Deep brain stimulation (DBS) is an effective treatment for motor symptoms of Parkinson's disease, but the magnitude of benefit varies between patients. A reliable tool that uses data available before surgery to forecast motor outcomes would support patient selection, expectation management and clinical decision-making. The review examines efforts to build predictive models that forecast post-DBS motor results using preoperative variables drawn from multiple domains.
The authors performed a systematic search of Web of Science, PubMed and Scopus to identify primary research articles that tested predictions from models based on pre-surgical data and focused specifically on motor outcomes after DBS for Parkinson's disease. From this search they identified 19 studies that matched these inclusion criteria. The review synthesises the design features, input types, outcome measures and validation strategies reported in those studies.
Across the 19 studies the candidate predictor types included:
The review notes that models using only clinical data tend to be better powered and more generalisable, whereas multimodal models incorporating imaging or other advanced measures sometimes achieve higher reported accuracy but are typically under-powered.
The body of work reviewed shows trade-offs between sample size, data modality and model performance. Studies with larger samples and broader applicability primarily used clinical data and produced limited accuracy, suggesting modest predictive utility from routinely collected clinical variables alone. In contrast, studies that included richer inputs such as MRI-derived features occasionally reported higher prediction accuracy, but these studies were often small and under-powered, limiting generalisability.
Other methodological limitations across the literature include a focus on a single DBS target, limited postoperative follow-up windows, and variable validation approaches. The authors emphasise that many studies restricted predictions to the first year after surgery and most targeted the subthalamic nucleus (STN), reducing the applicability of findings to other targets or longer-term outcomes.
Predictions in the reviewed studies were mostly constrained to the first postoperative year and typically assessed motor outcome using sum scores of motor performance rather than predicting individual motor domains. This practice limits interpretability for specific functions (for example, gait, bradykinesia or tremor) and for longer-term trajectories. The authors recommend developing models that predict domain-specific scores over a range of postoperative time points and that provide uncertainty estimates such as confidence intervals.
The review outlines several key recommendations intended to advance predictive modelling toward clinical readiness:
The authors present these recommendations as a framework for producing predictive models that can be reliably translated to clinical practice.
At present, predictive models that are broadly generalisable and clinically useful for preoperative counselling in DBS are lacking. Existing studies either lack sufficient power or focus narrowly on specific modalities or short-term, aggregated motor outcomes. To move toward clinically deployable tools, investigators should prioritise larger, multi-centre datasets, harmonised outcome definitions across motor domains, robust validation methods and transparent sharing of models and data. Such steps are necessary to create predictions that clinicians can reasonably apply across different clinics and patient populations.
The review concludes that while the concept of pre-surgical prediction of motor benefit from deep brain stimulation is promising, substantive methodological improvements and collaborative data-sharing are required before these models can be recommended for routine clinical use. The recommendations in this systematic review provide a roadmap for the field to develop predictive models that could meaningfully inform DBS candidate selection and postoperative expectations.