Treatment design for musculoskeletal disorders using in silico patient-specific dynamic simulations is becoming a clinical possibility. However, these simulations are sensitive to model paramet
Treatment design for musculoskeletal disorders using in silico patient-specific dynamic simulations is becoming a clinical possibility. However, these simulations are sensitive to model parameter values that are difficult to measure experimentally, and the influence of uncertainties in these parameter values on the accuracy of estimated knee contact forces remains unknown. This study evaluates which musculoskeletal model parameters have the greatest influence on estimating accurate knee contact forces during walking. We performed the evaluation using a two-level optimization algorithm where musculoskeletal model parameter values were adjusted in the outer level and muscle activations were estimated in the inner level. We tested the algorithm with different sets of design variables (combinations of optimal muscle fiber lengths, tendon slack lengths, and muscle moment arm offsets) resulting in nine different optimization problems. The most accurate lateral knee contact force predictions were obtained when tendon slack lengths and moment arm offsets were adjusted simultaneously, and the most accurate medial knee contact force estimations were obtained when all three types of parameters were adjusted together. Inclusion of moment arm offsets as design variables was more important than including either tendon slack lengths or optimal muscle fiber lengths alone to obtain accurate medial and lateral knee contact force predictions. These results provide guidance on which musculoskeletal model parameter values should be calibrated when seeking to predict in vivo knee contact forces accurately.