Evaluating the Use of Digital Biomarkers to Test Treatment Effects on Cognition and Movement in Patients with Lewy Body Dementia

INTRODUCTION

Lewy body dementia (LBD), which encompasses both Parkinson’s disease dementia (PDD) and dementia with Lewy bodies (DLB), is a neurodegenerative disorder that predominantly affects dopaminergic and cholinergic neurons [1, 2]. The formation and propagation of Lewy bodies accompanied by progressive neurodegeneration lead to an overlap of motor and cognitive impairments in LBD and a resultant range of symptoms, including dementia, parkinsonism, and behavioral and sleep disturbances [2–4]. Given the multifaceted nature of LBD, it is a challenging disease to manage therapeutically, and balancing symptomatic efficacy can often lead to worsening of other symptoms or an increased burden of side effects [5–7].

Utilizing wearable sensors or applications to assess motor function, cognition, and behavior opens the possibility of including these to develop digital biomarkers and apply them in both clinical trial and community settings. These devices have shown great promise in discriminating individuals with Parkinson’s disease (PD) from healthy controls and measuring PD symptom severity [8] as well as discriminating between healthy individuals and individuals at risk of progressing to dementia within 3 years [9]. Within clinical trials, digital devices allow for continuous and objective data monitoring for several weeks or months at a time, and once validated could potentially alleviate reliance on clinical scales or questionnaires that depend on a patient’s ability to recall information or complete diaries [10, 11]. In addition, digital devices may be less disruptive to a patient’s lifestyle or normal behavior than traditional clinical visits, less influenced by “white coat syndrome”, and, with adequate compliance, allow for long-term symptom monitoring or treatment tracking in patients’ natural environments [12, 13].

Mevidalen (LY3154207) is a selective D1 receptor positive allosteric modulator (D1PAM) which enhances the affinity of the D1 receptor for endogenous and exogenous dopamine thereby amplifying its response to dopamine [14, 15]. Because of its novel mechanism, mevidalen has the potential to improve cognitive performance by enhancing frontal dopaminergic neurotransmission, activating cortical neurons, and enhancing synaptic plasticity and D1-mediated enhanced acetylcholine release and is hypothesized to also improve motor function and daytime sleepiness [15, 16]. Mevidalen was recently assessed in a Phase 2 randomized, placebo-controlled trial (PRESENCE; NCT03305809) designed to test the safety and efficacy of mevidalen in patients with mild-to-moderate LBD. In PRESENCE, mevidalen did not meet its primary endpoint of improvement in cognition measured by change in the Continuity of Attention (CoA) composite score of the Cognitive Drug Research Computerized Cognition Battery (CDR-CCB) from baseline to Week 12; however, it did produce robust global improvement as measured by the Alzheimer’s Disease Cooperative Study-Clinical Global Impression of Change (ADCS-CGIC) [17]. Mevidalen treatment also resulted in statistically significant and clinically meaningful benefits in motor function on top of stable treatment while improving or not worsening non-motor complications commonly reported in current therapies [17].

In the PRESENCE trial, which assessed the impact of mevidalen for the symptomatic treatment of LBD, participants were asked to wear an actigraphy watch and were also invited to partake in a sub-study using an iPad trial app. This work aimed to evaluate whether the wrist-worn actigraphy watch for continuous activity and sleep tracking and frequent cognitive digital symbol substitution (DSST), spatial working memory (SWM), and finger tapping tests using the iPad app could detect mevidalen treatment effects in patients with mild-to-moderate dementia due to LBD.

METHODS

Patient population and trial design

PRESENCE (NCT03305809: ClinicalTrials.gov) was a Phase 2, multicenter, randomized, double-blind, parallel-group, placebo-controlled, 12-week clinical trial assessing mevidalen (LY3154207) for symptomatic treatment of LBD (PDD and DLB). Participants (n = 344) were randomized 1:1:1:1 to receive daily doses (10, 30, or 75 mg) of mevidalen or placebo. Digital devices were used in a subpopulation (all US-based and English-speaking participants (n = 238) for actigraphy and as an optional sub-study for US-based participants (n = 160) for the iPad app) to evaluate the cognitive, motor, activity, and sleep effects of mevidalen. The trial design can be visualized in Fig. 1. The study included a Screening Period of 7 to 14 days (Visit 1–2), a Pre-intervention Period of ≥11 days and ≤17 days (Visit 2–3), a 12-week Intervention Period (Visits 3–11), and a 14-day Safety Follow-up period that included a 2-week follow-up visit (Visit 801). Actigraphy participants wore the actigraphy devices continuously for three separate 2-week periods: pre-intervention, during intervention, post- intervention. iPad trial app participants performed 2 out of 3 tests (DSST, SWM, and finger tapping) on an alternating basis twice daily during pre- and post- intervention, and once daily during intervention.

OPEN IN VIEWER

Fig. 1

Schematic showing the PRESENCE trial period and the use of digital devices. Wk, week.

The main inclusion/exclusion criteria have been described in full elsewhere [17]. Enrolled participants were aged 40 to 85 years (inclusive) with dementia, as defined by a decline in cognitive function and with a Montreal Cognitive Assessment (MoCA) score of 10–23 (inclusive) [18]. Enrolled participants met the diagnostic criteria for PD [19] or DLB [20] with categorized Modified Hoehn and Yahr Stages 0–4 [21].

RESULTS

Compliance

During the intervention phase, overall actigraphy compliance was > 90% (Fig. 2A). Similar actigraphy compliance was also observed in the pre-intervention and post-intervention phases (Supplementary Figure 1). Note that the reduction in compliance on days 13 and 14 in the graph can be explained by participants returning devices at different times. Trial app compliance was > 60% among those who completed the study (Fig. 2C). There was a high rate of completed tests throughout the study period with a reduction toward the end of the study due to the return of devices (Supplementary Figure 2).

OPEN IN VIEWER

Fig. 2

Digital biomarker compliance. A) Actigraphy compliance during the intervention period. The number of compliant participants on each day is shown within the bars. B) Overall iPad Trial app compliance including data by all participants, whereas compliance removing data from participants who discontinued the study is shown in C).

Digital biomarkers detect mevidalen treatment effects

Overall, treatment with mevidalen resulted in a statistically significant reduction in actigraphy-measured daytime sleep from baseline to week 8 compared to placebo (Fig. 3A. 10 mg: p < 0.01, effect size = 0.24; 30 mg: p < 0.05, effect size = 0.23; and 75 mg: p < 0.001, effect size = 0.39), without having significant effects on night-time sleep in the same period (Fig. 3B. 10 mg: p = 0.135, effect size = 0.14; 30 mg: p = 0.902, effect size = 0.01; and 75 mg: p = 0.618, effect size = 0.05). Treatment with 75 mg mevidalen resulted in a statistically significant increase in the walking minutes per day from baseline to week 8 compared to placebo (Fig. 3C. p < 0.001, effect size = 0.39); however, increase in the walking minutes per day was not statistically significant with 10 mg or 30 mg mevidalen (Fig. 3C. 10 mg: p = 0.059, effect size = 0.17; and 30 mg: p = 0.282, effect size = 0.10). With regard to the iPad trial app, mevidalen treatment (10, 30, and 75 mg) did not increase the number of correct responses in the DSST (Fig. 3D. 10 mg: p = 0.694, effect size = 0.04; 30 mg: p = 0.527, effect size = 0.07; and 75 mg: p = 1.000, effect size = 0.00) or performance in the finger tapping test (Fig. 3F. 10 mg: p = 0.326, effect size = 0.12; 30 mg: p = 0.808, effect size = 0.03; and 75 mg: p = 0.690, effect size = 0.05) from baseline to week 12 compared to placebo. Treatment with 30 mg mevidalen statistically significantly increased SWM from baseline to week 12 compared to placebo (Fig. 3E. 30 mg: p = 0.003, effect size = 0.35), but there were no statistically significant changes with 10 mg or 75 mg mevidalen (Fig. 3E. 10 mg: p = 0.663, effect size = 0.05; and 75 mg: p = 0.462, effect size = 0.08).

OPEN IN VIEWER

Fig. 3

Key actigraphy (A–C) and trial app (D–F) outcomes. A) Actigraphy measurements detected a dose-dependent reduction in daytime sleep using the actigraphy watch during intervention (visit 9) returning to pre-intervention levels post-intervention. B) Nighttime sleep showed minimal changes. iPad trial assessments were performed throughout the study. C) A mevidalen-induced increase in a daily activity in walking minutes per day was observed during intervention (75 mg group). D) Mevidalen treatment showed minimal changes to DSST performance. E) A mevidalen-induced increase (30 mg group) was observed in SWM. F) Minimal treatment effects were observed in the finger tapping tests. Type 1 errors were not controlled for, and p-values were not adjusted. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001.

Digital biomarkers for continuous data capture

One potential advantage of utilizing digital biomarkers such as actigraphy in a clinical trial is that they offer continuous data monitoring and capture the period between study visits (actigraphy) or the entire study period (iPad trial app) rather than intermittent assessments at site visits. Continuous data capture for actigraphy allowed for both daily and hourly assessment of sleep and activity. A mevidalen-induced reduction was detected in daytime sleep from baseline throughout intervention as well as when hourly actigraphy-measured sleep was shown over a period of 24 h (when each actigraphy-measured sleep day was averaged from the full on-treatment period) (Fig. 4A and B, respectively). In the same respect, mevidalen-induced increases in daytime activity from baseline were observed throughout the intervention as well as when hourly actigraphy-measured activity was shown over a period of 24 h (when each actigraphy-measured activity day was averaged from the full on-treatment period) (Fig. 4C and D, respectively). The mevidalen-induced changes in daily daytime sleep (Fig. 4A) and daily daytime activity (Fig. 4C) returned to baseline levels in the post-intervention phase, whereas when plotting over a 24-h period, reduced daytime sleep (Fig. 4B) and increased daytime activity (Fig. 4D) can each clearly be observed in the treated groups beginning at 9am until 4pm.

OPEN IN VIEWER

Fig. 4

Digital biomarkers such as actigraphy allow for continuous data capture. A) Daily sleep. Reduction in total daytime sleep every day at all doses relative to placebo at steady state, with return to baseline post-intervention. B) Hourly sleep. Reduction in daytime sleep/inactivity from 9am to 3pm at all doses relative to placebo. C) Daily activity. Increase in average acceleration (m-g) every day at all doses relative to placebo at steady state, with return to baseline post-intervention. D) Hourly activity. Increase in average acceleration (m-g) from 9am to 7pm at all doses relative to placebo. Night in this study was defined as 7pm to 9am. X-axis break in A) and C) indicates the several weeks between intervention and post-intervention as per study schematic showing the PRESENCE trial period and the use of digital devices in Fig. 1.

Correlation between digital and clinical measures

Although digital biomarkers can detect treatment effects of mevidalen on sleep, activity, and digital cognitive assessments such as SWM, it is important to relate these findings to clinical assessments obtained at site visits in order to establish clinical relevancy of digital biomarker both at baseline (Fig. 5) and change from baseline (Supplementary Figure 3). At baseline, actigraphy-measured daytime sleep positively correlated with the ESS score (Fig. 5A, R = 0.19, p < 0.01), whereas actigraphy-measured daytime activity correlated with lower (and, therefore, better) scores on MDS-UPDRS Part II (Fig. 5B, R = –0.31, p < 0.001) which assesses the motor experiences of daily living. Better scores of SWM and DSST correlated with lower (less cognitively impaired) scores on the ADAS-Cog₁₃ scale (Fig. 5C, SWM: R = –0.54, p < 0.001; Fig. 5D, DSST: R = –0.52, p < 0.001). Change from baseline correlations revealed a weak but statistically significant relationship between actigraphy-measured daytime sleep and the ESS score (Supplementary Figure 3A, R = 0.2, p = 0.018), but non-significant relationship between daytime activity and MDS-UPDRS Part II Score (Supplementary Figure 3B, R = –0.13, p = 0.12), or SWM and DSST with ADAS-Cog₁₃ scores (Supplementary Figure 3C, SWM: R = 0.073, p = 0.46; Supplementary Figure 3D, DSST: R = 0.0093, p = 0.92).

OPEN IN VIEWER

Fig. 5

Clinical relevance of digital measures at baseline. Correlations between clinical assessments and Actigraphy (A and B) and iPad trial (C and D) assessments. A) Actigraphy-measured daytime sleep positively correlates with Epworth Sleepiness Scale score. B) Physical activity correlates with improvement on MDS-UPDRS Part II. C) Better scores of SWM correlate with lower ADAS-Cog₁₃. D) Better scores of DSST correlate with lower ADAS-Cog₁₃. ADAS-Cog₁₃, Alzheimer Disease Assessment Scale –Cognitive 13-Item Scale; DSST, digital symbol substitution; MDS, Movement Disorder Society–owned rating scales; SWM, spatial working memory; UPDRS, Unified Parkinson Disease Rating Scale.

DISCUSSION

The results of the digital assessments included in the PRESENCE trial indicate that the use of iPad apps and wearable devices is feasible and results in adequate compliance in participants with LBD. Digital metrics were sensitive to detecting treatment-related changes in cognition, motor activity and daytime sleep. Further supporting the symptomatic benefits of mevidalen that were assessed with standard outcomes, digital assessments may be a meaningful addition to traditional clinical assessments of activity and sleep.

The high rate of compliance to digital measures throughout this study indicates that the use of digital measures is feasible even in a cognitively and motorically impaired population, though compliance was less for the trial app than the actigraphy device. Regarding the trial app, compliance was similar across all tasks, likely given that the app was designed so that each task required completion before one could proceed to subsequent tasks, and the duration of tasks was limited to 5 min to minimize patient burden. Elements of the actigraphy device that may have increased compliance include that the device was largely passive, did not require charging and was both durable and water-resistant allowing for continuous wear during the 2-week period of use.

At baseline, scores of SWM and DSST correlated with ADAS-Cog₁₃ scores, actigraphy-measured daytime sleep correlated with the ESS score, and actigraphy-measured daytime activity correlated with MDS-UPDRS Part II scores. These baseline correlations between clinical measures of cognition, sleep and motor function and relevant digital biomarkers support the clinical validation of these digital measures. In contrast, change from baseline correlations between digital measures and clinical measures revealed mixed results. For example, we observed a non-significant relationship of SWM and DSST with ADAS-Cog₁₃ scores, a modest but statistically significant relationship between actigraphy-measured daytime sleep and ESS score, and a non-significant relationship between daytime activity and MDS-UPDRS Part II Score. Such discrepancy underscores the disparity between digital and clinical measurements in terms of assessment timing, cadence, sensitivity and that they likely measure slightly different disease status. The daytime activity used for change correlation was summarized from two weeks of continuous actigraphy data, while the MDS-UPDRS Part II Score was from one in-clinic. In addition, continuous activity from actigraphy sensors captures different aspects of physiological state from physician rated activity of daily living associated with motor impairment. For broader adoption of digital measures in clinical development or practice, it will be valuable for the community to align on standardized ways to compare digital and clinical measures.

The primary endpoint of mevidalen treatment in the PRESENCE trial was improvement in cognition assessed by the CDR-CoA, which was not met, nor was a secondary cognitive endpoint of improvement in ADAS-Cog₁₃ from baseline [17]. Supportive of the apparent lack of mevidalen-induced improvements in cognition were the limited treatment-related improvements in digital measures of cognition; however, treatment with 30 mg mevidalen did result in a statistically significant improvement in the SWM from baseline to Week 12 compared to placebo and the improvement during the last 6 weeks of treatment averages 50% from baseline. Although preclinical data did indicate that mevidalen treatment had the potential for improvement in SWM [15], the lack of cognitive benefit observed in clinical measures prompts caution in the interpretation of the improvement in SWM here; furthermore, given that analyses were exploratory and no correction for multiple comparisons was made, some results (e.g., improvement in SWM) may be spurious. Indeed, the improvement in SWM in the 30 mg mevidalen group may be partially driven by a 3-point lower score in this group at baseline.

Using traditional clinical assessments, mevidalen treatment (at the 30 and 75 mg doses) significantly improved MDS-UPDRS Part II scores and therefore motor function [17]. Mevidalen treatment (at the 75 mg dose) significantly increased actigraphy-measured daytime activity from baseline to week 8 compared to placebo, indicating that the actigraphy-measured increases in activity can predict the treatment-induced motor improvements. There was also a lack of noted mevidalen effect on finger tapping despite published evidence of its sensitivity [28]; however, the test was modified in this trial to use an iPad, and not a smartphone that was used in validating this task in other studies. Despite the inclusion of multiple daily tests and two weeks of pre-intervention data collection in an effort to overcome the potential learning effect, there still appears to be a continued learning effect for the duration of the trial. Indeed, it was observed in an observational study using mPower iPhone app that patients with PD required more finger tapping tests to reach a steady-state performance than healthy volunteers [28]. The participants with LBD in the current study were at an advanced stage and likely had more performance variability therefore two weeks of baseline period may not be sufficient to reach a steady state.

In the PRESENCE trial, potentially clinically-meaningful improvements in the ESS were observed with mevidalen treatment (75 mg mevidalen dose) [17], paralleled by statistically significant reductions in actigraphy-measured daytime sleep from baseline to week 8 compared to placebo at all doses of mevidalen (10, 30, and 75 mg). This reduction indicates that the actigraphy-measured reduction in daytime sleep can not only predict treatment-induced increases in wakefulness but can also detect changes in wakefulness earlier and with a smaller sample size than its clinical assessment counterpart. Although the actigraphy-measured sleep appeared to be more sensitive than clinical measurements, it is important to note that sleep was measured using an actigraphy-based algorithm [22, 33]; in future studies, the use of devices that measure other physiological measures related to sleep, such as heart rate variability, may be helpful to further differentiate and improve the accuracy of sleep detection.

Taken together, the results suggest that digital biomarkers may allow the identification of domain-specific treatment effects to inform future development using smaller sample sizes and over shorter durations than traditional clinical measures; however, there were several limitations associated with this study. First, the data here do not represent a full data set and rely on convenience sampling regarding participants and timepoints utilized. The population examined was limited by geographical location for actigraphy availability and the optional aspect of the iPad trial app sub-study, which may result in bias and limit the generalization of these findings to the broader PRESENCE trial cohort. While no clear differences in demographic or disease specific measures were observed between the digital biomarker cohort relative to the entire PRESENCE trial cohort nor across treatment arms in the digital biomarker cohort, it remains possible that unmeasured factors may have further biased these findings. Second, there were a disproportionate number of male versus female participants included in the digital component of the trial, which, although representative of the parent trial population and not unusual in clinical research [34], may also highlight a sex difference in the ease and willingness to use digital devices. In addition, the trial participants also lacked diversity in race [17]; additionally, given the English-speaking requirement for actigraphy participants which would result in a less-diverse cohort, these results would need to be further validated in patients from a more diverse background. Third, correlations between digital measurements and clinical measurements are often weak, and, although clinical validation is critical, the multitude of variables that each measure can have on the accuracy of results of clinical scales and associated correlations with digital measurements must be recognized (e.g., subjective versus objective assessment, real time versus recall, continuous versus sporadic assessments as well as the known limitations of current clinical measurements). In addition, despite the correlation of digital and clinical measurements related to activity and sleep shown here, further research would be required to determine if the changes observed are clinically meaningful to participants. To address these evolving challenges, numerous groups and efforts, such as the Parkinson’s Disease Digital Biomarker DREAM Challenge [35], the Critical Path for Parkinson’s Consortium, the Michael J Fox Foundation, the Digital Medical society, the Clinical Trials Transformation Initiative [36] and the open wearable initiatives, are collaboratively evaluating algorithms and developing benchmark datasets to facilitate wearable device and sensor signal verification, digital biomarker development, and clinical validation [37].

Finally, given the ongoing coronavirus pandemic and associated travel limitations imposed around the world, the use of telemedicine and digital measures has dramatically increased and has been suggested as a rare silver lining of the pandemic [38–40]. Over the course of the pandemic, digital devices have allowed for the continuity of care and clinical research and have perhaps accelerated the adoption of decentralized clinical trial (DCT) approaches. The use of decentralized intervention strategies using high-frequency, unsupervised, home-based assessments may shape the future of clinical trials, and digital devices play an integral role in these innovative clinical trials [39]. DCT and the associated use of digital devices may also provide increased diversity, inclusion, and representation in terms of sex, race, and socioeconomics in clinical trial populations.

Clinical measures for PD have largely remained unchanged and are limited by their episodic and subjective nature. These traditional assessments may not accurately capture symptoms that occur between visits, or the regular fluctuations associated with disease. Digital measures are objective, often passive, or minimally burdensome, and may have greater sensitivity than traditional measures and allow for more frequent assessments; therefore, they may allow for detection of drug effect in small sample sizes to inform drug development. Although the digital biomarker results in this Phase 2 study should be validated with an independent sample of patients either through another Phase 2 or a Phase 3 study, these results highlight the unique potential of digital devices to detect treatment effects over shorter durations and smaller sample sizes than traditional clinical outcomes. These results indicate that digital biomarkers offer great promise to empower smaller, more efficient, and better patient-centric clinical trial designs with a view of overall improved patient care.