Smartphone motor testing to distinguish idiopathic REM sleep behavior disorder, controls, and PD

Abstract

Objective: We sought to identify motor features that would allow the delineation of individuals with sleep study-confirmed idiopathic REM sleep behavior disorder (iRBD) from controls and Parkinson disease (PD) using a customized smartphone application.

Methods: A total of 334 PD, 104 iRBD, and 84 control participants performed 7 tasks to evaluate voice, balance, gait, finger tapping, reaction time, rest tremor, and postural tremor. Smartphone recordings were collected both in clinic and at home under noncontrolled conditions over several days. All participants underwent detailed parallel in-clinic assessments. Using only the smartphone sensor recordings, we sought to (1) discriminate whether the participant had iRBD or PD and (2) identify which of the above 7 motor tasks were most salient in distinguishing groups.

Results: Statistically significant differences based on these 7 tasks were observed between the 3 groups. For the 3 pairwise discriminatory comparisons, (1) controls vs iRBD, (2) controls vs PD, and (3) iRBD vs PD, the mean sensitivity and specificity values ranged from 84.6% to 91.9%. Postural tremor, rest tremor, and voice were the most discriminatory tasks overall, whereas the reaction time was least discriminatory.

Conclusions: Prodromal forms of PD include the sleep disorder iRBD, where subtle motor impairment can be detected using clinician-based rating scales (e.g., Unified Parkinson's Disease Rating Scale), which may lack the sensitivity to detect and track granular change. Consumer grade smartphones can be used to accurately separate not only iRBD from controls but also iRBD from PD participants, providing a growing consensus for the utility of digital biomarkers in early and prodromal PD.

Figure 1. Schematic diagram illustrating the major steps involved in data acquisition of 7 smartphone tasks assessing voice, balance, gait, finger tapping, reaction time, rest tremor, and postural tremor

For the voice task, using the inbuilt microphone, we recorded the sustained phonation “aaah”; the participants were instructed to “Hold the phone to your ear, take a deep breath, and say “aaah” at a comfortable and steady, tone and level, for as long as you can.” For the balance task, using the smartphone inertial measurement units (IMUs), we collected triaxial accelerometer sensor data; the participants were instructed to “Stand up straight and place the phone in your pocket. When the buzzer vibrates, stay standing until the buzzer vibrates again.” For the gait task, using the smartphone IMUs, we collected triaxial accelerometer sensor data; the participants were instructed to “Stand up and place the phone in your pocket. When the buzzer vibrates, walk forward 20 yards. Then, stop, turn around, and walk back again.” For the finger tapping task, using the touch screen sensors and timer, we recorded time and location (x-y screen coordinate position) of finger touch; the participants were instructed to “Tap the buttons below with the index and middle fingers of 1 hand alternatively, in a regular rhythm.” For the reaction time task, using the touch screen sensors and timer, we recorded the time of stimulus onset (appearance/disappearance of a screen button) and response (press/release the screen button) along with location (x-y screen coordinate position) of finger touch; the participants were instructed to “Press the screen button below as soon as it appears; release as soon as it disappears.” For the rest tremor task, using the smartphone IMUs, we collected triaxial accelerometer sensor data; the participants were instructed to “Sit upright, hold the phone in your tremor dominant hand and rest it lightly in your lap, and close your eyes and count backward from 100.” For the postural tremor task, using the smartphone IMUs, we collected triaxial accelerometer sensor data; the participants were instructed to “Sit upright and hold the phone in your tremor dominant hand, with the arm outstretched in front of you.”

Figure 2. Schematic diagram illustrating the major steps involved in the analysis of smartphone sensor data from 7 smartphone tasks assessing voice, balance, gait, finger tapping, reaction time, rest tremor, and postural tremor in the smartphone app used in this study

GSO = Gram-Schmidt orthogonalization; LASSO = least absolute shrinkage and selection operator; LLBFS = local learning-based feature selection; mRMR = minimum redundancy maximum relevance; PD = Parkinson disease; iRBD = idiopathic REM sleep behavior disorder; VFER = vocal fold excitation ratio.

Figure 3. Discrimination accuracies as a function of the number of salient features used in the machine learning discrimination analysis, for the 3 pairwise comparisons: controls vs Parkinson disease (PD) (panels A and B), controls vs idiopathic REM sleep behavior disorder (iRBD) (panels C and D), and iRBD vs PD (panels E and F)

The above accuracies were computed using all available recordings from the 3 clinical groups, using all 998 features computed from the 7 tasks, using 10-fold cross-validation (10 repetitions). The rankings of the most salient features were obtained using a majority voting scheme (using 5 feature selection algorithms). The feature rankings were obtained separately for each of the above 3 pairwise comparisons. Features were added into the machine learning classifier (random forest) in increments of 2 (starting from 2 and going up to 30), whereby higher ranked features were added first. The whole process of training and validation was repeated each time 2 new features were included. Sensitivity and specificity values (in %) were reported as mean (denoted by gray circles) and SD (vertical bars).

Figure 4. Graphical illustration of the most salient discriminatory tasks for the 3 pairwise comparisons: controls vs Parkinson disease (PD) (charts A–C, top horizontal panel), controls vs idiopathic REM sleep behavior disorder (iRBD) (charts D–F, middle horizontal panel), and iRBD vs PD (charts G–I, lower horizontal panel)

The above pie charts were generated using the 30 most salient features computed from the smartphone sensor recordings. The rankings of the most discriminatory features were obtained using a majority voting scheme (using 5 feature selection algorithms). The feature rankings were obtained separately for each of the above 9 pairwise comparisons. For a given pairwise comparison, charts were generated by computing the percentage of features that were selected for each of the 7 smartphone tasks. A larger pie segment corresponds to smartphone tasks that were identified as being relatively more discriminatory for the pairwise comparison under consideration. For each comparison, task rankings were computed using all available recordings (denoted by All, leftmost vertical panel), along with a subgroup analysis performed using pooling of observations for females and males, denoted by Female (middle vertical panel) and Male (rightmost vertical panel), respectively.