some of the subjects repeated the occasion, or that the
subjects missed some occasions.
By examining the learning progress during the train-
ing period, the results show that the weighted linear
kappa value has a high median value throughout all
occasions, and the baseline value was kappa = 0.84
(> 0.8; considered very good (30)). However, there
is a distinct improvement up until the third training
occasion (kappa = 0.92) (Fig. 5). The kappa value
then stabilizes over the rest of the period; occasion 11
median kappa = 0.96, and continues to be high during
the 2 last occasions; occasion 17 median kappa = 0.96
and occasion 18 median kappa =0.96. The improve-
ments between baseline (occasion 1) and the chosen
follow-ups (occasions 11, 17 and 18) were significant
(p < 0.001); baseline compared with the 1st day of week
2, 2 nd follow-up week 4 and 3 rd follow-up week 5. The
agreement between actual stimulation and interpreta-
tion of stimulation (learning curve) peaked at occasion
12 (median kappa = 0.98). Outliers presented in Fig. 5
were unique individuals at every occasion of different
age and sex, and did not follow a pattern that could be
used for analysis.
The subjects’ ability in distinguishing which finger
was stimulated is shown in Fig. 6. It was easiest to dis-
tinguish the middle finger, where 95% of the answers
were correct. Hardest to distinguish were the ring and
little finger (84% correct answers), and most errors
occurred when the stimulus was on the little finger and
the response was the ring finger (16%).
Touch on the forearm can be associated with specific fingers
213
Fig. 6. The confusion matrix shows correct answers (in %). It was
easiest to distinguish the area for the middle finger and it was as easy
to interpret stimulation on a predefined area for the little finger as for
the ring finger. The sum of the numbers shown in the matrix does not
add up to exactly 100% due to rounding.
Comparing men and women, the 2-tailed Mann–
Whitney U test showed there was no statistically sig-
nificant difference in kappa value between the sexes.
Of those participating in the study only 6 were men and
25 were women. The same statistical test also showed
that age did not have any influence on learning, when
comparing the kappa values of the different age groups.
Among the comments from the subjects a different
sensation was described in the predefined area that the
subject should associate with the middle finger. The
sensation was reported to be perceived as tingling or
as a stronger stimulation compared with stimulations
of the other areas. Some subjects also reported minor
differences in the sound from the servo motors during
the stimulation.
DISCUSSION
Fig. 5. The box plot shows improvement (in median kappa values) in
learning during the 18 occasions, with 95% confidence interval (95% CI).
The learning and evaluation was completed twice a day during occasion
1–10, once a day during occasion 11–15 was done once a day and once
per week during occasion 16–18. The underlined occasions show the
follow-ups. The improvement was statistically significant (***p < 0.001)
between the baseline and the 3 follow-ups.
This study shows that it is possible to induce an as-
sociation between stimuli on the skin of the forearm
with specific fingers following a structured training
programme and that the association remain after 2
weeks. The results also show that it is easy to learn to
interpret the stimuli on the skin of the forearm, and al-
ready after 3 training occasions the agreement between
the actual stimuli and the response can be considered
very good (30). The excellent agreement remains after
1 week of no training and still after 2 weeks after the
end of the training programme. The fast learning that
is shown in our group of 31 subjects is comparable
with results presented by Chai et al. (27) who reported
a “3-day-effect” in their study of 11 subjects during
7 consecutive days. The subjects in our study had a
J Rehabil Med 51, 2019