Analiza faktorjev tveganja in napovedovanje pojava globoke sternalne okužbe po posegu kirurške revaskularizacije srca

Kamenšek, Tina (author), Žibert, Janez (mentor), Kališnik, Jurij Matija (comentor)

Globoka sternalna okužba (angl. deep sternal wound infection ali DSWI), ki se lahko pojavi po posegu na odprtem srcu, povzroča večjo obolevnost in dvakrat višjo stopnjo smrtnosti, po zadnjih podatkih kar 15 %. Globoka sternalna okužba se pojavi pri 1–4% bolnikov po kirurški revaskularizaciji srca ter je povezana z znižanim preživetjem, podaljšano hospitalizacijo, višjo stopnjo reoperacij in povečano porabo finančnih in materialnih virov. Obstaja že nekaj točkovalnih sistemov s ciljem prepoznavanja visoko ogroženih bolnikov za pojav globoke sternalne okužbe, ki pa so različno sestavljeni in kompleksni za uporabo. Namen raziskave je ugotoviti pomembnost posameznih faktorjev tveganja za končno postoperativno oceno ogroženosti bolnikov za pojav globokega sternalnega infekta in s tem izboljšanje zgodnje detekcije rizičnih bolnikov, med drugimi tudi z uporabo metod umetne inteligence. Kohorto naše študije je sestavljalo 5221 bolnikov, po posegu kirurške revaskularizacije srca z uporabo zunajtelesnega obtoka. Iz povzetka referenčnih točkovalnih sistemov in novih predlogov dejavnikov tveganja smo izdelali tri različne modele faktorjev tveganja, ki smo jih validirali in med seboj primerjali z logistično regresijo in ROC (angl. receiver operating characteristic) analizo. Kasneje so bili uporabljeni še novejši pristopi umetne inteligence, extreme gradient boosting (XGB) in globoke nevronske mreže (angl. deep neural network - DNN), ki so pripomogli še k boljši napovedi v primerjavi s pogosteje uporabljeno metodo logistične regresije. Predstavljena raziskava je validirala obstoječe faktorje tveganja in odkrila nekaj novih. Končni model logistične regresije je sestavljalo 12 neodvisnih faktorjev tveganja, ki so obsegali tudi nabor kratkoročnih pooperativnih zapletov, in so občutno izboljšali napoved tveganja pojava globokega sternalnega infekta. Tudi metode umetne inteligence so v primerjavi s tradicionalnimi metodami to napoved izboljšale. To je dobra osnova za nadaljnji razvoj aplikacij, ki bodo omogočile lažjo implementacijo tovrstnih napovednih modelov v klinično uporabo.

• Language: Slovenian
• Keywords: kirurška revaskularizacija srca, globok sternalni infekt, strojno učenje, napovedni modeli
• Work type: Doctoral dissertation
• Organization: MF - Faculty of Medicine
• Year: 2025
• PID: 20.500.12556/RUL-167567
• Publication date in RUL: 28.02.2025

Secondary language

• Language: English
• Title: Analysis of risk factors and prediction of deep sternal wound infections after surgical revascularization of the heart

Abstract

Deep sternal wound infection (DSWI) which can occure after open heart surgery is associated with a significant increase in morbidity and doubled mortality by up to 15% according to recent data. DSWI, occurring in 1–4% of patients after surgical revascularization of the heart (CABG), is additionally associated with decreased survival, prolonged hospitalization, higher reoperation rates and resource utilization. Several scoring systems have been developed to identify high-risk patients for DSWI with varying complexity and performance characteristics. The objective of this study is to ascertain the significance of each contributing factor in the overall risk of DSWI following surgery and to enhance the early identification of patients at elevated risk for DSWI after CABG, utilizing artificial intelligence (AI) based methodologies. The cohort of the study comprised 5221 patients after CABG with cardiopulmonary bypass. By employing reference scoring systems and novel risk factor proposals, three distinct models/sets of risk factors were constructed, which were subsequently validated and compared with one another through logistic regression and receiver operating characteristic (ROC) analysis. Subsequently, we employed more contemporary AI methodologies, specifically extreme gradient boosting and deep neural network, to ascertain whether these could enhance the forecast further in comparison to the conventional logistic regression approach. The present study validated the current risk factors and identified some new ones. The final logistic regression model consisted of 12 independent risk factors, complemented with short-term postoperative complications, which significantly improved DSWI risk prediction. AI approaches outperformed traditional statistics-based methods in DSWI prediction. This is a good basis for the further development of applications that would enable the implementation of such models in clinical use.

• Keywords: coronary artery bypass grafting, deep sternal wound infection, machine learning, prediction models