Modernising tactile acuity assessment; clinimetrics of semi-automated tests and effects of age, sex and anthropometry on performance

Design and procedure

A test-retest study investigated the inter-rater reliability and measurement error of the iTAD and TPDT. Three assessors each performed both assessments in a private, quiet, room. The order of assessments was randomised between participants, but consistent between assessors for each participant. The order of assessors was randomised between participants. Participants received a 10-minute break between assessors and were blinded to their scores. Assessors were blinded to all iTAD scores and TPDT scores of the other assessors. The anatomical location of both assessments was standardised but independently determined by each assessor.

Participants

Using a convenience sample, individuals without current pain and neurological symptoms as well as without a history of persistent (i.e., >3 months) pain and neurological symptoms in the past five years were recruited from the general public.

Assessors

Three final year physiotherapy Masters students each performed iTAD and TPDT assessments after receiving approximately one hour of training for each procedure. Test instructions and performance were standardised using a testing protocol. Training included approximately 20–30 min of instructions on the testing protocol and about 30–40 min of practicing the protocol. Each assessor practised each protocol five times and underwent each assessment at least once.

TPDT assessment

A digital caliper (Renegade industrial, carbon fibre, RCFVC150) was used to establish TPDT, utilizing the two arms and pressure of its own weight as tactile stimuli. Contact area of the tip of each arm was approximately 0.25 mm × 0.5 mm and stimulus duration <1 s. During TPDT assessment, participants were seated with their forehead resting on a table in front of them. On the dominant side, TPDT was measured in a cranio-caudal direction with the caudal arm of the caliper stationary at 15 mm lateral to the spinous process of C7.

Similar to a previously published procedure (Luedtke et al., 2018), a two-alternative forced-choice (one or two points) staircase method was used, alternating two ascending and two descending runs (see Fig. 1). The caliper distance started at 15 mm, increasing with five mm steps. Steps were reduced to two mm for the subsequent runs. To avoid guessing, three consecutive reports of either one or two points indicated a reversal. A 10 mm step was added in the direction of the completed run, before running the reversed direction. The TPDT was calculated by averaging the scores of the four reversals, expressed in millimetres, with larger distances indicating poorer tactile acuity.

iTAD assessment

iTAD prototype.

The iTAD consists of a wearable neoprene collar containing twelve vibrotactile stimulators (∼200 Hz with 0.75 g vibration amplitude), arranged in three rows of four (see Fig. 2). A wirelessly connected tablet operates the stimulators and records user’s responses. After fitting and familiarisation, two tactile acuity tests were performed: the localisation test, which measures the ability to localise the vibrations (one second stimulus duration), and the orientation test, which measures the ability to determine the orientation of two successive adjacent vibrations relative to each other (0.7 s stimulus duration each). Accuracy scores (i.e., percentage correct) for each test, and the overall score (i.e., mean of both test), were calculated with higher scores indicating better tactile acuity. For a full description of the iTAD prototype and its assessment procedures, see elsewhere (Olthof et al., 2021).

Statistical analyses

Inter-rater reliability was assessed by calculating the intraclass correlation coefficient, model 2,1 (ICC_(2,1)) (i.e., two-way random, absolute agreement, single measures). Values were interpreted as poor (<0.5), moderate (0.5–0.74), good (0.75–0.89) or excellent (≥0.9) (Portney & Watkins, 2009).

The standard error of measurement (SEM) was calculated using variance components from the ANOVA table ($\sqrt{\left({\sigma }_{\mathrm{observer}}^2+{\sigma }_{\mathrm{residual}}^2\right)}$) to assess measurement error (De Vet et al., 2006). As TPDT and iTAD use different metric units, the SEM was additionally converted to the coefficient of variation (CoV) (i.e., SEM as a percentage of mean score) allowing direct comparison of measurement error (Hopkins, 2000). Since both <10% and <20% have been suggested as a good CoV (Atkinson & Nevill, 1998; Quan & Shih, 1996), results were categorised into: <10%, 10–20% and >20%.

To assist clinical interpretation of the SEM, the smallest detectable change with a 95% confidence interval (SDC_95) was calculated (1.96*$\sqrt{2}$*SEM) (De Vet et al., 2006). Additionally, SDC values with smaller confidence intervals (SDC_80, SDC_85 and SDC_90) were computed (replacing 1.96 with respective z-values).

Sample size calculation

To detect a hypothesized ICC_(2,1) of 0.7 with three repeated measurements (recommended to optimise sample size (Walter, Eliasziw & Donner, 1998)), while α = 0.05, β = 0.2 and ICC_(2,1) = 0.5 for the null hypothesis, a minimum of 40 participants is required (Walter, Eliasziw & Donner, 1998). Therefore, we aimed to recruit 40 participants.

Design and procedure

Using a cross-sectional design, the internal consistencies of the localisation and orientation accuracy score were investigated. Additionally, floor and ceiling effects of all iTAD scores were assessed, as well as their relationship with age, sex, BMI and neck surface area. In one session, participants performed both iTAD tests after age, sex, and anthropometric measures (see ‘Anthropometric measurements’) were recorded. Measures were taken by a single assessor in a private, quiet room. The assessor had several hours of prior experience performing iTAD assessments.

Participants

Recruitment and selection criteria were identical to experiment one. However, participants represent a different cohort without overlap.

iTAD tests

For procedure of the iTAD tests, see ‘iTAD assessment’.

Anthropometric measurements

After weight (kg) and height (m) were recorded, BMI was calculated (weight/height²). Using a tape measure, the distance from the caudal aspect of the external occipital protuberance to the spinous process of C7 was measured to quantify neck length (cm). Neck circumference (cm) was measured at half-way of the neck length measurement placing the tape measure horizontally around the neck. Using these measurements, the posterior neck surface area was estimated (neck length*(neck circumference/2)).

Statistical analysis

Internal consistency was assessed by calculating the inter-relatedness of accuracy scores between series within each test, using ICC model 2,k (i.e., two-way random, absolute agreement, average measures). The ICC_(2,k) was chosen over the Cronbach’s alpha, the equivalent of the ICC_(3,k) (i.e., two-way random, consistency, average measures), to include absolute differences between series. Although various cut-offs are proposed, most recommend 0.7−0.9 for high internal consistency (Taber, 2018).

For the accuracy scores, floor and ceiling effects were considered present if >15% of the participant scored within either the highest or lowest 20% of the scale (Terwee et al., 2007). However, such assessment would not be adequate for response times or rate correct scores, as their scales have no limit on one end and scores are (near) impossible at the other. Therefore, floor and ceiling effects for all iTAD scores were assessed by calculating z-scores for the skewness of their distribution (i.e., the skew value divided by its standard error) (Kim, 2013; Ho & Yu, 2015). Floor and ceiling effects may be present with a z-score >±1.96 in small (n<50) or >±3.29 in medium (50<n<300) sized samples (Kim, 2013; Ho & Yu, 2015).

In order to study the direct multivariate relationships of age, sex, BMI and neck surface area with the localisation and orientation accuracy score, multiple linear regressions (enter models) were performed. Furthermore, to estimate the potential indirect effects of age and sex through BMI and/or neck surface area, sequential mediation analyses were performed using the SPSS extension PROCESS (model 6; 5000 bootstrapped samples) as a secondary analysis. Mediation analyses for all other iTAD scores were performed as exploratory analyses. Regression models are expressed in (adjusted) explained variance (R²_adjusted). For all relationships, both the mean unstandardized regression coefficient (b) and the semi-partial correlation (sr) are provided. Indirect effects are expressed in percentage mediation (P_m).

Sample size

To examine internal consistency, a minimum of 100 participants is recommended (Terwee et al., 2007). To explore potential floor and ceiling effects, a minimum of 50 participants is recommended (Terwee et al., 2007). For the multiple linear regressions, 100 participants were needed to find a medium sized (f² = 0.15) prediction model using four predictors with a Bonferroni corrected p-value of 0.025 and 80% power. Taken together, the sample size was set for 100 participants, with 10 participants of both sexes in each of five age brackets (18–30, 31–40, 41–50, 51–60 and 61–70).

Internal consistency

The ICC_(2,6)(95% CI) for the localisation accuracy score was 0.84 (0.79−0.89) and the ICC_(2,4) (95% CI) for the orientation accuracy score was 0.86 (0.81−0.90).

Floor and ceiling effects

None of the accuracy scores had >5% of participants scoring in either the highest or lowest 20%. Only localisation response time had a skewness z-score >±1.96 (z =+2.73), which was still <±3.29.

Relationship with age, sex and anthropometry

Multiple linear regression analyses showed significant prediction models for both the localisation accuracy score (F(4,95) =4.92, p < 0.01, R²_adjusted = 0.14) and orientation accuracy score (F(4,95) =5.82, p < 0.01, R²_adjusted = 0.16).

For the localisation accuracy score, both age (b =−0.20, sr =−0.20, p = 0.03) and sex (b = 8.04, sr =0.25, p = 0.01) had a significant contribution to the model, whereas BMI (p = 0.11) and neck surface area (p = 0.89) did not. This indicates that, on average, men scored 8.04% higher than women, and that scores decreased by 0.20% for each year of age.

For the orientation accuracy score, both age (b =−0.39, sr =−0.36, p = 0.00) and sex (b = 6.69, sr =0.19, p = 0.04) contributed significantly to the model, whereas BMI (p = 0.71) and neck surface area (p = 0.94) did not. This indicates that, on average, men scored 6.69% higher than women, and that scores decreased by 0.39% for each year of age.

The mediation analyses indicated several significant relationships between demographic and anthropometric variables (see Fig. 3). However, for the localisation accuracy score, the total indirect effects of age (P_m = 0.16, p > 0.05), and sex (P_m = 0.06, p > 0.05), through BMI and neck surface area were non-significant. Similarly, the total indirect effects of age (P_m = 0.02, p > 0.05), and sex (P_m = 0.04, p > 0.05) were also non-significant for the orientation accuracy score. Additionally, all individual indirect effects were non-significant for both tests. This indicates that BMI and/or neck surface area did not significantly mediate the effects of age or sex for either accuracy score. Scatterplots of localisation and orientation accuracy scores as a function of age and sex are displayed in Fig. 4.

Exploratory mediation analysis of the other iTAD scores showed similar patterns, other than sex not significantly predicting the localisation or orientation response time and rate correct score. Additionally, rate correct scores were more strongly predicted by age. Figures of all mediation analyses can be found in Supplemental files (Figs. S1–S3). Scatterplots of all iTAD scores as a function of age and sex can be also be found in Supplement files (Figs. S4, S5).