Skin temperature as a predictor of on‐the‐road driving performance in people with central disorders of hypersomnolence

Excessive daytime sleepiness is the core symptom of central disorders of hypersomnolence (CDH) and can directly impair driving performance. Sleepiness is reflected in relative alterations in distal and proximal skin temperature. Therefore, we examined the predictive value of skin temperature on driving performance. Distal and proximal skin temperature and their gradient (DPG) were continuously measured in 44 participants with narcolepsy type 1, narcolepsy type 2 or idiopathic hypersomnia during a standardised 1‐h driving test. Driving performance was defined as the standard deviation of lateral position (SDLP) per 5 km segment (equivalent to 3 min of driving). Distal and proximal skin temperature and DPG measurements were averaged over each segment and changes over segments were calculated. Mixed‐effect model analyses showed a strong, quadratic association between proximal skin temperature and SDLP (p < 0.001) and a linear association between DPG and SDLP (p < 0.021). Proximal skin temperature changes over 3 to 15 min were predictive for SDLP. Moreover, SDLP increased over time (0.34 cm/segment, p < 0.001) and was higher in men than in women (3.50 cm, p = 0.012). We conclude that proximal skin temperature is a promising predictor for real‐time assessment of driving performance in people with CDH.


| INTRODUCTION
Excessive daytime sleepiness is the core symptom in people with central disorders of hypersomnolence (CDH; i.e., narcolepsy and idiopathic hypersomnia [IH]).As sleepiness behind the wheel is reported as one of the leading causes of automotive collisions, people with narcolepsy and IH are at increased risk of impaired fitness to drive (Philip et al., 2010;Pizza et al., 2015).Driving fitness in people with CDH is usually evaluated by a qualified physician.In some countries this is supported by a self-reported sleepiness questionnaire and/or an objective evaluation of daytime sleepiness using the Maintenance of Wakefulness Test (MWT; Pisarek et al., 2019).Inconsistencies in evaluation methods between countries are due to on-going debate about the MWT as a valid predictor of impaired fitness to drive (Bijlenga, Overeem, et al., 2022;Wise, 2006).We showed a low correlation between the MWT and objective driving performance in a previous study and concluded that the predictive performance of the MWT and other widely used vigilance tests was, at best, debatable (Bijlenga, Urbanus, et al., 2022).However, accuracy of assessment is critical, as implications for driving can have a psychosocial impact on people (Raggi et al., 2019).The development of more accurate measures to predict driving fitness in CDH is therefore much needed.
Body and skin temperature changes may be real-time assessors of sleepiness behind the wheel as they correlate highly with subjective and objective sleepiness.Studies investigating circadian changes in body temperature have reported a decrease in core body temperature preceding sleep onset due to peripheral vasodilation.The subsequent blood flow to the skin increases overall skin temperature (Gradisar & Lack, 2004;Kräuchi et al., 1999;Kräuchi & Wirz-Justice, 1994).However, skin temperature varies between distal (i.e., hands and feet) and proximal (i.e., abdomen, mid-thighs, and infraclavicular upper chest) skin areas (Kräuchi et al., 2000).Distal skin temperature (T dist ) is generally lower than proximal skin temperature (T prox ) during wakefulness, resulting in a lower and negative distalto-proximal gradient (DPG).A relatively large increase in T dist , compared to T prox , precedes sleep-onset which brings both temperatures closer together.This results in an increase of the DPG, which is associated with individuals being more prone to fall asleep.Increased peripheral skin temperature per se is also associated with shorter sleep onset latencies (SOLs; Kräuchi et al., 2000).
Compared to healthy controls, people with narcolepsy type 1 (NT1) have higher T dist and DPG throughout the day, which is related to objective sleepiness (Fronczek, Overeem, et al., 2006).
Treatment with sodium oxybate may partially normalise skin temperature in people with NT1 to a level similar to control participants (van der Heide et al., 2015).Even though baseline skin temperature seems to be affected in people with narcolepsy, sleep-wake thermoregulation is similar to healthy control participants.Manipulation of skin temperature influences sleep propensity of both healthy individuals and individuals with NT1 (Fronczek et al., 2008;Raymann et al., 2005).This effect of skin temperature manipulation on sleepiness also indicates the direction and causality of the relationship between the two.Skin temperature, sleepiness and sustained attention are interrelated.Relatively high T prox due to natural fluctuations during the day is associated with worse performance on a sustained attention to response task (SART) and vice versa (Romeijn & Van Someren, 2011).Worsening of sustained attention over time is an essential factor in driving fitness, which occurs naturally even in healthy individuals (Ting et al., 2008).
However, people with CDH perform worse than healthy controls on vigilance tasks and in driving performance tests (Findley et al., 1995;Fronczek, Middelkoop, et al., 2006;Thomann et al., 2014).This supports the necessity of accurately assessing driving fitness in people with CDH.
As both sleepiness and vigilance have been linked to fluctuations in skin temperature and driving performance, we hypothesised that skin temperature would be an accurate measure in evaluating driving fitness.We, therefore, examined (1) whether there is an association between skin temperature and on-the-road driving performance and, if so; (2) whether skin temperature can predict on-the-road driving performance in people with CDH.(Bijlenga, Urbanus, et al., 2022).

| Procedures
The test day consisted primarily of a 1-h highway driving test.Sensors to measure skin temperature were attached to different body areas, after which the participants were driven to the starting point on the highway.
After practising for 15 min, the driving test officially started.Participants were never allowed to practice for longer than 15 min, to maintain equal time-on-task effects.Halfway through the driving test, participants exited the highway to re-enter in the opposite direction and drive back to the starting point.Before and after the driving test, changes in symptom severity were assessed.As described in our previous study, all participants completed the SART, an objective measure to assess vigilance in CDH and the Karolinska Sleepiness Scale (KSS), a subjective measure of sleepiness that is sensitive to fluctuations (Bijlenga, Urbanus, et al., 2022).A timeline of all procedures is depicted in Figure 1.

| On-the-road driving test
Driving performance was assessed using a standardised on-the-road highway-driving test (O'Hanlon, 1984;Ramaekers, 2017).In this test, participants drove a specially instrumented car for $1 h over a 100-km (61-mile) primary highway circuit with a return halfway, accompanied by a licensed driving instructor having access to dual controls (brakes and accelerator).The participants' task was to drive with a steady lateral position between the delineated boundaries of the slower (right) traffic lane, while maintaining a constant speed of 95 km/h (58 mph).Participants were allowed to deviate from those instructions only to pass a slower vehicle, and to leave and re-enter the highway at the mid-circuit turnaround point.Participants were instructed beforehand to terminate the test by stopping the car on the road shoulder if they have doubts about their competence to continue safely.However, if they failed to do this and the driving instructor judged their performance to become unsafe, the participants could be ordered to stop the vehicle.The speed of the vehicle and the lateral position were constantly measured by a camera that was mounted on the roof of the car and saved on board for offline processing.After removing disturbances and possible overtaking manoeuvres, a mean and standard deviation of the lateral position (SDLP) in cm were derived over segments of 5 km (equal to $3 min of driving).This resulted in 20 SDLP values per participant.SDLP is a well-known measure of 'weaving' or road tracking error and is widely used in studies to measure driving performance under specific conditions, e.g., when under the influence of drugs or alcohol (Brooks-Russell et al., 2021;Jongen et al., 2017).A more detailed explanation and visualisation of the technique is published by Ramaekers (2017).
In our previous study, a cut-off SDLP value of 19.09 cm was set to differentiate between drivers who were or were not considered to be at increased risk of impaired driving (Bijlenga, Urbanus, et al., 2022).For this study, the same cut-off value was used.Participants who were asked to end the driving test prematurely by the driving instructor, were classified as being at increased risk of impaired driving, irrespective of their mean SDLP value.Participants with SDLP values below the cut-off value, who terminated the driving test based on their own judgement, were not classified as drivers at increased risk of impaired driving, as they eliminated the risk by stopping the vehicle.

| Skin temperature
Skin temperature was measured using wireless Thermochron iButtons (type DS1921H-F5), which store temperature data between 15 and 46 C with an accuracy of ±1 C, a resolution of 0.125 C and a sampling rate of 1/min.Nine iButtons were placed on the body before the test started.To measure T prox , buttons were attached to the bilateral infraclavicular areas, bilateral mid-thighs and at the abdomen, 1 cm above the navel.A weighted average of these five buttons was calculated according to the procedures of Raymann et al. (2005).T dist was calculated by averaging measurements of four other iButtons, positioned on the back of both hands (at the thenar webspace) and on the arches of the feet.

| Visualisation of the association between temperature and driving performance
The data were arranged following the approach of Romeijn and van Someren (2011) to visualise the association between temperature and driving performance.For each participant who did not terminate the test prematurely, temperature measures of all 20 segments were sorted from the lowest to the highest measured temperature along with the corresponding SDLP value in that segment.This was done for all three temperature measures separately (T prox , T dist and DPG).
Temperature measures and SDLP values were then averaged over subjects for each sorted segment and plotted with corresponding standard errors of the means.
The original dataset, including data from participants who terminated the test prematurely, was used to analyse further the association between temperature and driving performance.Averages of T prox , T dist and DPG were compared between participants considered at increased risk and not at increased risk of impaired driving and between men and women using two-sample t-tests.

| Association confirmation and predictive value of temperature for driving performance
To quantify the associations and evaluate the predictive value of temperature on driving performance, multiple mixed-effect model F I G U R E 1 An overview of the design and outcome measures.Placement of iButtons differentiating proximal and distal areas, a timeline of the test day and a simplified representation of standard deviation of lateral position (SDLP) calculation.KSS, Karolinska Sleepiness Scale; SART, sustained attention to response task.
analyses were performed with SDLP as a dependent outcome variable.First, possible associations between temperature and driving performance found by visualisation were examined.To this end, temperature outcomes measured in the same segment and one segment prior to each SDLP outcome were included in the model as fixed effects.Baseline skin temperatures, as measured during the practice drive, were subtracted from all 20 temperature averages for each participant to account for individual differences in skin temperature.Segments were put into the model as repeated measures with a covariance structure (first-order autoregressive) to account for any association between measurements of the same participant.All three temperature parameters were included separately as fixed effects, with segment (time), diagnosis, age, medication use and sex as covariates.Intercepts for participants were added as a random effect to account for individual differences in driving performance.
To evaluate predictive value of skin temperature, similar mixedeffect model analyses were performed using different fixed-effect variables.Instead of including temperature measurements from the same segment and one segment prior to SDLP outcomes, changes in skin temperature over two, three, four and five segments prior to each measured SDLP outcome were calculated.The calculation of these four temperature variables for each of the three temperature measures, resulted in 12 temperature parameters.
For each model, possible interaction terms were included following initial evaluation of any significant main effect.All analyses were performed using the IBM Statistical Package for the Social Sciences (SPSS), version 24.0.Statistical significance was indicated at an α-value of 0.05 and values between 0.05-0.10were considered a trend.

| RESULTS
Data of 44 participants were included, as driving data of one participant were lost due to a technical issue.Patient characteristics, temperature and performance outcomes are summarised in Table 1.
Because of the small number of participants with narcolepsy type 2 (NT2) and IH and considering the substantial overlap in phenotype Narcolepsy type 2 7 ( 16) 5 ( 18) 2 ( 12) Idiopathic hypersomnia 6 ( 14) 2 ( 7) 4 (25) Years of driving experience, mean (SD, range) 19.9 (14. between the diseases (Fronczek et al., 2020), these groups were combined.On average, participants had 19.9 years of driving experience and drove 9855 km/year with large variability between subjects.Two participants reported not to drive at all anymore.One of which had always been a professional driver who quit not long before the test day occurred.The other participant had not driven for 2 years but adjusted to the car quickly and was considered fit to drive, as reported afterwards by the driving instructor.Women on average had significantly lower T dist and DPG and higher T prox than men during the driving test.This significant difference already existed during the baseline segment for T prox (men: 34.0 C, women: 34.7 C, p = 0.015) and DPG (men: À4.2 C, women: À2.5 C, p = 0.001).For T dist a trend was observed ( p = 0.059) in the difference between men (31.5 C) and women (30.5 C) at baseline.MWT SOL was significantly higher in women compared to men ( p = 0.012).KSS scores were significantly increased after the driving test (4.2) compared to before the driving test (3.1)(p = 0.006), which was primarily seen in men (3.3 to 4.6, p = 0.021).There were no significant differences in SART performance prior to and after the driving test or between men and women.

| Visualisation of the association between temperature and driving performance
Of the 44 participants, 17 had an SDLP value above the set cut-off value of 19.09 cm from our previous study (Bijlenga, Urbanus, et al., 2022).Average DPG and T dist values were slightly higher in drivers with increased risk of impaired driving (mean [SD] DPG: À3.0 Visualisation of the association between driving performance and skin temperature, based on the ranked dataset resulted in Figures 2a-c and 3a-c.Figure 2b shows that there is an optimum T prox at which SDLP is lowest, with higher or lower T prox values associated with higher SDLP values.Due to large variance in T dist between participants, this association with SDLP seems absent (Figure 2a).An association between DPG and SDLP was observed but less clear compared to T prox (Figure 2c).Because of significant differences in skin temperature between men and women (Table 1), similar plots were made to visualise the association differences between sexes.Higher DPG values (Figure 3b) and lower T prox (Figure 3c) in men compared to women are associated with slightly higher SDLP values.The quadratic relationship between T prox and SDLP remains in both sexes (Figure 3c).The mean SDLP values were not significantly different between men and women, but men had a mean SDLP value above and women below the cut-off value (Table 1).Moreover, sex differences in the relationship between T dist and SDLP were again less clear due to large variation (Figure 3a).The highest SDLP values measured in each segment were almost all attributed to men (Figure 3d, based on the original dataset).

| Association confirmation and predictive value of temperature for driving performance
The expected association of both T prox and DPG with SDLP were confirmed by mixed-model analyses (Table 2).Due to the non-linear relationship between T prox and SDLP, a quadratic term was included into the model.Both the quadratic and linear term of T prox in the same segment and one prior segment affected SDLP significantly.For DPG, the best model fit was found using only a linear term.An increase of 1 C in DPG in the segment prior to any measured SDLP value was associated with a significant increase in SDLP of 0.80 cm ( p = 0.013).T dist in one segment prior was not significantly associated with SDLP (p = 0.898).
The predictive value of T prox and DPG were further investigated, based on their confirmed association with SDLP.Non-linear mixedeffect model analyses showed significant predictive value of T prox changes over two, three and five segments prior to the measured SDLP value.DPG did not show any predictive value for SDLP, based on linear mixed-effect model analyses.Parameter estimates of each model are shown in Table 2, without estimates of the covariates that were included in every analysis.
There were main effects of diagnosis, sex, and segment (time) but not of medication use and age.Based on the lowest Akaike Information Criterion value, the best-fit model included changes in T prox over five prior segments, diagnosis, sex, and segment (time) without interaction effects (Table 2).Interaction effects between sex Â temperature and segment Â temperature did not increase the model's accuracy.An increase in segment (time) was highly associated with an increased SDLP of 0.34 cm/segment ( p < 0.001 in all models), showing that SDLP was also affected by time spent behind the wheel.
Sex differences were confirmed by the model, with on average 3.50 cm higher SDLP values for men than for women ( p = 0.012).

| DISCUSSION
We investigated the predictive value of real-time skin temperature measurements for on-the-road driving performance in people with CDH.T prox was the best predictor for SDLP in this group.An optimum T prox was found and a relative decrease or increase preceded worse driving performance, most likely caused by increased sleepiness and/or a decline in sustained attention.Driving performance could be best predicted by looking at change in T prox over five preceding segments (i.e. 15 min).These findings do not entirely align with previous results of decreased T prox being associated with increased sleepiness (Fronczek, Overeem, et al., 2006;Kräuchi et al., 1999).The present analyses point towards an optimum proximal temperature and suggest a quadratic rather than a linear relationship.While previous studies also found an association between T dist and sleepiness, we did not find this association concerning driving performance.
Another important finding is the difference between men and women, not only in average skin temperature but also in the relationship between skin temperature and SDLP.The SDLP values were slightly higher in men than in women, but not significantly different based on a two-sample t-test.The prediction model showed a higher slope for men, meaning that men performed worse in the driving test.
All the SDLP values above cut-off were attributed to men, which may be due to differences in driving styles (Taubman-Ben-Ari & Skvirsky, 2022).It is well-known that thermoregulation is different in men and women (Kim et al., 1998).However, differences in driving performance between men and women with sleep-wake disorders have not been confirmed in previous studies (Aldrich, 1989;Liu et al., 2018).A systematic review showed sex differences in the effect of sleep medication on driving performance for some hypnotic drugs, but baseline differences in driving performance were not reported (Verster & Roth, 2012).We found significantly longer MWT SOL in women compared to men, suggesting differences in baseline symptom severity between the groups.The difference in driving performance between men and women with CDH should be further investigated to draw definitive conclusions.
No conclusions can currently be drawn about differences in the relation between skin temperature and driving performance between diagnostic groups.The small sample sizes of people with NT2 and IH are a limitation of our study.Another limitation is the accuracy of iButton measurements given the small temperature changes observed.
As previously described, the accuracy of iButton measurements is 1 C, which can be even more negatively influenced by larger skinenvironment temperature gradients and thermal inertia (Van Marken Lichtenbelt et al., 2006).Our study's intra-and inter-person variation in T dist was high, which might be a consequence of less accurate measurements.In most laboratory studies investigating sleep, the environmental temperature is constant (Fronczek, Overeem, et al., 2006;Kräuchi et al., 1999).However, in our study, the temperature of the environment surrounding distal areas seems to be more different from As opposed to T dist measurements, T prox outcomes seem to be minimally influenced by measurement error.This might be because proximal skin areas were covered by clothing, reducing the skinenvironment temperature gradient.We found no association between baseline temperature and variation in T prox (results not shown), which confirms that ambient temperature did not significantly affect the observed changes in T prox .Moreover, we have looked at changes in skin temperature over at least 3 min, which is much higher than the reported 19 s response time of the iButtons.Therefore, it is expected that there is negligible effect of thermal inertia.This is substantiated by the significant differences in skin temperature we found between men and women, which indicate that despite the possible measurement error of the iButtons, the measurements were accurate enough to detect groups differences.The field set-up of this study is a limitation, and, at the same time, it suggests an important practical issue for future applications.
Our findings give rise to the idea that manipulation of T prox may counter sleepiness and decreased vigilance behind the wheel in people with CDH.Skin temperature manipulation but also temperature measurements as a warning system could possibly become valuable additions to existing applications after careful consideration of the limitations currently encountered.Eye tracking sensors and other advanced driver assistance systems, such as lane change detection and forward collision warning, are already available (Singh et al., 2010;Song et al., 2018;Su et al., 2017).However, skin temperature measurements seem to benefit from being informative for longer before unsafe situations occur and manipulation could eliminate the possible risk all together.However, further research is necessary to look into optimal timing and temperature, considering the effect of external temperature, individual and sex differences, and the U-shaped relationship we found before future applications can be developed.
Participants were 45 consecutive individuals aged 18-75 years with a diagnosis of CDH as defined by the third edition of the International Classification of Sleep Disorders (ICSD-3, 2014), who were referred to the Dutch sleep-wake centres of Kempenhaeghe and Stichting Epilepsie Instellingen Nederland (SEIN) to evaluate their fitness to drive between June 2015 and January 2017.Of note, driving tests performed as part of this study were not used for an official evaluation of fitness to drive.MWT SOL and self-reported Epworth Sleepiness Scale (ESS) outcomes were used for this evaluation at the time of this study.However, the ESS scores were not used as part of this research as they can easily be manipulated by participants to qualify as eligible to drive.Participants either did not have any changes in treatment in the 6 weeks before the study or did not use any medication at the time of the study.All had a valid driver's licence.The study was approved by the Medical Ethics Committee of Maastricht University and Maastricht Academic Hospital (NL50579.068.14).All participants signed informed consent before enrolment.Other outcomes of this study have been previously reported

Figure 1
Figure 1 displays the placement of iButtons and differentiation between distal and proximal areas.Temperature measurements were averaged over each driving segment, resulting in 20 T dist and 20 T prox values per participant.A DPG was calculated by subtracting T prox from T dist for each of the 20 driving segments.

[ 1
.7] C; T dist : 31.6 [2.0] C) than in non-increased risk drivers (mean [SD] DPG: À3.6 [1.9] C; T dist : 31.1 [1.8] C), but not significantly different ( p = 0.340 and p = 0.477, respectively).The T prox was very similar and not significantly different between increased risk (mean [SD] 34.5 [1.1] C) and non-increased risk (mean [SD] 34.7 [0.9] C) drivers (p = 0.633).Average years of driving experience and kilometres driven per year between increased risk (mean [SD] 21.3 [17.0]years and 8.7 [9.5] km Â 10 3 /year, respectively) and non-increased risk drivers (mean [SD] 18.9 [12.4] years and 10.5 [10.9] km Â 10 3 / year respectively) were not significantly different (p = 0.598 and p = 0.586 respectively).Two participants were asked to terminate the driving test prematurely by the driving instructor; they were both also classified as drivers at increased risk according to the set cut-off value (average SDLP of 23.1 and 30.4).Four participants terminated the driving test prematurely based on their own judgement, of whom one with a mean SDLP value above the cut-off value (24.7 cm), one with a mean SDLP value of exactly 19.09 cm and two with a mean SDLP value below the cut-off value (15.8 cm and 18.7 cm).
) Visualisation of the association between temperature measures and standard deviation of lateral position (SDLP) based on the edited dataset of ranked segments (mean ± SEM).Vertical error bars indicate the between subject variability of SDLP; horizontal error bars represent the between subject variability for temperature measures.Associations between (a) distal skin temperature (T dist ) and SDLP, (b) proximal skin temperature (T prox ) and SDLP, and (c) the distal-to-proximal gradient (DPG) and SDLP are depicted separately.

F
I G U R E 3 (a-d) Visualisation of sex differences in the association between skin temperature and standard deviation of lateral position (SDLP).Average SDLP and skin temperature values per segment based on the edited dataset of ranked segments (mean ± SEM) for (a) distal skin temperature (T dist ) (b) proximal skin temperature (T prox ) and (c) the distal-to-proximal gradient (DPG) are depicted separately.Vertical error bars indicate the between subject variability of SDLP; horizontal error bars represent the between subject variability for temperature measures.(d) A representation of SDLP values at each segment for each participant separately.skintemperature due to the lack of coverage and frequent movement of the iButtons on the hands and feet.Furthermore, even though tests were scheduled to start at the same time of day (3:00 p.m.) for each participant, they were randomly scheduled throughout the year with different outside temperatures.This may have affected the interperson variability of the temperature within the vehicle despite our efforts to maintain constant ambient temperatures.We did not measure ambient temperature during the tests, but we did correct for individual differences in baseline temperature.
T A B L E 2 Mixed model analyses of the association between skin temperature and standard deviation of lateral position (SDLP) and changes in skin temperature as predictors of SDLP (n = 44), including one best-fit model reported with estimates of covariates.