Risk factor analysis and nomogram model construction for mortality in patients following colonic perforation surgery

Abstract

BACKGROUND

Colonic perforation is a surgical emergency with high mortality due to rapid progression to septic shock. Early identification of high-risk patients is critical for improving outcomes, yet existing predictive tools are often complex and lack clinical practicality.

AIM

To identify the risk factors for postoperative mortality in patients with colonic perforation and develop and validate a predictive model.

METHODS

A retrospective analysis was conducted on patients who underwent surgery for colonic perforation at the Department of Critical Medicine and General Surgery, Anqing Municipal Hospital, between January 2020 and July 2025. Patients were selected on the basis of inclusion and exclusion criteria and were classified into two groups according to postoperative outcomes: Death and survival. General demographics, laboratory results, and imaging data were collected and compared between the two groups. Univariate analysis was performed initially, followed by multivariate logistic regression analysis for variables with significant differences in the univariate analysis. A predictive model for postoperative mortality was constructed on the basis of the multivariate results. Internal validation was conducted using the bootstrap resampling method. The clinical optimal threshold was identified through decision curve analysis (DCA), and the operability of the dual cut-off strategy was demonstrated using a full-sample confusion matrix and a funnel-type pathway diagram. A nomogram was developed as a personalized prediction tool.

RESULTS

A total of 134 patients were included in the study, with 21 patients in the postoperative death group and 113 in the survival group, yielding a mortality rate of 15.6%. Moreover, no significant differences were found between the two groups concerning sex, history of hypertension, diabetes, cerebrovascular sequelae, cardiac history, haemoglobin level, albumin concentration, intraperitoneal free gas presence, intraperitoneal free fluid presence, perforation site, cause of perforation, operation time, or intraoperative blood loss volume. However, significant differences in age; American Society of Anaesthesiologists classification; Acute Physiology and Chronic Health Evaluation II (APACHE II) score; preoperative peripheral blood white blood cell (WBC) count; platelet count; serum total bilirubin level; serum creatinine level; lactate level; C-reactive protein level; procalcitonin level; the presence of portal venous gas (PVG); and time from onset to surgery (whether > 24 hours) were detected (all P < 0.05). Multivariate logistic regression analysis revealed that the APACHE II score [odds ratio (OR) = 1.24, 95% confidence interval (CI): 1.03-1.48], lactate level (OR = 2.40, 95%CI: 1.34-4.31), and presence of PVG (OR = 20.32, 95%CI: 1.89-218.45) were risk factors for postoperative mortality, whereas an elevated WBC count (OR = 0.68, 95%CI: 0.55-0.85) served as a protective factor. The constructed receiver operating characteristic curve revealed an area under the curve of 0.852 (95%CI: 0.791-0.913), with a Brier score = 0.072 and a slope ≈ 1. DCA demonstrated that within the 1%-40% threshold interval, the net benefit of the three model cut-off points surpassed those of the “full intervention” and “no intervention” scenarios. The full-sample pathway validation showed that the funnel-type process achieved 95.2% sensitivity for mortality screening with a 53% relative reduction in intensive care unit admissions.

CONCLUSION

The predictive model for postoperative mortality in patients with colonic perforation, which is based on four indicators (APACHE II score, lactate level, WBC count, and PVG presence), demonstrates strong predictive performance. The dual cut-off pathway (0.020 high-sensitivity screen + 0.121 resource-efficient confirmation) markedly reduced intensive care unit resource use while maintaining high sensitivity (95.2%). This investigation offers a replicable decision-making tool that can be integrated into information systems for future prospective studies.

Key Words: Colonic perforation; Mortality; Nomogram; Risk factors; Septic shock

Core Tip: A nomogram model incorporating the Acute Physiology and Chronic Health Evaluation II score, lactate level, white blood cell count, and presence of portal venous gas was developed and internally validated to predict mortality after colonic perforation surgery (area under the curve = 0.852). The dual-threshold “0.020 high-sensitivity screen + 0.121 resource-efficient confirmation” funnel strategy reduced the need for high-level intervention by 53% in this cohort while maintaining high sensitivity (95.2%). This visual tool offers a practical, system-compatible decision aid for early risk stratification and intensive care unit resource optimization, although external validation in multicentre, prospective studies is needed before clinical implementation.

INTRODUCTION

Colonic perforation is a surgical condition that is associated with severe complications, and its incidence has been increasing[1]. Typically, colonic perforation manifests as sudden and severe abdominal pain, and subsequent leakage of intestinal contents into the peritoneal cavity leads to rapid development of intra-abdominal infection, which can progress swiftly to septic shock and, in severe cases, result in death[2,3]. The most common aetiologies of colonic perforation include tumour perforation, diverticulitis, chronic constipation, and iatrogenic injury[4]. Colonic perforation is diagnosed through abdominal physical examination and imaging studies, with surgical exploration being the primary treatment option. Surgical approaches include open exploration or laparoscopic assessment, with bowel resection or repair performed on the basis of intraoperative findings, and the creation of a stoma when deemed necessary[5,6]. Therefore, colonic perforation is considered a serious intra-abdominal infectious condition with a high mortality rate, making the identification of risk factors for postoperative death following colonic perforation critical. Timely and effective interventions targeting these risk factors are essential for reducing mortality rates. Although research on postoperative intra-abdominal infections following colonic perforation remains limited, the existing assessment indicators are numerous and complex, which impedes their clinical applicability[7,8]. This study aims to analyse the risk factors for postoperative mortality in patients with colonic perforation and to develop a nomogram prediction model, facilitating the early identification of critically ill patients and the implementation of personalized treatment strategies during clinical diagnosis and management, ultimately contributing to a reduction in mortality rates.

MATERIALS AND METHODS

Patients

A retrospective analysis of the medical records of 134 patients with colonic perforation who underwent surgical treatment at the Department of Critical Medicine and the Department of General Surgery at Anqing Municipal Hospital between January 2020 and July 2025 was performed. The inclusion criteria were as follows: (1) Patients with intraoperatively confirmed colonic perforation; (2) Patients older than 18; and (3) Patients with complete clinical data. The exclusion criteria included the following: (1) Patients who did not undergo surgical treatment; (2) Patients with concurrent infections in other organs; (3) Patients with end-stage tumours; (4) Patients who experienced accidental deaths due to other causes; and (5) Patients whose clinical data were incomplete.

All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Declaration of Helsinki 1964 and later versions. This study was approved by the Ethics Committee of Anqing Municipal Hospital under ethical approval number Medical Ethics Review (2025) No. 173. Informed consent was obtained from all patients enrolled in the study.

Data collection

Clinical data: Clinical data for patients with colonic perforation were obtained from the medical records database of Anqing Municipal Hospital and included the following: (1) General information: Sex, age, comorbidities (e.g., diabetes mellitus, hypertension, cerebrovascular sequelae, and cardiac diseases), American Society of Anaesthesiologists (ASA) classification, and Acute Physiology and Chronic Health Evaluation II (APACHE II) score; (2) Perioperative laboratory data: Preoperative peripheral blood white blood cell (WBC) count, haemoglobin level, platelet count, albumin level, serum total bilirubin level, serum creatinine level, lactate level, C-reactive protein (CRP) level, and procalcitonin level; (3) Perioperative imaging data: Results of abdominal computed tomography examination [presence of intraperitoneal free gas, intraperitoneal free fluid, and portal venous gas (PVG)]; and (4) Perioperative clinical data: Time from onset to surgery, perforation site, cause of perforation, operation time, and amount of intraoperative blood loss.

Outcome determination: (1) Patients were split into the death and survival groups on the basis of 30-day postoperative survival outcomes; (2) Cardiac diseases included myocardial infarction and chronic heart failure; and (3) Intraperitoneal free fluid was defined as fluid depth exceeding 2 cm in the hepatorenal recess, paracolic gutters, or pelvis.

Missing data handling: The extent of missing data was assessed for all the candidate variables. Variables with a high percentage of missingness (> 15%) were excluded from the analysis. For the remaining variables, the percentage of missing data was very low (< 2% for any individual variable). Given the minimal and sporadic nature of the missing values, a complete-case analysis was deemed the most appropriate method. Consequently, only patients with complete data for all variables included in the final univariate and multivariate analyses were retained in the study cohort.

Statistical analysis

Statistical analyses were conducted utilizing SPSS 27.0 and R software (version 4.3.1). Continuous variables following a normal distribution are denoted as the mean ± SD, and intergroup comparisons were performed using t tests. For nonnormally distributed continuous variables, data are presented as medians (interquartile ranges), and group comparisons were performed using the Mann-Whitney U test, with results reported as the Z statistic. Categorical variables are represented as n (%), with between-group comparisons carried out using χ² tests or Fisher’s exact tests. Variables for which P < 0.05 in univariate analysis were incorporated into the multivariate logistic regression model (using a stepwise backwards elimination method with a P > 0.1 criterion for exclusion). Multicollinearity among variables was evaluated using the variance inflation factor (VIF < 10). A nomogram model was developed employing the rms package in R, with internal validation performed through the bootstrap method (1000 repetitions). Receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA) were generated to assess the model’s discrimination, calibration, and clinical utility. Statistical significance was defined as P < 0.05.

RESULTS

Baseline characteristics and univariate analysis of the study population

A total of 134 patients with colonic perforation were included in the study; these patients were divided into a death group (n = 21) and a survival group (n = 113) on the basis of survival outcomes, resulting in a mortality rate of 15.6%. Moreover, no statistically significant differences were detected between the groups in terms of sex, history of hypertension, diabetes, cerebrovascular sequelae, cardiac disease history, haemoglobin level, albumin level, free intraperitoneal gas presence, free intraperitoneal fluid presence, perforation site, cause of perforation, operation time, intraoperative blood loss volume or surgical approach. Statistically significant differences in age, ASA classification, APACHE II score, preoperative peripheral blood WBC count, platelet count, serum total bilirubin level, serum creatinine level, lactate level, CRP level, procalcitonin level, PVG presence, and time from onset to surgery (all > 24 hours) were detected (all P < 0.05) (Table 1 and raw patient data in Supplementary material).

Table 1 Baseline characteristics of the study cohort, n (%).

Variable	Mortality group (n = 21)	Survival group (n = 113)	t/χ2/Z	P value
Demographics
Female	8 (38.1)	53 (46.9)	χ2 = 0.53	0.46
Age (year)	82.00 (67-88)	71.00 (25-89)	Z = 4.27	< 0.001
Comorbidities
Hypertension	10 (47.6)	51 (45.1)	χ2 = 0.04	0.83
Diabetes	9 (42.9)	32 (28.3)	χ2 = 1.77	0.18
Cerebrovascular disease	7 (33.3)	29 (25.7)	χ2 = 0.52	0.47
Heart disease	5 (23.8)	29 (25.7)	χ2 = 0.03	0.85
Disease severity
ASA classification
II	7 (33.3)	91 (80.5)
III	12 (57.1)	22 (19.5)	χ2 = 20.57	< 0.001
IV	2 (9.5)	0 (0.0)
APACHE II score	8.95 ± 2.61	10.52 ± 3.80	t = 2.318	0.022
Laboratory findings
WBC (× 109/L)	7.52 (3.33-12.32)	9.93 (4.63-14.58)	Z = 3.73	< 0.001
Haemoglobin (g/L)	104.37 ± 10.63	100.24 ± 13.23	t = 1.38	0.169
Platelet (× 109/L)	83.00 (44-112)	197.00 (96-354)	Z = 4.68	< 0.001
Albumin (g/L)	30.8 ± 4.2	30.4 ± 2.3	t = 1.24	0.2176
Total bilirubin (μmol/L)	18.80 (14.00-28.80)	14.50 (7.60-28.80)	Z = 3.89	< 0.001
Serum creatinine (μmol/L)	128.00 (88-207)	95.00 (63-180)	Z = 3.88	< 0.001
Lactate (mmol/L)	1.90 (0.60-4.00)	1.70 (0.70-2.80)	Z = 2.07	0.038
CRP (mg/L)	247.00 (159-322)	156.00 (98-256)	Z = 4.12	< 0.001
PCT (μg/L)	45.50 (18.30-69.90)	15.60 (3.50-35.10)	Z = 4.18	< 0.001
Imaging findings
Pneumoperitoneum	15 (71.4)	77 (68.1)	χ2 = 0.09	0.76
Ascites	12 (57.1)	57 (50.4)	χ2 = 0.32	0.57
Portal venous gas	4 (19.0)	0 (0.0)	-	< 0.0011
Perforation details
Onset to surgery > 24 hours	12 (57.1)	36 (31.9)	χ2 = 5.21	0.022
Perforation site
Ascending colon	2 (9.5)	15 (13.3)
Transverse colon	1 (4.8)	3 (2.7)	χ2 = 0.63	0.96
Descending colon	1 (4.8)	7 (6.2)
Sigmoid colon	17 (81.0)	88 (77.9)
Perforation aetiology
Tumour	4 (19.0)	23 (20.4)
Diverticulum	6 (28.6)	30 (26.5)
Hernia	3 (14.3)	17 (15.0)	χ2 = 0.31	0.99
Chronic constipation	2 (9.5)	14 (12.4)
Foreign body	2 (9.5)	9 (8.0)
Iatrogenic	2 (9.5)	10 (8.8)
Spontaneous	2 (9.5)	10 (8.8)
Surgical data
Operative time (minutes)	178.57 ± 15.48	178.76 ± 18.23	t = -0.04	0.96
Intraoperative blood loss (mL)	38.57 ± 9.15	37.96 ± 9.96	t = 0.25	0.8
Surgical approach
Primary anastomosis	2 (9.5)	15 (13.3)
Hartmann's procedure	17 (81.0)	88 (77.9)	χ2 = 0.224	0.894
Simple repair	2 (9.5)	10 (8.8)

¹Fisher’s exact test.

ASA: American Society of Anaesthesiologists; APACHE II: Acute Physiology and Chronic Health Evaluation II; WBC: White blood cell; CRP: C-reactive protein; PCT: Procalcitonin.

Multivariate analysis of postoperative mortality in patients with colonic perforation

Variables found to be significant in the univariate analysis were incorporated into the multivariate analysis, with collinearity assessed (all VIFs < 5). The results revealed that the APACHE II score, lactate level, and PVG presence were independent risk factors, while an elevated WBC count served as a protective factor (Table 2).

Table 2 Multivariable logistic regression model.

Variable	β coefficient	SE	χ²	P value	OR (95%CI)	VIF
APACHE II	0.214	0.092	5.41	0.020	1.24 (1.03-1.48)	1.02
WBC	-0.382	0.112	11.62	< 0.001	0.68 (0.55-0.85)	1.18
Lactate	0.876	0.298	8.64	0.003	2.40 (1.34-4.31)	1.04
PVG	3.012	1.214	6.15	0.013	20.32 (1.89-218.45)	1.05

APACHE II: Acute Physiology and Chronic Health Evaluation II; WBC: White blood cell; PVG: Portal venous gas; VIF: Variance inflation factor; OR: Odds ratio; CI: Confidence interval.

Construction and internal validation of a nomogram model for postoperative mortality in patients with colonic perforation

The four independent variables (APACHE II score, lactate level, PVG presence, and WBC count), which were identified via multivariate logistic regression, were incorporated into the R language rms package to develop a visual nomogram (Figure 1). Internal validation was performed using the bootstrap method with 1000 repeated samples. Discrimination: The area under the curve (AUC) value was 0.852 [95% confidence interval (CI): 0.791-0.913] (Figure 2A). Calibration: The calibration curve exhibited a slope close to 1 (0.98) and an intercept near 0 (-0.02). The Brier score was 0.072. To account for overoptimism, bootstrap-corrected calibration estimates were obtained from 1000 resamples: The optimism-corrected slope was 0.95, the intercept was -0.03, and the Brier score was 0.075 (Figure 2B). Clinical utility: DCA demonstrated that within the 1%-40% threshold range, the model’s net benefit consistently surpassed that of both the “treat all” and “treat none” strategies (Figure 2C). Threshold sensitivity analysis revealed that both the Youden index and the F1 score reached their peak values at a threshold of 0.121 (Youden: 0.719; F1: 0.667), while at a threshold of 0.020, the sensitivity approached 100%, with the specificity reduced to 25%, representing the “zero missed diagnosis” entry threshold (Figure 2D).

Figure 1 Nomogram. A visual nomogram for predicting 30-day post-operative mortality in patients with colonic perforation. To determine the probability of death, the patient’s Acute Physiology and Chronic Health Evaluation II score, lactate level, white blood cell count, and presence (yes) or absence (no) of portal venous gas on their respective axes were located; a vertical line was drawn up to the points axis to score each variable. The points are summed, and the vertical line is dropped from the total points axis to the lower risk axis to determine the individual mortality probability. APACHE II: Acute Physiology and Chronic Health Evaluation II; WBC: White blood cell; PVG: Portal venous gas.

Figure 2 Model performance. A: Receiver operating characteristic curve of the four-variable nomogram. The area under the curve (AUC) value was 0.852 (95% confidence interval: 0.791-0.913), indicating excellent discrimination for 30-day mortality; B: Calibration plot comparing predicted vs observed mortality probabilities. The dashed 45° line represents perfect calibration; the solid line represents the model-based fit. A Brier score = 0.072 and slope ≈ 1 indicate high agreement across the entire risk range, supporting direct clinical use without further probability recalibration; C: Decision curve analysis. Decision curve for the nomogram across clinically relevant threshold probabilities (1%-40%). The net benefit of the model exceeds that of the “treat-all” (orange dashed) and “treat-none” (green dashed) strategies over the entire range, confirming clinical utility; D: Threshold sensitivity plot. Sensitivity, specificity, accuracy and F1 score vs probability threshold. The vertical dotted lines mark the two selected cut-offs: 0.020 (high-sensitivity screen) and 0.121 (optimal Youden/F1 point), supporting the dual-threshold funnel strategy. ROC: Receiver operating characteristic.

Validation of the dual-cut-off “funnel-type” pathway

The strategy of employing a “0.020 high-sensitivity red line + 0.121 resource-efficient screening” was implemented. The confusion matrix for the first cut-off (≥ 0.020) revealed a sensitivity of 95.2% (20/21), a specificity of 20.4%, and a negative predictive value of 95.8%. A review of the second cut-off (≥ 0.121) demonstrated a sensitivity of 83.3% and a specificity of 88.6%, representing the optimal balance point (Figure 3A). The full-sample pathway diagram indicated that 92 of 134 cases (68.7%) were flagged as high risk by the first tier. After the second-tier review, only 17 cases continued to require high-level intervention, while 75 cases were downgraded to standard monitoring, resulting in a 53% reduction in intensive care unit admissions (Figure 3B).

Figure 3 Dual-cut-off funnel validation. A: Confusion matrix (high-sensitivity cut-off of 0.020). Confusion matrix for the entire cohort (n = 134) at the screening threshold of 0.020. The model correctly identified 20 of 21 deaths (sensitivity = 95.2%) with only one false-negative case, achieving a negative predictive value of 95.8% and supporting its use as a high-sensitivity screening criterion; B: Funnel pathway diagram. Flow chart of the dual-cut-off funnel strategy. Step 1 (≥ 0.020) flagged 92 high-risk patients (68.7%). Step 2 (≥ 0.121) confirmed 17 for intensive intervention while downgrading 75 to standard care, reducing advanced resource use by 53% in this cohort.

Individualized prediction tool

The patient’s APACHE II score, WBC count, lactate level, and the presence or absence of a PVG were entered into the corresponding points axis at the top. These values were then summed to determine the total score, which was used to determine the total score axis, allowing for the individualized mortality risk of the patient to be calculated, thereby facilitating real-time bedside assessment.

DISCUSSION

Acute colonic perforation represents a prevalent surgical condition within the spectrum of acute abdominal disorders. Common aetiological factors include tumour-related factors, diverticulitis, chronic constipation, incarcerated inguinal hernia, and iatrogenic procedures. Neoplastic perforation can result from tumour-induced erosion of the normal bowel wall or from closed-loop obstruction caused by the tumour[9]. Spontaneous colonic perforation refers to perforation occurring in the absence of organic lesions or external trauma, most frequently affecting the antimesenteric border of the sigmoid colon and the rectosigmoid junction. This accounts for approximately 65% of cases, with chronic constipation serving as a primary precipitating factor[1]. Iatrogenic perforations are typically associated with endoscopic procedures. Despite relatively high bowel cleanliness in these patients, the onset is often insidious and frequently overlooked, leading to delays in treatment[10]. While the fluid content of the colonic lumen is relatively low, chemical irritation is less pronounced in these cases than in upper gastrointestinal perforation cases. However, the colonic contents are rich in pathogenic bacteria. Once perforation occurs, bacterial spread and extensive toxin absorption ensue, which can rapidly progress to septic shock and potentially result in death[8]. In the present study, the postoperative mortality rate following colonic perforation surgery was 15.6%, which aligns with findings from previous research[11].

This study revealed that the APACHE II score is the most robust single predictor of patient mortality risk, with an AUC value of 0.8055 when applied independently, thereby demonstrating exceptional discriminative ability. These results align with findings from numerous prior studies, further validating the use of the APACHE II score as the gold standard tool for assessment in critical care medicine[12]. The APACHE II score serves as an objective measure for the comprehensive evaluation of disease severity by integrating factors such as age, acute physiological parameters, and chronic health conditions. Its superior predictive performance derives from its multisystem, multidimensional evaluation framework, which captures the overall pathophysiological status, an aspect often missed by individual physiological indicators. In the multivariate logistic regression model of this study, the APACHE II score exhibited a significant positive predictive value [odds ratio (OR) = 1.24, 95%CI: 1.03-1.48], indicating a corresponding increase in patient mortality risk with higher scores, which is consistent with both its design and clinical relevance. Notably, although the APACHE II score alone demonstrated substantial predictive value, its performance was further enhanced when it was combined with additional indicators such as the lactate level and WBC count (the AUC value increased from 0.8055 to 0.852), emphasizing the synergistic benefit of multimodal biomarker assessment. In clinical settings, the APACHE II score should be employed as a foundational assessment tool and integrated with other critical indicators to guide clinical decision-making rather than relying solely on it[13].

Numerous studies have established that the lactate level acts as an independent prognostic predictor of mortality in patients with sepsis[14]. Elevated blood lactate concentrations are considered early indicators of multiple organ failure and have demonstrated superior accuracy over sequential organ failure assessment in predicting sepsis-related mortality[15]. Sepsis-3 guidelines designate serum lactate levels exceeding 2 mmol/L as the primary criterion for the clinical diagnosis of septic shock[16]. Lactate concentrations above 4 mmol/L have been reported to have 96% specificity for predicting in-hospital mortality in nonhypotensive patients[17]. In this study, an elevated lactate level (OR = 2.40; 95%CI: 1.34-4.31) was identified as a significant risk factor for postoperative mortality in patients with colonic perforation, which aligns with findings from previous studies[18].

WBCs are integral components of the human immune system and play pivotal roles in defending against infections and inflammation. Upon infection, cytokines such as granulocyte colony-stimulating factor, interleukin-1, and tumor necrosis factor-α stimulate the proliferation of bone marrow haematopoietic stem cells, increase the release of neutrophils into the bloodstream, and increase peripheral blood WBC counts[19]. However, in the case of severe bacterial infections, endotoxins impair the ability of the bone marrow to release neutrophils, while excessive consumption of WBCs occurs in conjunction with inadequate bone marrow compensation, resulting in reduced WBC counts, which serve as a prognostic marker for poor outcomes[20]. The findings of this study indicated that an elevated WBC count (OR = 0.68, 95%CI: 0.55-0.85) was a protective factor against postoperative mortality in patients with colonic perforation. A lower WBC count was associated with an increased risk of postoperative mortality in these patients. A preoperative reduction in the WBC count reflects a suppressed immune state and poor overall health, indicating impaired immune function, which is correlated with adverse postoperative outcomes.

The theoretical mechanisms underlying the occurrence of PVG include both mechanical and bacterial theories. Intestinal mucosal injury, bowel wall necrosis, or increased intraluminal gas pressure, which results in mucosal tears, allow gas to enter the portal venous system through the bowel wall. The bacterial theory posits that, in the presence of microbial infection, gas produced within the intestine escapes through the intestinal mucosa into the submucosal layer and blood vessels, subsequently entering the portal venous system[21]. Koizumi et al[22] analysed 1590 PVG cases from the Japanese national inpatient database and identified aetiologies such as intestinal ischaemia, gastrointestinal obstruction or distension, and gastrointestinal perforation. Given a reported mortality rate of 27.3%[22], aggressive exploratory laparotomy has been strongly advocated as a means to improve survival outcomes. In this study, the OR for PVG occurrence was 20.32, reflecting a robust positive correlation with mortality, indicating that the risk of death following PVG occurrence was approximately 20 times greater than that of the control group. However, the 95%CI was notably wide (1.89-218.45), potentially because of limited statistical power due to its inherently low incidence rate[23]. A sensitivity analysis confirmed that while PVG presence contributes to the model’s predictive power, the core combination of the APACHE II score, lactate level, and WBC count itself forms a stable and significant predictive foundation (AUC = 0.832). Therefore, the value of PVG presence in our model may lie in its role as a high-risk “flag”-when present, it dramatically elevates the predicted risk, but its absence does not diminish the utility of the model, which relies on more commonly available variables. This characteristic enhances the model's potential generalizability across settings where the prevalence of PVG may vary. Therefore, in clinical practice, it may be better to view PVG presence as an independent risk factor of colonic perforation. Clinicians should prioritize patients who present with PVG signs and promptly initiate aggressive interventional measures[24].

The final model developed in this study achieved an AUC value of 0.852 (95%CI: 0.791-0.913) for patients with colonic perforation, markedly outperforming both the APACHE II score alone and traditional lactate cut-off values. This suggests that the integration of APACHE II, WBC, lactate, and PVG data enhances the performance of the model in predicting mortality[25]. The calibration curve exhibited a Brier score of 0.072 and a slope approaching 1, indicating high concordance between the predicted probabilities and actual outcomes, thereby enabling direct bedside risk communication without the need for additional probability calibration. DCA demonstrated that, across the entire threshold range of 1%-40%, the net benefit of all three model cut-off points surpassed that of the “treat all” or “treat none” strategies[26]. The theoretical maximum net benefit of 0.1392 was achieved at a cut-off point of 0.010, whereas the clinically selected 0.020 cut-off point resulted in a marginal decrease of 0.006 while reducing the number of false-positive results by approximately 15%. Therefore, the 0.020 cut-off point was selected as the primary threshold. During the full-sample pathway validation, which included 134 patients, a first-tier score ≥ 0.020 identified 92 cases (68.7%) as high risk, with a sensitivity of 0.952. A second-tier score ≥ 0.121 maintained only 17 cases for high-level intervention but downgraded 75 cases. Compared with the Smith 2022 study, this model maintained exceptionally high sensitivity while improving specificity to 0.886 through a second-tier review, thereby avoiding the common pitfall of “high sensitivity → high false positives”[27-29].

The translation of this dual-cut-off model into clinical practice requires a structured protocol. We propose a two-stage implementation: First, upon patient admission, the nomogram score is calculated. Those with a risk probability ≥ 0.020 are immediately flagged for 'high-risk' status, triggering intensified initial monitoring (e.g., closer vital sign observation, senior surgical review, and point-of-care lactate testing). Second, within a short timeframe (e.g., 2-4 hours), this high-risk group undergoes a comprehensive reassessment. This reassessment incorporates the refined model prediction (using the ≥ 0.121 cut-off) alongside critical clinical judgement, response to initial resuscitation, and resource availability to make the final decision regarding intensive care unit (ICU) admission. This funnel approach leverages the model's speed for initial triage while ensuring that final, high-stakes decisions are augmented by clinical expertise.

This study has several limitations that should be acknowledged. First, the sample size, particularly the number of mortality events (n = 21), is limited. This results in a suboptimal event-per-variable ratio, which increases the risk of model overfitting and may affect the stability of the coefficient estimates. Although we employed bootstrap internal validation and maintained a parsimonious model to mitigate these concerns, the wide CI for PVG presence as a predictive factor of mortality is a direct manifestation of this statistical constraint.

The model was developed and validated using a single-center, retrospective dataset. The lack of external validation in an independent cohort limits definitive conclusions about its generalizability to other populations or healthcare settings. Therefore, future multicenter, prospective studies are essential for external validation prior to any clinical application. Finally, a formal cost-effectiveness analysis of the dual-cutoff pathway, while important for future implementation research, was beyond the scope of this study.

CONCLUSION

A postoperative mortality prediction model for patients with colonic perforation, developed using four key indicators, APACHE II score, lactate level, WBC count, and PVG presence, demonstrated robust predictive performance. In this cohort, the dual-cut-off pathway, consisting of a “0.020 zero-missed-diagnosis threshold” and a “0.121 resource-efficient screening” cut-off, allowed for a reduction in ICU resource utilization of more than 50%. This approach offers a replicable, system-embeddable decision framework for future prospective implementation studies. The nomogram prediction model, designed as a visualized graphical tool with a straightforward calculation process, exhibited high predictive accuracy during validation, further confirming its effectiveness. This model can facilitate the early identification of high-risk patients with colonic perforation, contributing to improved patient outcomes and reduced mortality rates.