Latent Class Analysis and Phenotypes Identification of Surgical Site Infection in Elderly Patients After Non-cardiac Surgery---Based on a Prediction Model Established by Two Centers Large Sample
Latent Class Analysis and Phenotypes Identification of Surgical Site Infection in Elderly Patients After Non-cardiac Surgery---Based on a Prediction Model Established by Two Centers Large Sample
This study utilizes Latent Class Analysis (LCA) to identify phenotypes of Surgical Site Infection (SSI) in elderly patients following non-cardiac surgery. By analyzing data from two large cohorts, the research establishes a predictive model that uncovers independent risk factors for SSI, including age, hyperlipidemia, and surgical characteristics. The model, with AUCs ranging from 0.753 to 0.791 across cohorts, offers a calibrated prediction of SSI risk. Furthermore, LCA delineates four distinct SSI subphenotypes, highlighting a critical subgroup with a higher infection rate. This subgroup presents a complex interplay of risk factors, indicating the need for tailored preventive strategies. The study's findings contribute to a nuanced understanding of SSI in elderly surgical patients and pave the way for more targeted infection control measures.
Backgrounds:
With the widespread use of prophylactic antibiotics during perioperative period and the continuous promotion of minimally invasive non-cardiac surgery type such as laparoscopic and thoracoscopic surgery, the incidence of superficial Surgical Site Infection (SSI) has been significantly reduced. Organ/deep SSI has become the dominant type of SSI. Currently, the classification of SSI is limited to the above location from shallow to deep, the epidemiological and clinical characteristics of SSI after non-cardiac surgery in elderly patients are still inadequately defined.
Objectives:
The investigators aimed to determine main risk factors for SSI after non-cardiac surgery among elderly patients in China and to further reveal the clinical attributes of those elderly patients afflicted with SSI.
Methods:
Potential risk factors for developing SSI were selected based on published data, clinical expertise, pathophysiological reasoning, and convenient considerations for future clinical applications. These SSI outcomes were rigorously calibrated by researchers complying with back-to-back principles following uniform diagnostic standards--European Perioperative Clinical Outcome (EPCO) definitions. According to the definitions, SSI in this study consists of three sites: superficial incision, deep incision, and organ/deep. The investigators define the occurrence of any of the above sites as SSI infection. Multivariable logistic regression analysis was used to identify risk factors for SSI. Data from population-based cohort of elderly patients undergoing non-cardiac and non-neurology surgery were used to derive the model. The risk prediction model was derived from the First Medical Center of the Chinese PLA General Hospital (January 2012 - August 2018). The investigators performed a nomogram, a complanation model based on the regression model with the graduated line segments as the main body. The discrimination was compared based on the AUC. The calibration was assessed by the calibration intercept and the slope. Decision curve analysis (DCA) was adopted to determine the nomogram's clinical usefulness and net benefit. Latent class analysis (LCA) was further used to explore the population features of SSI. LCA combines the latent variable theory with classified variables to explore the category latent variables behind statistically related classified explicit variables. Utilizing LCA, patients were classified into distinct cluster classifications, with each cluster's traits explained based on clinical factors. All continuous variables in the prediction model were treated as categorical variables before the LCA analysis. The number of categories was ascertained via the Bayesian information criterion (BIC). The lower BIC was also the elbow point which indicates better model fit. When determining the number of latent classes, clinical interpretations were taken into account. To demonstrate the combination of different risk factors in the prediction model, a chord chart and a characteristic data distribution map of subphenotypes were devised.
Inclusion Criteria:
Exclusion Criteria: