Content uploaded by Cas Smits
Author content
All content in this area was uploaded by Cas Smits on Jul 16, 2019
Content may be subject to copyright.
Copyright © 2017 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
<zdoi; 10.1097/AUD.0000000000000522>
0196/0202/17/XXXX-00/0 • Ear & Hearing • Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved • Printed in the U.S.A.
1
Objectives: This study determined the effect of hearing loss and English-
speaking competency on the South African English digits-in-noise hear-
ing test to evaluate its suitability for use across native (N) and non-native
(NN) speakers.
Design: A prospective cross-sectional cohort study of N and NN English
adults with and without sensorineural hearing loss compared pure-
tone air conduction thresholds to the speech reception threshold (SRT)
recorded with the smartphone digits-in-noise hearing test. A rating scale
was used for NN English listeners’ self-reported competence in speak-
ing English. This study consisted of 454 adult listeners (164 male, 290
female; range 16 to 90 years), of whom 337 listeners had a best ear four-
frequency pure-tone average (4FPTA; 0.5, 1, 2, and 4 kHz) of ≤25 dB HL.
Results: A linear regression model identified three predictors of the
digits-in-noise SRT, namely, 4FPTA, age, and self-reported English-
speaking competence. The NN group with poor self-reported English-
speaking competence (≤5/10) performed significantly (p < 0.01) poorer
than the N and NN (≥6/10) groups on the digits-in-noise test. Screening
characteristics of the test improved with separate cutoff values depend-
ing on English-speaking competence for the N and NN groups (≥6/10)
and NN group alone (≤5/10). Logistic regression models, which include
age in the analysis, showed a further improvement in sensitivity and
specificity for both groups (area under the receiver operating character-
istic curve, 0.962 and 0.903, respectively).
Conclusions: Self-reported English-speaking competence had a sig-
nificant influence on the SRT obtained with the smartphone digits- in-
noise test. A logistic regression approach considering SRT, self-reported
English-speaking competence, and age as predictors of best ear 4FPTA
>25 dB HL showed that the test can be used as an accurate hearing
screening tool for N and NN English speakers. The smartphone digits-
in-noise test, therefore, allows testing in a multilingual population famil-
iar with English digits using dynamic cutoff values that can be chosen
according to self-reported English-speaking competence and age.
Key words: Digits-in-noise, Hearing loss, Hearing screening, Hearing
test, Smartphone, Speech-in-noise.
(Ear & Hearing 2017;XX;00–00)
INTRODUCTION
An important part of maintaining health and well-being for
older adults is to screen for and treat hearing loss (Bushman
et al. 2012). Nevertheless, adult hearing screening programs are
very scarce. Hearing screening tests will become increasingly
important as the adult population is continuously growing and
life expectancy escalates. It is expected that the world’s adult
population aged 60 years and older will almost double from
12% to 22% by 2050 (World Health Organization 2015). The
incidence of hearing loss increases as the adult population ages
with approximately one-third of adults aged 65 years and older
affected by a disabling hearing loss (World Health Organization
2013). The latest Global Burden of Disease study (Global Bur-
den of Disease 2016) indicates that 1.33 billion people suffer
from hearing loss, making it the second most common impair-
ment evaluated. Unfortunately, only about 20% of adults with
hearing loss seek help (Smits et al. 2006; Davis et al. 2007).
An untreated hearing loss negatively impacts communica-
tion abilities and cognitive, physical, and psychological func-
tioning and general quality of life (Nachtegaal et al. 2009; Lin
2011; Davis et al. 2016). Communication difficulties related
to hearing loss can lead to poor social engagement resulting
in restricted socialization, impaired relationships with friends
and family with loneliness as a consequence, especially in the
elderly (Davis et al. 2016). Persons with hearing loss demon-
strate greater cognitive decline that may be associated with an
increased risk of dementia (Lin 2011; Lin & Ferrucci 2012;
Davis et al. 2016). Hearing loss is also related to physical
impairment in older adults with an increased likelihood to fall
due to impaired auditory and vestibular cues that limit envi-
ronmental awareness, attention, and postural control (Lin &
Ferrucci 2012). The communication, physical, and cognitive
effects of hearing loss have also been linked to psychological
impairments and feelings of depression, anxiety, frustration,
and fatigue resulting in poor quality of life (Davis et al. 2007).
The physical impairments associated with a hearing loss can
furthermore cause an added financial burden on the elderly due
to increased healthcare costs (Simpson et al. 2016).
Early hearing loss intervention and counseling are impor-
tant services that may prevent or forestall cognitive decline,
dementia, and the negative psychological and physical effects
associated with hearing loss and save future health-related
costs (Simpson et al. 2016). Hearing screening programs are
important for early detection of hearing loss to maximize hear-
ing rehabilitation and quality-of-life outcomes. Various hear-
ing screening tests exist, of which standard hearing screening
options usually include self-administered questionnaires and
pure-tone audiometry. Self-administered questionnaires are
an affordable method to detect hearing loss and could be uti-
lized by any healthcare professional (Swanepoel et al. 2013). In
recent years, more accessible hearing screening methods have
been developed, which individuals can access directly without a
healthcare professional. Many countries including the Nether-
lands, United States, Australia, Germany, Poland, Switzerland,
and France now offer landline telephone hearing screening tests
The South African English Smartphone
Digits-in-Noise Hearing Test: Effect of Age,
Hearing Loss, and Speaking Competence
Jenni-Marí Potgieter,1 De Wet Swanepoel,1–3 Hermanus Carel Myburgh,4 and Cas Smits5
1Department of Speech-Language Pathology and Audiology, University
of Pretoria, South Africa; 2Ear Sciences Centre, School of Surgery,
University of Western Australia, Nedlands, Australia; 3Ear Science Institute
Australia, Subiaco, Australia; 4Department of Electrical, Electronic and
Computer Engineering, University of Pretoria, Pretoria, South Africa; and
5Department of Otolaryngology – Head and Neck Surgery, Section Ear and
Hearing, and Amsterdam Public Health Research Institute, VU University
Medical Center, Amsterdam, The Netherlands.
17
Copyright © 2017 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
2 POTGIETER ET AL. / EAR & HEARING, VOL. XX, NO. XX, 00–00
based on the recognition of digits in noise. These self-admin-
istered tests measure the signal to noise ratio (SNR) where a
listener recognizes 50% of the digit triplets correctly (i.e., the
speech reception threshold [SRT]; Smits et al. 2004; Jansen et
al. 2010; Watson et al. 2012; Zokoll et al. 2012). These digits-in-
noise hearing screening tests are fully automated, which makes
the tests appealing because they can be self-administered. The
tests are also quick to administer and can be completed in only
a few minutes. Furthermore, the digits-in-noise hearing screen-
ing tests mimic everyday speech-in-noise environments and are
sensitive to detect hearing loss (Smits et al. 2004, 2013; Jansen
et al. 2010; Zokoll et al. 2012; Williams-Sanchez et al. 2014).
In countries like South Africa, where landline telephone
penetration is less than 13% of households (Statistics South
Africa 2013), a digits-in-noise hearing test over the land-
line telephone is inadequate to reach the general population.
To provide access to ear and hearing healthcare services, an
alternative platform was considered. A smartphone-based
digits-in-noise hearing test for end-users was developed and
validated in South African English (Potgieter et al. 2015). The
test can be downloaded in South Africa (www.hearZA.co.za)
as an application and on a smartphone or other iOS or Android
device. Low-cost smartphone penetration is approaching 80%
of households, making widespread access to the test possible
for people living in rural and urban areas (Ericsson Mobility
Report 2015; Potgieter et al. 2015). The test enables users to
conduct a self-test in the comfort of a home setting using the
application downloaded to a smartphone. The smartphone-
based digits-in-noise hearing test provided equivalent results
across earphones and headphone types. Contrary to landline
telephone hearing tests that are limited to the bandwidth of the
telephone network (approximately 300 to 3400 Hz), the App-
based smartphone test uses broadband digital quality signals
(30 to 20,000 Hz; Potgieter et al. 2015).
Employing an English-based smartphone digits-in-noise
hearing test in South Africa presents its own challenge consid-
ering the multilingual population, with 11 official languages.
Estimates indicate that only 9.6% of the population is native
(N) English speaking (Statistics South Africa 2011). Non-native
(NN) language listeners typically perform worse on standard
speech-in-noise tests compared with N listeners (van Wijn-
gaarden et al. 2002; Zokoll et al. 2013). The speech material
may contain unfamiliar vocabulary and complex grammatical
structures that influence NN language listeners’ performance on
speech recognition tasks (van Wijngaarden et al. 2002; Zokoll
et al. 2013). In addition to age of NN language acquisition,
amount of NN language use, and linguistic background, age
itself may also influence speech recognition (Rogers et al. 2006;
Rimikis et al. 2013).
An English-based digits-in-noise test has several advantages
compared with speech-in-noise tests that are based on open-set
sentence or word recognition that makes this test more ame-
nable for use in a multilingual setting. First, digits-in-noise tests
use simple speech material with low linguistic demands. Sec-
ond, the speech material is presented as a closed set (i.e., digits
between zero and nine). Third, English digits are mostly famil-
iar and often used by speakers of other languages (Branford &
Claughton 2002). Finally, Kaandorp et al. (2016) have shown
that normal-hearing NN listeners only needed a 0.8 dB higher
SNR than N listeners to recognize 50% of digit triplets cor-
rectly. These advantages provide the potential for an English-
based smartphone digits-in-noise hearing screening test to be
used as a national screening test in a multilingual country like
South Africa.
The aim of this study was to evaluate the South African
digits-in-noise hearing test’s suitability for use as a hearing
screening test. The study hypothesis posited that the digits-in-
noise SRT would be poorer in NN listeners with poor English-
speaking competence than in N listeners or NN listener with
good English-speaking competence but would be sufficiently
accurate for screening purposes.
MATERIALS AND METHODS
Listeners
Three private hearing healthcare practices, three public hos-
pital audiology units, and the University clinic at the Depart-
ment of Speech-Language Pathology and Audiology, University
of Pretoria, were involved in data collection. A convenience non-
probability sampling procedure was followed with participants
at clinical data collection sites who were available and willing
to participate in the research study. All listeners provided writ-
ten informed consent to participate. The group of listeners com-
prised participants who represented all 11 official languages
in South Africa (Table 1). The 11 official languages in South
Africa are English, Afrikaans, Northern Sotho, Zulu, Sotho,
Tswana, Xhosa, Tsonga, Swazi, Ndebele, and Venda (Statistics
South Africa 2011). A total of 458 listeners (166 male and 292
female) participated in this study. Four listeners with a mixed
hearing loss were excluded from all analyses, resulting in 454
TABLE 1. Characteristics of subjects according to their native language, sex, and age
Native Language Subjects (n) Male (n) Female (n) Age Range (yrs) Mean Age (yrs) SD
English 134 43 91 16–89 35 24
Afrikaans 109 40 69 16–90 49 24
Northern Sotho 60 23 37 16–79 34 19
Zulu 32 9 23 16–63 34 18
Sotho 25 15 10 16–67 33 17
Tswana 22 6 16 16–46 19 8
Xhosa 20 13 7 16–83 24 16
Tsonga 18 7 11 16–64 30 18
Swazi 10 5 5 16–25 18 3
Ndebele 9 1 8 19–59 29 16
Other 15 2 13 16–65 27 13
Copyright © 2017 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
POTGIETER ET AL. / EAR & HEARING, VOL. XX, NO. XX, 00–00 3
adult listeners (164 male and 290 female). Eleven of the 454
listeners did not have an English-speaking competence score.
The 11 listeners were excluded in analyses where the English-
speaking competence scores were used. The mean age was
36 years (±22 years) with a range of 16 to 90 years (Table 1).
Material and Apparatus
Test procedures included otoscopy, diagnostic pure-tone air
and bone conduction audiometry, the South African English
smartphone-based digits-in-noise test, and a self-administered
English language competence questionnaire. An otoscopic
evaluation was performed for observation of any obstruction in
the external auditory meatus. Excessive cerumen was removed
by a qualified audiologist or medical practitioner before testing
commenced.
Procedures
At all test sites, calibrated audiometers with supra-aural
headphones or insert earphones were used to conduct stan-
dard audiometry. Bone conduction audiometry was addition-
ally conducted on participants with average thresholds at 500,
1000, 2000, and 4000 Hz (four-frequency pure-tone average
[4FPTA]) of more than 25 dB HL. The modified Hughson–
Westlake method was used to seek pure-tone air and bone con-
duction thresholds (Hughson & Westlake 1944). The hearing
loss was categorized as conductive or mixed when the differ-
ence between the air and bone conduction 4FPTA was >15 dB
in the best ear (Margolis & Saly 2007).
The digits-in-noise test was administered binaurally on
Vodafone Kicka smartphones or a Samsung Trend smartphone.
Intraconchal Vodafone earphones were used to present the
stimuli from the Vodafone Kicka smartphones and a HD202II
Sennheiser supra-aural headphone was used with the Samsung
Trend smartphone. A study conducted by Potgieter et al. (2015)
observed no difference between the digits-in-noise SRTs for
these headphone types. The digits-in-noise hearing test con-
sisted of five screens. The first screen opened to a quick tutorial
screen that instructed the listener on how to use the application.
The second screen allowed the listener to select his/her sex. The
third screen asked the listener to select his/her date of birth. The
fourth screen instructed the listener to place either earphones
into the ears or supra-aural headphones on the ears. The listener
was presented with digits being repeated. A scroll bar allowed
the listener to adjust the volume to a comfortable level. On the
final screen, the listener entered his/her initials and surname. A
“start test” button commenced the test.
The test material is selected from a list of 120 unique digit
triplets stored in the application (Potgieter et al. 2015). In the
application, the sound files for the digits zero to nine were
stored separately in Ogg vorbis compressed audio file format
(Potgieter et al. 2015). The bi-syllabic digits zero and seven
were also used as speech tokens to minimize a possible learn-
ing effect (Smits et al. 2013, 2016). When the test started, the
digit triplets were assembled by concatenating the appropriate
digits with silent intervals of 500 msec at the beginning and end
of each triplet. Subsequent digits were followed by 200 msec
silences with 100 msec of uniform jitter between each digit to
add some uncertainty in the listening task for when the next
digit will be presented. The digit triplet files were mixed with
broadband speech-shaped noise at the required SNR to form a
stimulus. When triplets with negative SNRs were presented, the
test operated with a fixed noise level and a varying speech level.
The speech level became fixed, and the noise level varied when
triplets with positive SNRs were presented. By following this
test procedure, a nearly constant overall level of the stimuli was
ensured (i.e., digit triplet mixed with the noise). The digits were
pronounced by a female native speaker of South African Eng-
lish. When the test started, the first stimulus set was presented at
the listener’s self-chosen comfortable listening level. A pop-up
keypad allowed the listener to enter the response. If the digit
triplet was entered 100% correctly, the next stimulus was pre-
sented at a 2 dB lower SNR than the previous digit triplet. When
the digit triplet was entered incorrectly, the next stimulus was
presented at a 2 dB higher SNR than the previous digit triplet.
Each test used 24-digit triplets to estimate the SNR correspond-
ing to the 50% correct recognition probability (i.e., the SRT).
All stimuli were presented binaurally. See Potgieter et al. (2015)
for further details.
A nonstandardized self-reported rating scale for English lan-
guage competence was completed by each listener. A facilita-
tor/translator was present to assist illiterate listeners or listeners
with poor English language competence to complete the ques-
tionnaire. The questionnaire consisted of one simple question.
The question asked the listeners to rate their English-speaking
competence in everyday communication. A simple scoring
method was used in the form of a scale between 1 (not compe-
tent at all) and 10 (perfectly competent).
RESULTS
The sample of 454 listeners represented a range of self-
reported English-speaking competences across language groups
and ages (Fig. 1). For illustrative purposes, the listeners were
categorized into 3 groups with approximately the same amount
of listeners in each group (30% N English, 24% Afrikaans, and
46% other languages). Figure 2 illustrates the effect of age on
hearing loss for these 3 groups of listeners.
Predictive Variables of the Digits-in-Noise SRT
Linear regression models were constructed for continu-
ous variables (age and best ear 4FPTA) and categorical vari-
ables (sex and English-speaking competence) to test whether
these variables significantly predicted the digits-in-noise SRT.
Final model selection was based on backward elimination of
nonsignificant variables (p > 0.05). The relative quality of the
models was measured by the Akaike information criterion or
Bayesian information criterion. The Akaike information crite-
rion and Bayesian information criterion are measures used to
assess model fit. These measures are based on the likelihood
function of the model and can be used to compare the fit of
nontested models for the same data set (Hox 2002). A linear
regression model for normal-hearing listeners with best ear
4FPTA ≤25 dB HL indicated that English-speaking competence
(β = −0.210; 95% confidence interval [CI] = −0.287 to −0.134;
p < 0.001) and age (β = 0.042; 95% CI = 0.033 to 0.051;
p < 0.001) were significant predictors of the digits-in-noise SRT
for normal-hearing listeners. The linear regression model for
listeners with best ear 4FPTA >25 dB HL indicated that English-
speaking competence (β = −0.294; 95% CI = −0.553 to −0.036;
p < 0.001) was a significant predictor of the digits-in-noise SRT.
Copyright © 2017 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
4 POTGIETER ET AL. / EAR & HEARING, VOL. XX, NO. XX, 00–00
Age (β = 0.018; 95% CI = −0.028 to 0.064; p = 0.44), however,
was not a significant predictor of the digits-in-noise SRT. Sex
was not a significant predictor of the digits-in-noise SRT in both
linear regression models.
English Competence Groups
Listeners were grouped based on the self-reported English-
speaking competence score to allow an even distribution of lis-
teners in each group to determine comparisons in the analysis.
The average SRT for normal-hearing (best ear 4FPTA ≤25 dB
HL) N listeners was compared with the average SRT of nor-
mal-hearing NN listeners within each group of self-reported
English-speaking competence (scores 1 to 10) using t tests
(no multiple comparison corrections). Significant differences
in SRTs were observed between N listeners and the 5 groups
of NN listeners with English-speaking competence scores ≤5
(all p values < 0.01). No significant differences in SRTs were
observed between N listeners and the 5 groups of NN listeners
with English-speaking competence scores ≥6 (p values between
0.116 and 0.589). As such, NN listener groups were categorized
into NN with English-speaking competence scores of ≤5 and ≥6.
Next, a 2-way analysis of covariance was used to evaluate dif-
ferences in the digits-in-noise SRT for groups of listeners with
best ear 4FPTA ≤25 dB HL based on their English-speaking
Fig. 1. English-speaking competence across age and language categories (native English, Afrikaans, and all other languages).
Fig. 2. Best ear four-frequency pure-tone average (0.5, 1, 2, and 4 kHz) across age and language categories (native English, Afrikaans, and all other languages).
Copyright © 2017 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
POTGIETER ET AL. / EAR & HEARING, VOL. XX, NO. XX, 00–00 5
competence rating. Age and best ear 4FPTA were selected as
covariates. A significant difference was observed [p < 0.01;
F(4) = 154.91; R2 = 0.579] between the digits-in-noise SRT
(corrected for age and best ear 4FPTA) for N listeners (adjusted
mean = −9.5 dB; SE = 0.17; 95% CI = −9.8 to −9.2 dB), NN ≥6
(adjusted mean = −9.3 dB; SE = 0.13; 95% CI = −9.5 to −9.0
dB), and NN ≤5 (adjusted mean = −7.9 dB; SE = 0.24; 95% CI
= −8.4 to −7.4 dB). Pairwise comparisons only demonstrated a
significant difference (p < 0.01, Bonferroni corrected) between
the NN ≤5 and the 2 other groups. There was no significant dif-
ference between the N and NN ≥6 groups. Thus, N and NN ≥6
were grouped (N and NN ≥6) and the NN ≤5 were kept as a
separate group for further analyses.
The digits-in-noise SRT and best ear 4FPTA correlation
was significant (p < 0.01) for the N and NN ≥6 listeners group
(0.763, Pearson correlation) and for the NN ≤5 listeners group
(0.690, Pearson correlation; Fig. 3).
Reference Scores
Reference scores were determined from normal-hearing lis-
teners (best ear 4FPTAs ≤25 dB HL). Table 2 shows the mean,
range, and standard deviation of the SRT for the whole group
of listeners and for the N and NN ≥6 group and NN ≤5 group
separately. The average normal-hearing SRT in the N and NN
≥6 group is approximately 1.7 dB better (lower) than in the NN
≤5 group.
Screening Characteristics
To determine the screening characteristics of the test, logistic
regression models were used to determine equations to discrim-
inate between listeners with best ear 4FPTA ≤25 dB HL and
best ear 4FPTA>25 dB HL. Logistic regression models were
constructed for all listeners grouped together and for the sub-
groups N and NN ≥6 and NN ≤5 separately. The SRT was used
as predictor, and the additional value of using age as a predictor
was determined (Table 3). Highest test accuracy was obtained
by including age and SRT for both groups. In both groups, the
addition of age as predictor increased the specificity of the test.
Receiver operating characteristic (ROC) curves were deter-
mined from the results of the logistic regression analyses for N
and NN ≥6 and NN ≤5 separately. The first set of ROC curves
were based on the SRT of the listeners; the second set of ROC
curves were based on the SRT and age of the listeners. The
ROC curves were used to determine the area under the ROC
curve (AUROC), the cutoff values, and the sensitivity (propor-
tion of correctly identified listeners with a hearing loss among
the listeners with a hearing loss) and specificity (proportion of
correctly identified listeners with normal hearing among the lis-
teners with normal hearing) of the digits-in-noise test. Figure 4
shows the ROC curve based on all listeners as one group, and
Figure 5 shows the ROC curves for the subgroups with SRT as
predictor and with SRT and age as predictors.
DISCUSSION
The recently developed English smartphone digits-in-noise
test (Potgieter et al. 2015) promises widespread access to hear-
ing screening in a country like South Africa where smartphone
penetration is approaching 80% of households (Ericsson Mobil-
ity Report 2015). Because of the multilingual population in
South Africa, it is important to consider the effect of NN listen-
ers’ performance on the digits-in-noise test. This study deter-
mined the performance of NN English listeners on the South
African English smartphone digits-in-noise hearing test, com-
pared with N English listeners.
The smartphone digits-in-noise SRTs and the best ear
4FPTA were significantly correlated (r = 0.76 for N and NN
≥6; r = 0.69 for NN ≤5). The correlation for the N and NN
≥6 group agrees with previous results reported for the Dutch
(r = 0.72), French (r = 0.77), and American-English (r = 0.74)
landline telephone digits-in-noise hearing screening tests
(Smits et al. 2004; Jansen et al. 2010; Watson et al. 2012).
The smartphone digits-in-noise test was conducted binaurally,
while the Dutch, French, and American-English tests were ear
specific. Another difference, compared with previous reports,
is the inclusion of 70.4% NN listeners (n = 320) in this study.
These findings indicate that comparable correlation between
Fig. 3. Smartphone digits-in-noise speech reception threshold correlation
with best ear four-frequency pure-tone average (0.5, 1, 2, and 4 kHz) for
N and NN ≥6 group (r = 0.763) and NN ≤5 group (r = 0.690). N, Native
speakers; NN, non-native speakers.
TABLE 2. Demographics and performance summary for
normal-hearing listeners according to self-reported English-
speaking competence (best ear 4FPTA ≤25 dB HL)
Description
All
Language N&NN ≥6* NN ≤5*
Subjects (n) 337 291 46
Mean Age (yrs) 27 26 36
Age Range (yrs) 16 to 81 16 to 81 16 to 67
SD Age (yrs) 16 14 17
SRT Range (dB) 0.0 to −13.0 0.0 to −13.0 −4.8 to −12.4
Mean SRT (dB) −10.2 −10.4 −8.7
SD SRT(dB) 1.6 1.5 1.9
*“How competent are you in speaking English?” Rating scale (1 = no competence;
10 = perfect competence).
N, Native speakers; NN, non-native speakers.
Copyright © 2017 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
6 POTGIETER ET AL. / EAR & HEARING, VOL. XX, NO. XX, 00–00
digits-in-noise SRTs and best ear 4FPTAs can be obtained
in a sample where the majority of listeners are NN English
listeners with some degree of self-reported English-speaking
competency.
The results of the linear regression models indicated that
English-speaking competence is a significant predictor of the
digits-in-noise SRT; age is only a significant predictor for lis-
teners with best ear 4FPTA ≤25 dB HL. A contributing fac-
tor could be the difference in distribution of age and hearing
loss between listeners with best ear 4FPTA ≤25 dB HL and
listeners with a best ear 4FPTA >25 dB HL. Results also sup-
port findings by Moore et al. (2014) who indicated that age
may possibly have an effect on the digits-in-noise SRT. They
showed that a decline in cognitive functioning, associated
with age, has an effect on the digits-in-noise SRT. Koole et
al. (2016) indicated a low correlation between age and the
digits-in-noise SRT after controlling for pure-tone thresholds.
In light of the above, it is important to consider age when
determining the result of the digits-in-noise test in normal-
hearing listeners because it may contribute to the accuracy of
the screening test outcome. Accuracy of the smartphone dig-
its-in-noise test was evaluated by determining the AUROC,
sensitivity, and specificity of the test to discriminate between
listeners with best ear 4FPTA >25 dB and those with best
ear 4FPTA ≤25 dB, for N and NN English listeners. Logistic
regression models and ROC curve analysis demonstrate
that subgroups based on English-speaking competence and
including age as predictor increase the AUROC, sensitivity,
and specificity of the test.
Test performance improved in the N and NN ≥6 group
(AUROC = 0.962) when self-reported English-speaking
competence and age were considered. The sensitivity
(0.95) and specificity (0.87) for N and NN ≥6 English lis-
teners (best ear 4FPTA >25 dB HL) compared well to the
Dutch (0.91 and 0.93, respectively) and American-English
(0.80 and 0.83, respectively) digits-in-noise tests (Smits et
al. 2004; Watson et al. 2012). The sensitivity and specific-
ity was poorer for the NN ≤5 group than for the N and NN
≥6 group. Possible reasons for this finding might be the fact
that the NN ≤5 group was more heterogeneous in English-
speaking competence; the group included a smaller number
of listeners; and varying distributions of hearing loss degrees
may have influenced the calculated test characteristics. Self-
reported English-speaking competence was a significant pre-
dictor of the digits-in-noise SRT. Results by Kaandorp et al.
(2015) also indicated that NN listeners did not perform as
well as N Dutch listeners on the Dutch digits-in-noise test.
Vocabulary size and educational level had a small effect
(0.8 dB SNR increase) on the performance of NN listeners
on digits-in-noise recognition (Kaandorp et al. 2015). This
small difference in the performance between N listeners and
NN listeners was measured to revalidate that digits-in-noise
depend minimally on top-down processing (e.g., linguistic
skills; Smits et al. 2013).
ROC curve analysis was used to determine cutoff SNR val-
ues for “pass” (4FPTA ≤25 dB HL) and “refer” (4FPTA >25
dB HL) for hearing loss for N and NN English listeners. The
cutoff SNR value for pass or refer for hearing loss for N and
NN ≥6 English was −9.55 dB and −7.50 dB for NN ≤5 English
listeners. The higher cutoff SNR for NN ≤5 English listeners
can be expected as the linear regression model demonstrates
that English-speaking competence has a negative effect on the
digits-in-noise SRT, and the mean SRT for normal-hearing lis-
teners is higher for the NN ≤5 group than for the N and NN ≥6
group (Table 2). The mean SRT of the normal-hearing N and
NN ≥6 English listeners (−10.4 dB) is similar to the diotically
measured average SRT for the Dutch and American-English
digits-in-noise test (−10.0 and −11.2 dB SNR, respectively;
Smits et al. 2016). This comparison indicates that the digits-
in-noise test provides close comparisons across NN listeners
with good language proficiency (Smits et al. 2016).
The present study demonstrates that a smartphone appli-
cation provides an opportunity to use the English digits-in-
noise hearing test as a national test for South Africans. The
fact that English digits are often used by speakers of other
TABLE 3. Logistic regression models for N and NN ≥6 and NN ≤5 listeners
Predictors Equation AUROC Cutoff Value Sensitivity Specificity
All subjects SRT — 0.925 SRT = −9.55 dB 0.94 0.77
N&NN ≥6SRT — 0.943 SRT = −9.55 dB 0.95 0.83
SRT, age p = 1/[1 + exp (−0.562∙SRT − 0.080∙age)] 0.962 p = 0.149 0.95 0.87
NN ≤5SRT — 0.873 SRT = −7.50 dB 0.84 0.74
SRT, age p = 1/[1 + exp (−0.478∙SRT − 0.054∙age)] 0.903 p = 0.263 0.84 0.77
AUROC, area under the receiver operating characteristic curve; N, Native speakers; NN, non-native speakers; SRT, speech reception threshold.
Fig. 4. Receiver operating characteristic curve for all listeners with speech
reception threshold as predictor.
Copyright © 2017 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
POTGIETER ET AL. / EAR & HEARING, VOL. XX, NO. XX, 00–00 7
languages in South Africa (Branford & Claughton 2002)
allows for the possibility to accommodate NN listeners by
adjusting reference scores based on a self-reported English-
speaking competence. More representative data from diverse
language groups and English competence levels would be
useful to improve the validity of the test as a nationally
used screening tool. The smartphone application could be
programmed to report the test results in a listener’s native
language to allow for correct interpretation of test results
across NN listeners. Providing these adjustments can ensure
adequate test performance across N English and NN English
listeners. It is important to note that this study was limited to
a small group (15.9%; 72/454 listeners) of NN English lis-
teners with poor self-reported English-speaking competence
(scores ≤5/10). Future studies should aim to expand data on
this group of listeners.
ACKNOWLEDGMENTS
This work was supported by the National Research Foundation under the
grant number 88803.
The second and third authors have a relationship with the hearX Group,
who owns the right to the IP, that includes equity, consulting and potential
royalties.
Address for correspondence: De Wet Swanepoel, Department of Speech-
Language Pathology and Audiology, University of Pretoria, Pretoria 0002,
Republic of South Africa. E-mail: dewet.swanepoel@up.ac.za
Received November 4, 2016; accepted September 27, 2017.
REFERENCES
Branford, W., & Claughton, J. S. (2002). 10 mutual lexical borrowings
among some languages of Southern Africa: Xhosa, Afrikaans and Eng-
lish. In R. Mestherie (Ed.), Language of South Africa (p. 199). New York,
NY: Cambridge University Press.
Bushman, L. A., Belza, B., Christianson, P. (2012). Older adult hearing loss
and screening in primary care. J Nurse Pract, 8, 509–514.
Davis, A., Smith, P., Ferguson, M., et al. (2007). Acceptability, benefit
and costs of early screening for hearing disability: A study of potential
screening tests and models. Health Technol Assess, 11, 1–294.
Davis, A., McMahon, C. M., Pichora-Fuller, K. M., et al. (2016). Aging
and hearing health: The life-course approach. Gerontologist, 56(Suppl
2), S256–S267.
Ericsson Mobility Report. (2015). Sub-Saharan Africa. Stockholm, Swe-
den: Ericsson.
Global Burden of Disease (GBD). (2016). Global, regional, and national
incidence, prevalence, and years lived with disability for 310 diseases
and injuries, 1990 – 2015: A systematic analysis for the global burden of
disease study 2015. Lancet, 388, 1545–1602.
Hox, J. (2002). Multilevel Analysis, Techniques and Applications. Mahwah,
NJ: Lawrence Erlbaum Associates Inc, Publishers.
Hughson, W., & Westlake, H. (1944). Manual for a program outline for
rehabilitation of aural casualties both military and civilian. Trans Am
Acad Opthalmol Otolaryngol, 48, 1–15.
Jansen, S., Luts, H., Wagener, K. C., et al. (2010). The French digit triplet
test: A hearing screening tool for speech intelligibility in noise. Int J
Audiol, 49, 378–387.
Kaandorp, M. W., Smits, C., Merkus, P., et al. (2015). Assessing speech
recognition abilities with digits in noise in cochlear implant and hearing
aid users. Int J Audiol, 54, 48–57.
Kaandorp, M. W., De Groot, A. M., Festen, J. M., et al. (2016). The influ-
ence of lexical-access ability and vocabulary knowledge on measures of
speech recognition in noise. Int J Audiol, 55, 157–167.
Koole, A., Nagtegaal, A. P., Homans, N. C., et al. (2016). Using the dig-
its-in-noise test to estimate age-related hearing loss. Ear Hear, 37,
508–513.
Lin, F. R. (2011). Hearing loss and cognition among older adults in the
United States. J Gerontol A Biol Sci Med Sci, 66, 1131–1136.
Lin, F. R., & Ferrucci, L. (2012). Hearing loss and falls among older adults
in the United States. Arch Intern Med, 172, 369–371.
Margolis, R. H., & Saly, G. L. (2007). Toward a standard description of
hearing loss. Int J Audiol, 46, 746–758.
Moore, D. R., Edmondson-Jones, M., Dawes, P., et al. (2014). Relation
between speech-in-noise threshold, hearing loss and cognition from
40-69 years of age. PLoS One, 9, e107720.
Nachtegaal, J., Smit, J. H., Smits, C., et al. (2009). The association between
hearing status and psychosocial health before the age of 70 years:
Results from an internet-based national survey on hearing. Ear Hear,
30, 302–312.
Potgieter, J. M., Swanepoel, d. e. W., Myburgh, H. C., et al. (2015). Devel-
opment and validation of a smartphone-based digits-in-noise hearing test
in South African English. Int J Audiol, 55, 405–411.
Rimikis, S., Smiljanic, R., Calandruccio, L. (2013). Nonnative English
speaker performance on the Basic English Lexicon (BEL) sentences.
J Speech Lang Hear Res, 56, 792–804.
Rogers, C. L., Lister, J. J., Febo, D. M., et al. (2006). Effects of bilingual-
ism, noise, and reverberation on speech perception by listeners of normal
hearing. Appl Psycholinguist, 27, 465–485.
Simpson, A. N., Simpson, K. N., Dubno, J. R. (2016). Higher health care
costs in middle-aged US adults with hearing loss. JAMA Otolaryngol
Head Neck Surg, 142, 607–609.
Smits, C. (2016). Comment on ‘International collegium of rehabilitative
audiology (ICRA) recommendations for the construction of multilingual
speech tests’, by Akeroyd et al. Int J Audiol, 55, 268–269.
Fig. 5. Receiver operating characteristic curves for N and NN ≥6 group (left) and NN ≤5 group (right). The improvements of test characteristics are illustrated
by the shift of the curves to the upper left corner with the inclusion of age as a predictor. N, Native speakers; NN, non-native speakers.
Copyright © 2017 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
8 POTGIETER ET AL. / EAR & HEARING, VOL. XX, NO. XX, 00–00
Smits, C., Kapteyn, T. S., Houtgast, T. (2004). Development and validation
of an automatic speech-in-noise screening test by telephone. Int J Audiol,
43, 15–28.
Smits, C., Kramer, S. E., Houtgast, T. (2006). Speech reception thresholds
in noise and self-reported hearing disability in a general adult popula-
tion. Ear Hear, 27, 538–549.
Smits, C., Theo Goverts, S., Festen, J. M. (2013). The digits-in-noise test:
Assessing auditory speech recognition abilities in noise. J Acoust Soc
Am, 133, 1693–1706.
Smits, C., Watson, C. S., Kidd, G. R., et al. (2016). A comparison between
the Dutch and American-English digits-in-noise (DIN) tests in normal-
hearing listeners. Int J Audiol, 55, 358–365.
Statistics South Africa. (2011). Census in brief, South Africa. Retrieved Nov 6,
2017 from http://www.statssa.gov.za/publications/P03014/P030142011.pdf.
Statistics South Africa. (2013). General household survey. Retrieved Nov
6, 2017 from http://www.statssa.gov.za/publications/P0318/P03182013.
pdf.
Swanepoel, d. e. W., Eikelboom, R. H., Hunter, M. L., et al. (2013). Self-
reported hearing loss in baby boomers from the Busselton Healthy Age-
ing Study: Audiometric correspondence and predictive value. J Am Acad
Audiol, 24, 514–521; quiz 529.
van Wijngaarden, S. J., Steeneken, H. J., Houtgast, T. (2002). Quantifying
the intelligibility of speech in noise for non-native listeners. J Acoust Soc
Am, 111, 1906–1916.
Watson, C. S., Kidd, G. R., Miller, J. D., et al. (2012). Telephone screening
tests for functionally impaired hearing: Current use in seven countries
and development of a US version. J Am Acad Audiol, 23, 757–767.
Williams-Sanchez, V., McArdle, R. A., Wilson, R. H., et al. (2014). Valida-
tion of a screening test of auditory function using the telephone. J Am
Acad Audiol, 25, 937–951.
World Health Organization. (2013). Millions of People in the World Have
Hearing Loss That Can Be Treated or Prevented. Geneva: World Health
Organization.
World Health Organization. (2015). Aging and health. Retrieved Nov 6,
2017 from http://www.who.int/mediacentre/factsheets/fs404/en/.
Zokoll, M. A., Wagener, K. C., Brand, T., et al. (2012). Internationally com-
parable screening tests for listening in noise in several European lan-
guages: The German digit triplet test as an optimization prototype. Int J
Audiol, 51, 697–707.
Zokoll, M. A., Hochmuth, S., Warzybok, A., et al. (2013). Speech-in-noise
tests for multilingual hearing screening and diagnostics1. Am J Audiol,
22, 175–178.
