Neuropsychiatric symptoms in Parkinson’s disease can reverse within a levodopa cycle, yet conventional psychometric summaries do not show whether the same questionnaire items retain a coherent directional organization across medication states. To determine whether the 20-item Neuropsychiatric Fluctuation Scale (NFS) forms a reproducible bipolar state axis and to identify items responsible for condition-specific instability. The analytic record contained a \(20\times3\) matrix of principal-component loadings for Parkinson’s disease OFF medication, Parkinson’s disease ON medication, and healthy-control conditions, together with paired score summaries from 101 patients. A polarity-aware condition-geometry procedure quantified cosine similarity, angular separation, Pearson and Spearman association, exact sign concordance, item-bootstrap intervals, permutation probabilities, leave-one-item-out influence, loading-threshold sensitivity, and paired standardized effects. One hundred thousand item-bootstrap samples and 200,000 label permutations were generated from a fixed seed. Every item reversed sign from OFF to ON, producing a cosine similarity of \(-0.923\) (95% item-bootstrap interval \(-0.980\) to \(-0.843\)) and an angle of \(157.4^{\circ}\). ON and control loadings were closely aligned (cosine \(0.960\), 95% interval \(0.932\) to \(0.982\); angle \(16.2^{\circ}\)), with identical signs for all 20 items. Items 5, 8, 13, and 19 were weak in one condition, but their joint removal strengthened all three pairwise similarities. The mean global NFS score moved from \(-15.0\) in OFF to \(16.8\) in ON; the paired standardized transition was \(d_z=2.09\) (95% confidence interval 1.74–2.44). The plus-score increase and minus-score decrease contributed nearly equally to the global displacement. The NFS loading structure is best characterized as a stable bipolar axis whose orientation reverses with medication state. Its state discrimination is distributed across the item set rather than dependent on a small subset, although four items merit targeted investigation in future patient-level studies.
Neuropsychiatric symptoms in Parkinson’s disease can reverse within a levodopa cycle, yet conventional psychometric summaries do not show whether the same questionnaire items retain a coherent directional organization across medication states. To determine whether the 20-item Neuropsychiatric Fluctuation Scale (NFS) forms a reproducible bipolar state axis and to identify items responsible for condition-specific instability. The analytic record contained a \(20\times3\) matrix of principal-component loadings for Parkinson’s disease OFF medication, Parkinson’s disease ON medication, and healthy-control conditions, together with paired score summaries from 101 patients. A polarity-aware condition-geometry procedure quantified cosine similarity, angular separation, Pearson and Spearman association, exact sign concordance, item-bootstrap intervals, permutation probabilities, leave-one-item-out influence, loading-threshold sensitivity, and paired standardized effects. One hundred thousand item-bootstrap samples and 200,000 label permutations were generated from a fixed seed. Every item reversed sign from OFF to ON, producing a cosine similarity of \(-0.923\) (95% item-bootstrap interval \(-0.980\) to \(-0.843\)) and an angle of \(157.4^{\circ}\). ON and control loadings were closely aligned (cosine \(0.960\), 95% interval \(0.932\) to \(0.982\); angle \(16.2^{\circ}\)), with identical signs for all 20 items. Items 5, 8, 13, and 19 were weak in one condition, but their joint removal strengthened all three pairwise similarities. The mean global NFS score moved from \(-15.0\) in OFF to \(16.8\) in ON; the paired standardized transition was \(d_z=2.09\) (95% confidence interval 1.74–2.44). The plus-score increase and minus-score decrease contributed nearly equally to the global displacement. The NFS loading structure is best characterized as a stable bipolar axis whose orientation reverses with medication state. Its state discrimination is distributed across the item set rather than dependent on a small subset, although four items merit targeted investigation in future patient-level studies.
