Application of multivariate logistic regression in predicting postoperative intestinal function recovery and lymph node metastasis in colorectal cancer

Abstract

BACKGROUND

Early identification of patients with colorectal cancer (CRC) at risk of lymph node metastasis (LNM) and delayed recovery of postoperative intestinal function may improve perioperative optimization, postoperative surveillance, and individualized treatment planning. Multivariate logistic regression is suitable for integrating heterogeneous perioperative variables and may provide practical support for early clinical risk stratification.

AIM

To identify risk factors and construct multivariate logistic regression models predicting delayed postoperative intestinal function recovery and LNM in CRC.

METHODS

This retrospective observational study included 197 patients with CRC who underwent radical resection. To analyze intestinal function recovery, 49 patients and 148 patients were assigned to the delayed and normal groups, respectively, based on delayed gastrointestinal function recovery. To analyze nodal status, 68 patients and 129 patients were assigned to the LNM-positive and LNM-negative groups, respectively, based on postoperative pathological findings. Univariate and multivariate analyses identified independent predictors, and area under the curve and decision curve analyses evaluated model performance.

RESULTS

The delayed group had significantly higher rates of preoperative hypoalbuminemia (30.6% vs 10.1%), blood loss > 50 mL (46.9% vs 18.2%), and T3-4 tumors (85.7% vs 58.1%) than the normal group. The delayed group also exhibited persistently higher postoperative C-reactive protein and lower albumin levels (all P < 0.05). The LNM-positive group had more poorly differentiated tumors (45.6% vs 17.1%), fewer harvested lymph nodes (< 12; 27.9% vs 10.1%), and higher fibrinogen levels than the LNM-negative group. Multivariate analysis confirmed that low albumin levels [odds ratio (OR) = 3.86], blood loss > 50 mL (OR = 2.77), and T3-4 stage (OR = 3.49) were independent risk factors for delayed gastrointestinal function recovery. Poor differentiation (OR = 4.39), < 12 lymph nodes harvested (OR = 4.10), and high fibrinogen (≥ 4 g/L, OR = 2.48) were independent risks for LNM.

CONCLUSION

The constructed multivariate logistic regression models may serve as useful adjuncts for perioperative risk stratification and individualized management, although their clinical generalizability requires validation in larger prospective multicenter studies.

Key Words: Multivariate logistic regression; Colorectal cancer; Postoperative intestinal function recovery; Lymph node metastasis; Laparoscopic surgery; Inflammation; Fibrinogen

Core Tip: Lymph node metastasis and delayed recovery of postoperative intestinal function are critical factors influencing short-term recovery and long-term prognosis in patients with colorectal cancer. Using outcome-based grouping, we constructed and validated multivariate logistic regression models that integrated readily available perioperative clinical, pathological, and laboratory variables. The models identified key independent risk factors including preoperative hypoalbuminemia, intraoperative blood loss, tumor stage, tumor differentiation, lymph node yield, and fibrinogen levels. With good discrimination and clinical benefit demonstrated by receiver operating characteristic and decision curve analyses, these models provide early risk stratification, individualized perioperative management, and optimized clinical decision-making in colorectal cancer surgery.

INTRODUCTION

Colorectal cancer (CRC) is one of the most common gastrointestinal malignancies worldwide and is associated with high morbidity and mortality[1]. Although surgical resection remains the core treatment for CRC, postoperative complications and prognosis vary significantly, and delayed recovery of intestinal function and lymph node metastasis (LNM) are key factors affecting the quality of life and long-term efficacy of patients[2,3]. Intestinal function recovery directly aids nutrition improvement to enhance the quality of life during recuperation after surgery; It accelerates the recovery process of body condition to help realise better results in post-operative treatment. Primary indicators for stage-specific LNM to help make treatment decisions and predict patients’ outcomes[4]. Therefore, to determine who is more likely to have such problems earlier can provide a foundation for developing targeted countermeasures separately.

Over the past few years, multivariate logistic regression has frequently appeared in oncology studies due to its ability to combine multiple clinical indicators, predict outcomes, and control confounders at once. The latter has more advantages in predicting postoperative outcomes of multiple factors affecting patients simultaneously, particularly nutritional status, immune dysfunction, surgical damage, or tumour biological characteristics involved together. CRC: Delayed postoperative intestinal recovery is associated with patients’ characteristics, the degree of trauma during surgery and nutritionalinflammatory condition[5]; LNM: It is very close to tumour aggressivity, infiltration depth and pathological behaviour[6]. Therefore, by adding these commonly accessible clinicopathological factors to the multivariate logistic regression model, it could be used as an important reference for the initial individualised evaluation before and after surgery.

Most existing research has concentrated mainly on predicting a single outcome event; Fewer comprehensive investigations exist concerning postoperative intestinal function recovery and LNM[7]. There are some deficiencies in the research models, including a small sample size, loose-variable-selection criteria, and lack of external validations that limit their generalisation properties[8,9]. Moreover, some scholars have conducted comparative analysis on surgical techniques; however, currently at our centre there are insufficiently powered open-versus-laparotomic differences due to limitations in this surgery. Therefore, to predict the results in clinical trials will become possible and reasonable soon enough.

Therefore, for the multivariate logistic regression analysis on a cohort of 197 patients with CRC, identify independent factors related to delayed post-operative intestinal function recovery and LNM according to different outcomes, and build clinical reference prediction models aimed at supporting early diagnosis recognitions and more targeted perioperative management.

MATERIALS AND METHODS

General data

This retrospective observational study included 197 consecutive patients who underwent radical resection for CRC at Xishan People’s Hospital, Wuxi City, from January 2021 to December 2023. Written informed consent was obtained from all participants and the study protocol was approved by the Ethics Committee of Xishan People’s Hospital.

The inclusion criteria: (1) Patients who met the diagnostic criteria of the Chinese protocol of diagnosis and treatment of CRC (2020 Edition) and had pathological confirmation; (2) Patients who underwent radical surgery and had complete perioperative and follow-up data; and (3) Patients with American Society of Anesthesiologists grade ≤ III.

The exclusion criteria: (1) Patients with other malignant tumors (12 cases excluded); (2) Patients with severe cardiac, hepatic, renal, or respiratory disease (15 cases excluded); and (3) Patients with incomplete clinical or pathological data or interrupted follow-up (18 cases excluded).

Study grouping and outcomes

The same cohort of 197 patients was analyzed for two predefined outcomes: (1) Delayed gastrointestinal function recovery (DGFR) was defined as a first exhaust time of > 72 hours or a first defecation time of > 96 hours after surgery. Accordingly, 49 patients were classified into the delayed group and 148 into the normal group; and (2) LNM was determined by postoperative pathological examination. Accordingly, 68 patients were classified into the LNM-positive group and 129 into the LNM-negative group.

Data collection

Baseline data included: (1) Demographic variables (age, sex, and body mass index); (2) Clinical characteristics (smoking, alcohol use, diabetes, hypertension, and other comorbidities); (3) Preoperative laboratory indicators, including nutritional and inflammatory markers [albumin, C-reactive protein (CRP), white blood cell (WBC) count, neutrophil percentage, platelet count, and fibrinogen] and liver function indices (alanine aminotransferase and aspartate aminotransferase); (4) Tumor characteristics (location, size, T stage, and differentiation); (5) Surgical variables (approach, operative time, and intraoperative blood loss); and (6) Pathological variables (number of lymph nodes harvested and LNM status).

Postoperative laboratory indicators (albumin, CRP, and WBC count) were also tracked on days 4 and days 7 and at one month to assess the dynamic changes.

Statistical analysis

Statistical analysis was performed using SPSS 26.0 and R software. Descriptive statistics are presented as mean ± SD, median [interquartile range (IQR)], or counts (percentages). Normality was assessed using the Shapiro-Wilk test. For each outcome, univariate analyses (Student’s t-test, Mann-Whitney U test, or χ² test) were used to compare variables between the groups. Variables with P < 0.10 in univariate analysis were entered into a multivariate binary logistic regression model (forward: Likelihood ratio method) to identify independent risk factors. The results are expressed as odds ratios with 95% confidence intervals (CIs). The discriminative abilities of the prediction models were evaluated using the area under the receiver operating characteristic curve (AUC). Calibration was assessed using the Hosmer-Lemeshow test. Clinical utility was determined using decision curve analysis to calculate the net benefit across a range of threshold probabilities. A two-sided P < 0.05 was considered statistically significant.

RESULTS

Patient characteristics and descriptive statistics

Table 1 summarizes the descriptive characteristics of the study cohort. Demographic, operative, and tumor-related variables were available for all 197 patients. Detailed intestinal recovery indicators were available for 33 patients, and an extended laboratory panel was available for 86 patients, which are presented separately in Table 1 to improve clarity. Overall, the mean age was 58.1 ± 11.3 years, 55.3% of the cohort were male, 82.7% underwent laparoscopic surgery, and 65.0% had T3-4 tumors.

Table 1 Descriptive statistics of the entire patient cohort (n = 197).

Variable	Count	Mean/n	Std Dev/%	Min	25%	Median	75%	Max
Demographic and clinical
Age (years)	197	58.1	11.3	34	50	59	67	78
Sex (male)	197	109	-55.33%	-	-	-	-	-
BMI (kg/m2)	197	23.8	3.1	16.2	21.6	23.9	25.8	32.5
Smoker	197	62	-31.50%	-	-	-	-	-
Diabetes	197	31	-15.70%	-	-	-	-	-
Hypertension	197	67	-34.00%	-	-	-	-	-
Surgical and tumor
Surgical approach (laparoscopic)	197	163	82.74%	-	-	-	-	-
Operation time (minute)	197	152.4	31.7	85	130	150	175	245
Intraoperative blood loss (mL)	197	132.8	45.2	50	100	120	150	350
Tumor location (colon/rectum)	197	118/79	(59.9%/40.1%)	-	-	-	-	-
Tumor size (cm)	197	4.6	1.4	1.5	3.6	4.5	5.5	9
T3-4 stage	197	128	-65.00%	-	-	-	-	-
Poor differentiation	197	53	-26.90%	-	-	-	-	-
Intestinal function (subset, n = 33)
First exhaust time (days)	33	3.36	1.06	2	3	3	4	6
First defecation time (days)	33	3.79	1.14	2	3	4	4	6
Bowel sound recovery (days)	33	2.64	0.74	1	2	3	3	4
Laboratory parameters (subset, n = 86)
Albumin (g/L)	86	40.22	6.31	26.3	36.82	41.4	43.88	73.4
Globulin (g/L)	86	28.38	5.79	20.1	24.85	27.4	31.2	66.4
WBC (× 109/L)	86	6.26	2.28	2.53	4.79	5.86	7.53	16.77
Neutrophil %	86	8.3	70.99	0.17	0.58	0.65	0.72	659
Lymphocyte %	86	0.27	0.17	0.06	0.18	0.26	0.31	1.47
Platelet (× 109/L)	86	233.96	91.46	0.31	179	226.5	273.75	482
Fibrinogen (g/L)	86	3.68	0.92	1.93	3.01	3.6	4.28	6.23
ALT (U/L)	86	17.72	11.07	2	11	14.5	22	74
AST (U/L)	86	21.66	8.92	6	17	20	24.75	78

Std Dev: Standard deviation; Min: Minimum; Max: Maximum; BMI: Body mass index; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; WBC: White blood cell.

Univariate analysis based on outcome groups

Patients were divided into the delayed group (n = 49) and the normal group (n = 148) based on DGFR. Univariate analysis (Table 2) showed that the delayed group had significantly higher proportions of preoperative hypoalbuminemia (< 35 g/L) (30.6% vs 10.1%, P = 0.001), intraoperative blood loss of > 50 mL (46.9% vs 18.2%, P < 0.001), and T3-4 stage tumors (85.7% vs 58.1%, P < 0.001) than the normal group. In addition, the delayed group had lower preoperative albumin levels, higher postoperative day 4 CRP levels, and longer postoperative hospital stays (all P < 0.05).

Table 2 Comparison of baseline characteristics by intestinal function recovery outcome, n (%)/mean ± SD.

Variable	Delayed group (n = 49)	Normal group (n = 148)	Statistic (χ2/t/z)	P value
Age (years)	59.8 ± 10.5	57.5 ± 11.5	1.251	0.213
Sex male	25 (51.0)	84 (56.8)	0.490	0.484
BMI (kg/m2)	23.5 ± 3.3	23.9 ± 3.0	-0.822	0.412
Preop albumin < 35 g/L	15 (30.6)	15 (10.1)	12.671	0.001
Blood loss > 50 mL	23 (46.9)	27 (18.2)	17.233	< 0.001
T3-4 stage	42 (85.7)	86 (58.1)	12.894	< 0.001
Poor differentiation	16 (32.7)	37 (25.0)	1.133	0.288
Laparoscopic surgery	32 (65.3)	111 (75.0)	1.851	0.174
Pre-op albumin (g/L)	37.1 ± 5.2	41.2 ± 4.8	-5.122	< 0.001
Post-op day 4 CRP (mg/L)	28.5 ± 6.1	20.3 ± 5.4	8.91	< 0.001
Hospital stay (days)	14.2 ± 3.8	10.1 ± 2.9	7.762	< 0.001

BMI: Body mass index; Pre-op: Preoperative; Post-op: Postoperative; CRP: C-reactive protein.

Patients were separately categorized into the LNM-positive group (n = 68) and the LNM-negative group (n = 129). As shown in Table 3, the LNM-positive group had a significantly higher proportions of poorly differentiated tumors (45.6% vs 17.1%, P < 0.001) and a lymph node yield of < 12 (27.9% vs 10.1%, P = 0.001). These patients also had larger tumors, a markedly higher proportion of T3-4 stage disease, and higher platelet and fibrinogen levels, suggesting an association between tumor aggressiveness, inadequate nodal yield, and a proinflammatory/prothrombotic state.

Table 3 Comparison of baseline characteristics by lymph node metastasis outcome, n (%)/mean ± SD.

Variable	LNM-positive (n = 68)	LNM-negative (n = 129)	Statistic (χ2/t/z)	P value
Age (years)	57.3 ± 11.8	58.5 ± 11.0	0.701	0.483
Sex male	39 (57.4)	70 (54.3)	0.172	0.678
Poor differentiation	31 (45.6)	22 (17.1)	19.87	< 0.001
Lymph nodes < 12	19 (27.9)	13 (10.1)	10.76	0.001
T3-4 stage	61 (89.7)	67 (51.9)	χ2 = 29.41	< 0.001
Tumor size (cm)	5.2 ± 1.5	4.3 ± 1.2	t = 4.52	< 0.001
Platelet (× 109/L)	258.3 ± 98.1	221.4 ± 84.5	t = 2.65	0.009
Fibrinogen (g/L)	3.95 ± 0.89	3.54 ± 0.88	t = 3.05	0.003
Laparoscopic surgery	46 (67.6)	97 (75.2)	χ2 = 1.33	0.249

LNM: Lymph node metastasis.

Multivariate logistic regression analysis of independent risk factors

Variables with P < 0.10 in the univariate analysis were included in the multivariate logistic regression models. The final models identified independent risk factors for each outcome, including detailed regression coefficients and adjusted odds ratios, as shown in Table 4 (DGFR model) and Table 5 (LNM model).

Table 4 Multivariate logistic regression for delayed intestinal function recovery.

Dependent variable	Independent risk factor	β	SE	OR (adjusted)	95%CI	P value
Delayed intestinal function	Low albumin (< 35 g/L)	1.35	0.43	3.86	1.66-8.97	0.002
	Blood loss > 50 mL	1.02	0.36	2.77	1.37-5.62	0.005
	T3-4 stage tumor	1.25	0.42	3.49	1.53-7.96	0.003

SE: Standard error; OR: Odds ratio; CI: Confidence interval.

Table 5 Multivariate logistic regression for lymph node metastasis.

Dependent variable	Independent risk factor	β	SE	OR (adjusted)	95%CI	P value
Lymph node metastasis	Poorly differentiated tumor	1.48	0.52	4.39	1.58-12.19	0.005
	Lymph nodes harvested < 12	1.41	0.49	4.1	1.57-10.72	0.004
	Elevated fibrinogen (≥ 4 g/L)	0.91	0.41	2.48	1.11-5.56	0.027

SE: Standard error; OR: Odds ratio; CI: Confidence interval.

Post-operative complication

The total complications among all subjects were 22.8 per cent (45/197). Most of the complications were incision infection (9.1%, 18/197) and pulmonary infection (5.1%, 10/197). The incidence of clinically severe anastomotic leakage (Clavien-Diagnosis classification II or higher) was as low as 2 per cent, equivalent to our institution’s annual standard. Delayed groups had significantly more complications compared with the normal group [complication rate: 38.8% vs 17.6%, P < 0.05].

Inflammatory and nutritional Trajectories across different outcomes

The postoperative trajectory showed that the inflammation-nutrition balance index (inflammatory marker level: CRP level; WBC count and neutrophil percentage; albumin level) was more significantly affected in the delayed group than in the early-operation group over time (P < 0.001); The results are displayed in Table 6. The dynamic inflammation and nutritional trajectory are shown in Figure 1. However, as shown by a somewhat less pronounced pattern for the LNM-positive group. The dynamic changes of inflammatory and nutritional marker during the process are illustrated as follows.

Figure 1 Dynamic inflammatory and nutritional trajectories by outcome groups. A: Albumin dynamics by recovery outcome; B: C-reactive protein dynamics by recovery outcome; C: White blood cell dynamics by recovery outcome; D: Neutrophil dynamics by recovery outcome. WBC: White blood cell; CRP: C-reactive protein; POD: Postoperative day.

Table 6 Dynamic changes in laboratory indicators by outcome groups, mean ± SD.

Time point	Indicator	Delayed group (n = 49)	Normal group (n = 148)	P value	LNM-positive (n = 68)	LNM-negative (n = 129)	P value
Preoperative	Albumin (g/L)	37.1 ± 5.2	41.2 ± 4.8	< 0.001	39.1 ± 5.9	40.8 ± 5.1	0.038
	CRP (mg/L)	10.5 ± 3.8	9.2 ± 3.1	0.012	10.1 ± 3.5	9.3 ± 3.2	0.089
	WBC (× 109/L)	6.8 ± 2.4	6.1 ± 2.1	0.025	6.5 ± 2.3	6.1 ± 2.2	0.187
POD 4	Albumin (g/L)	31.2 ± 3.8	35.8 ± 3.2	< 0.001	33.5 ± 3.9	35.1 ± 3.5	0.004
	CRP (mg/L)	28.5 ± 6.1	20.3 ± 5.4	< 0.001	24.8 ± 6.3	21.5 ± 5.8	< 0.001
	WBC (× 109/L)	11.5 ± 2.3	9.2 ± 1.9	< 0.001	10.3 ± 2.2	9.6 ± 2.0	0.019
	Neutrophil %	81.5 ± 6.8	72.8 ± 6.1	< 0.001	77.2 ± 7.1	74.1 ± 6.5	0.002
POD 7	Albumin (g/L)	33.8 ± 3.5	38.4 ± 3.4	< 0.001	36.1 ± 3.8	37.8 ± 3.6	0.002
	CRP (mg/L)	19.8 ± 5.1	12.1 ± 4.0	< 0.001	16.5 ± 5.3	13.2 ± 4.5	< 0.001
1 month	Albumin (g/L)	37.5 ± 3.9	40.9 ± 3.5	< 0.001	39.0 ± 4.1	40.5 ± 3.8	0.011
	CRP (mg/L)	13.5 ± 3.9	8.5 ± 2.8	< 0.001	11.2 ± 3.8	9.1 ± 3.2	< 0.001

LNM: Lymph node metastasis; CRP: C-reactive protein; WBC: White blood cell; POD: Postoperative day.

Lymph node yields and their occurrence of metastases

Lymph-node-based data are analyzed as follows in detail below. The median number of lymph node harvested across all subjects was 15 (IQR: 11-20). The LNM-negative group had a significantly higher median lymph node yield (16, IQR: 12-22) than the LNM-positive group (13, IQR: 9-18, P = 0.003). Further divided by means of lymph node involvement (< 12 vs ≥ 12), and by whether metastasis occurred (> 0 vs = 0). Based on this, it was found that those in the low-dissection group and high-positive metastasis were at the lowest risk; they exhibited both large tumour sizes and a higher frequency of poorly differentiated types. There was also a significant negative correlation coefficient between the number of lymph node dissection and preoperative CRP. The more inflammation, the lower it correlated to the difficulty in collecting affected lymph nodes after operation. The distribution of lymph node production, correlation with tumor size, and subgroup differences based on lymph node production and metastasis status are shown in Figure 2.

Figure 2 Analysis of lymph node yield and metastasis status. LN: Lymph node; LNM: Lymph node metastasis.

Predictive model evaluation and clinical applicability

To evaluate the performance of the multivariate logistic regression models in detail. The AUC for the model predicting delayed intestinal function was 0.82 (95%CI: 0.76-0.88), which showed excellent discriminatory capability. The LNM model performed relatively well as indicated by an AUC of 0.79 (95%CI: 0.72-0.85). Hosmer-Lemeshow test showed that the two model prediction probabilities’ deviation from real values was not statistically significant at P = 0.42 and P = 0.56 (Figure 3).

Figure 3 Performance of the multivariate logistic regression prediction models. The P values from the Hosmer-Lemeshow test were greater than 0.05, indicating good model calibration for both outcomes. The area under the curve was 0.82 for the intestinal function model and 0.79 for the lymph node metastasis model. Calibration plots showing agreement between the predicted probabilities and actual observed outcomes. The dotted line represents the ideal fit (Figure 3C and D). Evaluation of models for predicting (Figure 3A and C) delayed intestinal function recovery and (Figure 3B and D) lymph node metastasis. A: Predicted probability distribution: Bowel function delay; B: Predicted probability distribution: Lymph node metastasis; C: Calibration curve: Bowel function delay; D: Calibration curve: Lymph node metastasis.

The quantification of the model’s clinical applicability used decision curve analysis. As shown in Figure 4, for the prediction of delayed intestinal function, the model showed that it had higher net benefits than the treatment-all-or-none strategy within approximately 10%-55% of the target probability. In addition, the models showed a positive net benefit for the threshold probability of LMN in this range. It can be seen from this that by guiding clinical decision-making through these models, intensifying the attention paid to patients at higher risks of delays or considering postoperative treatment options in cases with unclear staging is necessary; it will contribute positively to patient outcomes.

Figure 4  Decision curve analysis.

Subgroup analysis according to surgical approach

Given the potential influence of surgical approach from each other at first, it can be concluded through post hoc subgroup analysis whether or not the type of surgery affects the outcome in this specific setting. Among patients with normal intestinal recovery (n = 148), those who underwent laparoscopic surgery (n = 111) had a significantly shorter median hospital stay (9 days, IQR: 8-11) than those who underwent open surgery (11 days, IQR: 10-13, P = 0.007). The surgical method itself was unable to eliminate these risks from independently associated with the outcome variables. Laparoscopic surgery found no significant difference in the amount of bleeding during surgery; it is possible that this phenomenon results from some modified postoperative factors among these influencing factors. To show the image features corresponding to different postsurgical results, typical preoperative computed tomography images of all subjects in this research have been demonstrated as shown below Figures 5 and 6.

Figure 5 Typical preoperative computed tomography imaging findings in a patient from the study cohort. These images illustrate the common tumor characteristics integral to the prediction models. A: Axial contrast-enhanced computed tomography image showing irregular sigmoid colon wall thickening (arrow) with a maximum thickness of approximately 8.2 mm, consistent with a T3 stage tumor (penetrating the muscularis propria); B: Coronal reconstructed image demonstrating the craniocaudal extent of the tumor measuring approximately 4.6 cm (double-headed arrow); C: The sagittal image clearly depicts the relationship between the tumor and adjacent structures; the presence of preserved fat planes suggests no invasion of the surrounding organs; D: Coronal enhancement image showing homogeneous enhancement of the tumor with no evidence of distant metastasis. These features (T stage and tumor size) were key variables in the multivariate analysis.

Figure 6 Representative imaging comparisons highlighting features associated with different outcomes. This figure compares patients with divergent postoperative outcomes, underscoring the imaging basis for risk stratification. A: Preoperative computed tomography (CT) image of a patient who subsequently developed delayed intestinal function recovery, showing T3N1M0 rectal cancer (arrow). The patient had preoperative hypoalbuminemia and significant intraoperative blood loss; B: Preoperative CT image of a patient with uncomplicated normal recovery of intestinal function showing T2N0M0 colon cancer (arrow). The patient had normal preoperative albumin levels and minimal blood loss; C: Preoperative CT image of a patient in the lymph node metastasis-positive group showing a bulky, poorly differentiated T4aN2 tumor (arrow) with extramural vascular invasion; D: Preoperative CT image of a patient from the lymph node metastasis-negative group showing a well-demarcated T3N0 tumor (arrow). The contrasting features of tumor stage, size, and morphology between these pairs visually represent the risk profiles identified by the prediction models.

DISCUSSION

Developed the target value-based forecasting model of postoperative DGFR and LNM in patients with CRC[10]. As shown in the Results, preoperative hypoalbuminaemia, postoperatively greater than 50 mL of blood loss and T3-4 stages, delays gastric passage; Intra-operatively poor differentiation, lymph nodes negative < 12 and increased preoperative fibrinogen were associated with LNM[11]. In addition to surgery, other potential factors also influenced the gastrointestinal function postoperatively; preoperative nutrition status of the patient and the condition before surgery were all related to this. Continuously lower albumin concentrations and elevated inflammatory indices were found in the delayed group at all postoperative times; thus, it can be concluded that a nutritional-inflammatory disorder impedes recovery from delayed intestinal coordination function reconstruction[12-15].

LNM may also have additional information for patients after surgery; it is similar to what pathology reports would list normally. An increased fibrinogen concentration reflects a systemic pro-inflammatory and pro-thrombotic condition, which can help tumour spread and therefore biologically possible as an adjuvant predictor[16-18]. Both methods are suitable for determining whether to increase the duration of follow-up after surgery or take other measures based on nutritional assessment, reduce blood loss during operation, handle pathological examination results in clinics promptly and provide targeted rehabilitation nursing[19-22]. Although the obtained AUC gains and decision-curve benefits should be considered an indication of effective risk classification; in isolation, they are not sufficient to change the treatment plan directly[23-26]. These models will be used in combination with overall clinical assessment.

Several limitations affect this study. Firstly, this was a single-centre retrospective observational study; therefore, selection bias and information incompleteness could not be fully ruled out. In addition, some expanded laboratory and recovery items were available only in part of the sample group; therefore, the reliability of these descriptions was impacted by this. Thirdly, there was no external validation cohort and mechanistic experimental proof available. Therefore, the identified models should be considered to be clinically suggestive but not yet validated instruments.

CONCLUSION

In summary, we constructed two multivariate logistic regression models to predict DGFR and LNM after CRC surgery. The models integrated readily available perioperative variables, including nutritional, operative, pathological, and fibrinogen-related indicators and showed favorable discrimination and potential clinical utility within this cohort. These findings support the value of early perioperative risk stratification; however, the models should be interpreted cautiously because this was a single-center retrospective study. Future work should include external validation in multicenter prospective cohorts and mechanistic studies to further improve predictive accuracy and clinical applicability.