Prediction of Nonunion After Nonoperative Treatment of a Proximal Humeral Fracture

Background: The prevalence of nonunion after a proximal humeral fracture (PHF) and the risk factors for its occurrence are poorly defined. We aimed to estimate the rate of nonunion in nonoperatively treated patients and to produce a clinical model for its prediction. Methods: Two thousand two hundred and thirty adult patients (median age, 72 years [range, 18 to 103 years]; 75.5% were female) with a PHF underwent assessment of fracture union using standard clinical evaluation and conventional radiographs. We assessed the prevalence of nonunion and measured the effect of 19 parameters on healing. Best statistical practices were used to construct a multivariate logistic regression model. The PHF assessment of risk of nonunion model (PHARON) was externally validated in a subsequent prospectively collected population of 735 patients, treated by the same protocol in our institution. Results: Overall, 231 (10.4%) of 2,230 patients developed nonunion. Only 3 (0.8%) of 395 patients with a head-shaft angle (HSA) of >140° developed nonunion; in this cohort, none of the measured candidate variables were independently predictive of nonunion on multivariate logistic regression analysis. In the larger cohort of 1,835 patients with an HSA of ≤140°, 228 (12.4%) developed nonunion. Decreasing HSA, increasing head-shaft translation (HST), and smoking were independently predictive of nonunion on multivariate analysis. The prevalence of nonunion was very low (1%) in the majority with both an HSA of >90° and HST of <50%, whereas the risk was much higher (83.7%) in the 8.3% with an HSA of ≤90° and HST of ≥50%. In both groups, the prevalence of nonunion was much higher in smokers. Conclusions: The prevalence of nonunion after PHF is higher than previously reported. Most patients have favorable risk-factor estimates and a very low risk of this complication, but a smaller subgroup is at much higher risk. The risk can be accurately estimated with PHARON, using standard clinical assessment tools. Level of Evidence: Prognostic Level III . See Instructions for Authors for a complete description of levels of evidence.