Computer navigation has been shown to improve the accuracy of total knee replacement (TKR) when compared to intra or extra osseous referencing. Currently the surgical transepicondylar axis (TEA) is used to help determine femoral component rotation. This relies on the surgeon identifying medial and lateral epicondyles intra-operatively. This process has been shown to have a high variability and operator dependency. The functional flexion axis (FFA) of the femur is a kinematically derived reference axis which has previously been shown in a cadaveric model to correspond well with the transepicondylar axis. This study was therefore designed to evaluate its accuracy in vivo. 50 patients undergoing total knee replacement under the care of the three senior authors were prospectively recruited. A preoperative CT scan was obtained and the TEA evaluated by 2 independent clinicians. TKR was undertaken in the standard fashion using Stryker navigation. The FFA was derived at 3 time points during the procedure: pre-incision, post osseous registration and following component implantation. The deviations of the FFA and surgical TEA (surTEA) to the CT-derived TEA (ctTEA) was calculated and comparisons drawn between the 2 methods with respect to validity, as well as within and between-patient reproducibility. While the FFA results were highly correlated between pre and post-arthrotomy (r = 0.89), the post-incision FFA (−1.60+/−3.7) was significantly internally rotated (p<0.01) relative to the pre-incision FFA (−2.50+/−3.4). In addition the surgical TEA (−0.40+/−3.6) was significantly internally rotated (p = 0.02) relative to the post-incision FFA (1.80+/−3.7) for the combined data from all 2 surgeons. However, when examined individually, 1 of the 2 surgeons showed no significant difference between the FFA and TEA. In addition, the two methods demonstrated comparable between-patient variability in the knee axis, although surgeon-dependent patterns remained. The FFA has been shown to be of equivalent accuracy to the surgical TEA but surprisingly does not avoid its operator-dependency. Further evaluation of the FFA method with possible adjustments to the algorithm is warranted.
