Studies were included in this review if they 1) investigated driving or aspects of driving as the exposure variable, in either an on-road or simulated setting, 2) assessed at least one physiological outcome in adults (18 years of age or older), and 3) included a control or comparison condition. There was no restriction regarding the research design (observational or intervention studies), study location, sample size, date of publication or duration of the driving intervention. Non-empirical manuscript types, including case reports, letters to the editor, editorials, and reviews as well as conference abstracts were excluded.

A systematic search for full-text research articles written in English was conducted in March 2016 using electronic databases: PsycINFO (1806-present), Medline (1946-present) and CINAHL, with no restriction on year of publication. Driving-related search terms (motor vehicle OR automobile OR commut* OR transport* OR travel OR driving OR driver) and stress response-related search terms (stress OR distress OR cortisol OR heart rate OR blood pressure OR anxiety OR inflammation OR affective OR cardiovascular OR physiological) were combined with an “AND” and searched within the abstract and/or title.

After removing the duplicates, the title and abstract of each study were assessed by MA and ineligible studies were removed. Backward referencing was then used to identify additional relevant studies. The full texts of remaining studies were assessed independently by two authors (MA, KE), and any disagreement was discussed involving a third author (DD) until consensus was reached among MA, KE and DD. The process of study selection is shown in the CONSORT diagram (Fig 2).

The following information from each paper was extracted (independently by MA and JS) and summarized in Table 1: authors, publication year, setting (where not specified, setting was assumed to be the first author’s affiliation), objectives of the study (e.g. investigating the physiological stress response to driving), study design (experimental vs. observational), participant characteristics (age, gender, and health status), sample size, protocol, physiological outcomes and results. We also extracted additional details on the measurement instruments, such as electrocardiography (ECG) for heart rate, and the timing of measurement. In order to demonstrate the overall patterns of associations, we further distilled study findings in Table 2, using symbols (↑ indicating a significant increase, ↓ for a significant decrease and a ↔ for no statistically significant change). Funding sources for each study were extracted. All data extracted were presented separately for simulator and on-road studies.

Due to heterogeneity in study design and outcome measures, a meta-analysis was neither possible nor appropriate, thus a systematic review detailing key findings of individual studies was conducted. Our review followed the PRIMSA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) reporting criteria.

The initial search generated 27,926 potentially relevant articles, as shown in Fig 2. After removing 631 duplicates, 27,295 papers remained. Of those, 27,182 were removed based on reading the title, 88 were removed based on reading the abstracts and an additional three papers were identified from backward referencing. Twenty-one papers were excluded after reading the full text, leaving a total of seven articles for inclusion in this review. Primary reasons for exclusion based on full text included lack of quantitative assessment of stress or comparison to a baseline or control. The majority of studies (5 of 7) were funded by government/research grants [18–22]. One study was sponsored by a Tobacco Research Council [23] and one did not specify funding sources [24].

The study by Ashton et al. [23] involved three levels of difficulty. Level 1 (the control) required participants to respond to light signals alone, Level 2 required participants to respond to light signals which were reinforced by movement of the car in the simulated film, and Level 3 involved participants responding to light signals, which at times conflicted with the movement of the car. Each session was 20 minutes in duration. The light signals corresponded to a certain action that the participant was required to complete, i.e. a green light signal indicated the participant should switch on their indicator, whilst a red light indicated that the brake pedal must be pressed. Subjects in the study by Seeman et al. [20] were seated in a simulated console, which resembled the driving seat of a car, with one of two films projected onto the wall in front of the subjects. The participants were required to ‘drive along’ with both films, with each sitting preceded by a two-hour “rest” or “recovery” period. The first film, titled “good driving strategies”, was 18 minutes long and involved steering, turning and changing lanes. The second film, titled “evasive action control”, was ten minutes long and involved breaking and rapid turning to avoid a crash. Yamaguchi et al. [21] incorporated two sub-studies, with Study 1 involving five minutes of baseline rest followed by 21 minutes of simulated driving and Study 2 including an additional task, which required input of a series of numbers into a touch panel, designed to replicate the use of a navigation device. The participants within the study by Yamakoshi et al. [22] were required to sit in a driving seat and rest for ten minutes. The participants were then instructed to drive the simulator monotonously for 120 minutes at a set speed, as if they were actually driving. The driving session ended prematurely if the participant left the specified lane.

The study by Aronsson et al. [18] required bus drivers to drive predetermined “moderately stressful” routes for their usual eight-hour shift. This session was then compared with eight hours spent viewing educational television programs. Participants within the Bellet et al. [19] study were required to drive two hours and this was compared with two hours of seated rest. The bus drivers in the Sluiter et al. [24] study were all given an identical route to drive, which consisted of three days of driving, and this was compared with two days of rest.

Cortisol was measured in four studies, of which three measured urinary cortisol [18, 19, 24] and one plasma [20]; catecholamine excretion was measured in three studies all of which were urinary measures [18, 19, 24]; blood pressure was measured in three [18, 22, 23] and heart rate in two [22, 23]. Less commonly used (in one study only) markers of stress included normalised pulse volume, total peripheral resistance, salivary Chromogranin-A [22], as well as adrenocorticotrophic hormone (ACTH) [20] and salivary amylase, a marker of the sympathetic nervous system response [21]. The results of the physiological outcomes from each study are summarised in Table 2.

It is important to acknowledge that the conclusions from the current systematic review are based on a small number of studies with heterogeneous study designs and outcomes measured. Several limitations apply, but more to the empirical studies than the review process. Specifically, simulated driving studies may lack validity when compared with on-road studies, due to the fact that they do not have the same degree of risk as on-road driving [27]. Simulators may also result in different stress responses from on-road driving due to “simulator sickness” [28], a common syndrome that involves nausea, dizziness and vomiting whilst performing simulated tasks. Despite better validity and resemblance to everyday driving situations, findings from the on-road studies should still be interpreted with caution. Two studies were conducted in long distance bus drivers, suggesting that the results may not be generalizable to everyday car driving [18, 24]. It has been shown that dealing with potentially dangerous customers, night shifts and potential health complications associated with bus driving could lead to stress [29]. All three on-road studies measured urinary cortisol and catecholamines, which may not capture an acute stress response to driving, as previously discussed. Despite the use of a randomized crossover trial by Bellet et al. [19], the study is dated. The current driving environments, including traffic congestion [30] and distractions including navigators [21], are distinctively different from those at the time of the study (1969), limiting the generalisability of the findings to the driving experience nowadays. On the other hand, given that it is reasonable to assume that the driving situation today may involve more “driving impedance” than the 1960s, one may conclude that the significant stress response found in this study could be underestimated.

Given the problem of car dependency and the continuous increase in car use in many societies, as observed in Australia [31, 32], it is critical to improve the current understanding of the health implications of driving. Given the complex nature of driving with health and wellbeing [6], studies should continue to investigate various health outcomes of driving and their related mechanisms, including stress response to driving. Both epidemiological and physiological studies are indispensable in this inquiry. Here, we provide several suggestions for future physiological studies that aim to quantify stress response to driving exposure as a key step towards understanding the mechanisms through which driving may impair cardiovascular health.

One major gap in the literature is the lack of state-of-the-art outcome measures whilst performing on-road driving, such as salivary cortisol, heart rate and heart rate variability (HRV). The use of salivary sampling for cortisol will be a major improvement from the current studies as it better captures “free cortisol” as compared with plasma and urinary measures, and it measures immediate physiological response to driving as an acute stressor. In addition, HRV is a commonly used biomarker that measures the flexibility and balance of the autonomic nervous system in dealing with stressors [33], such as mental stress tasks and work stress [34]. Second, as synthesised results from the current review suggest a difference between short bouts of driving and long hours of driving, it raises the question regarding whether a “threshold effect” exists in terms of driving-induced stress, for example, how many hours of uninterrupted driving would make driving a significant stressor? Third, characteristics such as age, genetic predisposition and various lifestyle behaviors can affect the automatic nervous system and reduce HRV permanently, which is in turn associated with a range of cardiovascular diseases [35]. Future studies may benefit from recruiting larger and more diverse samples to examine whether HRV responses to driving differ by the individual characteristics listed above. Moreover, to further elucidate the potential link between driving and CVD outcomes, additional studies are needed to examine intermediate outcomes between acute stress response and CVD, such as maladaptation and dysregulation induced by repeated exposures to the stressor.