# A feasibility open-label clinical trial utilizing second-generation artificial intelligence based on the constrained-disorder principle in patients with Parkinson's disease

## TL;DR

A clinical trial tested AI-driven Levodopa dosing for Parkinson's disease, showing some improvement in symptoms and high app adherence.

## Contribution

This study introduces a novel AI approach based on the constrained-disorder principle to personalize Levodopa treatment in Parkinson's disease.

## Key findings

- 80% of patients showed UPDRS improvement, nearing clinically significant thresholds.
- 60% of patients reported subjective improvement on the PGI-I scale.
- 80% of patients used the AI app daily, indicating strong adherence.

## Abstract

Parkinson's disease (PD) is a neurodegenerative disorder treated with Levodopa, but long-term use often causes motor complications like "wearing-off," "on-off" effects, and Levodopa-induced dyskinesias, requiring careful management to balance benefits and risks. The Constrained Disorder Principle (CDP) defines biological systems by their inherent variability. CDP-based second-generation artificial intelligence (AI) systems introduce controlled variability into treatment regimens to counteract compensatory mechanisms that underlie drug loss of effectiveness.

This open-label, proof-of-concept feasibility clinical trial aimed to assess the technical feasibility and preliminary evidence of improved response to Levodopa by implementing algorithm-controlled therapeutic regimens.

In this 14-week, open-label, single-center study, five PD patients used an app that randomized their Levodopa dosing times and dosages within pre-defined ranges. Primary outcomes were changes in the Unified Parkinson's Disease Rating Scale (UPDRS) and the Patient Global Impression of Improvement (PGI-I) scale. Statistical analysis was performed using the Wilcoxon signed-rank test with effect size calculations.

80% of patients demonstrated clinical improvement on the UPDRS, with a mean improvement of 4.4 points (p = 0.063, Cohen's d=0.82, 95% CI: −0.3–9.1), approaching but not exceeding the established minimal clinically significant difference threshold. Additionally, 60% reported a subjective improvement on the PGI-I scale. Furthermore, 80% of patients used the app daily, indicating high adherence.

The results of this feasibility trial provide preliminary, hypothesis-generating evidence that CDP-based second-generation AI-driven personalized Levodopa dosing regimens may be technically feasible and potentially associated with clinical improvements in PD patients. However, the open-label design, small sample size, and absence of control conditions necessitate cautious interpretation. Adequately powered, randomized, double-masked controlled trials are needed to confirm these findings and rigorously evaluate efficacy and long-term effects.