Age, emotional burden and deep brain stimulation electrode location shape Parkinson’s disease quality of life

npj | digital medicine Article Published in partnership with Seoul National University Bundang Hospital https://doi.org/10.1038/s41746-026-02828-7 Age, emotional burden and deep brain stimulation electrode location shape Parkinson’s disease quality of life Check for updates 1 1 1,2 2 1,3,4,5,6,7 1234567890():,; 1234567890():,; Shiva Khoshnoud , Farzin Negahbani , Idil Cebi , Daniel Weiss & Alireza Gharabaghi Postoperative quality-of-life (QoL) outcomes after subthalamic deep brain stimulation in Parkinson’s disease vary widely. Previous studies based on PDQ-39 summary scores have reported opposing relationships between baseline and postoperative QoL, reflecting analytic variability, measurement noise, and limited feature scope. To address these inconsistencies, we analyzed 130 patients using an explainable random-forest classifier with SHAP analysis trained to predict QoL changes exceeding minimal clinically important difference thresholds. Baseline variables included PDQ-39 subscores, along with demographic, motor, cognitive, and affective measures, and electrode coordinates derived from imaging. Predictors of QoL improvement included younger age, greater preoperative diseaserelated emotional burden, and electrode placement at the motor-associative transition in the right subthalamic nucleus. The model achieved an area under the curve of 0.70 on the held-out test set, with balanced sensitivity and specificity. Identifying interpretable cut-offs for age, emotional burden and electrode location supports individualized counseling and treatment planning, advancing outcome prediction in neuromodulation. Enhancing health-related quality of life (QoL) is a central goal in the management of Parkinson’s disease (PD), where QoL is often markedly reduced1. Unlike clinician-rated motor evaluations, QoL metrics such as the PDQ-39 capture the broader physical, emotional, and social burden of PD; however, identifying consistent predictors of QoL change has proven difficult, because QoL reflects the complex interplay of motor and non-motor symptoms, psychological state, and disease progression2. Deep brain stimulation (DBS) of the subthalamic nucleus (STN) effectively alleviates motor symptoms in advanced PD, yet its impact on QoL remains variable3. Although many patients experience meaningful improvement, others show minimal benefit or even decline4–8. Studies investigating predictors of QoL outcomes have reported conflicting results9. Some have linked greater preoperative QoL burden to stronger postoperative gains6,10–12, whereas others found the opposite relationship13,14. A closer review of prior work indicates that these apparent contradictions may arise at least in part from analytic approaches. Frameworks for QoL outcome modeling after DBS generally fall into two categories: changebased and end-state models, each with distinct methodological and clinical implications. Change-based models4,6,7,11,15–17 quantify improvement relative to baseline and align with the clinical goal of predicting postoperative benefit. However, they are statistically prone to mathematical coupling and regression to the mean, because baseline values are embedded within both predictor and outcome terms18,19. Consequently, such models often suggest that patients with poorer baseline QoL show greater postoperative improvement, a pattern that may reflect analytic dependency rather than genuine differential treatment response. End-state models5,13,14 circumvent this coupling by using postoperative QoL as the dependent variable, yet they make it difficult to disentangle baseline functioning from the treatment effect, since postoperative QoL remains strongly influenced by preoperative status. Consequently, these models tend to favor patients who already perform well before surgery and provide limited insight into which individuals are likely to experience substantial DBS-related benefit, thereby reducing their utility for preoperative decision support. The present study adopts an alternative framework that combines the conceptual strengths of both approaches: (1) by defining change as externally anchored and categorical rather than continuous, and (2) by incorporating multimodal predictors to distribute explanatory variance across independent domains. 1 Institute for Neuromodulation and Neurotechnology, University Hospital Tübingen, Faculty of Medicine, University Tübingen, Tübingen, Germany. 2Center for Neurology, Department for Neurodegenerative Diseases, and Hertie Institute for Clinical Brain Research, University Tübingen, Tübingen, Germany. 3Center for Digital Health (CDH), Tübingen, Germany. 4Cognitive Science Center (CSC), Tübingen, Germany. 5Center for Bionic Intelligence Tübingen Stuttgart (BITS), Tübingen, Germany. 6German Center for Mental Health (DZPG), Tübingen, Germany. 7Max Planck-University of Toronto Centre for Neural Science & Technology e-mail: (MPUTC), Toronto, ON, Canada. npj Digital Medicine | (2026)9:434 1 Article https://doi.org/10.1038/s41746-026-02828-7 Fig. 1 | Framework for analysis and model development. Schematic of the analytical workflow, including feature selection, data preprocessing, model training, validation, and interpretation. (1) Applying minimal clinically important difference (MCID) thresholds, i.e., externally validated criteria independent of baseline scores20–23, reduces coupling artifacts and aligns the model with clinically meaningful, patient-centered outcomes. Although MCID classification is applied at the individual level, it derives from group-based validation studies, constituting an externally anchored endpoint rather than a direct mathematical transformation of baseline and postoperative scores19. This approach minimizes, though does not entirely eliminate, the coupling effect inherent in continuous change-score models while focusing on change that is clinically perceptible19,22. (2) Incorporating a multimodal feature set spanning demographic, motor, cognitive, and affective domains24,25 further mitigates analytic dependency and unmeasured confounding arising from incomplete representation of interacting factors26. This design distributes explanatory variance across independent dimensions27, preventing baseline QoL from dominating the model28. Moreover, by analyzing preoperative PDQ-39 subscores rather than the global summary index, the analysis reduces aggregation bias and partial circularity with the postoperative PDQ-39 total. When the same composite score is used at both time points, shared measurement structure can inflate associations between pre- and postoperative values. Using domain-specific subscores preoperatively while retaining the total postoperative score limits this dependency, preserves meaningful variance across QoL dimensions, and allows each baseline domain to contribute independently to outcome prediction14,19. This modified change-based design aims to provide a biologically grounded and clinically interpretable framework for pre (...truncated)