Factors associated with complication of cranioplasty: CT-Based Risk assessment for Early Failure of Autologous-Bone Cranioplasty

Purpose To determine whether preoperative noncontrast CT features predict early revision after autologous bone cranioplasty and to develop a simple CT-based risk framework. Methods We retrospectively studied adults undergoing autologous cranioplasty at a single center (2010–2020). Preoperative CT variables and clinical indicators were evaluated as candidate predictors in the regression and machine-learning models. The primary outcome was unplanned revision surgery within 30 days. Associations were tested with multivariable logistic regression. A feature-importance derived from the random forest model was used to assess the relative contributions of risk factors. Results Among 376 patients (mean age 48.9 years; 71.0% male), 52 (13.8%) underwent revision surgery within postoperative 30 days. Encephalomalacia (EM) area was the dominant predictor (adjusted OR 4.311). A prespecified cut-off of ≥ 620 mm² yielded sensitivity 90.38% and specificity 59.88% (OR 14.03). Sinking skin flap (OR 6.17), hydrocephalus (OR 3.35), and EM involving ≥ 2 lobes (OR 4.37) were also significant. All patients (10/10) who underwent concurrent VP shunting during cranioplasty required revision surgery, resulting in a 100% failure rate for this group. A logistic regression achieved AUC 0.83 and accuracy 88.0% (specificity 97%, sensitivity 22%); random forest achieved AUC 0.85 and accuracy 88.2% (specificity 96%, sensitivity 33%). EM area ranked highest in feature importance. Conclusions Preoperative CT provides independent prognostic information for early revision after autologous cranioplasty. A simple thresholded EM-area metric can serve as a high-sensitivity screening tool for predicting early failure after cranioplasty. When EM burden is substantial, the reconstructive strategy may warrant modification.