Nomogram to predict postoperative infectious complications after surgery for colorectal cancer: a retrospective cohort study in China

Background Postoperative infectious complications (ICs) after surgery for colorectal cancer (CRC) increase in-hospital deaths and decrease long-term survival. However, the methodology for IC preoperative and intraoperative risk assessment has not yet been established. We aimed to construct a risk model for IC after surgery for CRC. Methods Between January 2016 and June 2020, a total of 593 patients who underwent curative surgery for CRC in Chengdu Second People’s Hospital were enrolled. Preoperative and intraoperative factors were obtained retrospectively. The least absolute shrinkage and selection operator (LASSO) method was used to screen out risk factors for IC. Then, based on the results of LASSO regression analysis, multivariable logistic regression analysis was performed to establish the prediction model. Bootstraps with 300 resamples were performed for internal validation. The performance of the model was evaluated with its calibration and discrimination. The clinical usefulness was assessed by decision curve analysis (DCA). Results A total of 95 (16.0%) patients developed ICs after surgery for CRC. Chronic pulmonary diseases, diabetes mellitus, preoperative and/or intraoperative blood transfusion, and longer operation time were independent risk factors for IC. A prediction model was constructed based on these factors. The concordance index (C-index) of the model was 0.761. The calibration curve of the model suggested great agreement. DCA showed that the model was clinically useful. Conclusion Several risk factors for IC after surgery for CRC were identified. A prediction model generated by these risk factors may help in identifying patients who may benefit from perioperative optimization.


Introduction
Colorectal cancer (CRC) is among the commonest malignancies worldwide [1][2][3]. Surgical resection is considered the best choice for a potentially radical cure [4][5][6][7]. Even with advances in surgical techniques and perioperative treatment in recent years, mortality and morbidity rates after CRC surgery remain considerable, mainly due to postoperative infectious complications (ICs) [8,9]. ICs after surgery for CRC have been demonstrated to increase cost, hospital stays, and delay the initiation of adjuvant treatments [10]. Importantly, multiple studies have shown that they are associated with decreased long-term survival [10][11][12][13].
Possible explanations for the relationship between postoperative ICs and oncological outcome include (1) escape of intraluminal neoplastic cells in patients with anastomotic leak [11], (2) local and systemic proliferation of proinflammatory cytokines and mediators [11,14], (3) the association of ICs with increased TNM stage [15], (4) delays in the initiation of adjuvant treatments [11], and (5) poor surgical technique, which may increase the incidence of ICs and tumor recurrence. A better understanding of risk factors associated with ICs after surgery for CRC can aid healthcare providers in preoperative counseling and surgical decision-making, suggest complication-reducing strategies, and help in considering preventative measures.
Therefore, we designed the study to identify risk factors for ICs after surgery for CRC. We also used the risk factors to generate a nomogram that can predict the probability of postoperative ICs. We chose to evaluate preoperative and intraoperative factors because this model would be more clinically friendly and useful than models based on postoperative factors when ICs would be imminent. To our knowledge, this is the first prediction model that could predict the possibility of postoperative IC after surgery for CRC.

Study population and ethical issues
Between January 2016 and April 2020, 593 consecutive patients who underwent surgery for primary CRC at Chengdu Second People's Hospital were enrolled in the study. The inclusion criteria were (1) histologically confirmed CRC, (2) patients underwent surgery for CRC with radical resection, (3) patients had resection with a primary anastomosis without a protecting stoma, and (4) patients over 18 years old. The exclusion criteria were (1) palliative surgery, (2) with local surgical treatment (such as trans-anal endoscopic microsurgery), (3) with a stoma (such as Hartmann's procedure, abdominoperineal resection, and anastomosis with a de-functioning stoma), (4) patients less than 18 years old, (5) with emergency surgery, (6) with evidence of infection or systemic autoimmune disease before surgery, and (7) with incomplete medical data. Patient data were extracted from a prospectively maintained CRC database. This study was approved by the Ethics Commission of the hospital (Chengdu Second People's Hospital).

Clinicopathological materials
Various preoperative and intraoperative variables were collected for risk factor selection as follows: basic information: sex, age, body mass index, smoking history, the American Society of Anesthesiologists (ASA) score, preexisting co-morbidities (including heart disease, hypertension requiring medication, chronic pulmonary disease, diabetes mellitus), previous abdominal surgery, neoadjuvant chemo-radiotherapy, and preoperative and/ or intraoperative blood transfusion; laboratory tests information: preoperative hemoglobin and albumin level; tumor information: preoperative TNM stage, tumor location, and tumor size; and surgical information: surgical approach, combined organ resection, intraoperative blood loss, and operation time.
Preoperative staging evaluation included digital rectal examination, rectal endosonography, colonoscopy, and MRI or CT scans. The indication for blood transfusion was a hemoglobin level below 80 g/L. When the hemoglobin level was between 80 and 100 g/L, blood transfusion was selected based on hemodynamics and oxygen saturation [16]. The operations in the study were performed by two surgeons (S.J., and B.J.). Both of them are attending doctors and have at least 14 years of experience in gastrointestinal surgery. Each of them performs at least 230 operations for gastric and colorectal cancer annually since 2015.

Definition of postoperative infectious complications
In the present study, ICs were graded according to the Clavien-Dindo surgical complication system [17]. When a patient had at least two ICs, the higher grade was adopted [18]. ICs were defined as Clavien-Dindo grade II or more severe. ICs included wound infection, anastomosis leakage, intra-abdominal abscesses and collections, cholecystitis, infectious diarrhea, and pneumonia.
(1) Wound infection was confirmed when it gets painful with pustular discharge and/or a positive culture, the opening of the wound, and antibiotic treatment was required. (2) Anastomotic leakage was considered if any of the following situations were observed: fecal or gas discharge from the drain tract, vagina, or the incisional wound; fecal peritonitis; or peritonitis along with anastomotic defect confirmed by rectal examination, endoscopy, laparotomy, or radiological findings [19]. (3) Intraabdominal abscesses and collections were confirmed by ultrasonography or computed tomography (CT) scans, accompanied by systemic inflammatory response lasting We used the least absolute shrinkage and selection operator (LASSO) method to find the optimal variables with non-zero coefficients as risk factors [21]. Then, based on the results of LASSO regression analysis, multivariable logistic regression analysis was used to establish a prediction model, and a nomogram was generated. Bootstraps with 300 resamples were performed for internal validation. The predictive performance was assessed by Harrell's concordance index (C-index). A calibration curve was plotted to evaluate the calibration of the nomogram. A decision curve analysis (DCA) was created to evaluate the clinical usefulness of the nomogram. P-value of < 0.05 was considered significant.

Risk factor selection
We used the LASSO regression analysis to evaluate the 24 variables (Fig. 1). Finally, we screened out 4 variables with nonzero coefficients as potential risk factors of postoperative ICs. These risk factors included chronic pulmonary disease, diabetes mellitus, preoperative and/or intraoperative blood transfusion, and longer operation time.

Nomogram and validation
We further conducted a multivariable logistic regression analysis and generated a prediction model to get a deep insight into the relationship between ICs and these risk factors. The results of the multiple logistic regression analysis are shown in Table 3 and visualized in the form of a nomogram to guide healthcare providers in the clinic (Fig. 2).   (Fig. 3). To use the nomogram, first, draw a vertical line to the top points row to assign points for each factor, and then, add the points from each factor together and drop a vertical line from the total points row to get the risk of IC.

Clinical usefulness
The decision curve analysis for the nomogram is shown in Fig. 3B. It showed that using the nomogram to predict ICs following surgery for CRC added more net benefit than the treat-all or treat-none strategies when the threshold probability is greater than 0.23.

Discussion
IC remains the most significant cause of early morbidity and it decreases long-term survival after surgery for CRC [10,22,23]. Therefore, early recognition and prevention of IC in high-risk patients is an important issue. In the present study, a considerable number of patients Fig. 1 Risk factors selection using the least absolute shrinkage and selection operator (LASSO) logistic regression model. Final risk factors include chronic pulmonary disease, diabetes mellitus, preoperative and/or intraoperative blood transfusion, and longer operation time. a Optimal parameter (λ) selection in the LASSO model used five-fold cross-validation and minimum criteria. The partial likelihood deviance (binomial deviance) curve was plotted versus log(λ). Dotted vertical lines were drawn at the optimal values by using the minimum criteria and the 1 SE of the minimum criteria (the 1-SE criteria). b LASSO coefficient profiles of the 26 features. A coefficient profile plot was plotted against the log (λ) sequence, and the 4 non-zero coefficients were chosen at the values selected using fivefold cross-validation. SE, standard error  [24][25][26][27]. Furthermore, chronic pulmonary diseases, diabetes mellitus, preoperative and/or intraoperative blood transfusion, and longer operation time were identified as independent risk factors for ICs. A satisfactory model for ICs was also constructed based on these risk factors. The model can be used to target IC prevention and monitor interventions beyond standard infection prevention in highrisk patients who are likely to benefit.
In the present study, patient-related factors (chronic pulmonary disease and diabetes mellitus) were identified as independent risk factors for IC after surgery for CRC, which is in well agreement with previous literature [28][29][30]. Therefore, special attention should be paid to patients with these co-morbidities and we believe that preoperative treatment of these co-morbidities is essential for postoperative recovery in CRC patients.
As an indicator of the complexity and difficulty of the operation [31], our data validate previous studies that longer operation time is an independent predictor for IC [25,32,33]. Longer operation time may increase susceptibility to infection, resulting in IC development after surgery for CRC [7,25]. Blood transfusion was another independent risk factor for IC. These findings were consistent with a previous study [34]. Although blood transfusion can improve oxygen delivery capacity and tissue perfusion in patients with severe anemia, it may also lead to systemic inflammation and other transfusion-related adverse events, particularly acute lung injury and infection [35,36]. Furthermore, preoperative and intraoperative blood transfusions may reflect the patient's poor Fig. 2 Nomogram for predicting IC following surgery for CRC. The nomogram was generated based on chronic pulmonary disease, diabetes mellitus, preoperative and/or intraoperative blood transfusion, and longer operation time systemic condition or complexity of the surgery [37]. Therefore, special attention should be paid to CRC patients who have a blood transfusion in the perioperative period.
In the present study, we constructed a model to predict the possibility of IC after surgery for CRC. Healthcare providers could make individualized predictions of the IC probability with this model, which aligns with the current concept of personalized medicine [38]. Knowledge of the risk factors for IC would allow intervening in two ways: prevention and rigorous follow-up in highrisk patients after surgery. Prevention can be achieved by preoperative optimization of some high-risk conditions and correcting risk factors such as chronic pulmonary disease using bronco-dilatator treatment before surgery. A rigorous postoperative follow-up could allow the early recognition of IC, thus enabling its early intervention.
The strengths of the study are that it included a wide range of potential risk factors for IC. The proposed model was created based on routinely collected perioperative information to maximize its application and generalizability. Furthermore, we used the LASSO regression to identify risk factors for IC. LASSO regression allows selecting factors to include in the regression model, avoiding the usual methods of automatic factor selection (such as forward, backward and stepwise method), which have been previously reported to give wrong results in some situations [21]. Our study also had some limitations. First, the retrospective nature of the study may introduce bias. Prospective studies are needed to validate the prediction model. Second, the study was only a single-center study and the results were internally validated, external validation is needed to determine whether the results can be applied to other institutions.

Conclusion
Several risk factors for IC after surgery for CRC were identified. A prediction model generated by these risk factors may help in identifying patients who may benefit from perioperative optimization.  Fig. 3 a Calibration curve of the nomogram for predicting IC following surgery for CRC. b Decision curve analysis for predicting IC following surgery for CRC. The x-axis shows the threshold probability. The y-axis represents net benefit. "None" to the assumption that no patient developed IC and "All" refers to the assumption that all patients developed IC. When the score is greater than 0.23, using the nomogram to predict IC adds more net benefit than the treat-none or treat-all strategies