Apathy and Anxiety in De Novo Parkinson's Disease Predict the Severity of Motor Complications

Abstract

Background

Neuropsychiatric and affective symptoms are prevalent in prodromal and clinical Parkinson's disease (PD). Some evidence suggests that they may also signify risk for motor complications (motor fluctuations and dyskinesias) of dopamine replacement therapy (DRT).

Objective

To determine whether neuropsychiatric symptoms present in de novo PD (ie, before DRT initiation) predict the severity of eventual motor complications of DRT.

Methods

We used clinical, demographic, neurobehavioral, and neuroimaging data from the Parkinson's Progression Markers Initiative (PPMI), a multicenter observational PD study. Participants were unmedicated at enrollment and 361 initiated DRT during PPMI follow‐up. We used Cox proportional hazard and multivariate ordinal mixed‐effects regression models to evaluate the relationship between baseline neuropsychiatric symptoms and motor complications as measured by the Movement Disorders Society‐revised Unified Parkinson's Disease Rating Scale (MDS‐UPDRS).

Results

The cumulative incidences of dyskinesias and motor fluctuations during follow‐up (6.0 ± 1.5 years) were 34.3% and 59.9%, respectively. Both apathy and high trait‐anxiety (top quartile) conveyed over two‐fold increases in hazard for dyskinesia onset and for adverse impact on activities of daily living caused by both dyskinesias and motor fluctuations. The longitudinal severity of motor fluctuations and dyskinesias was significantly predicted by baseline trait‐anxiety and apathy, but not depression. Models were adjusted for dimensionally related symptoms (eg autonomic dysfunction) and potential confounding variables (eg DRT dose).

Conclusions

Apathy and anxiety levels in de novo PD may be neuropsychiatric biomarkers of vulnerability to earlier and more disabling motor complications of DRT.

Parkinson's disease (PD) causes death of dopamine (DA)‐producing cells in the substantia nigra of the ventral midbrain, thereby engendering the cardinal motor signs of the disease. DA replacement therapy (DRT) restores DA levels and ameliorates motor impairment; this has been the core strategy of symptomatic control in PD for decades. ¹ However, DRT is complicated over time by motor fluctuations and levodopa‐induced dyskinesias (LID). ² The onset of these motor complications of therapy (MCT) is determined by increasing dopaminergic pathology and synaptic changes (ie disease burden and duration), rather than the duration of DRT exposure. ³ Nonetheless, higher DRT doses hasten MCT onset, indicating they originate from interactions between underlying pathology and DRT. ² , ³ This suggests the possibility of forecasting the course of MCT with data from before DRT initiation (de novo PD), by which time dopaminergic degeneration is well underway. ⁴ However, it remains unclear which clinical, demographic, or neuroimaging measures in early PD are most predictive of MCT.

Although DA dysfunction and DRT directly underlie MCT, non‐dopaminergic mechanisms can modify their onset and severity. For example, altered balance between serotonergic and dopaminergic terminals in the striatum appears to facilitate LIDs, possibly due to mishandling of exogenous DA by serotonergic synaptic machinery. ⁵ Serotonergic pathology is extensive throughout the prodromal PD brain and correlates with motor impairment in early clinical PD. ⁶ Animal models also support a role for noradrenaline loss in exacerbating motor impairment ⁷ and LID onset. ⁸

Like MCT, neuropsychiatric symptoms in PD (PD‐NPS) arise from both DA and non‐DA changes. PD‐NPS are variable, complex, prevalent, and disabling, which likely reflects the pathological heterogeneity of PD and multifactorial etiologies. ⁹ , ¹⁰ , ¹¹ These properties, along with their symptomatic dominance in prodromal PD, make them intriguing biomarkers of latent pathophysiology. ¹² For example, DA agonists and acetylcholinesterase inhibitors are clinically efficacious for reducing apathy in PD, which could be explained by overlapping contributions of motivational and cognitive deficits to apathy. ¹³ , ¹⁴ Depression and anxiety are often associated with serotonergic and/or noradrenergic changes; the raphe nuclei and locus coeruleus are heavily affected even in prodromal PD. ¹⁵ , ¹⁶ Interestingly, pramipexole and venlafaxine most consistently show clinical efficacy for reducing depression in PD, ¹³ reinforcing the hypothesis that both dopaminergic and non‐dopaminergic mechanisms underlie various PD‐NPS. Selective neurotransmitter deficits also shape PD‐NPS subtypes. For example, cholinergic PD psychosis manifests in relation to cognitive decline and anticholinergic medications, in contradistinction to the drug‐induced dopaminergic hallucinosis caused by DA agonists. ¹⁷

Given these observations, we reasoned that PD‐NPS prior to DRT initiation may be useful for predicting MCT severity. A recent exploratory study ¹⁸ suggested that pre‐DRT anxiety may predict a more rapid onset of LIDs, but did not analyze PD‐NPS specifically or in a dissociable fashion (eg no adjustments for correlated features such as depression or dysautonomia ¹⁹ were made). We therefore sought to investigate which PD‐NPS in pre‐DRT disease are most predictive of MCT severity. To this end, we used data from the Parkinson's Progression Markers Initiative (PPMI), a multicenter observational study of de novo PD.

Methods

Study Design

PPMI (www.ppmi-info.org) is a multi‐center observational PD study that enrolls participants with a recent PD diagnosis who have not yet initiated antiparkinsonian therapy (de novo PD). Enrollment has been active since June 2010 and data are available upon reasonable request and approval via the PPMI data portal (http://www.ppmi-info.org/data). Research documents, including the protocol, planned measures, and evaluation timeline is available on the website as well. Information is available in published form. ²⁰ All PPMI centers obtain IRB approval and written informed consent from all participants. Notable exclusion criteria include a Hoehn & Yahr stage above stage II at screening, atypical parkinsonian syndromes, use of PD medication at time of screening or expectation to initiate treatment <6 months, clinician judgment of dementia, and recent (last 6 months) use of medications thought to interfere with DaTscan imaging. Full inclusion/exclusion criteria are provided in the PPMI protocol via the links above. Data for 494 participants were available as of June 6, 2020. Based on the availability of variables of interest for the present study, we identified 361 participants for inclusion (see next section for details).

Clinical Measures

Our primary outcomes of interest were the onset and severity of motor complications of therapy (MCT) among participants for whom data were available at baseline and after DRT initiation. The Movement Disorders Society‐revised Unified Parkinson's Disease Rating Scale (MDS‐UPDRS) Motor Complications (part IV) section was used to track the primary outcome variables of interest, dyskinesias and motor fluctuations. Each item is scored on a non‐linear scale from 0 to 4. We extracted data for the following part IV items: Daily time spent with dyskinesias (q4.1), functional impact of dyskinesias (q4.2), daily time spent in the “off” state of motor fluctuations (q4.3), and functional impact of motor fluctuations (q4.4). Items 4.1 and 4.3 measure the average proportion of the waking day where symptoms are present (0 = 0%, 4 = >75%). Items 4.2 and 4.4 measure their functional impact (0 = No impact on daily activities or social interactions, 4 = symptoms preclude most activities or social interactions when present).

Neuropsychiatric and non‐motor scales used in our investigation included the 15‐item Geriatric Depression Scale (GDS‐15), State–Trait Anxiety Inventory (STAI), Montreal Cognitive Assessment (MoCA), the apathy item of the MDS‐UPDRS part I scale, the Epworth Sleepiness Scale (ESS), the Scales for Outcomes in Parkinson's disease ‐ Autonomic Dysfunction (SCOPA‐AUT), and the REM Sleep Behavior Questionnaire. Scores for all scales were calculated in accordance with PPMI recommendations.

Other variables selected for analysis based on prior studies ³ , ¹⁸ , ²¹ included motor impairment (MDS‐UPDRS part III score), sex, levodopa equivalent daily dose (LEDD), age at diagnosis, disease duration, and DaTScan striatal binding ratios (SBRs) in caudate and putamen. LEDD was calculated in accordance with convention. ²² We also tabulated the total LEDD at the time of DRT initiation (“Starting LEDD dose”) and for each visit (“Current LEDD”) as separate variables. Results of dopamine transporter (DaT) imaging with (Ioflupane‐¹²³I) single photon emission computed tomography (SPECT) were used as calculated by PPMI. After image processing, voxel‐wise standardized uptake value (SUV) averages were used to calculate SBRs for the left/right caudate and putamen, where SBR is calculated with the formula: (target region/occipital lobe reference)‐1. ²³

Data for 133 of the 494 PPMI participants were excluded due to missing or unavailable data. In brief, 80 participants did not initiate DRT during available follow‐up. Another 25 participants were excluded due to lack of any MDS‐UPDRS part IV evaluations. Finally, totals of 18, 4, 3, 2, and 1 participants were excluded due to lack of baseline/screening data for the STAI, REM Sleep Behavior Questionnaire, SCOPA‐AUT, and ESS scales. We have provided a description of the clinical characteristics of these participants and statistical comparisons to the included group (n = 361) in supplemental Table S3.

Modeling and Statistical Analysis

All analyses were conducted in R and statistical significance was set at (α = 0.05). ²⁴ We used separate approaches to dissociate risk factors for the onset of MCT and the eventual severity of MCT. Onset was investigated using Cox proportional‐hazards regression of the incidence of dyskinesias or fluctuations as a function of disease duration and baseline NPS. MCT events were defined as the first PPMI visit where participants endorsed a score ≥ 1 on the corresponding MDS‐UPDRS IV item. Participants who did not experience the event during their available follow‐up were coded as not having an event (right‐censorship). We used the R packages survival ²⁵ and survminer ²⁶ for hazard analyses. Severity of MCT symptoms was investigated using cumulative link mixed models, a form of ordinal regression modeling that supports random effects, analogous to linear mixed‐effects models. ²⁷ Individual MDS‐UPDRS IV item scores are not necessarily linear and are more appropriately modeled as ordinal variables. Proportional odds assumptions for all models were checked using the nominal effects test. All models achieved acceptable convergence (ie maximum likelihood estimation by gradient descent optimization terminated with max gradient <0.0001). Participant ID was entered as the only random effect, all others were fixed variables.

Variables of the PPMI inventory were selected for regression if they measured neuropsychiatric symptoms or were previously associated with motor complications in PD. ³ , ¹⁸ , ²¹ In total, 16 variables of interest were identified for the saturated starting model. Neuropsychiatric scales included: depression (GDS‐15), trait anxiety (STAI), state anxiety (STAI), daytime sleepiness (ESS), apathy (MDS‐UPDRS I), and cognition (MoCA). Other scales or measures included: age, age at diagnosis, sex, motor impairment (MDS‐UPDRS III), mean caudate DaT SBR, mean putamen DaT SBR, SCOPA‐AUT score, LEDD at DRT initiation (starting LEDD), and the presence/absence of REM behavior disorder (RBD) symptoms. RBD symptoms were considered present for a score ≥ 5. ²⁸ All 16 variables were used to describe the study population at baseline (Table 1) and for univariate ordinal regression models. Disease duration was used to track time‐to‐event in proportional‐hazards and therefore not included as an independent variable. Similarly, age is collinear with disease duration and was excluded, leaving 14 variables for proportional‐hazards analysis. Results of all univariate ordinal and proportional‐hazards regression analyses are presented in full in Tables S1 and S2. To correct for the multiple comparisons (16 or 14 variables), we applied the Holm method to adjust P‐values. Univariate models with adjusted P‐values <0.05 were then used together for multivariate analyses. All models that included an anxiety variable were also adjusted for depression (GDS‐15) and autonomic dysfunction (SCOPA‐AUT) regardless of their univariate P‐values due to the need to adjust for these confounding effects. Graphs of the cumulative incidence of motor complications (presence and functional impact) are overlaid with hazards ratios, 95% confidence intervals, and P‐values from univariate proportional‐hazards regression.

TABLE 1.

Clinical characteristics and therapeutic trajectories in PPMI (n = 361)

	Variable	Mean (SD)	Min	Max
Baseline clinical measures	Age (y)	61.6 (9.6)	34.0	85.0
	Age at diagnosis (y)	61.0 (9.6)	32.1	84.9
	Disease duration (y)	0.5 (0.5)	0	3.0
	Duration of PPMI follow‐up (y)	6.0 (1.5)	1.1	11.7
	Hoehn & Yahr stage*
	1	157 (43.5%)
	2	201 (55.7%)
	3	3 (0.8%)
	Male sex	238 (65.9%)
	RBD symptoms present (RBD score ≥ 5)	136 (37.7%)
	MDS‐UPDRS Part III (“off” PD meds)	20.9 (8.9)	4	51
	Caudate DaTScan SBR	1.97 (0.54)	0.39	3.71
	Putamen DaTScan SBR	0.82 (0.30)	0.24	2.48
	MoCA (total)	27.1 (2.4)	17	30
	GDS‐15 total	2.4 (2.5)	0	14
	Depression (GDS‐15 total > 4)	53 (14.7%)
	Apathy (n scoring ≥1 on MDS‐UPDRS q1.5)	61 (16.9%)
	State–trait anxiety inventory
	State subscore	33.2 (10.4)	20	76
	Trait subscore	32.6 (9.5)	20	63
	SCOPA‐AUT total	8.4 (5.3)	0	28
Therapeutic measures	DRT delay from diagnosis (y)	1.5 (1.0)	0.2	4.7
	DRT delay from baseline PPMI evaluations (y)	1.0 (0.8)	0.03	4.0
	LEDD at DRT initiation	215 (180)	10	1200
	Cumulative dyskinesia prevalence (MDS‐UPDRS dyskinesia daily duration ≥1)	124 (34.3%)
	Cumulative fluctuations prevalence (MDS‐UPDRS fluctuations daily duration ≥1)	217 (59.9%)
	Time to initial dyskinesias from diagnosis (y)	5.1 (1.7)	1.2	10.6
	Time to initial fluctuations from diagnosis (y)	4.4 (1.6)	1.2	9.3

Summary statistics for the PPMI participants in this analysis. Mean values and range are presented for continuous variables, with standard deviations in parentheses. Count data are presented for ordinal and nominal variables.

^*Three participants transitioned to Hoehn & Yahr stage III in the interval between screening and baseline evaluation.

Results

Baseline demographic and clinical characteristics of the cohort included in the present analysis are provided in Table 1. Briefly, among 361 participants identified for inclusion, duration of PPMI follow‐up was 6.0 ± 1.5 (mean ± standard deviation) years. At baseline, 53 (14.7%) met GDS‐15 criteria ²⁹ for a depressive disorder and 61 (16.9%) endorsed apathy. Throughout follow‐up, the cumulative prevalence of motor fluctuations and dyskinesias was 59.9% and 34.3%, with onset at 4.4 ± 1.6 years and 5.1 ± 1.7 years after PD diagnosis, respectively. DRT initiation was similar across participants, starting 1.5 ± 1.0 years after diagnosis at an initial dosage of 215 ± 180 mg/day. PPMI participants excluded (n = 133) from the present analysis were comparable to the 361 included cases in most characteristics (Table S3), though they exhibited higher DaTscan SBRs in the caudate and putamen and slightly less motor impairment.

We first ascertained whether certain neuropsychiatric symptoms—measured prior to DRT initiation—predicted a more rapid onset of MCT. Results of multivariate proportional‐hazards regression with baseline clinical variables (Table 2) showed that after adjusting for depression, trait‐anxiety and apathy were independently associated with the onset of dyskinesias, as well as the start of functional impact for both motor fluctuations and dyskinesias. Baseline depression, starting LEDD, and autonomic dysfunction were not significantly associated with the onset of any MCT effect in multivariate models. A full list of univariate associations is provided in Table S1.

TABLE 2.

Multivariate modeling of dyskinesia and fluctuation onset with cox regression

Outcome variable (score 0–4)	Baseline variable	Hazard ratio	95% CI	P
Dyskinesias: Presence (MDS‐UPDRS 4.1 ≥ 1)	Depression (GDS‐15)	1.01	0.91–1.11	0.894
	Trait‐anxiety (per Δ10 pts)	1.28	1.02–1.61	0.034
	Apathy score	1.45	1.06–1.99	0.020
Dyskinesias: Functional impact (MDS‐UPDRS 4.2 ≥ 1)	Starting LEDD (per 100 mg/day)	1.13	0.99–1.29	0.068
	Depression (GDS‐15)	1.04	0.91–1.19	0.579
	Trait‐anxiety (per Δ10 pts)	1.47	1.05–2.06	0.024
	Apathy score	1.75	1.20–2.54	0.003
	SCOPA‐AUT	1.04	0.99–1.09	0.089
Fluctuations: Presence (MDS‐UPDRS 4.3 ≥ 1)	No significant univariate predictors after correction for multiple comparisons
Fluctuations: Functional impact (MDS‐UPDRS 4.4 ≥ 1)	Trait‐anxiety (per Δ10 pts)	1.26	1.04–1.54	0.020
	Apathy score	1.39	1.05–1.85	0.022

One multivariate proportional‐hazards regression model was constructed for each MCT outcome variable (MDS‐UPDRS items 4.1–4.4). Variables were selected based on statistical significance (P < 0.05 after Holm adjustment to correct for 14 multiple comparisons) in univariate modeling. Results for all univariate models are provided in TABLE S2 Hazard ratios were calculated as the exponentiated coefficients of the proportional‐hazards model Events were defined as a score ≥ 1 on the corresponding MDS‐UPDRS IV and individuals were right‐censored if this score was never reached. Statistically significant P‐values (P < 0.05) are entered as bold text.

To visualize the predictive value of baseline apathy and trait‐anxiety for MCT onset, we created cumulative incidence plots (Fig. 1). Participants were stratified using the presence/absence of apathy (≥ 1 on the MDS‐UPDRS apathy item) and using the bottom and top quartiles of the STAI trait‐anxiety score at baseline (score ≤ 25 and ≥37, respectively). The figure panels are overlaid with the hazard ratios (with 95% CI and P‐values) for each group. Baseline apathy and high trait‐anxiety (top quartile) conveyed an approximately 2‐fold increase in the hazard for the incidence of dyskinesias and over 3‐fold increased hazard for functional impact of dyskinesias. Motor fluctuation onset was not significantly associated with either baseline measure in multivariate analyses and thus is not graphically depicted, but apathy and high trait‐anxiety nonetheless hastened the onset of functional impact from motor fluctuations (2‐fold hazard increase for either).

FIG 1.

Cumulative incidence of motor complications predicted by baseline apathy and anxiety. Proportional‐hazards models of MCT were constructed for baseline apathy (left column) and trait‐anxiety (right column). Time‐to‐event cumulative incidence graphs of dyskinesia incidence (top row), dyskinesia functional impact (middle row) and fluctuation functional impact (bottom row) are overlaid by median hazard indicators (dashed vertical lines). Hazard ratios (HR) for each variable are also presented in text, along with 95% confidence intervals and P‐values. HRs correspond to the exponentiated coefficients of proportional‐hazards regression. Each event is defined as a score ≥ 1 on the corresponding MDS‐UPDRS IV item.

We considered that different patterns of DRT onset could confound the relationship between neuropsychiatric symptoms and MCT onset (eg if higher anxiety correlated with earlier or higher DRT doses). To address this possibility of bias, we analyzed DRT onset patterns based on anxiety (low, middle, or high) and apathy at baseline (Table 3). Baseline apathy was not associated with the interval between diagnosis and DRT initiation, total DRT duration, starting LEDD, or use of dopamine agonists at LEDD initiation. Interestingly, individuals with low trait‐anxiety at baseline (STAI‐T score ≤ 25) delayed DRT initiation by about 0.4 years, a statistically significant difference. No differences in total DRT duration, starting dose, or agonist use were observed among the anxiety groups.

TABLE 3.

DRT use in relation to baseline apathy and anxiety symptoms

DRT measure	No baseline apathy (n = 300)	Baseline apathy (n = 61)	W/χ2	P
Disease duration at enrollment (years)	0.5 (0.5)	0.6 (0.6)	8198	0.200
DRT initiation delay (years after diagnosis)	1.5 (1.0)	1.6 (1.1)	8941	0.778
DRT duration (years)	4.5 (1.5)	4.3 (1.9)	9950	0.281
Starting LEDD	214 (185)	218 (152)	8415	0.316
Agonist at DRT initiation	86 (29%)	17 (28%)	0	1.000

DRT measure	Trait anxiety ≤ 25(n = 93)	Trait anxiety 25–37(n = 169)	Trait anxiety ≥ 37(n = 99)	χ2	P
Disease duration at enrollment (years)	0.6 (0.7)	0.5 (0.5)	0.5 (0.5)	2.66	0.265
DRT initiation delay (years after diagnosis)	1.8 (1.2) a	1.4 (0.8)	1.4 (0.9)	9.88	0.007
DRT duration in PPMI	4.3 (1.7)	4.6 (1.5)	4.6 (1.7)	3.29	0.193
Starting LEDD	213 (184)	217 (183)	213 (173)	0.114	0.945
Agonist at DRT initiation	25 (27%)	53 (31%)	25 (25%)	1.310	0.520

Baseline apathy and anxiety measurements as applied to proportional‐hazards regression (Table 1, Fig. 1) were investigated for potential confounding patterns of DRT use. Apathy was defined as a score ≥ 1 on the MDS‐UPDRS apathy item (q1.5). Participants were grouped by trait‐anxiety as high (top quartile, score ≥ 37), low (bottom quartile, score ≤ 25), or middle trait‐anxiety severity levels (25th–75th percentile). Continuous variables were compared with the Mann–Whitney U (apathy) and Kruskal‐Wallis test (anxiety), for which test‐statistics and P‐values are presented. Agonist use was compared using the Chi‐square (χ²) test. Statistically significant P‐values (P < 0.05) are entered as bold text.

^aPost‐hoc Mann–Whitney U test significant (P < 0.05) after Holm correction when compared to other anxiety levels.

To determine whether baseline neuropsychiatric features predict the eventual severity of MCT in addition to the rapidity of their onset, we modeled MCT variables with longitudinal multivariate ordinal regression (Table 4). These results consistently identified baseline trait anxiety and apathy—but not depression— as significant predictors of MCT severity; however, as was observed for proportional‐hazards models, we did not identify any non‐motor or neuropsychiatric predictors of motor fluctuation daily duration. We also found that autonomic dysfunction was an independent predictor of the functional impact of motor fluctuations. Current LEDD and disease duration were also consistently predictive of MCT severity.

TABLE 4.

Multivariate models of dyskinesia and fluctuation severity

Outcome variable (score 0–4)	Baseline variable (except *)	Odds ratio	95% CI	P
Daily time spent with dyskinesias (MDS‐UPDRS 4.1)	*PD duration (y)	2.46	2.13–2.84	<0.001
	*Current LEDD (per 100 mg/day)	1.07	1.03–1.11	0.001
	Trait‐anxiety (per Δ10 pts)	1.96	1.15–3.36	0.014
	Apathy score	2.20	1.01–4.82	0.048
Functional impact of dyskinesias (MDS‐UPDRS 4.2)	*PD duration (y)	1.81	1.52–2.15	<0.001
	*Current LEDD (per 100 mg/day)	1.04	1.00–1.08	0.058
	Depression (GDS‐15)	1.02	0.83–1.26	0.828
	Trait‐anxiety (per Δ10 pts)	1.75	1.04–2.95	0.034
	Apathy score	2.41	1.21–4.79	0.012
	SCOPA‐AUT total	1.07	0.99–1.14	0.071
Daily time spent in the “off” state (MDS‐UPDRS 4.3)	*PD duration (y)	1.74	1.60–1.91	<0.001
	*Current LEDD (per 100 mg/day)	1.05	1.02–1.08	0.002
Functional impact of fluctuations (MDS‐UPDRS 4.4)	*PD duration (y)	1.95	1.75–2.17	<0.001
	*Current LEDD (per 100 mg/day)	1.06	1.02–1.09	<0.001
	Trait‐anxiety (per Δ10 pts)	1.57	1.09–2.25	0.015
	Apathy score	1.88	1.10–3.23	0.022
	SCOPA‐AUT total	1.06	1.01–1.12	0.016

One multivariate ordinal mixed‐effects regression model was constructed for each MCT outcome variable (MDS‐UPDRS items 4.1–4.4). Variables were selected based on statistical significance (P < 0.05 after Holm adjustment to correct for 16 multiple comparisons) in univariate modeling. Results for all univariate models are provided in Table S3 Odds ratios were calculated as the exponentiated coefficients of the ordinal regression model. All models satisfied the proportional odds assumption of ordinal regression. Statistically significant P‐values (P < 0.05) are entered as bold text.

Discussion

We report that pre‐DRT levels of trait anxiety and apathy—but not depression—were powerful predictors of the onset and severity of MCT after DRT is initiated. This conclusion is supported by complementary longitudinal analyses—proportional‐hazards regression and ordinal mixed‐effects regression—that adjusted for previously identified MCT risk factors. This finding supports the notion that pre‐DRT levels of anxiety and apathy may be useful for assessing individual patient risk for motor fluctuations and dyskinesias, which may be useful from a prognostic standpoint. Additionally, our findings indicate the non‐motor features of early PD may be useful biomarkers of latent neurobiological changes or factors that predispose to MCT. ³⁰

The STAI anxiety scale used in the present study distinguishes between trait‐anxiety—levels of long‐standing anxiety felt in general—and state‐anxiety—current anxiety “in the moment.” Thus, whereas state‐anxiety is responsive to somatic symptoms or psychological and environmental stressors, trait‐anxiety is more closely related to neuroticism. ³¹ The stronger association we observe between MCT and trait‐anxiety is therefore noteworthy because anxious, pessimistic, and neurotic personality traits are also associated with increased risk for PD itself. ³² Thus, these constructs may be measuring phenomena related to the neurobiological changes in prodromal PD, such as early noradrenergic or serotonergic dysfunction. Furthermore, trait‐anxiety reports may be less likely than state‐anxiety to be confounded by aspects of anxiety related to participation in research evaluation or fluctuant autonomic function (which is more comprehensively measured by the SCOPA‐AUT in PPMI). The lack of association between depression and MCT after adjusting for trait‐anxiety is also noteworthy. The GDS‐15 scale is biased toward a focus on psychological, social, and emotional elements of depression (eg fearfulness, hopelessness, withdrawal) as opposed to somatic symptoms. ³³ This is advantageous for use in PD, as somatic depressive symptoms overlap substantially with core features of parkinsonism (eg fatigue, cognitive changes). Thus, our findings are unlikely to be confounded by the misattribution of parkinsonian symptoms to depressive symptomatology and provide strong support for anxiety as a better predictor of MCT onset and severity.

We also found that apathy was a strong predictor of MCT after adjustment for depression. While these constructs may be difficult to distinguish in PD, they are nonetheless distinct entities and can be attributed to different neurotransmitter pathologies. ¹⁴ Dopamine is a key substrate of motivation, ³⁴ making the hypodopaminergic states of PD a likely source of some apathetic sentiments. Accordingly, dopamine agonists reduce feelings of apathy in PD ³⁵ and worse dopaminergic denervation predicts apathy during LEDD reduction after deep‐brain stimulation (DBS) operations. ³⁶ , ³⁷ However, executive cognitive impairment can contribute to the presentation of apathy by interfering with the planning and execution of normal activities. ³⁸ Thus, apathy is unlikely to be a purely dopaminergic phenomenon—indeed, rivastigmine is clinically efficacious for reducing PD apathy, suggesting a more complex pathological basis. ¹³ , ³⁹ Nonetheless, the dopaminergic elements of apathy may be important for our observation that pre‐DRT apathy is predictive of more rapid and severe MCT onset.

Some previous studies have investigated predictors of MCT in PD, but with much less emphasis on neuropsychiatric and non‐motor features. For example, one recent study ²¹ of dyskinesia and motor fluctuation onset reported univariate correlations with depression and anxiety scales but did not pursue a multivariate characterization of their utility for predicting MCT. Additionally, this study did not enroll participants prior to DRT initiation and had 18‐month visit intervals (vs. 6‐month in PPMI), which reduced temporal resolution for MCT onset. Another study did use PPMI to investigate predictors of dyskinesia onset, ¹⁸ but did not investigate apathy or examine predictors of MCT severity, as we have done here. Although the authors did note the association between dyskinesia and trait‐anxiety, they did not adjust for depression or autonomic dysfunction in multivariate analyses. Thus, we believe that our analysis builds upon these reports by specifically evaluating neuropsychiatric predictors and providing novel evidence for predictors of MCT severity (as opposed to onset alone).

We do acknowledge several limitations to our analysis. Firstly, we recognize that apathy was assessed using only a single item of the MDS‐UPDRS (q1.5). Whereas prior estimates have placed the prevalence of apathy in early PD at approximately 25% when using the neuropsychiatric inventory (NPI) or semi‐structured interviews, ⁴⁰ , ⁴¹ we found that 17% of PPMI enrollees endorsed apathy with this instrument. Therefore, it is possible that the MDS‐UPDRS item is slightly less sensitive than the NPI, but this may also indicate a somewhat lower risk of false‐positive endorsements. The MDS‐UDPRS apathy item also correlates strongly (0.67) with the PD‐validated Lille Apathy Rating Scale. We also provided some evidence that low trait‐anxiety may lead to delayed DRT initiation. Due to perfect collinearity between time‐variables, we cannot specifically adjust for this factor while also accounting for disease duration. However, there is strong evidence that disease duration is a much more important determinant of MCT onset than cumulative DRT duration. ³ Thus, this particular finding may simply constitute an interesting observation of how trait‐anxiety can shape decisions around the initiation of treatment. Future work may address whether this is the case or if another latent correlated factor is at play. Finally, 133 PPMI participants were excluded from the present analysis, primarily due to lack of DRT initiation or MCT data (MDS‐UPDRS part IV) at present. These participants had notably higher DaT SBRs in both the caudate and putamen, which could be a factor that impacts PD symptoms and the DRT initiation timeline. Thus, our analysis may include a selection bias such that participants with relatively higher DaT SBR soon after diagnosis were not able to be analyzed. This suggests our study may be less generalizable to PD patients with higher DaT preservation at the time of diagnosis, but also supports further targeted studies as these participants initiate DRT.

Altogether, the present study provides several unique insights into the value of neuropsychiatric features—particularly anxiety and apathy—as indicators of MCT risk in PD. In addition to suggesting prognostic value for these symptoms that could assist clinicians when discussing treatment trajectories with patients, our findings indicate that anxiety and apathy could be biomarkers for neuropathological changes that underlie these complications. ³⁰ Since anxiety, apathy, and other non‐motor features are common in prodromal PD, they could also become increasingly important prognostic biomarkers for understanding disease trajectory. This assumes a significant degree of overlap in the neurobiological origins of these symptoms in prodromal versus early PD, which remains to be clarified. Nonetheless, neuropsychiatric features may be useful for investigating the myriad changes in the prodromal PD brain that indicate vulnerability to experiencing severe therapeutic complications, which may in turn accelerate research on strategies to abrogate their occurrence.

Author Roles

(1) Research Project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript Preparation: A. Writing of the first draft, B. Review and Critique.

J.T.H.: 1B, 1C, 2A, 2B, 3A

K.P.: 2C, 3B

L.L.G.: 2C, 3B

K.A.M.: 2C, 3B

G.M.P.: 1A, 2C, 3B

Disclosures

Ethical Compliance Statement: All PPMI centers obtain approval from their respective institutional review boards and written informed consent is obtained from all participants. Research documents, protocols, and specifics are available at www.ppmi-info.org. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this work is consistent with those guidelines.

Funding Sources and Conflicts of Interest: The authors report no funding pertinent to this manuscript, nor any conflicts of interest, financial or otherwise. PPMI—a public‐private partnership—is funded by the Michael J. Fox Foundation for Parkinson's Research and funding partners, including Abbvie, Allergan, Amathus Therapeutics, Avid Radiopharmaceuticals, Biogen, Biolegend, Bristol‐Myers Squibb, Celgene, Denali, GE Healthcare, Genentech, GlaxoSmithKline, Janssen Neuroscience, Lilly, Lundbeck, Merck, Meso Scale Discovery, Pfizer, Piramal, Prevail Therapeutics, Roche, Sanofi Genzyme, Servier, Takeda, Teva, UCB, Verily, and Voyager Therapeutics. Golub Capital is a philanthropic funding partner.

Financial Disclosures for the Previous 12 Months: J.T.H.: Receives tuition and stipend support through the Medical Scientist Training Program at the Johns Hopkins School of Medicine (NIH/NIGMS T32GM007309) and through the National Institute on Aging (F30AG067643). K.P.: No disclosures to report. L.L.G.: No disclosures to report. K.M.: Dr. Mills receives salary support through the NIH NCATS (KL2TR001077, PI Daniel Ford) and the NIH/NINDS (5K23NS101096). G.P.: Dr. Pontone receives funding through the NIH/NIA as part of a K23 award (5K23AG044441).

Supporting information

Table S1. Univariate models of dyskinesia and fluctuation onset with Cox regression.

For each of the 14 potential predictors of MCT onset suitable for proportional‐hazards regression, one univariate model was constructed per MCT outcome variable (MDS‐UPDRS items 4.1–4.4). Hazard ratios were calculated as the exponentiated coefficients of the proportional‐hazards model. Events were defined as a score > 1 on the corresponding MDS‐UPDRS IV and individuals were right‐censored if this score was never reached. P‐values were corrected for 14 comparisons using the Holm method.

Table S2. Univariate models of dyskinesia and fluctuation severity. For each of the 16 potential predictors of MCT onset suitable for mixed‐effects ordinal regression, one univariate model was constructed per MCT outcome variable (MDS‐UPDRS items 4.1–4.4). Odds ratios were calculated as the exponentiated coefficients of the ordinal regression model. All models satisfied the proportional odds assumption of ordinal regression. P‐values were corrected for 16 comparisons using the Holm method.

Table S3. Clinical characteristics PPMI participants excluded from analysis (n = 133). Summary statistics for the PPMI participants excluded from this analysis (n = 133). Mean values and range are presented for continuous variables, with standard deviations in parentheses. Count data are presented for ordinal and nominal variables. *n = 132; **n = 129; ***n = 108. Statistical tests for a significant difference between the excluded (n = 133) and included (n = 361) groups are provided in the final column. Results are summarized with t‐statistics for t‐test comparisons, W‐statistics for Mann–Whitney U non‐parametric comparisons, and χ² values for Pearson's Chi‐squared test. Df = degrees of freedom.