You do realize predicting failure to recover doesn't do a damn bit of good in getting survivors recovered? Or are you that fucking clueless? Survivors want recovery, do the research that gets there!
Long-term forecasting of a motor outcome following rehabilitation in chronic stroke via a hierarchical bayesian dynamic model
Journal of NeuroEngineering and Rehabilitation volume 20, Article number: 83 (2023)
Abstract
Background
Given the heterogeneity of stroke, it is important to determine the best course of motor therapy for each patient, i.e., to personalize rehabilitation based on predictions of long-term outcomes. Here, we propose a hierarchical Bayesian dynamic (i.e., state-space) model (HBDM) to forecast long-term changes in a motor outcome due to rehabilitation in the chronic phase post-stroke.
Methods
The model incorporates the effects of clinician-supervised training, self-training, and forgetting. In addition, to improve forecasting early in rehabilitation, when data are sparse or unavailable, we use the Bayesian hierarchical modeling technique to incorporate prior information from similar patients. We use HBDM to re-analyze the Motor Activity Log (MAL) data of participants with chronic stroke included in two clinical trials: (1) the DOSE trial, in which participants were assigned to a 0, 15, 30, or 60-h dose condition (data of 40 participants analyzed), and (2) the EXCITE trial, in which participants were assigned a 60-h dose, in either an immediate or a delayed condition (95 participants analyzed).
Results
For both datasets, HBDM accounts well for individual dynamics in the MAL during and outside of training: mean RMSE = 0.28 for all 40 DOSE participants (participant-level RMSE 0.26 ± 0.19—95% CI) and mean RMSE = 0.325 for all 95 EXCITE participants (participant-level RMSE 0.32 ± 0.31), which are small compared to the 0-5 range of the MAL. Bayesian leave-one-out cross-validation shows that the model has better predictive accuracy than static regression models and simpler dynamic models that do not account for the effect of supervised training, self-training, or forgetting. We then showcase model’s ability to forecast the MAL of “new” participants up to 8 months ahead. The mean RMSE at 6 months post-training was 1.36 using only the baseline MAL and then decreased to 0.91, 0.79, and 0.69 (respectively) with the MAL following the 1st, 2nd, and 3rd bouts of training. In addition, hierarchical modeling improves prediction for a patient early in training. Finally, we verify that this model, despite its simplicity, can reproduce previous findings of the DOSE trial on the efficiency, efficacy, and retention of motor therapy.
Conclusions
In future work, such forecasting models can be used to simulate different stages of recovery, dosages, and training schedules to optimize rehabilitation for each person.
Trial registration This study contains a re-analysis of data from the DOSE clinical trial ID NCT01749358 and the EXCITE clinical trial ID NCT00057018
Introduction
Recent modeling work has sought to predict the long-term spontaneous recovery of individuals post-stroke from baseline clinical or neural data, e.g., [1,2,3,4]. Whereas such predictions are useful for clinical and research stratification, the neurorehabilitation clinician needs to accurately predict the long-term changes in motor outcomes in response to specific treatments. With the predicted responses, the clinician could then determine the best course of motor therapy for each patient, i.e., personalize rehabilitation [5].
A difficulty is that stroke is heterogeneous and exhibits considerable variability, including in response to motor therapy [6]. It is known that the integrity of the corticospinal tract predicts gains in functional outcomes due to rehabilitation, e.g., [4, 7, 8]. However, multiple other factors are also likely to affect these gains, as well as the retention of these gains following rehabilitation. For instance, we have previously shown that the integrity of visuospatial working memory modulated the effect of blocked, but not distributed, training schedules in chronic stroke [9]. In addition, in re-analyses of the data of the EXCITE [10] and DOSE [11] trials, we have shown that approximately one-fourth of participants continued to see improvements in upper extremity (UE) function following training; conversely, another fourth lost most gains in UE function that resulted from therapy [12, 13].
Given this variability in response to therapy, we need a paradigm shift in predictive modeling in neurorehabilitation that, in addition to clinical and lesion data, incorporates repeated measurements of motor outcomes as soon as they become available during motor therapy. Predictive models that primarily consider such repeated measurements indexed in time order (i.e., time-series) are forecasting models. For example, a recent model can more accurately forecast spontaneous recovery 6 months post-stroke when incorporating repeated measurements than only baseline data [14]. Here, we extend such an approach to forecast the effect of rehabilitation in chronic stroke.
What should be the form of forecasting models in neurorehabilitation? Since neurorehabilitation is based on the premise that sensorimotor activity improves motor recovery via brain plasticity, i.e., “changeability”, the models need to account for the changes in outcomes both during movement therapy, when an increase in performance is expected, and outside of therapy, when both a decrease in performance due to forgetting and an increase in performance are possible. Previously, we proposed a piece-wise linear model of changes in a motor outcome in the DOSE clinical trial, in which the periods of therapy marked the limit between the different linear segments [13]. Although this model well accounted for positive and negative changes both during and following therapy, a model of this type cannot generalize to other datasets because it depends on the timing of training and measurements.
We propose a state-space modeling approach to predict motor outcomes during and following rehabilitation post-stroke. The model has a compact representation and an adjustable time resolution, allowing generalization to different data sets and even to different schedules of therapy for individual patients. The model extends a previous non-linear, first-order state-space model that explained the long-term changes, and the variability in these changes, in arm use following training in the EXCITE trial [12]. This previous model uses a retention term to account for the performance decay often observed post-training, at least in subgroups of patients [12, 13] and a “self-training” term to account for the change in spontaneous use of the paretic limb outside of training when UE function is above a threshold post-therapy [12, 13, 15, 16], which further increases future use and function. In the present model, we further account for the response to therapy via an input term proportional to the dose of motor training, as in our previous piece-wise model [13]. Indeed, animal studies, meta-analyses, and recent clinical trials with large doses, including the DOSE trial, showed that large training doses improve UE function, e.g., [11, 17,18,19].
Previous models in neurorehabilitation typically predict the mean of the future outcome, e.g., [8, 20, 21]. However, such point estimation of a future outcome is insufficient for clinical decision-making in neurorehabilitation because clinicians need to account for the uncertainty of the forecast when assessing different treatment options.Footnote 1 To provide interval estimation, we utilize the Bayesian approach, which extends our previous work [12], as Bayesian models naturally deal with uncertainties by focusing on the probability distributions of all parameters.
A final difficulty for accurate long-term forecasting in neurorehabilitation, however, is that for each new patient, there is initially no or little data on the effect of motor therapy. A hierarchical Bayesian model [22] can, in theory, refine the initial predictions by incorporating prior information from similar patients, via “hyper-parameters.” Crucially, these hyper-parameters can be used as individual prior parameters when predicting the response of a new individual when little outcome data are available, i.e., early in therapy.
Here, we therefore propose and test a novel hierarchical Bayesian dynamic modeling (HBDM) framework that can accurately forecast a clinical measure following rehabilitation in chronic stroke. As a testbed of our model, we use the Motor Activity Log (MAL) data from both the DOSE trial [11] and the EXCITE trial [10]. We test whether a minimal model with three terms, accounting for retention, response to external training, and self-learning, respectively, can better predict the MAL than reduced dynamical models and non-dynamical regression models for these two datasets. Then, using the DOSE data, we simulate the model to forecast the MAL of “new” patients up to 8 months ahead and study the change in the long-term accuracy of the forecasts as additional training data becomes available. We compare the prediction accuracy for models with and without a hierarchical structure for different ranges of forecasting. Finally, we validate the model by testing whether it can account for our previous results on the DOSE dataset on the efficacy, efficiency, and retention of motor training in chronic stroke.
More at link.
No comments:
Post a Comment