International Journal of Epidemiology, 2017, 266–277

doi: 10.1093/ije/dyw141
Advance Access Publication Date: 27 July 2016
Original article

Road Accidents

Usefulness of overnight pulse oximeter as the

sleep assessment tool to assess the 6-year risk
of road traffic collision: evidence from the
Taiwan Bus Driver Cohort Study
Wei-Te Wu,1 Su-Shan Tsai,2 Hui-Yi Liao,1 Yu-Jen Lin,3 Ming-Hsiu Lin,4
Trong-Neng Wu,1,5 Tung-Sheng Shih4 and Saou-Hsing Liou1,6,7,*
1
National Institute of Environmental Health Sciences, National Health Research Institutes, Miaoli,
Taiwan, 2Graduate Institute of Life Sciences, National Defense Medical Center, Taipei, Taiwan,
3
Institute of Occupational Medicine and Industrial Hygiene, National Taiwan University, Taipei, Taiwan,
4
Institute of Labor, Occupational Safety and Health, Ministry of Labor, New Taipei, Taiwan,
5
Department of Healthcare Administration, Asia University, Taichung, Taiwan, 6Department of
Occupational Safety and Health, China Medical University, Taichung, Taiwan and 7Department of
Public Health, National Defense Medical Center, Taipei, Taiwan
*Corresponding author. National Institute of Environmental Health Sciences, National Health Research Institutes,
35 Keyan Road, Zhunan Town, Miaoli County, 35053 Taiwan, R.O.C. E-mail: shliou@nhri.org.tw
Accepted 13 May 2016

Abstract
Background: In order to support health service organizations in arranging a system for
prevention of road traffic collisions (RTC), it is important to study the usefulness of sleep
assessment tools. A cohort study was used to evaluate the effectiveness of subjective
and objective sleep assessment tools to assess for the 6-year risk of both first RTC event
only and recurrent RTC events.
Methods: The Taiwan Bus Driver Cohort Study (TBDCS) recruited 1650 professional driv-
ers from a large bus company in Taiwan in 2005. The subjects were interviewed in per-
son, completed the sleep assessment questionnaires and had an overnight pulse oxim-
eter survey. Moreover, this cohort of drivers was linked to the National Traffic Accident
Database (NTAD) and researchers found 139 new RTC events from 2005 to 2010. Primary
outcomes were traffic collisions from NTAD, nocturnal oxygen desaturation index (ODI)
from pulse oximeter, Pittsburg sleeping quality score, Epworth daytime sleepiness score,
Snore Outcomes Survey score and working patterns from questionnaires. A Cox propor-
tional hazards model and an extended Cox regression model for repeated events were
performed to estimate the hazard ratio for RTC.
Results: The RTC drivers had increased ODI4 levels (5.77 6 4.72 vs 4.99 6 6.68 events/h;
P ¼ 0.008) and ODI3 levels (8.68 6 6.79 vs 7.42 6 7.94 events/h; P ¼ 0.007) in comparison
with non-RTC drivers. These results were consistent regardless of whether ODI was eval-
uated as a continuous or a categorical variable. ODI4 and ODI3 levels increased the

C The Author 2016; all rights reserved. Published by Oxford University Press on behalf of the International Epidemiological Association
V 266

Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194

by guest
on 04 May 2018
 International Journal of Epidemiology, 2017, Vol. 46, No. 1 267

6-year RTC risks among professional drivers even after adjusting for age, education, his-
tory of cardiovascular disease, caffeine intake, sleeping pills used, bus driving experience
and shift modes. Moreover, there was an increased trend for ODI between the stratifica-
tion of the number of RTCs in comparison with the non-RTC group. In the extended Cox
regression models for repeated RTC events with the Anderson and Gill intensity model
and Prentice-Williams-Petersen model, measurement of ODI increased hazards of the
subsequent RTC events.
Conclusion: This study showed that an increase in the 6-year risk of RTC was associated
with objective measurement of ODI for a sign of sleep-disordered breathing (SDB), but
was not associated with self-reported sleeping quality or daytime sleepiness. Therefore,
the overnight pulse oximeter is an effective sleep assessment tool for assessing the risk
of RTC. Further research should be conducted regarding measures to prevent against
SDB among professional drivers.

Key words: Portable pulse oximeter, oxygen desaturation index, sleep-disordered breathing, road traffic collision,
professional drivers

Key Messages

• We used a large professional driver cohort to assess the effectiveness of sleep assessment tools for the prediction of

the subsequent 6-year RTC risk.

• We found that the nocturnal oxygen desaturation index is effective for assessing the risk of both first RTC event only

and recurrent RTC events.

• This study is the first comprehensive assessment that includes: validated sleep assessment; systematic traffic colli-

sions data collection; controls for confounding factors; and consideration of recurrent traffic collision events. This
supports evidence that suggests that health service organizations should arrange a screening system for prevention
of sleep-related traffic crashes.
• This study provides evidence for future research about portable devices or mobile health applications which can suc-

cessfully limit drivers’ RTC and thus improve their health and safety.

Introduction There is approximately a 3–7% prevalence of obstruct-

Road traffic injuries are estimated to be the eighth leading
cause of death globally and will become the fifth leading
cause of death by 2030.1 Approximately 1.24 million peo-
ple died on the world’s roads in 2010, and another 20 to
50 million sustained non-fatal injuries as a result of RTC.1
Sleepiness and sleep disorders in drivers are increasingly
recognized as an important factor contributing to the bur-
den of traffic-related morbidity and mortality.2–5
Epidemiological studies estimate that the percentage of
RTC attributable to sleepiness ranges from 1–3% for the
USA6 and 1.9–3.9% in Norway7 to 10% in France8 and
19% for New Zealand.4 The degree of performance im-
pairment due to severe chronic sleepiness was similar to
that of regular cellular phone use while driving9 or driving
over the legal blood alcohol limit with a 0.05 % blood al-
cohol concentration.10
ive sleep apnoea (OSA) in the general population.11 In con-
trast, previous studies identified that about 28.2% of
professional drivers from the USA,12 15.8% of profes-
sional drivers from Australia,13 and 10% of professional
drivers from both the UK14 and Hong Kong15 suffered
sleep apnoea. Sleep disorders have been shown to be more
common among professional drivers than the general
population, and impaired alertness due to sleep disorders is
more dangerous to human life and property in professional
drivers. Therefore, effective screening tools for sleepiness-
related accidents and provision of relevant prevention and
treatment among professional drivers are in urgent need.
Studies on the usefulness of sleep assessment tools in re-
lation to RTC in professional drivers is important to sup-
port health service organizations in arranging a system of
screening for sleep-disordered breathing (SDB). SDB is a

Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194

by guest
on 04 May 2018
 268 International Journal of Epidemiology, 2017, Vol. 46, No. 1

sleep disorder that involves cessation of or a significant
decrease in airflow in the presence of breathing effort,
and causes recurrent oxyhaemoglobin desaturations and
arousals from sleep. Overnight laboratory-based polysom-
nography (PSG) has been regarded as the gold standard for
the diagnosis of SDB, which is mainly quantified by the
apnoea-hypopnoea index defined as the sum of apnoea
and hypopnoea events per hour during sleep.16,17
However, this method is expensive, technically difficult
and not easily accessible to professional drivers. Review of
previous papers revealed several problems in this research
field, including incomplete sleep problems definition, ab-
sence of systematic traffic collisions data collection and
poor control for confounding factors.18 It also did not
show clear causal association between results from sleep
assessment tools and RTC. Thus, we performed a large co-
hort study of male professional drivers in order to evaluate
the effectiveness of subjective and objective sleep assess-
ment tools to assess for the risk of first RTC event only
and of recurrent RTC events. We hypothesized that the
objective measurement of overnight pulse oximeter, and
self-reports including sleeping quality score, daytime
sleepiness and snore outcomes, would predict RTC risk
and that the associations would persist after adjusting for
RTC risk factors.
Methods
Participants
The study procedures are presented in Figure 1. The
Taiwan Bus Driver Cohort Study (TBDCS) included 1650
professional drivers from the largest transportation compa-
nies in Taiwan. First, we used this cohort to link the
Driving Hours Dataset (total number of re-
cords ¼ 1 518 350 person-times) based on the Event Data
Recorder from 2005 to 2007. We only selected 1037 pro-
fessional drivers whose total driving period exceeded 100
days during the 3 years after undergoing an assessment
questionnaire interview. There were 23 subjects who did
not have a completed assessment questionnaire, and they
were excluded. The remaining 1014 drivers completed the

Taiwan Bus Driver Cohort Study

(TBDCS) (N=1650)

Driving Hours Dataset Excluded:

Driving period < 100
2005/09-2007/11
days (n=613)

Basic and working paerns quesonnaire

Pisburg sleeping quality quesonnaire (PSQI) Uncompleted:
Epworth dayme sleepiness quesonnaire (ESS) (n=23)
Snore outcomes survey quesonnaire (SOS)

Overnight pulse oximeter survey

(N=1014)

Taiwan Naonal Death

Registry (2005-2012):
2005-2010 Naonal Traffic
Death cases=8
Accident Database (NTAD)
Criteria: the main culprit
Excluded:
Traffic collisions were
related to alcohol used,
Road Traffic Collision
cell phone used, vehicle case (RTC):
breakdowns, and bad (N=139)
sight distance
Final
(n=32)
Finally RTC drivers: Non-RTC drivers:
(N=107) (N=875)

Figure 1. Study flow diagram.

Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194

by guest
on 04 May 2018
 International Journal of Epidemiology, 2017, Vol. 46, No. 1 269

overnight pulse oximeter survey for evaluating the risk of
SDB. The 1014 subjects were linked to the National
Traffic Accident Database (NTAD) and 139 non-fatal
RTC cases (culprit) were found from 2005 to 2010. We
further excluded 32 RTC cases that were related to alcohol
use, cell phone use, vehicle breakdowns such as brake fail-
ure, directional control failure, lighting system failure, tyre
blow-out, vehicle parts falling off and poor vision ahead
such as curved road, steep hill and parked vehicles on the
road. Finally, 107 RTC drivers and 875 non-RTC drivers
who did not experience a traffic collision in the 6-year
follow-up period were included in the subsequent analysis.
The institutional review board of the National Health
Research Institutes and Tri-Service General Hospital,
Taiwan, approved this study. Informed consent was ob-
tained from each of the subjects after a detailed explan-
ation of the nature of the study and the interviewer had
explained possible consequences of the study, on the day of
the personal interviews.
Data sources for RTC
In the present study, the deterministic record linkage strat-
egy was used to pick a Personal Identification Number
(PIN) in Taiwan which is a unique identifier, and records
sharing the same value were used to identify the same per-
son. Using each worker’s PIN, researchers were able to
link 139 new RTC cases from professional drivers to the
NTAD from 1 January 2005 to 31 December 2010.
Information on new cases of RTC was obtained from
the NTAD, which was established by the National Police
Agency, Ministry of the Interior, Taiwan, to monitor the
deaths and injuries from traffic accidents in Taiwan.
Taiwan uses a two-stage process to evaluate the complete-
ness and reliability of the RTC report, and whether this
can enter the registration system of NTAD. The NTAD
contains all of the police-reported RTC data in the Taiwan
area, such as the location, time and day of the crash, the
weather condition under which the crash occurs, speed
limit, pavement surface conditions (e.g. wet or dry) and il-
lumination conditions at the time of the RTC. The charac-
teristics of culpability are also available, such as age,
gender, seat belt and helmet use, severity of injury, alcohol
involvement, cell phone use and type of licence.
Sleep assessment questionnaires
After written informed consent was obtained from individ-
ual participants, each study participant responded to
a structured interview which collected information on
socio-demographic characteristics (age, ethnicity, marital
and education status), work conditions (year of first
employment and bus driving experience) and lifestyle hab-
its (smoking, drinking, exercising and amounts of refresh-
ing drinks and medicine used). Then, the following
questionnaires were applied:
i. The Pittsburgh Sleep Quality Index (PSQI) is a self-
rated questionnaire which assesses sleep quality and
disturbances over a 1-month period.19 Nineteen indi-
vidual items generate seven component scores: subject-
ive sleep quality, sleep latency, sleep duration, habitual
sleep efficiency, sleep disturbances, use of sleeping
medication and daytime dysfunction. The sum of the
seven component scores yields one global score of sub-
jective sleep quality (range 0–21); higher scores repre-
sent poorer subjective sleep quality.19,20
ii. The Snore Outcomes Survey questionnaire was ini-
tially developed by the Clinical Outcomes Research
Unit of the Massachusetts Eye and Ear Infirmary to as-
sess snoring and SDB.21 The Snore Outcomes Survey
contains eight items that evaluate the duration, sever-
ity, frequency and consequences of problems associ-
ated with SDB, and each item has five to six response
options. The Snore Outcomes Survey scores is normal-
ized on a scale ranging from 0 (worst) to 100
(best).21,22 This study used the Chinese version of the
Snore Outcomes Survey questionnaire that demon-
strated good test-retest reliability to evaluate adults
with SDB among the Chinese-speaking population.22
iii. The Epworth sleepiness questionnaire (ESS) is a simple
questionnaire measuring the general level of daytime
sleepiness (i.e. chronic sleepiness).23 We used the
Chinese version of the ESS, a self-rating scale of eight
items, with which the participants indicated their
probability of falling asleep in eight different daily situ-
ations.24 The scale scores range from 0 (best) to 24
(worst). Generally, a score greater than 11 indicates
impairment of alertness.23,24
Overnight SpO2 monitoring
Overnight peripheral capillary oxygen saturation (SpO2)
was monitored at home using a high-resolution pulse ox-
imeter wristwatch (PULSOX-300i, Konica Minolta
Sensing, Inc., Osaka, Japan). The sampling frequency of
the oximeter PULSOX-300i is 1 Hz on memory interval
and has an average time of 3 s. All participants were asked
to perform the operation again if it was recorded that: (i)
the total sleep time was less than 4 h or more than 9 h; (ii)
the pulse waveform stopped for more than 1 h; or (iii) the
self-recorded sleep time and pulse waveform testing time
did not match. This was used to minimize the bias from
the recorded oximetry data.

Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194

by guest
on 04 May 2018
 270
The data were downloaded onto a personal computer
the following morning and were processed with a specific-
ally designed computer program, Profox (Profox
Associates, Escondido, CA). Oxygen desaturation index
(ODI), cumulative time percentage with SpO2 < 88%
(CT88), and the lowest and average SpO2 were extracted
from the oximetry data. ODI is the hourly average number
of desaturation episodes, which are defined as at least 3%
or 4% decrease in saturation from the average saturation
in the preceding 120 s and lasting > 10 s.
Statistical analysis
If the level of indices exhibited skewed distributions, the
original data were transformed by using a natural loga-
rithm to approximate a normal distribution. The means
and standard deviations were used to describe the distribu-
tions of continuous variables. The percentages were used
to describe the distributions of categorical variables. The
chi-square test was used to compare the distribution of
demographic characteristics between RTC and non-RTC
drivers. The independent t test was used for testing differ-
ences with PSQI, Snoring score, ESS score and ODI level
between RTC and non-RTC drivers. One-way analysis of
variance (ANOVA) and linear trend testing were used to
determine whether there were any significant differences
between the means of three groups including the non-RTC
group, one-time traffic collision group and more than two
traffic collisions group. This study used a stepwise regres-
sion analysis of a Cox proportional hazards model in
which the choice of predictive variables was carried out by
an automatic procedure. A variable has to be at the 0.15
level before it can be entered into the initial model,
whereas a variable in the final model has to be at the 0.15
level for it to remain in the final model. A Cox propor-
tional hazards model was performed in order to estimate
the hazard ratio (HR) for RTC among professional drivers
with an objective measurement of overnight pulse oximeter
and self-reports including sleeping quality score, daytime
sleepiness and snore outcomes.
Drivers who survived and did not have traffic collisions
before the cut-off date (31 December 2010) contributed to
the person-year time between work start date and cut-off
date. Drivers who survived and had traffic collisions con-
tributed to the person-year time between work start date
and RTC date. Meanwhile, those known to have died and
who did not have traffic collisions before the cut-off date
contributed to the person-year time between their work
start date and their date of death. Moreover, extended Cox
regression models for repeated events were carried out uti-
lizing a counting process approach (Anderson and Gill in-
tensity model) and stratified Cox model approach
Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194
by guest
on 04 May 2018
International Journal of Epidemiology, 2017, Vol. 46, No. 1
[Prentice-Williams-Petersen PWP) model] to assess hazards
of subsequent RTC events. A counting process approach
for each RTC event established essentially independent in-
crements. The inference method was based on the model-
based covariance estimate and the robust sandwich covari-
ance estimate. The survival time layout was the time from
work start date to the first RTC event (or time from work
start date until cut-off date), and the time from the previ-
ous event to the next event (or time from the previous
event until cut-off date). Additionally, the PWP total time
model and the PWP gap time model with common effects
were used to further explore the effect on recurrent events.
For the PWP total time model, the survival time layout was
the same as the counting process approach. For the PWP
gap time model, the start time of each RTC event was time
from work start date, and the stop time was the length of
the time interval to next event (or the time until cut-off
date). The analysis was performed using SAS software (ver-
sion 9.3; SAS Institute).
Furthermore, shift work modes were defined as day
shift only (days of work schedule between 6am and 5 pm /
total driving days  80%), second shift only (days of work
schedule between 5 pm and 12 pm /total driving day-
s  65%), third shift only (days of work schedule between
12 pm and 6am /total driving days  65%) and irregular
shift based on the Driving Hours Dataset.
Results
By the end of the follow-up period (31 December 2010),
more than half of the cohort subjects were less than 40
years old at the time of their first employment, and almost
half of the subjects worked irregular shifts (Table 1). The
descriptive statistics of the RTC and non-RTC drivers in
terms of demographic characteristics, lifestyle behaviour,
diet and sleep patterns are presented in S1 Table (available
as Supplementary data at IJE online). There was no differ-
ence in the distribution of age, body mass index (BMI)
level, marriage status, education status, smoking, drinking,
chewing betel nut, exercise habit or between the two
groups (P > 0.05). Meanwhile, these two groups were simi-
lar in terms of history of sleep disease, snore status, sleep-
ing hours, time to falling asleep and daytime sleep (P > 0.
10). However, the distribution of caffeine drinking and
sleeping pill use was found to be different between the
RTC and non-RTC groups (P < 0.05).
Sleep assessment between RTC and non-RTC
drivers
In comparison with non-RTC drivers, the RTC group was
higher in ODI4 (5.77 6 4.72 vs 4.99 6 6.68 events/h;
 International Journal of Epidemiology, 2017, Vol. 46, No. 1 271

Table 1. Work patterns of the professional drivers dose-response gradients for PSQI, Snore Outcomes Survey
scores, ESS and ODI in relation to the number of RTC. The
Subjects Person-years
results showed that the subjects with more than two traffic
n (%) Sum (%) collisions had higher average ODI4 and ODI3 levels and
than the RTC group (only one traffic collision) and non-
Total subjects 1014 9701.2
RTC group (ODI4: 6.52 6 5.17, 5.57 6 5.31, 4.99 6 6.68
Non-RTC drivers 875 (86.3) 8469.8 (87.3)
RTC drivers 139 (13.7) 1231.4 (12.7) events/h; P-value for trend ¼ 0.041; ODI3: 10.15 6 7.16,
Selected RTC driversa. 107 (10.6) 943.7 (9.7) 8.29 6 6.68, 7.42 6 7.94 events/h; P-value for trend ¼ 0.026)
Year of first employment (S2 Table, available as Supplementary data at IJE online).
1989–2002 384 (37.9) 5175.7 (53.4)
2003–05 340 (33.5) 2759.5 (28.4)
2006–08 290 (28.6) 1766.0 (18.2) Associations between RTC and risk factors
Age at first employment (years)
As shown in the univariate Cox regression analysis, univer-
 32 306 (30.2) 2331.9 (24.0)
sity and college [HR ¼ 2.752; 95% confidence interval
33–38 348 (34.3) 3519.8 (36.3)
 39 360 (35.5) 3849.6 (39.7) (CI) ¼ 1.430–5.295; P ¼ 0.002), cardiovascular disease
Bus driving experience (years) (CVD) history (HR ¼ 2.605; 95% CI ¼ 1.059–6.406;
2 223 (22.0) 1471.3 (15.2) P ¼ 0.037), caffeine drinks used (HR ¼ 1.659; 95%
2.1–5 307 (30.3) 2453.0 (25.3) CI ¼ 1.030–2.600; P ¼ 0.038) and sleeping pill use
5.1–8 201 (19.8) 1948.9 (20.1) (HR ¼ 3.478; 95% CI ¼ 1.280–9.452; P ¼ 0.015) were
>8 283 (27.9) 3828.1 (39.5)
associated with increased RTC risks. Age (HR ¼ 0.962;
Shift work modesb.
95% CI ¼ 0.936–0.988; P ¼ 0.005) and bus driving experi-
Day shifts only 434 (42.8) 4339.6 (44.7)
Irregular shift 485 (47.8) 4360.5 (44.9)
ence were shown to be associated with decreased RTC
Second & third shift only 95 (9.4) 1001.1 (10.3) risks (Figure 2). Moreover, in a stepwise regression ana-
lysis of the Cox model, the predictive variables including
a
Excluded traffic collisions related to alcohol used, cell phone used, vehicle
age, education, CVD history, caffeine drinking, use of
breakdowns and poor visibility.
b
Based on Driving Hours Dataset from 2005 to 2007. sleeping pills, bus driving experience and shift work, were
selected in the subsequent model.

Table 2. Comparison of PSQI score, Snore Outcomes Survey

score, ESS score and ODI level between RTC and non-RTC RTC and sleep assessment
drivers
After adjusting for age, education, CVD history, caffeine
Variables RTC group Non-RTC group drinking, use of sleeping pills, bus driving experience and shift
(n ¼ 107) (n ¼ 875) modes, regardless of whether ODI was evaluated as a con-
Mean (SD) Mean (SD) P-value tinuous or a categorical variable, the ODI4 and ODI3 levels
were found to be associated with increased RTC risks among
Global PSQI score 5.36 (2.83) 5.23 (2.79) 0.626 professional drivers (as a continuous variable: HR ¼ 1.328;
Snore Outcomes 71.46 (14.92) 73.99 (13.83) 0.077
95% CI ¼ 1.087–1.622; P ¼ 0.005 and HR ¼ 1.398; 95%
Survey score
CI ¼ 1.109–1.761; P ¼ 0.005, respectively). Although we
ESS score 6.68 (4.55) 6.48 (4.31) 0.653
found that the Snore Outcomes Survey score  60 was
ODI4 (events/h)a 5.77 (4.72) 4.99 (6.68) 0.008 slightly associated with an increase in RTC risks in compari-
ODI3 (events/h)a 8.68 (6.79) 7.42 (7.94) 0.007 son with score > 60 (HR ¼ 1.558; 95% CI ¼ 0.997–2.434;
CT88 (%)a 1.71 (4.42) 1.64 (6.51) 0.063 P ¼ 0.052), we did not find an association with the Snore
Outcomes Survey score as a continuous (HR ¼ 0.989; 95%
SD, standard deviation.
a
Assess the difference in mean natural log (Ln)-transformed markers CI ¼ 0.976–1.002; P ¼ 0.096) and different cut-point of 55
(HR ¼ 1.199; 95% CI ¼ 0.667–2.154; P ¼ 0.545) (Table 3).
P ¼ 0.008) and ODI3 (8.68 6 6.79 vs 7.42 6 7.94 events/ In Table 4, the RTC case (only one traffic collision) and
hr; P ¼ 0.007) (Table 2). Moreover, the RTC group had a RTC case (more than two traffic collisions) groups showed
slight decrease in Snore Outcomes Survey score in com- an increase of ODI4 (HR ¼ 1.251; 95% CI 1.000–1.564;
parison with the non-RTC group (71.46 6 14.92 vs P ¼ 0.05; and HR ¼ 1.650; 95% CI 1.055–2.581;
73.99 6 13.83 events/h; P ¼ 0.077) (Table 2). P ¼ 0.028, respectively) and ODI3 (HR ¼ 1.296; 95% CI
The researchers also performed one-way ANOVA and 1.000–1.679; P ¼ 0.05; and HR ¼ 1.831; 95% CI 1.093–
tests for trend in order to examine the association of the 3.067; P ¼ 0.022, respectively) in comparison with the non-

Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194

by guest
on 04 May 2018
 272
Figure 2. Hazard ratio, 95% CI and P-value for RTC associated with various risk factors in bus drivers from 2005 to 2010.
RTC group, after adjustment for age, education, CVD his-
tory, caffeine drinking, use of sleeping pills, bus driving ex-
perience and shift work. It was found that there was a
positive trend of ODI levels when stratified by the number
of traffic collisions experienced; participants with more traf-
fic collisions were more likely to have higher ODI levels.
In the extended Cox regression models used to assess
hazards of recurrent RTC events, the counting process
Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194
by guest
on 04 May 2018
International Journal of Epidemiology, 2017, Vol. 46, No. 1
approach and stratified Cox model approach both found
that ODI4 and ODI3 levels were associated with increased
risk of recurrent RTC events among professional drivers
(proportional means model: HR ¼ 1.310; 95% PCI 1.094–
1.564; P ¼ 0.003; and HR ¼ 1.364; 95% CI 1.103–1.686;
P ¼ 0.004, respectively; PWP gap time model: HR ¼ 1.334;
95% CI 1.101–1.617; P ¼ 0.003; and HR ¼ 1.412; 95%
CI 1.131–1.763; P ¼ 0.002, respectively) (Table 5).
 International Journal of Epidemiology, 2017, Vol. 46, No. 1 273

Table 3. Hazard ratio and 95% CI for RTC by PSQI score, Snore Outcomes Survey score, ESS score and ODI level in bus drivers
from 2005 to 2010

Dependent variable Univariate analysis Model 1b

HR 95% CI P-value HR 95% CI P-value

Global PSQI score

1 As a continuous variable 1.028 0.962 1.099 0.414 0.986 0.919 1.058 0.696
2 As a categorical variable (ref. < 5)
5–9.9 1.205 0.808 1.798 0.361 1.026 0.683 1.542 0.901
 10 1.471 0.739 2.929 0.271 1.010 0.489 2.089 0.978
Snore Outcomes Survey score
3 As a continuous variable 0.988 0.975 1.001 0.065 0.989 0.976 1.002 0.096
4 As a categorical variable (ref. > 60)
 60 1.603 1.030 2.494 0.037 1.558 0.997 2.434 0.052
5 As a categorical variable (ref. > 55)
 55 1.258 0.704 2.246 0.439 1.199 0.667 2.154 0.545
ESS score
6 As a continuous variable 1.030 0.987 1.076 0.177 1.005 0.959 1.053 0.832
7 As a categorical variable (ref. < 11)
 11 1.153 0.693 1.919 0.583 0.997 0.593 1.674 0.990
ODI4
8 As a continuous variablea 1.257 1.035 1.526 0.021 1.328 1.087 1.622 0.005
9 As a categorical variable (ref. < 6.0)
6.0–9.9 (cut points: 75th) 1.859 1.153 2.996 0.011 1.951 1.204 3.161 0.007
 10 (cu -points: 90th) 1.275 0.706 2.305 0.421 1.554 0.850 2.841 0.152
ODI3
10 As a continuous variablea 1.311 1.047 1.642 0.018 1.398 1.109 1.761 0.005
11 As a categorical variable (ref. < 9.0)
9.0–14.9 (cut points: 75th) 2.047 1.285 3.260 0.003 2.155 1.346 3.449 0.001
 15.0 (cut points: 90th) 1.595 0.911 2.794 0.102 1.950 1.099 3.458 0.022
CT88
12 As a continuous variablea 1.101 0.993 1.222 0.069 1.113 0.998 1.241 0.055
13 As a categorical variable (ref. < 0.5)
0.5–3.4 (cut points: 75th) 1.424 0.876 2.315 0.154 1.362 0.819 2.264 0.233
 3.5 (cut points: 90th) 1.382 0.765 2.495 0.283 1.750 0.962 3.184 0.067

a
We used Cox models to relate natural log (Ln)-transformed continuous variables.
b
Model 1: adjusted for age, education, caffeine drinks used, sleeping pills used, bus driving experience, shift work and CVD history.

Table 4. Hazard ratio and 95% CI for number of RTC events by ODI level in bus drivers from 2005 to 2010a

ODI4b ODI3b CT88b

HR 95% CI P-value HR 95% CI P-value HR 95% CI P-value

RTC drivers (only one traffic collision; n ¼ 85) 1.251 1.000 1.564 0.050 1.296 1.000 1.679 0.050 1.120 0.992 1.265 0.068
vs non-RTC drivers (n ¼ 875)
RTC drivers (more than two traffic collisions; 1.650 1.055 2.581 0.028 1.831 1.093 3.067 0.022 1.097 0.858 1.402 0.460
n ¼ 22) vs non-RTC drivers (n ¼ 875)

a
Adjusted for age, education, caffeine drinks used, sleeping pills used, bus driving experience, shift work and CVD history.
b
We used Cox models to relate natural log (Ln)-transformed continuous variables.

Discussion consideration of recurrent traffic collisions events, in order

This study is new within sleeping accident research. It to predict which subset of drivers are at an increased RTC
adopted a comprehensive and cohort study design, includ- risk. The major findings of this study are that there was a
ing the validated sleep assessment, systematic traffic colli- positive association among professional drivers between
sions data collection, control for confounding factors and both the 6-year risk of first RTC event only and recurrent

Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194

by guest
on 04 May 2018
 274
Table 5. Analysis of survival analysis with recurrent RTC events by ODI level in bus drivers from 2005 to 2010a
ODI4b
HR 95% CI
Counting process approach
The intensity model (the model-based 1.310 1.074 1.298 0.008
covariance estimate)
The proportional means model (the robust 1.310 1.094 1.564 0.003
sandwich covariance estimate)
Stratified Cox model approach (conditional model)
PWP total time model 1.334 1.101 1.615 0.003
PWP gap time model 1.334 1.101 1.617 0.003
a
Adjusted for age, education, caffeine drinks used, sleeping pills used, bus driving experience, shift work and CVD history.
b
We used Cox models to relate natural log (Ln)-transformed continuous variables.
RTC events, and ODI levels. Furthermore, there was a
trend of increased ODI levels when stratified by the num-
ber of RTC experienced in comparison with the non-RTC
group. Additionally, the determinants of educational sta-
tus, sleeping pill use and irregular shift work were associ-
ated with increased RTC risks, but age and bus driving
experience were associated with decreased RTC risks.
A persistent challenge was how to identify those at risk
for drowsy-driving road crashes in an easy, reliable and
cost-effective way. According to articles on the topic of
screening tools for SDB, overnight pulse oximetry is the
most important instrument to use for identifying SDB in
many portable multichannel sleep apnoea screening de-
vices. ODI had a strong correlation with the parameters
measuring SDB from PSG.25–27 A recent study using
ODI3 ¼ 12.5 and ODI4 ¼ 6.5 events/h as cutting points for
screening in Taiwan professional drivers with Apnea-hypo-
pnea index  15 events/h, showed that these provide excel-
lent sensitivities (0.95 and 0.95, respectively) and good
specificities (0.85 and 0.82, respectively).27 For partici-
pants with a cut-off of ODI4 > 10, there were high sensitiv-
ity (93.3%) and specificity (74.6%) to detect moderate and
severe OSA (AHI > 15 events per hour).25 Meanwhile,
overnight pulse oximetry is cost-effective and shows sub-
stantial accuracy and applicability in the general popula-
tion and among workers.25–28
Drivers with OSA frequently complain of excessive day-
time fatigue and sleepiness because of non-restorative and
continuously disrupted sleep. A previous study showed
that sleep restriction or sleepiness induced substantial per-
formance degradation even though time awake (8 h) and
session driving times (105 min) were relatively short.2
Three meta-analysis studies have clearly demonstrated
increased RTC risk for individuals with sleep ap-
noea.18,29,30 Moreover, previous research has also shown
that treatment of sleep apnoea with continuous positive
Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194
by guest
on 04 May 2018
International Journal of Epidemiology, 2017, Vol. 46, No. 1
ODI3b CT88b
P-value HR 95% CI P-value HR 95% CI P-value
1.364 1.084 1.715 0.008 1.093 0.980 1.218 0.110
1.364 1.103 1.686 0.004 1.093 0.986 1.210 0.090
1.411 1.132 1.760 0.002 1.102 0.993 1.613 0.067
1.412 1.131 1.763 0.002 1.103 0.994 1.225 0.065
airway pressure devices consistently improved driver per-
formance and reduced the number of accidents.31–33
The common sleep assessment questionnaires, such as
the PSQI, Snore Outcomes Survey scores and ESS, were
also used in this study, even though self-assessment scales
were not always applicable to everyone. Snoring is caused
by partial obstruction of the upper airways, and is con-
sidered a diagnostic marker or precursor of SDB.21,22 In
the present study, snoring as measured by the Snore
Outcomes Survey questionnaire shows a slight relationship
to risk of RTC. However, the results were not consistent
with Snore Outcomes Survey scores as continuous data
and within different cut-points.
We still recommend that the use of objective measure-
ment of overnight pulse oximeter is better than self-report
snoring outcomes. By using the PSQI, Sabbagh-Ehrlich
et al.34 found that truck drivers with moderate to severe
sleep quality problems were 2.9 times more likely to be
involved in severe crashes.34 In another study, drivers’ in-
somnia (poor sleep quality) tended to be associated with a
high risk of sleep-related car accidents.35 However, the
present study did not find an association between PSQI
and RTC. Another previous study showed that an increase
of 1 unit of ESS was associated with a covariate-adjusted
4.4% increase of having at least one accident
(P < 0.001).36 In a study with large samples of drivers, the
sleepiness of drivers (ESS  11) was associated with a
higher risk of multiple vehicle accidents (OR ¼ 1.31; 95%
CI 1.14–1.51), and of a single vehicle accident (OR ¼ 1.29;
95% CI 1.14–1.45).13 In a large survey of Thai commercial
bus or truck drivers, drivers with an ESS score  11 were
more likely to have accidents (OR ¼ 1.68; 95% CI 1.41–
2.00).37 However, other authors and the present study
have not found any relationship between ESS and RTC
risk.35,38–40 The disparity between these reports of the ESS
correlating or not correlating with RTC may also lie in the
 International Journal of Epidemiology, 2017, Vol. 46, No. 1
methodology of the individual studies. Additionally, based
on observations about the drivers’ behaviours during this
study period, these sleep assessment questionnaires such as
PSQI and ESS are not eligible for effective SDB screening.
This is due to some of the following reasons: (i) short-term
naps often occurred in bus drivers’ waiting rooms, leading
to excessive daytime sleepiness; (ii) some physiological
adaptation may be developed to cope with their special
working patterns, especially for experienced professional
drivers; and (iii) these drivers during irregular duty periods
often slept alone to avoid disturbing other members of the
family. All of these behaviours could weaken the discrimi-
nating power of the questionnaires.
The nature of the work of professional drivers is charac-
terized by irregular work shifts, long hours of driving, sed-
entary restricted postures, rare exercise, long-term sleep
deficiency and a unique working environment,41–43 which
may have lasting effects on sleep problems and lead to an
increased risk of RTC. Our results indicate that the charac-
teristics such as age and bus driving experience may be
associated with traffic crashes in drivers. In a recent study
among regular highway drivers, the populations at the
highest risk of accidents were those of young age (18–30
years) (OR ¼ 1.42; P < 0.001) and professional drivers
(OR ¼ 1.15; P < 0.05).44 Additionally, a study which
updated and expanded a model of factors associated with
sleepy driving suggested that younger, less experienced
drivers who drive long distances appear to be particularly
at risk from the effects of sleepiness on driving.7 Another
relevant driver study was performed in Taiwan. It investi-
gated at-fault accident risk by driving experience, display-
ing a U-shaped outcome.45 Novice drivers whose driving
experience was less than 3 years had the highest at-fault ac-
cident rate (12.4%), and those whose driving experience
was from 6 to 14 years possessed a relatively low at-fault
accident risk.45 Driving experience of drivers appears to in-
fluence accident risk with a positive correlation with the
number of driving accidents.
The current findings have several health policy and mar-
ket implications. Given the large numbers of professional
drivers recruited to this study, the TBDCS study clearly
provides results that will have important implications in
road safety. The usefulness of a portable pulse oximeter in
relation to RTC is important to support health service or-
ganizations in arranging a system for screening for SDB in
professional drivers. Additionally, our results can provide
evidence for future research about portable devices or mo-
bile health applications that can successfully limit drivers’
RTC and thus improve their health and safety.
We fully acknowledge that the methodology of this in-
vestigation has some limitations and unavoidable uncer-
tainties. First, PSG examination is the gold standard for
Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194
by guest
on 04 May 2018
275
the diagnosis of SDB. However, due to the high cost and a
limited availability of sleep PSG labs, it would be difficult
to provide screening for SDB for most professional drivers.
Thus, this study used practical alternative screening tools
such as overnight pulse oximetry and self-reported sleep
time to make a diagnosis of SDB. Second, clinicians find
patients with severe chronic lung disease where the oxim-
eter reading may be reduced. This may have affected our
goals of monitoring for SDB. Although this study did not
find subjects with severe lung disease from the self-
questionnaire, it is still possible that under-reporting re-
sults happened. Third, these sleep assessment tools were
tested only once, at the beginning of the study. Meanwhile,
we do not know whether treatments for sleep disorders
were received among these participants during the follow-
up periods. This could reduce the consequence of sleeping-
related RTC and cause underestimation in our results.
Fourth, this study could not find appropriate methods of
assessing sleep quality and daytime sleepiness via both sub-
jective and objective tools. Finally, the inclusion of solely
male professional drivers in the study restricts the inter-
pretation of the results with females. Moreover, our out-
comes should not be fully extrapolated to the general
population.
In conclusion, there is an increased risk of RTC associ-
ated with ODI for a sign of SDB with professional drivers.
No association was found using sleep assessment question-
naires with sleeping quality and daytime sleepiness out-
comes. The findings of this investigation suggest that the
overnight pulse oximeter is an effective sleep assessment
tool to assess for the risk of RTC. A regular screening sys-
tem of overnight pulse oximeters to protect against sleep
disorders among professional drivers should be supported
by all health service organizations in the future.
Supplementary Data
Supplementary data are available at IJE online.
Funding
This study was partly supported by the National Health Research
Institutes of Taiwan (98-EO-PP01, 99-EO-PP01, 00-EO-PP01, EO-
101-PP-01, EO-102-PP-01 and EO-103-PP-01) and Institute of
Labor, Occupational Safety and Health (IOSH96-M102 and
IOSH97-M102), Taiwan. The funders had no role in study design,
data collection and analysis, decision to publish or preparation of
the manuscript.
Acknowledgments
The authors thank the administrators and drivers in the Bus
Company for their participation and cooperation. The current ana-
lysis was based on data provided by the Collaboration Center of
 276
Health Information Application, Ministry of Health and Welfare,
Executive Yuan, Taiwan
Conflict of interest: The authors have declared that no competing
interests exist.
References
1. WHO. WHO Global Status Report On Road Safety 2013:
Supporting A Decade Of Action: World Health Organization;
2013.
2. Philip P, Sagaspe P, Moore N, et al. Fatigue, sleep restriction and
driving performance. Accident; analysis and prevention 2005;
37: 473–8.
3. Connor J, Whitlock G, Norton R, Jackson R. The role of driver
sleepiness in car crashes: a systematic review of epidemiological
studies. Accident; analysis and prevention 2001; 33: 31–41.
4. Connor J, Norton R, Ameratunga S, et al. Driver sleepiness and
risk of serious injury to car occupants: population based case
control study. BMJ 2002; 324: 1125.
5. Smolensky MH, Di Milia L, Ohayon MM, Philip P. Sleep dis-
orders, medical conditions, and road accident risk. Accident;
analysis and prevention 2011; 43: 533–48.
6. Lyznicki JM, Doege TC, Davis RM, Williams MA. Sleepiness,
driving, and motor vehicle crashes. Council on Scientific Affairs,
American Medical Association. JAMA 1998; 279: 1908–13.
7. Phillips RO, Sagberg F. Road accidents caused by sleepy drivers:
Update of a Norwegian survey. Accident; analysis and preven-
tion 2013; 50: 138–46.
8. Philip P, Vervialle F, Le Breton P, Taillard J, Horne JA. Fatigue,
alcohol, and serious road crashes in France: factorial study of na-
tional data. BMJ 2001; 322: 829–30.
9. Laberge-Nadeau C, Maag U, Bellavance F, et al. Wireless tele-
phones and the risk of road crashes. Accident; analysis and pre-
vention 2003; 35: 649–60.
10. Williamson AM, Feyer A-M, Mattick RP, Friswell R, Finlay-
Brown S. Developing measures of fatigue using an alcohol com-
parison to validate the effects of fatigue on performance.
Accident Analysis & Prevention 2001; 33: 313–26.
11. Punjabi NM. The epidemiology of adult obstructive sleep apnea.
Proceedings of the American Thoracic Society 2008; 5: 136–43.
12. Pack AI, Maislin G, Staley B, et al. Impaired performance in
commercial drivers: role of sleep apnea and short sleep duration.
Am J Respir Crit Care Med 2006; 174: 446–54.
13. Howard ME, Desai AV, Grunstein RR, et al. Sleepiness, sleep-
disordered breathing, and accident risk factors in commercial ve-
hicle drivers. Am J Respir Crit Care Med 2004; 170: 1014–21.
14. Vennelle M, Engleman HM, Douglas NJ. Sleepiness and sleep-
related accidents in commercial bus drivers. Sleep Breath 2010;
14: 39–42.
15. Hui DS, Ko FW, Chan JK, et al. Sleep-disordered breathing and
continuous positive airway pressure compliance in a group of
commercial bus drivers in Hong Kong. Respirology 2006; 11:
723–30.
16. Chesson A, Ferber RA, Fry JM, et al. Practice parameters for the
indications for polysomnography and related procedures. Sleep
1997; 20: 406–22.
17. Qaseem A, Dallas P, Owens DK, Starkey M, Holty J-EC,
Shekelle P. Diagnosis of obstructive sleep apnea in adults: a
Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194
by guest
on 04 May 2018
International Journal of Epidemiology, 2017, Vol. 46, No. 1
clinical practice guideline from the American College of
Physicians. Annals of internal medicine 2014; 161: 210–20.
18. Zhang T, Chan AH. Sleepiness and the risk of road accidents for
professional drivers: A systematic review and meta-analysis of
retrospective studies. Safety Science 2014; 70: 180–8.
19. Buysse DJ, Reynolds CF, 3rd, Monk TH, Berman SR, Kupfer DJ.
The Pittsburgh Sleep Quality Index: a new instrument for psychi-
atric practice and research. Psychiatry research 1989; 28:
193–213.
20. Tsai PS, Wang SY, Wang MY, et al. Psychometric evaluation of
the Chinese version of the Pittsburgh Sleep Quality Index
(CPSQI) in primary insomnia and control subjects. Quality of
life research : an international journal of quality of life aspects of
treatment, care and rehabilitation 2005; 14: 1943–52.
21. Gliklich RE, Wang PC. Validation of the snore outcomes survey
for patients with sleep-disordered breathing. Archives of
otolaryngology–head & neck surgery 2002; 128: 819–24.
22. Chen NH, Li HY, Gliklich RE, Chu CC, Liang SC, Wang PC.
Validation assessment of the Chinese version of the Snore
Outcomes Survey. Quality of life research : an international jour-
nal of quality of life aspects of treatment, care and rehabilitation
2002; 11: 601–7.
23. Johns MW. Daytime sleepiness, snoring, and obstructive sleep
apnea. The Epworth Sleepiness Scale. Chest 1993; 103: 30–6.
24. Chen NH, Johns MW, Li HY, et al. Validation of a Chinese ver-
sion of the Epworth sleepiness scale. Quality of life research : an
international journal of quality of life aspects of treatment, care
and rehabilitation 2002; 11: 817–21.
25. Chung F, Liao P, Elsaid H, Islam S, Shapiro CM, Sun Y. Oxygen
desaturation index from nocturnal oximetry: a sensitive and spe-
cific tool to detect sleep-disordered breathing in surgical patients.
Anesth Analg 2012; 114: 993–1000.
26. Niijima K, Enta K, Hori H, Sashihara S, Mizoue T, Morimoto Y.
The usefulness of sleep apnea syndrome screening using a port-
able pulse oximeter in the workplace. J Occup Health 2007; 49:
1–8.
27. Ting H, Huang RJ, Lai CH, et al. Evaluation of candidate meas-
ures for home-based screening of sleep disordered breathing in
Taiwanese bus drivers. Sensors 2014; 14: 8126–49.
28. Netzer N, Eliasson AH, Netzer C, Kristo DA. Overnight pulse
oximetry for sleep-disordered breathing in adults: a review.
Chest 2001; 120: 625–33.
29. Tregear S, Reston J, Schoelles K, Phillips B. Obstructive sleep
apnea and risk of motor vehicle crash: systematic review and
meta-analysis. Journal of clinical sleep medicine : JCSM : official
publication of the American Academy of Sleep Medicine 2009;
5: 573–81.
30. Ellen RL, Marshall SC, Palayew M, Molnar FJ, Wilson KG,
Man-Son-Hing M. Systematic review of motor vehicle crash risk
in persons with sleep apnea. Journal of clinical sleep medicine :
JCSM : official publication of the American Academy of Sleep
Medicine 2006; 2: 193–200.
31. Krieger J, Meslier N, Lebrun T, et al. Accidents in obstructive
sleep apnea patients treated with nasal continuous positive air-
way pressure: a prospective study. The Working Group
ANTADIR, Paris and CRESGE, Lille, France. Association
Nationale de Traitement a Domicile des Insuffisants
Respiratoires. Chest 1997; 112: 1561–6.
 International Journal of Epidemiology, 2017, Vol. 46, No. 1
32. Engleman HM, Asgari-Jirhandeh N, McLeod AL, Ramsay CF,
Deary IJ, Douglas NJ. Self-reported use of CPAP and benefits of
CPAP therapy: a patient survey. Chest 1996; 109: 1470–6.
33. Komada Y, Nishida Y, Namba K, Abe T, Tsuiki S, Inoue Y.
Elevated risk of motor vehicle accident for male drivers with ob-
structive sleep apnea syndrome in the Tokyo metropolitan area.
Tohoku J Exp Med 2009; 219: 11–6.
34. Sabbagh-Ehrlich S, Friedman L, Richter E. Working conditions
and fatigue in professional truck drivers at Israeli ports. Injury
Prevention 2005; 11: 110–4.
35. Philip P, Chaufton C, Orriols L, et al. Complaints of poor sleep
and risk of traffic accidents: a population-based case-control
study. PloS one 2014; 9: e114102.
36. Powell NB, Schechtman KB, Riley RW, Guilleminault C, Chiang
RP-y, Weaver EM. Sleepy driver near-misses may predict acci-
dent risks. Sleep 2007;30:331–42.
37. Leechawengwongs M, Leechawengwongs E, Sukying C,
Udomsubpayakul U. Role of drowsy driving in traffic accidents:
a questionnaire survey of Thai commercial bus/truck drivers.
J Med Assoc Thai 2006;89:1845–50.
38. Masa JF, Rubio M, Findley LJ. Habitually sleepy drivers have a
high frequency of automobile crashes associated with respiratory
disorders during sleep. Am J Respir Crit Care Med 2000;162:
1407–12.
Downloaded from https://academic.oup.com/ije/article-abstract/46/1/266/2617194
by guest
on 04 May 2018
277
39. Young T, Blustein J, Finn L, Palta M. Sleepiness, driving and ac-
cidents: sleep-disordered breathing and motor vehicle accidents
in a population-based sample of employed adults. Sleep
1997;20:608–13.
40. Amra B, Dorali R, Mortazavi S et al. Sleep apnea symptoms and
accident risk factors in Persian commercial vehicle drivers. Sleep
Breath 2012;16:187–91.
41. Bunn TL, Slavova S, Struttmann TW, Browning SR. Sleepiness/
fatigue and distraction/inattention as factors for fatal vs nonfatal
commercial motor vehicle driver injuries. Accid Anal Prev
2005;37:862–69.
€ B, Ozkan
42. Oz € T, Lajunen T. An investigation of the relationship
between organizational climate and professional drivers’ driver
behaviours. Saf SCi 2010;48:1484–89.
43. Taylor AH, Dorn L. Stress, fatigue, health, and risk of
road traffic accidents among professional drivers: the contribu-
tion of physical inactivity. Annu Rev Public Health 2006;27:
371–91.
44. Philip P, Sagaspe P, Lagarde E et al. Sleep disorders and acciden-
tal risk in a large group of regular registered highway drivers.
Sleep Med 2010;11:973–79.
45. Tseng C-M. Social-demographics, driving experience and yearly
driving distance in relation to a tour bus driver’s at-fault accident
risk. Tour Manag 2012;33:910–15.