Reliability of some tremor measurement outcome variables in field testing situations

Many summary measures of data obtained from tremor measurement procedures are commonly reported. The reliability of many of these summary measures of tremor measurements made in field testing situations is unknown. The purpose of the present investigation was to assess the reliability of a number of summary measures produced by the software of a widely used, commercially available tremor measurement instrument using data collected in three field epidemiologic studies.

Tremor data were obtained from 689 participants in 3 previously conducted studies of groups exposed to elemental mercury or arsenic. A widely used, commercially available tremor measurement instrument was used. Two-axis accelerometer measurements were obtained on 2 or more trials from each hand for each participant. Estimates of trial-to-trial and internal consistency reliability were calculated for 5 summary measures calculated by instrument manufacturer's software and 5 additional summary measures calculated from data output by the software.

An RMS acceleration measure had the highest reliability in all 3 studies. The average over 4 trials of RMS acceleration and its logarithm had high reliability (>0.9). Recalculation of a tremor summary index and a harmonicity index as suggested by Edwards and Beuter (1999) resulted in measures with higher reliability and better distributional shape than the corresponding measures provided by the instrument manufacturer's software. The results in all three studies were similar.

For the tremor measurement instrument and testing procedure that we employed, we recommend using the common logarithm of the RMS accelerations and recalculated tremor index as summary measures. We also recommend employing multiple trials of each type (e.g., with each hand) and averaging summary measures from those trials to derive outcome measures of tremor for use in epidemiologic studies. We recommend at least 2 trials for RMS acceleration measures and more for less reliable measures, particularly for designs employing repeated measurements of individuals. Summary measures averaged over at least 4 trials for mean frequency, dispersion of frequency, and power in the 3-6.5 and 6.6-10 Hz frequency ranges have sufficiently high reliability for use in epidemiologic studies.