Treatment outcomes among children with cerebral palsy are mediocre, unpredictable, and stagnant over several decades. We use a nearest-neighbors matching algorithm to predict outcomes from individuals. The algorithm allows clinician input regarding the relevant matching parameters, treatment of choice, and outcome of interest. The algorithm was tested on 1092 limbs that underwent single-event multi-level surgery. Predictions compared favorably to previous regression-based approaches, producing smaller root mean squared errors across the spectrum of kinematics.