Abstract
Early detection of pharyngocutaneous fistula (PCF) after total laryngectomy (TL) could prevent severe complications such as major vessel rupture. We aimed to develop prediction models for detecting PCF in the early postoperative period. We retrospectively analyzed patients (N = 263) who received TL between 2004 and 2021. We collected clinical data for fever (>38.0 °C) and blood tests (WBC, CRP, albumin, Hb, neutrophils, lymphocytes) on postoperative days (POD) 3 and 7, and fistulography on POD 7. Clinical data were compared between fistula and no fistula groups, and significant factors were selected using machine learning. Using these clinical factors, we developed improved prediction models for PCF detection. Fistula occurred in 86 (32.7%) patients. Fever was significantly (p < 0.001) more common in the fistula group, and ratios (POD 7 to 3) of WBC, CRP, neutrophils, and neutrophils-to-lymphocytes (NLR) were significantly higher (all p ≤ 0.001) in the fistula group than in the no fistula group. Leakage on fistulography was more common in the fistula group (38.2%) than in the no fistula group (3.0%). The area under curve (AUC) of fistulography alone was 0.68, but predictive models using a combination of fistulography, WBC at POD 7, and neutrophil ratio (POD 7/POD 3) showed better diagnostic performance (AUC of 0.83). Our predictive models may detect PCF early and accurately, which could reduce fatal complications following PCF.