Developing a calculable risk prediction model for sternal wound infection after median sternotomy: a retrospective study

Abstract

Background: Diagnosing sternal wound infection (SWI) following median sternotomy remains laborious and troublesome, resulting in high mortality rates and great harm to patients. Early intervention and prevention are critical and challenging. This study aimed to develop a simple risk prediction model to identify high-risk populations of SWI and to guide examination programs and intervention strategies.

Methods: A retrospective analysis was conducted on the clinical data obtained from 6715 patients who underwent median sternotomy between January 2016 and December 2020. The least absolute shrink and selection operator (LASSO) regression method selected the optimal subset of predictors, and multivariate logistic regression helped screen the significant factors. The nomogram model was built based on all significant factors. Area under the curve (AUC), calibration curve and decision curve analysis (DCA) were used to assess the model's performance.

Results: LASSO regression analysis selected an optimal subset containing nine predictors that were all statistically significant in multivariate logistic regression analysis. Independent risk factors of SWI included female [odds ratio (OR) = 3.405, 95% confidence interval (CI) = 2.535-4.573], chronic obstructive pulmonary disease (OR = 4.679, 95% CI = 2.916-7.508), drinking (OR = 2.025, 95% CI = 1.437-2.855), smoking (OR = 7.059, 95% CI = 5.034-9.898), re-operation (OR = 3.235, 95% CI = 1.087-9.623), heart failure (OR = 1.555, 95% CI = 1.200-2.016) and repeated endotracheal intubation (OR = 1.975, 95% CI = 1.405-2.774). Protective factors included bone wax (OR = 0.674, 95% CI = 0.538-0.843) and chest physiotherapy (OR = 0.446, 95% CI = 0.248-0.802). The AUC of the nomogram was 0.770 (95% CI = 0.745-0.795) with relatively good sensitivity (0.798) and accuracy (0.620), exhibiting moderately good discernment. The model also showed an excellent fitting degree on the calibration curve. Finally, the DCA presented a remarkable net benefit.

Conclusions: A visual and convenient nomogram-based risk calculator built on disease-associated predictors might help clinicians with the early identification of high-risk patients of SWI and timely intervention.

Keywords: Median sternotomy; Nomogram; Postcardiac surgical complications; Prediction model; Sternal wound infection.

Figure 1

Flowchart showing the process of selecting patients for the study. HIV human immunodeficiency virus, SWI sternal wound infection

Figure 2

Discriminative features selection by the least absolute shrink and selection operator (LASSO) model. By identifying the optimal penalization coefficient (lambda) in the LASSO model with 10-fold cross-validation, the partial likelihood deviance (binomial deviance) curve was plotted against log(lambda). Dotted vertical lines were drawn at the value with the minimum criteria (Left) and 1 standard error of the minimum criteria (Right)

Figure 3

Nomogram to predict the risk of sternal wound infection (SWI) after median sternotomy. The nomogram offered a visual and quantifiable scoring system baesd on a combination of clinical characteristics to estimate the probability of developing SWI. COPD chronic obstructive pulmonary disease, CPT chest physiotherapy

Figure 4

Receiver operator characteristic curve of the nomogram model. CI confidence interval, AUC area under the curve

Figure 5

Internal calibration curves for the nomogram model. The predicted probability of sternal wound infection (SWI) was plotted on the x-axis and the actual observed probability of SWI was plotted on the y-axis. The diagonal line (45° line) meant a perfect prediction by an ideal model. The apparent calibration curve (dotted line) represented the calibration of the nomogram model. The bias-corrected curve (solid line) represented the actual predictive performance after correcting the optimism with 1000-times bootstrap resampling. A closer fit to the diagonal line represented a better prediction

Figure 6

Internal decision curve analysis for the nomogram model. The vertical axis (y-axis) represented the net benefit. The black solid horizontal line represented the assumption that no patient had sternal wound infection (SWI). The grey solid line represented the assumption that all patients had SWI. This decision curve analysis could provide a potential net benefit within the defined range (0–22%)