Radiographic parameter-driven decision tree reliably predicts aseptic mechanical failure of compressive osseointegration fixation

Background and purpose — Compressive osseointegration fixation is an alternative to intramedullary fixation for endoprosthetic reconstruction. Mechanical failure of compressive osseointegration presents differently on radiographs than stemmed implants, therefore we aimed to develop a reliable radiographic method to determine stable integration. Patients and methods — 8 reviewers evaluated 11 radiographic parameters from 29 patients twice, 2 months apart. Interclass correlation coefficients (ICCs) were used to assess test–retest and inter-rater reliability. We constructed a fast and frugal decision tree using radiographic parameters with substantial test–retest agreement, and then tested using radiographs from a new cohort of 49 patients. The model’s predictions were compared with clinical outcomes and a confusion matrix was generated. Results — 6 of 8 reviewers had non-significant intra-rater ICCs for？≥？one parameter; all inter-rater ICCs were highly reliable (p < 0.001). Change in length between the top of the spindle sleeve and bottom of the anchor plug (ICC 0.98), bone cortex hypertrophy (ICC 0.86), and bone pin hypertrophy (ICC 0.81) were used to create the decision tree. The sensitivity and specificity of the training cohort were 100% (95% CI 52–100) and 87% (CI 74–94) respectively. The decision tree demonstrated 100% (CI 40–100) sensitivity and 89% (CI 75–96) specificity with the test cohort. Interpretation — A stable spindle length and at least 3 cortices with bone hypertrophy at the implant interface predicts stable osseointegration; failure is predicted in the absence of bone hypertrophy at the implant interface if the pin sites show hypertrophy. Thus, our decision tree can guide clinicians as they follow patients with compressive osseointegration implants