Can smartphone motor tasks track progression in iRBD and early Parkinson's disease?

Testing whether phone tasks track change over time

The study asked a question that cross-sectional designs cannot answer: do repeated smartphone motor tasks detect worsening over 24 months in people with isolated REM sleep behavior disorder (iRBD) and early Parkinson's disease, compared with healthy controls? iRBD is treated here as a prodromal stage of synucleinopathy, so a measure that changes in iRBD would be a candidate marker for trials run before overt parkinsonism appears.

Twenty-six controls, 22 patients with iRBD, and 23 with early-stage, treated Parkinson's disease contributed data. Participants used a midrange Android phone with a preinstalled app and were prompted every 14 days to complete an 8–10-minute set of speech, psychomotor, hand tremor, and U-turn tasks, seated, with the phone on a table and low ambient noise. The analysis covered 4,633 sessions, with a mean session duration of 8.65 minutes.

Progression was estimated with linear mixed-effects models that compared the rate of change between groups over the study period, using group-by-month interaction terms. Because the study was exploratory with respect to progression sensitivity, no correction for multiple comparisons was applied and significance was set at p < 0.05. That choice matters when reading the small number of significant progression findings.

Only three features separated progression from controls

Over the full 24 months, the models found significant slope differences between controls and the iRBD group in exactly two features: monopitch during reading and mean turn speed with the non-dominant hand, both p < 0.05. Monopitch refers to reduced pitch variation in speech, one of the perceptual hallmarks of hypokinetic dysarthria that the authors selected features to capture.

Between controls and the Parkinson's disease group, the only progression difference was in time to first voice break during sustained phonation, p < 0.05. This measure shortens as the interval before the voice first breaks during a prolonged vowel becomes shorter; the authors note it has not been reported as a progression marker before and hypothesize it may reflect abnormal muscle tone or vocal fold hyper-adduction, possibly a non-dopaminergic component of progression.

The contrast with the cross-sectional results is the paper's central point. Several speech, psychomotor, and U-turn measures showed adequate sensitivity for distinguishing iRBD or Parkinson's disease from controls at a given time, yet almost none of those same measures detected change over two years. The authors state this supports the view that cross-sectional performance does not reliably predict the ability to track progression.

Cross-sectional separation did not imply longitudinal sensitivity

For controls versus iRBD, significant group differences appeared in monoloudness, mean reaction time in the non-dominant hand, tapping frequency in the non-dominant hand, mean word period in both hands, and tremor energy in both hands. For controls versus Parkinson's disease, differences appeared in monopitch, monoloudness, mean word period in both hands, tremor energy in both hands, and mean turn speed in both hands.

These are group-level differences averaged across the study, not slopes. A feature can sit at a different level between groups while changing at the same rate, which is precisely what most of these measures did. The paper's summary table makes the split explicit: monopitch separated Parkinson's disease from controls but not iRBD from controls, while it was the only speech feature to track progression in iRBD and did not do so in Parkinson's disease.

The authors attribute the absence of monopitch progression in Parkinson's disease to a ceiling effect, suggesting the measure may be most applicable in prodromal stages. That is an author interpretation rather than a tested mechanism, and it bounds how the iRBD result should be read: the same feature is not interchangeable across the two patient groups.

Practice effects persisted for up to a year

A significant learning effect in the first year affected diadochokinetic irregularity, diadochokinetic rate, mean reaction time in the dominant hand, tapping frequency in both hands, and hand tremor energy in both hands. In other words, performance improved with repetition on tasks that involve simple, repeated motor actions, which can mask or mimic the slow decline a progression marker is meant to capture.

Some learning slopes were steeper in patients than in controls. During the first year, mean reaction time in the non-dominant hand improved more in both iRBD and Parkinson's disease than in controls; tapping frequency in the non-dominant hand improved more in iRBD; and tremor energy in the non-dominant hand improved more in Parkinson's disease. Across the whole study, iRBD showed a greater learning slope than controls in mean word period in the dominant hand.

The authors conclude that task and feature selection should depend on study design, and that learning effects can likely be controlled by choosing the measurement period. For a reader planning a remote trial, this means a feature that looks sensitive in a single visit may be unusable for longitudinal endpoints unless the early practice phase is excluded or modeled.

Subthreshold parkinsonism sharpened the iRBD signal

Fourteen of the 22 iRBD participants (64%) had subthreshold parkinsonism at baseline, defined as an MDS-UPDRS part III score greater than 6 after excluding action and postural tremor. Within iRBD, monopitch progressed significantly more in this subgroup than in iRBD without subthreshold parkinsonism (p < 0.01), while no such effect appeared for U-turn. The authors caution that the subgroup with subthreshold parkinsonism had better average speech performance at baseline, so the comparison is not clean.

Clonazepam treatment did not significantly affect progression of either monopitch or U-turn in iRBD. Stability was high over the full 24 months for all features, with intraclass correlation coefficients above 0.75. Short-term stability between the first and second month was weaker for some features: monoloudness, diadochokinetic irregularity, and jitter fell at or below 0.50, while articulation rate, voice onset time, and dominant-hand mean reaction time fell between 0.50 and 0.75.

Age influenced all psychomotor features, mean turn speed in both hands, articulation rate, and diadochokinetic rate. For the primary progression outcomes, the only significant group-dependent age effect was in time to first voice break for both patient groups. These dependencies mean a feature's raw slope is not the whole story; the models adjusted for age as a global fixed effect.

Phone slopes correlated weakly with clinical scales

Almost all features showed weak correlations with the total MDS-UPDRS part III score in both patient groups, with absolute Spearman coefficients below 0.40. The exception was mean word period in the dominant hand for Parkinson's disease participants, at a coefficient of 0.55 (p < 0.01).

Correlations with levodopa equivalent dose in the Parkinson's disease group were also weak, with absolute coefficients below 0.35, except for the relationship between decreased time to first voice break and increased levodopa equivalent dose, at −0.45 (p < 0.05). The authors note that Parkinson's disease participants received gradually increasing dopaminergic doses during the study, which could have offset progression of some markers, and that this design does not match a neuroprotective trial in which treatment would typically be withheld or held stable.

This is a scope condition, not a footnote. The Parkinson's disease progression result comes from a treated cohort whose medication roughly doubled in levodopa equivalent dose from baseline to the 24-month retest, so the single significant finding there should be read against that background.

Choose features by stage and by study duration

The paper's practical priority follows from the split between cross-sectional and longitudinal performance. For a longitudinal trial in prodromal iRBD, monopitch during standardized reading and non-dominant-hand turn speed are the two features with demonstrated progression sensitivity over 24 months, and monopitch appears stronger in iRBD participants closer to phenoconversion. For early treated Parkinson's disease, time to first voice break is the only feature that separated progression from controls, and it is a new candidate rather than an established one.

For short-term evaluation, the cross-sectional separators are the relevant set, but the same features should not be assumed to work as progression endpoints. The authors' own recommendation is to integrate multiple motor domains and to add more naturalistic, passive monitoring of voice and gait, since active repetitive tasks carry learning effects that persisted for up to a year.

The most decision-relevant reading is therefore a matching rule rather than a single best measure: pick monopitch or non-dominant turn speed for prodromal longitudinal work, treat time to first voice break as a Parkinson's disease candidate needing confirmation, and avoid tapping, diadochokinesis, and hand tremor features as progression endpoints unless the first-year learning phase is handled explicitly.

Sources used in this answer