Abstract
Background
Colorectal cancer patients requiring perioperative intensive care unit (ICU) admission represent a high-risk subgroup with substantial short-term mortality. This study aimed to develop and internally validate a nomogram for estimating 3-month all-cause mortality in colorectal cancer patients requiring perioperative ICU care.
Methods
This single-center retrospective cohort study screened patients hospitalized for colorectal cancer surgery between January 2014 and December 2018. The final cohort included 415 patients requiring perioperative ICU admission. The primary endpoint was all-cause mortality within 3 months after ICU admission. The intended prediction time point was the end of the first 24 h after ICU admission. Candidate predictors were obtained from the preoperative period, the index surgery, and the first 24 h after ICU admission. Multivariable logistic regression was used to identify independent predictors and construct the nomogram. Model performance was evaluated using receiver operating characteristic curve analysis, bootstrap internal validation, calibration analysis, and decision curve analysis.
Results
Among 415 included patients, 58 (14.0%) died within 3 months after ICU admission. The final nomogram incorporated preoperative albumin, APACHE II score, CRRT within the first 24 h after ICU admission, mechanical ventilation within the first 24 h after ICU admission, and laparoscopic surgery. Higher albumin level and laparoscopic surgery were associated with lower mortality risk, whereas higher APACHE II score, CRRT, and mechanical ventilation were associated with increased mortality risk. The nomogram showed good discrimination, with an AUC of 0.882 (95% CI, 0.829–0.927) and bootstrap internal validation yielded an optimism-corrected C-index of 0.872. Calibration analysis showed good agreement between predicted and observed risk, and decision curve analysis suggested potential clinical utility across clinically relevant threshold probabilities. The nomogram showed better discrimination than APACHE II alone.
Conclusions
We developed and internally validated a nomogram for estimating 3-month mortality at the end of the first 24 h after ICU admission in colorectal cancer patients requiring perioperative ICU care. The model may support early risk stratification in this high-risk population, but external validation is required before routine clinical implementation.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s12957-026-04495-0.
Keywords: Colorectal cancer, ICU, Mortality, Nomogram, Prediction model
Background
Colorectal cancer remains one of the leading causes of cancer-related morbidity and mortality worldwide [1]. Surgical resection is the main curative treatment for patients with resectable colorectal cancer, and advances in minimally invasive techniques, perioperative management, and enhanced recovery after surgery (ERAS) protocols have improved outcomes for many patients [2, 3]. However, in clinical practice, a subset of patients still requires perioperative intensive care unit (ICU) admission because of advanced age, limited physiological reserve, extensive surgical stress, severe postoperative complications, infection, or acute clinical deterioration. These patients represent a particularly high-risk subgroup, in whom early mortality may be driven not only by tumor burden but also by acute physiological derangement and organ dysfunction [4].
Existing prognostic tools have important limitations in this setting. The tumor-node-metastasis (TNM) staging system is widely used for assessing long-term oncological prognosis, but it does not fully capture the acute physiological disturbances that determine short-term outcomes in critically ill surgical patients [5].Conversely, commonly used ICU severity scores, such as the Acute Physiology and Chronic Health Evaluation II (APACHE II) score and Sequential Organ Failure Assessment (SOFA) score, reflect acute illness severity and organ dysfunction, but they do not incorporate cancer- and surgery-related factors such as nutritional reserve or operative approach [6, 7].Therefore, a prediction model that combines baseline patient reserve, surgical characteristics, and early ICU severity indicators may provide more individualized short-term risk assessment for colorectal cancer patients requiring perioperative critical care.
Serum albumin is a readily available marker of nutritional and inflammatory status and has been associated with postoperative morbidity and mortality in colorectal surgery [8]. In addition, early organ-support requirements after ICU admission, including mechanical ventilation and continuous renal replacement therapy (CRRT), may reflect severe respiratory or renal dysfunction during the early phase of critical illness. However, few studies have focused specifically on individualized prediction of 3-month mortality in colorectal cancer patients requiring perioperative ICU admission.
To address this clinical need, we retrospectively analyzed colorectal cancer patients requiring perioperative ICU admission at our institution. The aim of this study was to develop and internally validate a nomogram for estimating 3-month all-cause mortality at the end of the first 24 h after ICU admission. By integrating preoperative albumin, surgical approach, APACHE II score, and early ICU organ-support indicators, this model may support early risk stratification in this high-risk population.
Methods
Study design and ethics
This was a single-center retrospective cohort study conducted at the Sixth Affiliated Hospital of Sun Yat-sen University. Patients hospitalized for colorectal cancer surgery between January 2014 and December 2018 were retrospectively screened. The study focused on patients who required perioperative intensive care unit (ICU) admission during the same hospitalization.
The study protocol was approved by the Medical Ethics Committee of the Sixth Affiliated Hospital of Sun Yat-sen University (Approval No. E2025324). Because this study used de-identified retrospective clinical data, the requirement for informed consent was waived by the ethics committee.
Inclusion and exclusion criteria
Patients were eligible for inclusion if they were aged 18 years or older, had a histologically confirmed diagnosis of colorectal cancer, underwent colorectal cancer surgery at our institution, and required perioperative ICU admission during the same hospitalization. Perioperative ICU admission was defined as ICU admission occurring before or after the index colorectal cancer surgery for perioperative monitoring, postoperative organ support, surgical complications, or acute clinical deterioration related to the index hospitalization.
Patients were excluded if they met any of the following criteria: ICU admission for reasons unrelated to the index colorectal cancer surgery; ICU admission after non-index surgery or surgery for recurrent disease; surgery or major treatment performed at another hospital; no surgical treatment during the index hospitalization; coexistence of other primary malignancies or other major surgical diseases that could confound the assessment of short-term prognosis; insufficient clinical records for eligibility assessment; or unavailable or uncertain 3-month outcome information.
The preoperative period was defined as the interval from hospital admission for the index colorectal cancer surgery to the start of the operation. Preoperative laboratory values were defined as the most recent available laboratory results before the index operation.
Patient selection
Between January 2014 and December 2018, 603 patients hospitalized for colorectal cancer surgery at our institution were initially screened. After applying the predefined inclusion and exclusion criteria, 188 patients were excluded for the following reasons: non-perioperative ICU admission (n = 81), ICU admission after non-index surgery or surgery for recurrent disease (n = 39), surgery or previous treatment performed at another hospital (n = 20), no surgery or ICU admission for non-postoperative reasons (n = 14), previous surgery or historical records inconsistent with the present study cohort (n = 13), concomitant non-colorectal malignancy or other major surgery (n = 11), and unavailable or uncertain 3-month outcome information due to death or discharge against medical advice without reliable confirmation (n = 10). Finally, 415 patients were included in the final cohort.
The patient selection process is summarized in Supplementary Figure S1.
Clinical data collection and outcome measures
Clinical data were retrospectively extracted from the institutional electronic medical record system. Baseline demographic variables included age, sex, body mass index (BMI), and comorbidity burden. The Charlson Comorbidity Index (CCI) was used to summarize the burden of coexisting diseases, including heart failure, diabetes, cerebrovascular disease, and pulmonary disease.
Tumor-related and treatment-related variables included tumor location, histological differentiation, TNM stage, vascular invasion, perineural invasion, extent of resection, neoadjuvant chemotherapy, and adjuvant chemotherapy. Surgical variables included the use of a laparoscopic approach during the index colorectal cancer surgery.
Preoperative laboratory variables included white blood cell count, platelet count, hemoglobin, albumin, total protein, and carcinoembryonic antigen. Perioperative clinical variables included bowel obstruction and infection.
Acute illness severity was evaluated using the Acute Physiology and Chronic Health Evaluation II (APACHE II) score and the Sequential Organ Failure Assessment (SOFA) score. Both scores were calculated using the worst physiological and laboratory values recorded during the first 24 h after ICU admission. Early ICU organ-support variables included vasoactive agent use, mechanical ventilation, and continuous renal replacement therapy (CRRT).
The primary outcome was all-cause mortality within 3 months after ICU admission. Survival time was calculated from the date of ICU admission to death or the last available follow-up within 3 months.
Prediction time point and predictor assessment window
The intended prediction time point of the nomogram was defined as the end of the first 24 h after ICU admission. Preoperative laboratory variables, including serum albumin, were obtained from the most recent laboratory tests before the index operation. Surgical approach was recorded according to the index colorectal cancer surgery.
Acute physiological severity was assessed using the worst physiological and laboratory values recorded during the first 24 h after ICU admission to calculate the APACHE II and SOFA scores. Mechanical ventilation and CRRT were defined as organ-support interventions required within the first 24 h after ICU admission. Therefore, all predictors included in the final nomogram were available at the intended time of model application.
Statistical analysis
Continuous variables were assessed for distribution and are presented as mean ± standard deviation for normally distributed data or median with interquartile range for skewed data. Between-group comparisons were performed using the independent Student’s t test for normally distributed continuous variables and the Mann–Whitney U test for skewed continuous variables. Categorical variables are presented as counts and percentages and were compared using Pearson’s chi-square test or Fisher’s exact test, as appropriate.
Among the 415 patients included in the final cohort, 3-month mortality outcome data were complete. Missingness in the final model predictors was low: albumin was missing in 4 patients, APACHE II score in 5 patients, and laparoscopic surgery in 2 patients, whereas CRRT and mechanical ventilation within the first 24 h after ICU admission had no missing values. Complete-case analysis was used for multivariable model development, resulting in 405 patients included in the final multivariable logistic regression model. The number and proportion of missing values for relevant variables are provided in Supplementary Table S1.
Candidate predictors were selected according to clinical relevance, univariate associations, event number, and potential collinearity. Multicollinearity among predictors in the final model was assessed using variance inflation factors. The correlation between APACHE II and SOFA was also examined because both scores reflect acute physiological derangement and organ dysfunction. Variables considered clinically relevant to short-term mortality included demographic characteristics, nutritional and laboratory parameters, tumor-related factors, surgical variables, acute physiological severity scores, and early ICU organ-support indicators. No automated stepwise selection procedure was used. To reduce the risk of overfitting, the number of predictors retained in the final model was restricted according to the number of outcome events.
Because APACHE II and SOFA scores both reflect acute illness severity and may introduce collinearity, only APACHE II was retained in the final multivariable model. Variables that were strongly associated with long-term oncological prognosis but less directly linked to acute early ICU deterioration were not included in the primary model. However, their potential contribution was further examined in sensitivity analyses.
A multivariable logistic regression model was fitted to identify independent predictors of 3-month mortality. The final predictors were then used to construct a nomogram using the rms package in R. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC) with 95% confidence interval and the concordance index. Internal validation was performed using 1,000 bootstrap resamples to estimate optimism-corrected model performance. Calibration was evaluated using calibration plots, calibration intercept, calibration slope, and mean absolute error. Clinical utility was assessed using decision curve analysis. The DeLong method was used to compare the AUC of the nomogram with that of APACHE II alone.
To assess model robustness and address the potential contribution of tumor burden and treatment-related predictors, sensitivity analyses were performed. First, TNM stage was added to the final model to evaluate its incremental prognostic value. Second, a reduced model excluding CRRT and mechanical ventilation was fitted to assess model performance without early organ-support variables.
All statistical analyses were performed using R software. The main packages used included rms, pROC, rmda, and dcurves. A two-sided P value < 0.05 was considered statistically significant. Model reporting was guided by the TRIPOD statement, and a completed TRIPOD checklist was submitted as a supplementary file.
Results
Baseline characteristics
Between January 2014 and December 2018, 603 patients hospitalized for colorectal cancer surgery were initially screened. After applying the predefined inclusion and exclusion criteria, 188 patients were excluded, and 415 patients were included in the final cohort. The patient selection process is shown in Supplementary Figure S1.
Among the 415 included patients, 58 patients (14.0%) died within 3 months after ICU admission, whereas 357 patients (86.0%) survived. There were no significant differences between survivors and non-survivors in age (P = 0.933) or body mass index (BMI; P = 0.052).
Patients who died within 3 months had significantly more severe acute physiological derangement during the first 24 h after ICU admission than survivors. The median APACHE II score was higher in non-survivors than in survivors [19.0 (IQR, 14.0–26.0) vs. 10.0 (IQR, 7.0–14.0); P < 0.001], as was the median SOFA score [11.0 (IQR, 6.0–15.0) vs. 4.0 (IQR, 3.0–7.0); P < 0.001]. Non-survivors also had lower preoperative albumin levels [33.2 (IQR, 28.5–39.2) vs. 37.9 (IQR, 33.2–41.1); P < 0.001] and lower total protein levels [61.8 (IQR, 54.8–66.8) vs. 66.9 (IQR, 61.4–72.3); P < 0.001].
With respect to surgical and treatment-related characteristics, non-survivors were less likely to undergo laparoscopic surgery (42.1% vs. 72.8%, P < 0.001), achieve radical resection (55.4% vs. 83.9%, P < 0.001), or receive adjuvant chemotherapy (5.2% vs. 31.4%, P < 0.001). Early ICU organ-support requirements were also more common among non-survivors, including mechanical ventilation within the first 24 h after ICU admission (84.5% vs. 35.9%, P < 0.001) and CRRT within the first 24 h after ICU admission (24.1% vs. 2.5%, P < 0.001). Length of ICU stay was longer among non-survivors than survivors [103.0 (IQR, 39.0–234.8) vs. 27.0 (IQR, 20.0–68.0); P < 0.001]; however, this variable was considered a descriptive outcome-related characteristic and was not used as a baseline predictor for model development.
Detailed baseline characteristics are presented in Table 1.
Table 1.
Baseline characteristics of colorectal cancer patients requiring perioperative ICU admission
| Variable | Categories | Total (N = 415) | Survival (N = 357) |
Death (N = 58) | p value |
|---|---|---|---|---|---|
| Age | Median (IQR) | 72.00 (63.00–79.00) | 72.00 (63.00–80.00) | 72.00 (64.25-79.00) | 0.933 |
| BMI | Median (IQR) | 22.29 (19.83–24.79) | 22.55 (19.98–24.80) | 21.11 (18.77–23.68) | 0.052 |
| APACHE II | Median (IQR) | 11.00 (7.00–15.00) | 10.00 (7.00–14.00) | 19.00 (14.00–26.00) | < 0.001 |
| SOFA | Median (IQR) | 5.00 (3.00–7.00) | 4.00 (3.00–7.00) | 11.00 (6.00–15.00) | < 0.001 |
| Total Protein | Median (IQR) | 66.42 (60.59-72.00) | 66.91 (61.38–72.25) | 61.80 (54.76–66.80) | < 0.001 |
| Albumin | Median (IQR) | 37.30 (32.59–40.86) | 37.89 (33.20–41.10) | 33.20 (28.48–39.19) | < 0.001 |
| Hemoglobin | Median (IQR) | 110.50 (92.00-126.00) | 111.40 (93.60-126.10) | 105.00 (82.30-120.70) | 0.043 |
| Platelet | Median (IQR) | 242.00 (188.05–300.00) | 244.00 (190.00-299.20) | 238.80 (183.00-303.75) | 0.679 |
| WBC | Median (IQR) | 6.45 (5.05–8.39) | 6.34 (5.02–8.18) | 7.26 (5.21–10.37) | 0.063 |
| CEA | Median (IQR) | 1.00 (0.00–1.00) | 0.00 (0.00–1.00) | 1.00 (0.00–1.00) | 0.012 |
| Length of ICU stay | Median (IQR) | 39.00 (20.00-85.50) | 27.00 (20.00–68.00) | 103.00 (39.00-234.75) | < 0.001 |
| CCI score | Median (IQR) | 7.00 (6.00–8.00) | 7.00 (6.00–8.00) | 7.00 (7.00–8.00) | 0.605 |
| Gender | 0.405 | ||||
| F | 148 (35.7%) | 124 (34.7%) | 24 (41.4%) | ||
| M | 267 (64.3%) | 233 (65.3%) | 34 (58.6%) | ||
| CCI Risk | 0.924 | ||||
| 2 | 77 (18.6%) | 67 (18.8%) | 10 (17.2%) | ||
| 3 | 338 (81.4%) | 290 (81.2%) | 48 (82.8%) | ||
| Tumor location | 0.047 | ||||
| Rectum | 185 (46.0%) | 167 (48.1%) | 18 (32.7%) | ||
| Colon | 217 (54.0%) | 180 (51.9%) | 37 (67.3%) | ||
| TNM | 0.004 | ||||
| Stage 0 | 8 (1.9%) | 6 (1.7%) | 2 (3.5%) | ||
| Stage I | 45 (10.9%) | 41 (11.5%) | 4 (7.0%) | ||
| Stage II | 163 (39.4%) | 146 (40.9%) | 17 (29.8%) | ||
| Stage III | 114 (27.5%) | 102 (28.6%) | 12 (21.1%) | ||
| Stage IV | 84 (20.3%) | 62 (17.4%) | 22 (38.6%) | ||
| Histological differentiation | 0.229 | ||||
| Poorly differentiated | 16 (4.3%) | 12 (3.7%) | 4 (9.1%) | ||
| Moderately differentiated | 260 (70.3%) | 231 (70.9%) | 29 (65.9%) | ||
| Well differentiated | 75 (20.3%) | 68 (20.9%) | 7 (15.9%) | ||
| Well-moderately differentiated | 3 (0.8%) | 3 (0.9%) | 0 (0.0%) | ||
| Moderately poorly differentiated | 15 (4.1%) | 11 (3.4%) | 4 (9.1%) | ||
| Undifferentiated | 1 (0.3%) | 1 (0.3%) | 0 (0.0%) | ||
| Vascular invasion | 0.930 | ||||
| Yes | 53 (13.4%) | 46 (13.2%) | 7 (14.9%) | ||
| No | 342 (86.6%) | 302 (86.8%) | 40 (85.1%) | ||
| Perineural invasion | 0.592 | ||||
| Yes | 53 (13.5%) | 45 (13.0%) | 8 (17.0%) | ||
| No | 341 (86.5%) | 302 (87.0%) | 39 (83.0%) | ||
| Laparoscopic surgery | < 0.001 | ||||
| No | 130 (31.5%) | 97 (27.2%) | 33 (57.9%) | ||
| Yes | 283 (68.5%) | 259 (72.8%) | 24 (42.1%) | ||
| Extent of resection | < 0.001 | ||||
| Radical | 327 (80.0%) | 296 (83.9%) | 31 (55.4%) | ||
| Palliative | 66 (16.1%) | 49 (13.9%) | 17 (30.4%) | ||
| Debulking | 16 (3.9%) | 8 (2.3%) | 8 (14.3%) | ||
| Adjuvant Chemotherapy | < 0.001 | ||||
| No | 300 (72.3%) | 245 (68.6%) | 55 (94.8%) | ||
| Yes | 115 (27.7%) | 112 (31.4%) | 3 (5.2%) | ||
| Neoadjuvant Chemotherapy | 1.000 | ||||
| No | 369 (88.9%) | 317 (88.8%) | 52 (89.7%) | ||
| Yes | 46 (11.1%) | 40 (11.2%) | 6 (10.3%) | ||
| Vasopressor use | 0.028 | ||||
| No | 254 (64.6%) | 225 (67.0%) | 29 (50.9%) | ||
| Yes | 139 (35.4%) | 111 (33.0%) | 28 (49.1%) | ||
| Infection | 0.026 | ||||
| No | 253 (64.4%) | 226 (66.7%) | 27 (50.0%) | ||
| Yes | 140 (35.6%) | 113 (33.3%) | 27 (50.0%) | ||
| Mechanical ventilation | < 0.001 | ||||
| No | 238 (57.3%) | 229 (64.1%) | 9 (15.5%) | ||
| Yes | 177 (42.7%) | 128 (35.9%) | 49 (84.5%) | ||
| CRRT | < 0.001 | ||||
| No | 392 (94.5%) | 348 (97.5%) | 44 (75.9%) | ||
| Yes | 23 (5.5%) | 9 (2.5%) | 14 (24.1%) | ||
| Bowel obstruction | 0.003 | ||||
| No | 253 (61.6%) | 228 (64.6%) | 25 (43.1%) | ||
| Yes | 158 (38.4%) | 125 (35.4%) | 33 (56.9%) | ||
| Pulmonary disease | 1.000 | ||||
| No | 383 (92.3%) | 329 (92.2%) | 54 (93.1%) | ||
| Yes | 32 (7.7%) | 28 (7.8%) | 4 (6.9%) | ||
| Mild Diabetes | 0.755 | ||||
| No | 341 (82.2%) | 292 (81.8%) | 49 (84.5%) | ||
| Yes | 74 (17.8%) | 65 (18.2%) | 9 (15.5%) | ||
| Cerebrovascular disease | 0.213 | ||||
| No | 393 (94.7%) | 340 (95.2%) | 53 (91.4%) | ||
| Yes | 22 (5.3%) | 17 (4.8%) | 5 (8.6%) | ||
| Heart failure | 1.000 | ||||
| No | 407 (98.1%) | 350 (98.0%) | 57 (98.3%) | ||
| Yes | 8 (1.9%) | 7 (2.0%) | 1 (1.7%) |
Continuous variables are presented as medians with interquartile ranges. Categorical variables are presented as counts and percentages. Group comparisons were performed using the Mann–Whitney U test, Pearson’s chi-square test, or Fisher’s exact test, as appropriate. Available case numbers are shown for variables with missing data. Missing values were not imputed for descriptive comparisons. ICU length of stay is presented descriptively and was not used as a baseline predictor for model development. APACHE II and SOFA scores, mechanical ventilation, and CRRT were assessed within the first 24 h after ICU admission
Abbreviations: Alb albumin, APACHE II Acute Physiology and Chronic Health Evaluation II, BMI body mass index, CCI Charlson Comorbidity Index, CEA carcinoembryonic antigen, CRRT continuous renal replacement therapy, ICU intensive care unit, IQR interquartile range, SOFA Sequential Organ Failure Assessment, WBC white blood cell count
Independent risk factor analysis
Complete-case analysis was used for multivariable model development. A total of 405 patients with complete data for the final predictors and outcome were included in the final multivariable logistic regression model, including 351 survivors and 54 non-survivors. With 54 outcome events and five final predictors, the events-per-variable ratio was 10.8. Missing data for key variables are summarized in Supplementary Table S1.
Based on clinical relevance, univariate associations, event number, and potential collinearity, five variables were included in the final model: preoperative albumin, APACHE II score, CRRT within the first 24 h after ICU admission, mechanical ventilation within the first 24 h after ICU admission, and laparoscopic surgery.
In multivariable logistic regression analysis, higher preoperative albumin level was independently associated with lower 3-month mortality risk (odds ratio [OR] = 0.930, 95% confidence interval [CI]: 0.883–0.980, P = 0.007). Higher APACHE II score during the first 24 h after ICU admission was associated with increased mortality risk (OR = 1.129, 95% CI: 1.071–1.189, P < 0.001). Requirement for CRRT within the first 24 h after ICU admission was also associated with increased mortality risk (OR = 4.980, 95% CI: 1.716–14.457, P = 0.003), as was the requirement for mechanical ventilation within the first 24 h after ICU admission (OR = 2.976, 95% CI: 1.192–7.429, P = 0.019). Laparoscopic surgery was associated with lower 3-month mortality risk compared with non-laparoscopic surgery (OR = 0.379, 95% CI: 0.182–0.789, P = 0.010).
The complete-case multivariable logistic regression results are summarized in Table 2.
Table 2.
Complete-case multivariable logistic regression analysis for 3-month mortality
| Variable | OR | 95% CI | P value | Raw_P |
|---|---|---|---|---|
| CRRT (Yes) | 4.98 | 1.716–14.457 | 0.003 | 0.003150642 |
| Mechanical ventilation (Yes) | 2.976 | 1.192–7.430 | 0.019 | 0.019482552 |
| Albumin | 0.93 | 0.883–0.980 | 0.007 | 0.006789331 |
| APACHE II | 1.129 | 1.071–1.189 | < 0.001 | < 0.001 |
| Laparoscopic surgery | 0.379 | 0.182–0.789 | 0.01 | 0.009501861 |
The multivariable model was fitted using complete-case analysis and included 405 patients with 54 outcome events. Mechanical ventilation and CRRT were defined as organ-support interventions required within the first 24 h after ICU admission
OR odds ratio, CI confidence interval, APACHE II Acute Physiology and Chronic Health Evaluation II, CRRT continuous renal replacement therapy
Construction and internal validation of the nomogram
A nomogram incorporating the five independent predictors was constructed to estimate the individual probability of 3-month all-cause mortality after ICU admission (Fig. 1).
Fig. 1.
Nomogram for predicting 3-month mortality in colorectal cancer patients requiring perioperative ICU admission. Note: The nomogram was constructed using five predictors identified in the complete-case multivariable logistic regression model: preoperative albumin, APACHE II score, CRRT within the first 24 h after ICU admission, mechanical ventilation within the first 24 h after ICU admission, and laparoscopic surgery. To use the nomogram, the value of each predictor is located on its corresponding axis and projected to the points scale. The points assigned to each predictor are summed to obtain the total score, which corresponds to the predicted probability of 3-month all-cause mortality after ICU admission
The apparent AUC of the nomogram was 0.882 (95% CI, 0.829–0.927), indicating good discrimination (Fig. 2A). Internal validation using 1,000 bootstrap resamples yielded an optimism-corrected C-index of 0.872.
Fig. 2.
Discrimination and calibration of the nomogram. Note: (A) Receiver operating characteristic curve of the nomogram for predicting 3-month all-cause mortality. The apparent AUC was 0.882 (95% CI, 0.829–0.927). B Calibration plot based on bootstrap internal validation. The x-axis represents the predicted probability of 3-month mortality, and the y-axis represents the observed probability. The diagonal dashed line represents perfect calibration. The calibration analysis showed good agreement between predicted and observed risk. The apparent calibration intercept and slope were 0.000 and 1.000, respectively. After bootstrap correction, the calibration slope was 0.923. The mean absolute error from bootstrap calibration was 0.012
Calibration analysis showed good agreement between predicted and observed 3-month mortality risk (Figure 2B). The apparent calibration intercept and slope were 0.000 and 1.000, respectively. After bootstrap correction, the calibration slope was 0.923. The mean absolute error from bootstrap calibration was 0.012.
Decision curve analysis
Decision curve analysis was performed to evaluate the potential clinical utility of the nomogram across a range of threshold probabilities (Fig. 3). The nomogram showed a higher net benefit than the default strategies of treating all patients or treating none across clinically relevant threshold probabilities, particularly between approximately 5% and 50%, suggesting that the model may provide clinical value for early risk stratification in this ICU population.
Fig. 3.
Decision curve analysis of the nomogram. Note: Decision curve analysis was used to evaluate the net benefit of the nomogram across different threshold probabilities. The nomogram showed higher net benefit than the treat-all and treat-none strategies across clinically relevant threshold probabilities, particularly between approximately 5% and 50%
Comparison of predictive performance between the nomogram and APACHE II score
To evaluate whether the combined nomogram improved discrimination beyond a general ICU severity score, its performance was compared with that of APACHE II alone. The nomogram achieved a higher AUC than APACHE II alone (0.882 vs. 0.830; Fig. 4). DeLong’s test indicated that this difference was statistically significant (P = 0.007). These findings suggest that integrating preoperative albumin, surgical approach, and early ICU organ-support variables with APACHE II was associated with improved discrimination for 3-month mortality in this cohort.
Fig. 4.
Comparison of receiver operating characteristic curves between the nomogram and APACHE II score. Note: The nomogram showed a higher AUC than APACHE II alone for predicting 3-month all-cause mortality after ICU admission (0.882 vs. 0.830). The difference was statistically significant according to DeLong’s test (P = 0.007), suggesting improved discrimination in this cohort
Sensitivity analyses
Sensitivity analyses were performed to evaluate the robustness of the final model and to address the potential contribution of tumor burden and treatment-related predictors.
First, TNM stage was added to the primary model. In this extended model, TNM stage was not independently associated with 3-month mortality (P = 0.664) and did not improve discrimination. The AUC of the extended model including TNM stage was 0.881, which was similar to that of the primary model.
Second, a reduced model excluding CRRT and mechanical ventilation was fitted to assess model performance without early organ-support variables. This reduced model retained acceptable discrimination, with an AUC of 0.854.
These findings suggest that TNM stage provided limited incremental prognostic value after accounting for acute physiological derangement and early ICU variables, whereas early organ-support indicators contributed additional prognostic information in this cohort. The results of the sensitivity analyses are summarized in Supplementary Table S2.
Discussion
Perioperative ICU admission after colorectal cancer surgery represents a clinically important high-risk scenario, in which short-term mortality is often driven not only by tumor-related factors but also by acute physiological derangement, organ dysfunction, infection, and the need for organ support. In the present study, we developed and internally validated a nomogram for estimating 3-month all-cause mortality in colorectal cancer patients requiring perioperative ICU admission. To our knowledge, this is among the first nomograms developed specifically for this critically ill surgical oncology population. The model was designed for use at the end of the first 24 h after ICU admission and integrated preoperative albumin, APACHE II score, CRRT, mechanical ventilation, and laparoscopic surgery. The nomogram showed good discrimination, calibration, and potential clinical utility in the internal cohort, and it outperformed APACHE II alone in discriminating 3-month mortality risk.
The clinical rationale for focusing on this population is that conventional oncological staging and general ICU scoring systems capture different aspects of risk. TNM stage is well established for long-term oncological prognosis, but early mortality after perioperative ICU admission is frequently influenced by acute systemic inflammation, sepsis, respiratory failure, renal dysfunction, and multiple organ dysfunction syndrome (MODS) [4]. Conversely, general ICU scores such as APACHE II and SOFA reflect acute physiological severity, but they do not fully incorporate cancer- and surgery-related information, including nutritional reserve and operative approach. The present model therefore attempts to bridge this gap by combining baseline patient reserve, surgical characteristics, and early ICU severity indicators.
Preoperative albumin was identified as an independent protective factor against 3-month mortality. This finding is consistent with previous studies showing that hypoalbuminemia is associated with adverse postoperative outcomes in colorectal and other gastrointestinal surgical populations [8–14]. Albumin is not only a marker of nutritional status but also reflects systemic inflammation, tumor-related catabolism, and overall physiological reserve [15, 16]. In critically ill surgical patients, low albumin may be associated with impaired tissue repair, reduced intravascular oncotic pressure, interstitial edema, and compromised microcirculation, which can adversely affect postoperative recovery and increase vulnerability to infection, anastomotic complications, and organ dysfunction [17–20]. In this context, preoperative albumin may serve as a simple and objective indicator of the patient’s ability to tolerate surgical trauma and subsequent critical illness.
The requirement for CRRT and mechanical ventilation within the first 24 h after ICU admission was independently associated with increased 3-month mortality. These findings are consistent with previous reports linking perioperative acute kidney injury requiring renal replacement therapy and postoperative respiratory failure with poor outcomes [21, 22]. Importantly, these variables should not be interpreted as causal treatment effects. Rather, they represent early markers of severe renal and respiratory dysfunction requiring organ support. Their inclusion allowed the model to capture early progression of critical illness beyond APACHE II score alone. At the same time, mechanical ventilation and CRRT are treatment-related variables, and their use may be influenced by institutional ICU admission policies, thresholds for organ-support initiation, and clinical practice patterns. Therefore, although these variables were available at the intended prediction time point, their inclusion may affect external transportability and should be considered when applying the nomogram in other settings.
Laparoscopic surgery was associated with lower 3-month mortality in the multivariable model. This finding is broadly consistent with the potential benefits of minimally invasive colorectal surgery, including reduced surgical trauma, lower inflammatory response, and faster postoperative recovery, as reported in previous colorectal surgery studies such as the COLOR II trial [23]. However, this association should be interpreted cautiously. In real-world surgical practice, patients selected for laparoscopic surgery are often more physiologically stable, have more favorable anatomy or disease extent, and are less likely to require emergency open procedures. Conversely, patients undergoing open surgery may have more advanced disease, severe obstruction or perforation, hemodynamic instability, or more complex operative conditions [24, 25]. Therefore, laparoscopic surgery in this model should be understood not only as a surgical approach but also as a marker of patient selection, disease extent, and physiological fitness. Residual confounding cannot be fully excluded in this retrospective cohort.
Several variables that were significant in univariate analysis, including bowel obstruction, TNM stage, extent of resection, adjuvant chemotherapy, infection, and vasopressor use, were not retained in the primary model. Bowel obstruction may contribute to intestinal barrier dysfunction, bacterial translocation, fluid imbalance, and increased surgical risk [26–29]. Advanced tumor burden, complex tumor location, systemic inflammation, and operative case selection may also increase perioperative risk or influence surgical decision-making [30–32]. Similarly, vasopressor use reflects severe circulatory compromise and is closely related to sepsis, vasoplegia, and shock management, but it may overlap with APACHE II score and other early organ-support variables [33, 34].
However, the primary aim of this model was to estimate short-term mortality at the end of the first 24 h after ICU admission, when acute physiological derangement and early organ-support requirements were expected to dominate prognosis. Given the limited number of outcome events, the final model was deliberately restricted to avoid overfitting. To address the concern regarding TNM stage, we performed a sensitivity analysis by adding TNM stage to the primary model. TNM stage was not independently associated with 3-month mortality in this extended model and did not improve discrimination, suggesting limited incremental value after accounting for APACHE II score, albumin, surgical approach, and early ICU organ-support indicators in this cohort.
The decision not to retain SOFA in the final model also requires explanation. Both SOFA and APACHE II were significantly higher in non-survivors than in survivors, reflecting the importance of acute illness severity. However, because both scores capture overlapping aspects of physiological derangement and organ dysfunction, including both in a model with a limited number of events could increase collinearity and reduce model stability. APACHE II was retained because it incorporates broader acute physiological information and chronic health-related factors during the early ICU period. No substantial multicollinearity was observed among predictors in the final model, with VIF values ranging from 1.032 to 1.213. APACHE II and SOFA were highly correlated (r = 0.836), supporting the decision to retain APACHE II and exclude SOFA from the final model.Previous studies have also shown that general ICU scoring systems have variable performance in oncology and surgical populations, supporting the need for disease- and context-specific risk models [35, 36].
The nomogram demonstrated good discrimination, with an AUC of 0.882 and an optimism-corrected C-index of 0.872 after 1,000 bootstrap resamples. Calibration analysis also suggested good agreement between predicted and observed risk. Decision curve analysis further indicated that the nomogram may offer a higher net benefit than the treat-all and treat-none strategies across clinically relevant threshold probabilities. These findings suggest that the combination of preoperative reserve, acute physiological severity, early organ-support requirements, and surgical approach may improve risk stratification in colorectal cancer patients requiring perioperative ICU care. Nevertheless, the model should be regarded as a preliminary risk-stratification tool rather than a model ready for direct clinical implementation.
This study has several limitations. First, it was a single-center retrospective study, which may limit generalizability. ICU admission criteria, perioperative management strategies, and thresholds for mechanical ventilation or CRRT may vary across institutions. Second, the number of outcome events was limited. Although the final complete-case model included 54 deaths for five predictors, corresponding to an events-per-variable ratio of 10.8, the relatively small number of events may still affect model stability and the adaptability of the nomogram to external populations [37, 38]. Third, external validation was not performed. Although bootstrap internal validation was used to reduce optimism and assess internal consistency, independent external validation is necessary before routine clinical use can be recommended. Fourth, missing data were handled using complete-case analysis. Although missingness in the final predictors was low, complete-case analysis may introduce selection bias. Fifth, mechanical ventilation and CRRT are treatment-related variables. Even though they were recorded within the first 24 h after ICU admission and were available at the intended prediction time point, confounding by indication and institutional practice variation may influence their prognostic meaning. Finally, variable selection was based on clinical relevance, univariate associations, event number, and collinearity considerations. This pragmatic approach was intended to reduce overfitting, but it may still introduce model instability. Future prospective multicenter studies with larger sample sizes, predefined predictor windows, and external validation cohorts are needed to confirm the robustness and clinical applicability of this nomogram.
Conclusions
In this single-center retrospective cohort study, we developed and internally validated a nomogram for estimating 3-month all-cause mortality in colorectal cancer patients requiring perioperative ICU admission. The model was designed for use at the end of the first 24 h after ICU admission and incorporated preoperative albumin, APACHE II score, CRRT, mechanical ventilation, and laparoscopic surgery. The nomogram showed good discrimination, calibration, and potential clinical utility in the internal cohort, and it performed better than APACHE II alone for short-term mortality prediction.
This model may support early risk stratification and communication among ICU physicians, surgeons, patients, and families. However, because the study was retrospective and single-center with a limited number of outcome events and no external validation, the nomogram should be regarded as a preliminary risk-assessment tool. Prospective multicenter validation is required before routine clinical implementation.
Supplementary Information
Acknowledgements
During the preparation of this work, the authors utilized a large language model (Gemini) strictly to polish the English language and improve readability. The authors independently reviewed and edited the content following the use of this tool and assume comprehensive responsibility for the ultimate integrity and content of this publication.
Abbreviations
- Alb
Albumin
- APACHE II
Acute Physiology and Chronic Health Evaluation II
- ARDS
Acute respiratory distress syndrome
- AUC
Area under the curve
- BMI
Body mass index
- CCI
Charlson Comorbidity Index
- CEA
Carcinoembryonic antigen
- CI
Confidence interval
- C-index
Concordance index
- CRRT
Continuous renal replacement therapy
- DCA
Decision curve analysis
- EPV
Events per variable
- ERAS
Enhanced Recovery After Surgery
- Hb
Hemoglobin
- ICU
Intensive Care Unit
- IQR
Interquartile range
- MAE
Mean absolute error
- MODS
Multiple organ dysfunction syndrome
- OR
Odds ratio
- PLT
Platelet
- ROC
Receiver operating characteristic
- SD
Standard deviation
- SIRS
Systemic inflammatory response syndrome
- SOFA
Sequential Organ Failure Assessment
- TNM
Tumor, Node, Metastasis
- VA
Veterans Affairs
- WBC
White blood cell
Authors’ contributions
CD: Study conception and design; Drafting of the manuscript.CD & DC: Data acquisition and analysis.QL & XZh: Critical revision of the manuscript for important intellectual content.All authors: Final approval of the version to be published.
Funding
Not applicable.
Data availability
The datasets generated and/or analysed during the current study are not publicly available due to patient privacy and institutional confidentiality policies, but are available from the corresponding author on reasonable request.
Declarations
Ethics approval and consent to participate
Ethical clearance for this investigation was granted by the Medical Ethics Committee of the Sixth Affiliated Hospital of Sun Yat-sen University (Approval No. E2025324). Given the retrospective design and the use of deidentified patient records, the necessity for obtaining written informed consent was officially exempted by the institutional committee.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Chengcheng Dong and Daiyin Cao contributed equally to this work and share first authorship.
Contributor Information
Xiaofei Zhang, Email: zhxiaof5@mail.sysu.edu.cn.
Qianxin Luo, Email: Luoqx23@mail.sysu.edu.cn.
References
- 1.Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209–49. [DOI] [PubMed] [Google Scholar]
- 2.Benson AB, Venook AP, Al-Hawary MM, Arain MA, Chen YJ, Ciombor KK, et al. Colon Cancer, Version 2.2021, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw. 2021;19(3):329–59. [DOI] [PubMed] [Google Scholar]
- 3.Gustafsson UO, Scott MJ, Hubner M, Nygren J, Demartines N, Francis N, et al. Guidelines for Perioperative Care in Elective Colorectal Surgery: Enhanced Recovery After Surgery (ERAS(®)) Society Recommendations: 2018. World J Surg. 2019;43(3):659–95. [DOI] [PubMed] [Google Scholar]
- 4.Puxty K, McLoone P, Quasim T, Kinsella J, Morrison D. Survival in solid cancer patients following intensive care unit admission. Intensive Care Med. 2014;40(10):1409–28. [DOI] [PubMed] [Google Scholar]
- 5.Amin MB, Greene FL, Edge SB, Compton CC, Gershenwald JE, Brookland RK, et al. The Eighth Edition AJCC Cancer Staging Manual: Continuing to build a bridge from a population-based to a more personalized approach to cancer staging. CA Cancer J Clin. 2017;67(2):93–9. [DOI] [PubMed] [Google Scholar]
- 6.Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: a severity of disease classification system. Crit Care Med. 1985;13(10):818–29. [PubMed] [Google Scholar]
- 7.Vincent JL, Moreno R, Takala J, Willatts S, De Mendonça A, Bruining H, et al. The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure. On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine. Intensive Care Med. 1996;22(7):707–10. [DOI] [PubMed] [Google Scholar]
- 8.Moghadamyeghaneh Z, Hwang G, Hanna MH, Phelan MJ, Carmichael JC, Mills SD, et al. Even modest hypoalbuminemia affects outcomes of colorectal surgery patients. Am J Surg. 2015;210(2):276–84. [DOI] [PubMed] [Google Scholar]
- 9.Christina NM, Tjahyanto T, Lie JG, Santoso TA, Albertus H, Octavianus D, et al. Hypoalbuminemia and colorectal cancer patients: Any correlation? A systematic review and meta-analysis. Med (Baltim). 2023;102(8):e32938. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Kang SC, Kim HI, Kim MG. Low Serum Albumin Level, Male Sex, and Total Gastrectomy Are Risk Factors of Severe Postoperative Complications in Elderly Gastric Cancer Patients. J Gastric Cancer. 2016;16(1):43–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Ahmetasevic E, Hasukic S, Agic M, Brkic S. Preoperative Hypoalbuminemia as a Predictor of Severe Postoperative Complications in Patients Undergoing Whipple Pancreatoduodenectomy. Med Arch. 2023;77(5):350–3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Geng L, Zong R, Shi Y, Xu K. Prognostic role of preoperative albumin-bilirubin grade on patients with hepatocellular carcinoma after surgical resection: a systematic review and meta-analysis. Eur J Gastroenterol Hepatol. 2020;32(7):769–78. [DOI] [PubMed] [Google Scholar]
- 13.Gibbs J, Cull W, Henderson W, Daley J, Hur K, Khuri SF. Preoperative serum albumin level as a predictor of operative mortality and morbidity: results from the National VA Surgical Risk Study. Arch Surg. 1999;134(1):36–42. [DOI] [PubMed] [Google Scholar]
- 14.Alajmi A, Almehari A, Alzahrani AR, Aljurays Y, Alzahrani N, Aladel AM, et al. Impact of Preoperative Serum Albumin Level on the Outcome of Colorectal Cancer Surgery. Cureus. 2024;16(4):e57655. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Nazha B, Moussaly E, Zaarour M, Weerasinghe C, Azab B. Hypoalbuminemia in colorectal cancer prognosis: Nutritional marker or inflammatory surrogate? World J Gastrointest Surg. 2015;7(12):370–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.McMillan DC. Systemic inflammation, nutritional status and survival in patients with cancer. Curr Opin Clin Nutr Metab Care. 2009;12(3):223–6. [DOI] [PubMed] [Google Scholar]
- 17.Bosmans JW, Jongen AC, Bouvy ND, Derikx JP. Colorectal anastomotic healing: why the biological processes that lead to anastomotic leakage should be revealed prior to conducting intervention studies. BMC Gastroenterol. 2015;15:180. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Golub R, Golub RW, Cantu R Jr., Stein HD. A multivariate analysis of factors contributing to leakage of intestinal anastomoses. J Am Coll Surg. 1997;184(4):364–72. [PubMed] [Google Scholar]
- 19.Sartelli M, Chichom-Mefire A, Labricciosa FM, Hardcastle T, Abu-Zidan FM, Adesunkanmi AK, et al. The management of intra-abdominal infections from a global perspective: 2017 WSES guidelines for management of intra-abdominal infections. World J Emerg Surg. 2017;12:29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Daams F, Luyer M, Lange JF. Colorectal anastomotic leakage: aspects of prevention, detection and treatment. World J Gastroenterol. 2013;19(15):2293–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Uusalo P, Hellman T, Järvisalo MJ. Mortality and associated risk factors in perioperative acute kidney injury treated with continuous renal replacement therapy. Perioper Med (Lond). 2021;10(1):57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Serpa Neto A, Hemmes SN, Barbas CS, Beiderlinden M, Fernandez-Bustamante A, Futier E, et al. Incidence of mortality and morbidity related to postoperative lung injury in patients who have undergone abdominal or thoracic surgery: a systematic review and meta-analysis. Lancet Respir Med. 2014;2(12):1007–15. [DOI] [PubMed] [Google Scholar]
- 23.van der Pas MH, Haglind E, Cuesta MA, Fürst A, Lacy AM, Hop WC, et al. Laparoscopic versus open surgery for rectal cancer (COLOR II): short-term outcomes of a randomised, phase 3 trial. Lancet Oncol. 2013;14(3):210–8. [DOI] [PubMed] [Google Scholar]
- 24.Verzaro R, Mattia S, Rago T, Casella F, Ferroni A, Gianfreda V, et al. Selection Bias in Colorectal Surgery in a Non-Tertiary Hospital: Laparoscopic Versus Open Surgery. J Laparoendosc Adv Surg Tech A. 2018;28(3):263–8. [DOI] [PubMed] [Google Scholar]
- 25.Pisano M, Zorcolo L, Merli C, Cimbanassi S, Poiasina E, Ceresoli M, et al. 2017 WSES guidelines on colon and rectal cancer emergencies: obstruction and perforation. World J Emerg Surg. 2018;13:36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Biondo S, Parés D, Frago R, Martí-Ragué J, Kreisler E, De Oca J, et al. Large bowel obstruction: predictive factors for postoperative mortality. Dis Colon Rectum. 2004;47(11):1889–97. [DOI] [PubMed] [Google Scholar]
- 27.Deitch EA. Bacterial translocation or lymphatic drainage of toxic products from the gut: what is important in human beings? Surgery. 2002;131(3):241–4. [DOI] [PubMed] [Google Scholar]
- 28.Tekkis PP, Kinsman R, Thompson MR, Stamatakis JD. The Association of Coloproctology of Great Britain and Ireland study of large bowel obstruction caused by colorectal cancer. Ann Surg. 2004;240(1):76–81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Iversen LH, Bülow S, Christensen IJ, Laurberg S, Harling H. Postoperative medical complications are the main cause of early death after emergency surgery for colonic cancer. Br J Surg. 2008;95(8):1012–9. [DOI] [PubMed] [Google Scholar]
- 30.Baracos VE, Martin L, Korc M, Guttridge DC, Fearon KCH. Cancer-associated cachexia. Nat Rev Dis Primers. 2018;4:17105. [DOI] [PubMed] [Google Scholar]
- 31.Watt DG, Horgan PG, McMillan DC. Routine clinical markers of the magnitude of the systemic inflammatory response after elective operation: a systematic review. Surgery. 2015;157(2):362–80. [DOI] [PubMed] [Google Scholar]
- 32.Miskovic D, Ni M, Wyles SM, Tekkis P, Hanna GB. Learning curve and case selection in laparoscopic colorectal surgery: systematic review and international multicenter analysis of 4852 cases. Dis Colon Rectum. 2012;55(12):1300–10. [DOI] [PubMed] [Google Scholar]
- 33.Levy B, Fritz C, Tahon E, Jacquot A, Auchet T, Kimmoun A. Vasoplegia treatments: the past, the present, and the future. Crit Care. 2018;22(1):52. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Evans L, Rhodes A, Alhazzani W, Antonelli M, Coopersmith CM, French C, et al. Surviving Sepsis Campaign: International Guidelines for Management of Sepsis and Septic Shock 2021. Crit Care Med. 2021;49(11):e1063–143. [DOI] [PubMed] [Google Scholar]
- 35.Ho KM, Lee KY, Williams T, Finn J, Knuiman M, Webb SA. Comparison of Acute Physiology and Chronic Health Evaluation (APACHE) II score with organ failure scores to predict hospital mortality. Anaesthesia. 2007;62(5):466–73. [DOI] [PubMed] [Google Scholar]
- 36.den Boer S, de Keizer NF, de Jonge E. Performance of prognostic models in critically ill cancer patients - a review. Crit Care. 2005;9(4):R458–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350:g7594. [DOI] [PubMed] [Google Scholar]
- 38.Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol. 1996;49(12):1373–9. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets generated and/or analysed during the current study are not publicly available due to patient privacy and institutional confidentiality policies, but are available from the corresponding author on reasonable request.