Healthy gait requires complex coordination between the neuromuscular and skeletal systems. In patients with movement disorders such as cerebral palsy, these coordination patterns may be disrupted and their gait patterns are inefficient and unstable. Understanding the interactions between the neuromuscular and skeletal systems is necessary for designing treatments to improve gait for individuals with impaired walking. This dissertation describes two classes of tools—physics-based dynamic simulation and statistical modeling—for understanding and improving human gait. First, we developed and validated a computer model of the full body that can be used to generate dynamic simulations of gait. Second, we developed and tested statistical models that used features extracted from clinical gait analysis data to predict surgical outcomes in children with cerebral palsy.

Physics-based musculoskeletal models provide a means to study human movement and predict the effects of interventions on gait. In this thesis, we developed and shared a three-dimensional musculoskeletal model with high-fidelity representations of the lower limb musculoskeletal geometry and force-generation capacity. This model can be used to compute muscle-tendon lengths and velocities in gait, characterize force-generating capacity over joint ranges of motion, and examine muscle-tendon force and length dynamics. We validated our model in context of its intended use—to quickly generate accurate dynamic simulations of movement—and freely shared the model on simtk.org with sample simulations for the community to use and extend. At the time of this thesis submission, our model has been downloaded by over 1,500 unique users worldwide since its release in July 2015.

Statistical models complement physics-based models in investigating relationships between observed features of movement and clinical outcomes following orthopedic treatment. We built a set of statistical models to predict the efficacy of orthopedic surgery in improving gait for cerebral palsy patients. While many children with cerebral palsy undergo surgeries to relieve abnormal muscle tone and correct bony malalignments, the rate of positive surgical outcomes is frustratingly low, and the marginal gains of surgery with respect to expected natural progression of the child is unknown. To address this, we developed multivariate regression models to predict by how much a patient’s gait would improve with surgery when controlling for the expected natural progression of gait without surgery. We analyzed data describing the motion of 2,333 limbs (of which about two-thirds had undergone orthopedic surgery), and found that only 37% of these limbs were expected to show a clinically meaningful improvement following surgery. We found good neural control, as assessed in the patient’s physical exam, was important for positive outcomes, along with other modifiable biomechanical features such as walking speed and muscle strength.

Among limbs that underwent orthopedic surgery, we analyzed the marginal effectiveness of a gastrocnemius lengthening surgery to correct equinus, a common gait pattern characterized by walking on the toes that is observed in many cerebral palsy patients. We used our physics-based musculoskeletal model to compute gastrocnemius lengths as a function of clinically measured ankle and knee flexion kinematics in gait for each limb. We showed that limbs with “short” gastrocnemius in gait are twice as likely to benefit from a gastrocnemius lengthening surgery as are the limbs whose gastrocnemii are “not short,” and are more likely to maintain the benefits from this surgery over time. We created and shared easy-to-use clinical worksheets that can be used as part of the pre-surgical evaluation when formulating a treatment plan for a patient.

This dissertation developed and shared open-source simulation and statistical models to study human gait. We demonstrated how computer simulations of gait can be used to study muscle coordination in walking and running, as well as provide clinically-relevant metrics that can be used to make treatment recommendations. Drawing on our understanding of muscle function in gait, we developed clinically-translatable statistical models to predict likelihood of good outcomes following orthopedic surgery for children with cerebral palsy. This work demonstrates the utility of combining biomechanical modeling and statistical learning to improve current state of clinical care for patients with gait pathology. Our musculoskeletal and statistical models are shared openly with the community to allow others to adopt and build upon our work.