer surgery is high, and they are prone to malnutrition. The reasons stem from two main aspects. The first is the influence of gastric cancer itself on the nutritional status of the body. Specifically, 1) gastric cancer affects the function of the stomach as a critical digestive organ, and severe gastric cancer damages the normal structure of the digestive tract; 2) cancer, as a consumption disease, accelerates the basal metabolism and nutrient consumption of patients; 3) the pain and psychological impact induced by gastric cancer leads to decreased appetite and intake of patients; and 4) the influence of gastric cancer on the body's microenvironment and intestinal flora causes the patient's absorption disorder. The second reason is related to the impact of surgery on nutritional status, as: 1) surgical removal of part or all of the gastric tissue inevitably harms the patient’s digestive function; 2) surgical stress and trauma worsen the patient’s psychological burden; 3) postoperative symptoms, such as abdominal distension and nausea, lead to loss of appetite; 4) postoperative chemotherapy and complications result in increased nutritional consumption and decreased intake, and boost incidences of malnutrition. Additionally, patients with malnutrition after gastric cancer surgery generally develop adverse body states such as low tissue protein levels, impaired immune function, and internal environment disturbances, which are prone to severe complications such as postoperative infection and poor anastomotic healing [14]. Malnourished patients have a decreased willingness and ability to exercise postoperatively, resulting in muscle atrophy and decreased intestinal motility [15]. All of the above tend to aggravate the malnutrition state of the patient, forming a vicious circle. Malnutrition is closely related to the survival rate and quality of life of patients with gastric cancer after surgery. Xiao et al. reported that nutritional deficiencies after gastric cancer surgery directly impact the treatment effect of patients and can easily lead to adverse clinical outcomes [16]. Hirahara et al. revealed that the peak period of malnutrition occurred three months after gastric cancer surgery [10]. Therefore, a comprehensive analysis of related clinical research and pathogenesis demonstrates that patients with gastric cancer have a higher risk of malnutrition after surgery and are prone to malnutrition, which in turn affects their mid-term and long-term prognosis. At present, the assessment of malnutrition in patients after gastric cancer surgery has remarkable lag and insufficient accuracy. Consequently, it is difficult to effectively screen high-risk patients during hospitalization or the perioperative period to achieve early warning and intervention.
The indicators affecting nutritional status were used as independent variables, and the nutritional status (whether or not malnutrition occurs) three months after surgery was the dependent variable. First, the difference variables were obtained by a simple difference test. Subsequently, variables statistically different and clinically believed to have a theoretical impact on the outcome were incorporated into multivariate logistic regression in combination with clinical theory. Finally, the variables for building the predictive model were derived. Statistical results and clinical mechanisms were considered in the entire model construction process. The variables in the model were not only those with relevant statistical results (all are independent risk factors for malnutrition after gastric cancer surgery, as determined by P < 0.05 using multivariate analysis), but also those that have a direct theoretical relationship with the outcome of malnutrition.
The TNM stage of gastric cancer was adopted as a standard to evaluate the degree of infiltration and metastasis of the tumor itself. A higher stage indicated a higher degree of malignancy. In patients with high TNM staging, digestion and absorption dysfunction occurs due to the severe invasion of the tumor into the stomach and surrounding lymph nodes. Cancer cells compete with normal cells in the body for nutrients and consume significant amounts of energy and protein. Besides, patients with a high TNM stage usually develop accompanying symptoms such as anorexia, pain, nausea, and vomiting, resulting in insufficient intake. Such patients require wider surgical resection and longer courses of postoperative chemotherapy, increasing the risk of postoperative malnutrition. Ravasco et al. suggested that the malignant degree of gastric cancer is a critical factor affecting the nutritional status of patients, and TNM staging is an effective clinical indicator to evaluate the degree of malignancy [17]. Lee et al. demonstrated that the malignant degree of gastric cancer is closely correlated with the occurrence of postoperative malnutrition and is an independent risk factor [18].
Cardiac function classification (NYHA) is an evaluation of patients' cardiac function from the perspective of clinical symptoms, especially for patients with chronic heart failure. Kinugawa et al. suggested that patients with chronic heart failure frequently suffer from malnutrition ascribed to changes in systemic metabolism and increased body consumption, with an incidence rate of 16%‒62% [19, 20]. Patients undergoing gastric cancer surgery are more likely to suffer from insufficient body intake, loss of appetite, and increased risk of postoperative malnutrition because of reduced intake, and exercise tolerance, especially if complicated with cardiac insufficiency. Sze et al. revealed that chronic heart failure aggravates the symptoms of gastrointestinal congestion and intestinal edema in patients with gastric cancer, impacts the absorption of nutrients, and increases the occurrence of malnutrition [21].
PAB is synthesized by hepatocytes and is so named because it is generally displayed in front of albumin by electrophoresis. Owing to its short half-life of only about 12 hours, it is more sensitive than albumin and transferrin in response to malnutrition. Aoyama et al. reported that prealbumin can be used as a representative indicator of postoperative nutritional status in patients with gastric cancer and is correlated with the recurrence rate and survival rate of gastric cancer patients [22]. Zu et al. confirmed that the level of prealbumin at admission is an independent risk factor for the long-term prognosis of gastric cancer patients [23].
The NLR is an indicator of the degree of inflammation in the body. From one perspective, the more intense the inflammatory response of gastric cancer patients, the greater the consumption of nutrients, and the more likely it is to aggravate stress trauma such as surgery, increase the chance of postoperative infection, and promote the occurrence of postoperative malnutrition status [24]. From another perspective, an increase in this index indicates an increase in neutrophils and/or a decrease in lymphocytes in gastric cancer patients. Neutrophils can drive tumor growth by producing soluble cytokines and various proteases [25]. Moreover, they boost tumor metastasis by inhibiting functions such as effector T cells and NK cells. A decrease in the number of lymphocytes implies a decrease in immune function and surveillance, making it easier for the tumor to metastasize [26]. The accelerated growth or metastasis of gastric cancer as a tumor directly affecting the digestive organs will directly aggravate the consumption of nutrients and obstruct digestion and absorption function.
Regarding anxiety grading, the SAS was employed to effectively evaluate and grade whether the patient has anxiety and its severity. Negative emotions such as anxiety can further affect the appetite of patients with gastric cancer through behavioral mechanisms, resulting in insufficient intake of nutrients [27]. Moreover, a state of anxiety can induce a state of stress in gastric cancer patients through neuroendocrine pathways, such as the hypothalamic-pituitary-adrenal axis [28]. Additionally, anxiety aggravates the body's inflammatory response, inhibits immunity, and influences the recovery of postoperative nutritional status in patients with gastric cancer [29]. Anxiety is an independent risk factor affecting the nutritional prognosis of cancer patients [30].
During the process of selecting variables for modeling in this study, the selected variables were required to meet multi-dimensional requirements, in addition to considering the statistical results and the mechanism of malnutrition after gastric cancer surgery. The selected variables can reflect not only the situation of the tumor itself (TNM staging) but also the cardiac function, nutritional reserve, the body’s inflammatory response and anxiety level of gastric cancer patients through related indicators (NYHA, PAB, NLR, and SAS scale grading), and the pathophysiological state of the whole body. Finally, joint prediction of malnutrition outcomes from multiple perspectives can improve model efficiency. With respect to the model display, the nomogram visually expresses the meaning of the model and quantifies risks concisely and effectively, which is convenient for clinical use.
In the design method of this study, the included patients were divided into the training set and validation set, and the internal and external joint verification after modeling was conducted with various methods to effectively evaluate the predictive ability and extrapolation of the model. The model discrimination was evaluated by the ROC curve, and the obtained AUC values of the two sets were 0.842 and 0.815, respectively. According to the standard of AUC prediction validity, the overall discrimination of the prediction model was excellent. Compared with the commonly used NRS2002 scale (AUC: 0.794), the model had higher validity in predicting malnutrition after gastric cancer surgery. The prediction accuracy of the model was confirmed by the calibration curve (visual observation, Brier value: 0.161, 0.195) and the Hosmer-Lemeshow goodness-of-fit test (training set and validation set, P > 0.05). The DCA curve demonstrated that the predicted probability of this model was in the range of 10%‒70%, the level of clinical net benefit was the highest, and that the model possessed acceptable clinical applicability. Meanwhile, the variables in the prediction model were clinical classification, laboratory indicators, or scoring scales commonly used in the clinic, which can be quickly mastered by doctors and nurses while being popularized and used by hospitals at all levels.
First, the variables in the model had a theoretical causal relationship with the outcome of malnutrition. Second, the pathophysiological states of the body represented by the cardiac function classification (NYHA), preoperative nutritional reserve index (PAB), inflammation degree (NLR), and anxiety state (SAS classification) in the model variables were clinically interventionable. The corresponding state was improved by lowering the cardiac load, increasing the preoperative nutritional reserve, reducing the body’s inflammatory response, and relieving anxiety, contributing to the increased risk of postoperative malnutrition. Therefore, the prediction model can provide not only early warning for high-risk patients with malnutrition after gastric cancer surgery but also theoretical support and practical guidance for early intervention in high-risk patients.
However, this study, as a retrospective analysis of a single center, still has some limitations. First, the number of cases is small, and the source is limited. Second, the established model needs to be verified by an in-depth prospective cohort study. Third, the selected indicators did not include an evaluation of the patients' families, including socioeconomic status, types of meals, and eating habits. Therefore, we expect to conduct a multi-center prospective study incorporating more innovative indicators to further improve the predictive performance of the model.
In this study, the statistical method of multivariate regression was used to identify the risk factors and establish the prediction model. In the selection of variables, the pathogenesis and clinical intervention of postoperative malnutrition of gastric cancer were considered. Moreover, a variety of statistical strategies were used to verify the model internally and externally; the results proved that the model has good prediction efficiency and clinical scalability, can effectively predict the occurrence of postoperative malnutrition in patients with gastric cancer, while guiding early intervention plans.
ALT Alanine Aminotransferase
AST Aspartate Aminotransferase
AUC Area Under Curve
BMI Body Mass Index
DCA Decision Curve Analysis
GLIM Global Leadership Initiative on Malnutrition
HIS Hospital Information System
MAMC Mid-Arm Muscle Circumference
NLR Neutrophil-to-Lymphocyte Ratio
NRS2002 Nutrition Risk Screening Scale
NYHA New York Health Association
OR Odds Ratio
PAB Prealbumin
ROC Receiver Operating Characteristic
SAS Self-Rating Anxiety Scale
TNM Tumor Node Metastasis
Ethical approval and consent to participate
This study is based on the information collected from the database of The Second Hospital of Anhui Medical University. The establishment of the database was approved by the hospital research ethics committee. This study is a retrospective analysis. Each case has a notification and consent form for the use of clinical data.
Consent for publication
Not applicable.
Availability of data and materialsThe datasets used and/or analyzed during the current study are available from the first author or corresponding author on reasonable request.
Competing interestsThe authors declare that they have no competing interests.
Funding
This study was supported by the 2021 Anhui Nursing Society Scientific Research Program Key Project (AHHL a202108), 2021 Anhui Nursing Society Scientific Research Program Youth Project (AHHL a202116), and 2020 Anhui Medical University Scientific Research Fund Youth Project (2020xkj136).
Authors’ contribution
TD conceptualized the study design; performed data collection, statistical analysis, and data interpretation; and contributed to manuscript preparation, literature search, and funds collection. DW conceptualized the study design, performed statistical analysis and data interpretation, and contributed to funds collection. JT performed data collection, statistical analysis, and data interpretation. ZL performed data collection, statistical analysis, and data interpretation, and contributed to literature search. MZ conceptualized the study design, performed statistical analysis and data interpretation, and contributed to funds collection.
Acknowledgments
The authors thank Ms. Qingquan Bi, School of nursing, Anhui Medical University, for providing research ideas. The authors also thank Editage for its language editing service.