Predictive model for surgical intervention in pediatric acute hematogenous osteomyelitis.

Abstract

The emergence of multidrug-resistant bacteria has resulted in more complicated disease courses and worsening prognoses for patients with acute hematogenous osteomyelitis (AHO), increasing the necessity for surgical intervention. This research attempts to identify the risk variables related to surgical patients and build prediction models. From December 2015 to December 2022, children admitted to a single quaternary care pediatric hospital with AHO had their charts retrospectively reviewed. Based on the therapy methods, the patients were divided into 3 cohorts: multiple surgery, single surgery, and conservative care. Multivariate logistic regression analysis was used to identify independent risk factors related to single and recurrent surgery. A nomogram was created to visually represent the various risk factors, and a calibration curve was plotted to evaluate the model's goodness of fit. The Hosmer-Lemeshow test and the area under the receiver operating characteristic (ROC) curve were used to assess how well the models matched. A total of 218 patients were included in the analysis, out of which 150 patients underwent surgical procedures, with 21 individuals undergoing multiple surgeries. The multivariate binary logistic regression revealed that an increase in absolute neutrophil counts (ANC) (adjusted odds ratio [aOR], 1.14 [95% confidence interval {CI}, 1.05-1.24]) and the presence of Methicillin-resistant Staphylococcus aureus (MRSA) (aOR, 6.97 [95% CI, 1.94-25.06]) were strong predictors of surgical intervention. The prediction model demonstrated an area under the curve (AUC) value of 0.76, while the Hosmer-Lemeshow test showed χ² = 7.3, P = 0.50. In another separated model, the C-reactive protein (CRP) level upon admission (aOR, 1.02 [95% CI, 1.00-1.03]) and the CRP level after the initial surgery (aOR, 1.04 [95% CI, 1.01-1.06]) strongly predict multiple surgeries, with the AUC value of 0.91 obtained and HosmerLemeshow test (χ² = 8.7, P = 0.36) yielded. The calibration curves of the two models were drawn separately, and it was observed that the slopes of both models were close to one. Two prediction models were developed by statistical analysis of clinical data. Their accuracy and discrimination were validated, indicating a promising potential for clinical application.