A Causal Framework for Precision Rehabilitation

TL;DR

This framework aims to identify the Optimal Dynamic Treatment Regimens (ODTR), or the decision-making strategy that takes in the range of available measurements and biomarkers to identify interventions likely to maximize long-term function.

Abstract

Precision rehabilitation offers the promise of an evidence-based approach for optimizing individual rehabilitation to improve long-term functional outcomes. Emerging techniques, including those driven by artificial intelligence, are rapidly expanding our ability to quantify the different domains of function during rehabilitation, other encounters with healthcare, and in the community. While this seems poised to usher rehabilitation into the era of big data and should be a powerful driver of precision rehabilitation, our field lacks a coherent framework to utilize these data and deliver on this promise. We propose a framework that builds upon multiple existing pillars to fill this gap. Our framework aims to identify the Optimal Dynamic Treatment Regimens (ODTR), or the decision-making strategy that takes in the range of available measurements and biomarkers to identify interventions likely to maximize long-term function. This is achieved by designing and fitting causal models, which extend the Computational Neurorehabilitation framework using tools from causal inference. These causal models can learn from heterogeneous data from different silos, which must include detailed documentation of interventions, such as using the Rehabilitation Treatment Specification System. The models then serve as digital twins of patient recovery trajectories, which can be used to learn the ODTR. Our causal modeling framework also emphasizes quantitatively linking changes across levels of the functioning to ensure that interventions can be precisely selected based on careful measurement of impairments while also being selected to maximize outcomes that are meaningful to patients and stakeholders. We believe this approach can provide a unifying framework to leverage growing big rehabilitation data and AI-powered measurements to produce precision rehabilitation treatments that can improve clinical outcomes.

Figures (4)

• Figure 3: High-level diagram of how the Optimal Dynamic Treatment Regimen (ODTR) would be applied longitudinally (iteratively) to determine the optimal intervention at any stage during rehabilitation. At each point in time, the phenotype of a patient would be established based on the set of measurements and biomarkers acquired to date. This phenotype would be passed to the decision policy, predicting the next intervention(s) most likely to maximize long-term function. This would be applied, and the process would be repeated. Critically, the ODTR is designed not only to make some short-term change in impairment but also to maximize function and participation over the patient's lifespan, aligned with the (potentially changing) goals of the patient.
• Figure 4: High-level diagram describing the process that generates rehabilitation data. At any given time, the patient has an underlying condition, reflecting the state of their body and support system. This gives rise to a range of measurements we can make, including biomarkers and clinical outcome assessments (COAs) that can be measured to gain insight into this, which are contextualized with the ICF. Then, interventions are selected based on the measurements and decision strategies, ranging from standard practice to clinical trials or even the ODTR. Ideally, these are precisely documented per the RTSS. As a result of these interventions, the patient has an updated health condition, and the process repeats.
• Figure 5: Causal diagram for EMG biofeedback program. A) shows a more minimal model that includes some of the targets of this therapy, including learned disuse and spinal sprouting. It captures how EMG is measured and provided as biofeedback to induce muscle activation that can alter these targets. It also captures how these latent variables influence other measurements, such as arm function and real-world independence. B) shows a more comprehensive model that includes additional measurements, such as kinematics or measures that reflect spinal cord injury lesions. C) includes images of some of these measurements, including the EMG measured during biofeedback or arm kinematics during upper extremity assessments and wheelchair propulsion.
• Figure 6: Example of causal framework applied to post-stroke gait impairments. A) results from sullivan_model_2011 with the causal interactions modeled more abstractly at the level of the ICF. Hashed elements were those excluded from the final model. B) is a more neuroanatomically motivated model that captures the multiple pathways and feedback loops responsible for gait impairments. This model also indicates how interventions like high-intensity gait training (HIGT), functional electrical stimulation (FES), and ankle-foot orthoses (AFOs) can be incorporated. C) shows an example of how computer vision can extract a great deal of information about changes in walking during inpatient rehabilitation for someone with a stroke, with our latest methods even inferring ground reaction forces and net joint torques. This shows increases in the independent modulation of the paretic limb joint torques.