Predicting Outcome in a Cohort of Isolated and Combined Dystonia within Probabilistic Brain Mapping

Background

Probabilistic brain mapping is a promising tool to estimate the expected benefit of pallidal deep brain stimulation (GPi‐DBS) in patients with isolated dystonia (IsoD).

Objectives

To investigate the role of probabilistic mapping in combined dystonia (ComD).

Methods

We rendered the pallidal atlas and the volume of tissue activated (VTA) for a cohort of patients with IsoD (n = 20) and ComD (n = 10) that underwent GPi‐DBS. The VTA was correlated with clinical improvement. Afterwards, each VTA was applied on the previously published probabilistic model (Reich et al., 2019). The correlation between predicted and observed clinical benefit was studied in a linear regression model.

Results

A good correlation between observed and predicted outcome was found for both patients with IsoD (n = 14) and ComD (n = 7) (r² = 0.32; P < 0.05). In ComD, 42% of the variance in DBS response is explained by VTA‐based outcome map.

Conclusion

A probabilistic model would be helpful in clinical practice to circumvent unpredictable and less impressive motor results often found in ComD.

Probabilistic mapping of the antidystonic effect of pallidal neurostimulation: a multicentre imaging study

Deep brain stimulation of the internal globus pallidus is a highly effective and established therapy for primary generalized and cervical dystonia, but therapeutic success is compromised by a non-responder rate of up to 25%, even in carefully-selected groups. Variability in electrode placement and inappropriate stimulation settings may account for a large proportion of this outcome variability. Here, we present probabilistic mapping data on a large cohort of patients collected from several European centres to resolve the optimal stimulation volume within the pallidal region. A total of 105 dystonia patients with pallidal deep brain stimulation were enrolled and 87 datasets (43 with cervical dystonia and 44 with generalized dystonia) were included into the subsequent 'normative brain' analysis. The average improvement of dystonia motor score was 50.5 ± 30.9% in cervical and 58.2 ± 48.8% in generalized dystonia, while 19.5% of patients did not respond to treatment (<25% benefit). We defined probabilistic maps of anti-dystonic effects by aggregating individual electrode locations and volumes of tissue activated (VTA) in normative atlas space and ranking voxel-wise for outcome distribution. We found a significant relation between motor outcome and the stimulation volume, but not the electrode location per se. The highest probability of stimulation induced motor benefit was found in a small volume covering the ventroposterior globus pallidus internus and adjacent subpallidal white matter. We then used the aggregated VTA-based outcome maps to rate patient individual VTAs and trained a linear regression model to predict individual outcomes. The prediction model showed robustness between the predicted and observed clinical improvement, with an r2 of 0.294 (P < 0.0001). The predictions deviated on average by 16.9 ± 11.6 % from observed dystonia improvements. For example, if a patient improved by 65%, the model would predict an improvement between 49% and 81%. Results were validated in an independent cohort of 10 dystonia patients, where prediction and observed benefit had a correlation of r2 = 0.52 (P = 0.02) and a mean prediction error of 10.3% (±8.9). These results emphasize the potential of probabilistic outcome brain mapping in refining the optimal therapeutic volume for pallidal neurostimulation and advancing computer-assisted planning and programming of deep brain stimulation.