The inclusion criteria for studies, which was defined a priori, were as follows: 1) English-language studies published on or after January 1, 1960, 2) licensed drivers 15 years of age and older, 3) studies published in a peer-reviewed journal or non-published (i.e. “grey literature”), which included master's theses, doctoral dissertations, and conference papers, 4) studies limited to randomized control trials, cohort studies, case-control studies, or case-control types of studies , i.e. case cross-over, case-time series, etc. 5) outcome measure reported for at least one specific medication, 6) outcome measure reported as the odds or risk of a motor vehicle collision or some affected aspect of driving ability during an on-road assessment or driving simulation (e.g. brake reaction time, weaving, standard deviation of lateral position, etc.). If the study reported outcome measures for both specific medications and illicit drugs or specific medications combined with alcohol, only outcome measures for specific medications alone were reported. A ‘medication’ was defined as a substance either available by prescription or over-the-counter to remedy a medical condition. Therefore, caffeine, nicotine, vitamins, and nutraceuticals were excluded. If the medication usage was combined with a medical procedure, then the study was excluded to avoid bias. While marijuana has been legalized for medicinal purposes in several states, it was not considered a medication in this analysis as it is still defined as an illegal substance by federal law. Because of the vast difference in the fidelity of driving simulators, a driving simulator must have consisted of a screen, pedals, and steering wheel. If the study did not specify the components of the simulator, an attempt was made to search the make and model of the simulator noted in the study to see if it was comprised of these constituents. The search date of January 1, 1960 was arbitrarily chosen as no DUID studies existed or were published prior to this time. Because of the complexity of the initial study question, it was decided post-hoc to only present the studies whose outcomes reported the association between a specific medication and the odds or risk of a motor vehicle collision.

Studies were acquired from the following fourteen databases: 1) Medline (within EBSCO host), 2) PubMed, 3) Scopus, 4) International Pharmaceutical Abstracts (IPA), 5) Cochrane Central Register of Controlled Trials (CENTRAL), 6) CINAHL (within EBSCO host), 7) AgeLine, 8) Web of Science (WOS), 9) PsychInfo, 10) Transportation Research Information Services (TRID), 11) Academic Search Complete, 12) EconLit 13) SafetyLit, and 14) ProQuest Dissertations and Theses (ProQuest). All searches were performed by TMR with the assistance of a Health Sciences Librarian from West Virginia University. The last search was performed in June 2014. All searches were conducted using Medical Subject Headings (MESH) terminology. Each search contained the phrases, “drug”, “medication”, “traffic collision”, and “motor vehicle”. An example search strategy (ProQuest) is included in Appendix A.1. In addition to the fourteen databases, studies from TMR's personal library were also reviewed for eligibility. Hand searches from the reference lists of included studies were also examined. Government websites, such as the National Highway Traffic Safety Administration, were also searched for relevant government-performed studies.

All included studies were independently selected by TMR and BR. Any discrepancies regarding the inclusion of studies were resolved through discussion. If a consensus could not be reached, MZ acted as the arbitrator. All studies, whether included or excluded in the review, were stored in EndNote, version X5, along with reasons for including or excluding the study.

A codebook was developed by TMR in Microsoft Excel 2007. The codebook included variables regarding study characteristics (i.e. study design, year when published, etc.), study population (i.e. country of origin, ages/sex of participants, etc.), medications investigated, outcome measures, and study quality. All studies were coded by TMR and CP independent of one another. After coding was completed, both authors met to compare the entries for accuracy and/or precision. Any discrepancies were resolved through discussion. If disagreements could not be rectified, MZ acted as the arbitrator.

The included studies were evaluated for quality by TMR and CP independent of one another. An abbreviated version of the checklist for measuring study quality designed by Downs and Black was used to score the studies for quality.⁶¹ The original checklist designed by Downs and Black, which has been deemed both valid and reliable, is a 27 question, qualitative check-list to evaluate a study for internal and external validity, quality of reporting, and power; for every positive attribute/response, a study receives one point.⁶¹ At the end of assessment, the points are tallied; the more points a study earns, the higher the study's quality.⁶¹ Since not all questions in the Downs and Black scale were applicable to the review, the abbreviated checklist contained 5 measures of quality (i.e. a total of 5 possible points). It was determined a priori that if a study had less than 3 points, it would not be included in the review. After all coding and quality assessments were completed, the authors met to compare their entries for accuracy and/or precision. Any discrepancies were resolved through discussion. If disagreements could not be reached, MZ acted as the arbitrator. No studies that met the eligibility criteria were excluded based on quality.

All studies were synthesized qualitatively. No meta-analysis was performed given the a priori assumption that excessive statistical and/or methodological heterogeneity would be observed.⁶² However, the overall odds or risk ratio for each medication and outcome was presented graphically and included the point estimate and corresponding 95% confidence intervals. Post hoc, it was realized that some included studies only presented crude estimates, ^63-82 while others adjusted their outcome measures for various potential confounders such as age, gender, miles driven, day of the week, etc.^83-89 For those studies that presented adjusted estimates or did not include 95% confidence intervals for crude estimates, ^83-89 then the adjusted point estimates and 95% confidence intervals for the population were reported. If a study's point estimate was greater than 1, this indicated that there was excess risk of collision; a point estimate less than 1 indicated that the medication use was associated with a decreased risk of collision (i.e. it was protective against MVC). Results were considered statistically significant if the 95% confidence intervals did not include 1.

The search processes for the selection of studies, as well as reasons for excluding studies, are presented in Figure 1. Of the 6,516 records obtained, 208 studies met the original study question. Of these 208 studies, 27 pertained to the association of specific medications and the odds or risk of motor vehicle collision, while the others pertained to the association of specific medications and affected driving ability determined through the use of driving simulators (n=90) or actual driving assessments (n=91).

The characteristics of the 27 studies included in this review are presented in Table 1. ^63-89 The included studies spanned from 1992-2013. Of these studies, eight (29.6%) were conducted in Norway,^{69,70,72,73,75,77,79,82} six (22.2%) in the United States,^{63,64,81,83-85} five (18.5%) from Canada,^65,66,86-88 three (11.1%) from England,^71,74,76 three (11.1%) from France,^67,78,80 one (3.7%) from the Netherlands,⁶⁸ and one (3.7%) from Taiwan.⁸⁹ Eleven (40.7%) of these studies were cohort designs, while the rest consisted of case-control or variations of a case-control design. Only one study was limited to females.⁶³ Nine (33.3%) of the studies were limited to older adult drivers typically 60 years of age and older, nine (33.3%) included drivers aged approximately 18-70 years, and the remaining nine (33.3%) covered all licensed drivers from the country of origin. Three studies (11.1%) were doctoral dissertations, ^63,66,81 one study (3.7%) was a conference paper, ⁶⁸ and the remaining 23 studies (85.2%) were published in peer-reviewed journals. Among these studies, 53 specific medications were evaluated. Diazepam, Zolpidem, Zopiclone, and Insulin were the most commonly studied medications. With respect to study quality, 23 of the 27 studies (85.2%) received a score of 5 (Table 1). The studies that were scored 4 out of 5 typically did not explain their statistical methodology in sufficient detail, i.e., did not mention or explain what types of regression they utilized to obtain study outcome measurements.^65,67,68

The medications that were investigated by the studies included in this review are presented by drug category in Figures 2-9. The drug categories consisted of analgesics, anticonvulsants, antidepressants, antihistamines, antihyperglycemics, benzodiazepines, and sleep-enhancing medications; if a drug could not be categorized into any of these groups, it classified as ‘Other Medications’. Studies were adjusted by the original study authors for age and gender,⁸³ age, sex, gender, and annual mileage,⁸⁴ age, sex, gender, and days driven per week,⁸⁵ and age, gender, residence in the country, previous injurious MVC, chronic disease score, exposure to antidepressants, anti-epileptics, benzodiazepines, antipsychotics, anti-migraine, muscle relaxants, and narcotic analgesics.⁸⁶ Another study was adjusted for cardiac or stroke events within the past year as well as the following drug classes in the previous 60 days: antidepressants, anti-epileptics, benzodiazepines, antipsychotics, anti-migraine, narcotic analgesics, muscle relaxants.⁸⁷ Another study was adjusted for age, gender, previous MVC, and place of residence,⁸⁸ while another was adjusted for concomitant use of the following medications: Zolpidem, Zopiclone, long and short acting benzodiazepines, antihistamines, anticonvulsants, antidepressants, other sedatives/hypnotics, other psychoactive drugs, muscle relaxants, and opioid analgesics.⁸⁹

The principal finding of this study is that among the 53 specific medications investigated by the 27 studies included in this review, 15 medications (28.3%) were associated with an increased risk of motor vehicle collision. The medications that were associated with an increased risk of collision were: Buprenorphine, Codeine, Dihydrocodeine, Methadone, Tramadol, Levocitirizine, Diazepam, Flunitrazepam, Flurazepam, Lorazepam, Temazepam, Triazolam, Carisoprodol, Zolpidem, and Zopiclone. Two (3.8%) of the 53 medications, Sulfonylurea and Warfarin, may be protective against MVC. All other 36 medications (67.9%) were not significantly associated with MVC. The findings of this review illustrate that certain medications, even within the same class or drug category, may be more associated with crash risk than others.

These findings are consistent with the current literature. With the exception of Carisoprodol, Dihydrocodeine, and Tramadol, 12 of the 15 medications that were associated with increased MVC have also been studied in either driving simulations or actual driving assessments. Both types of studies use common measures to assess one's ability to safely navigate a vehicle; these measures include the ability to maintain position within the lane [i.e. standard deviation of lateral position (SDLP)], tracking (i.e. following an object), missed instructions or directions, actual collisions with cars or objects, the ability to maintain constant speed, reactions or reaction time to stimuli, the ability to keep the vehicle on the road, steering wheel angle, divided attention (i.e. performing a task, such as acknowledging a sign or symbol, while driving), and gap acceptance (i.e. ability to judge distance between objects). There is evidence from both driving simulation and driving assessment studies that eight of these medications may also affect some aspects of driving ability; this includes Diazepam, Flunitrazepam, Flurazepam, Lorazepam, Temazepam, Triazolam, Zolpidem, and Zopiclone.

Diazepam, an anxiolytic, has been studied more extensively than any other of the medications. Six driving assessments^90-95 and ten simulation studies ^96-105 have investigated its effects on driving ability. The majority of studies have been randomized, double-blind, and cross-over by design. Participants in these studies have involved both genders of different age ranges. Professional drivers have also been used to study its effects.⁹³ Diazepam has been shown to negatively affect driving ability in virtually all the studies whether the source is an actual driving assessment or driving simulation. Collectively, Diazepam has affected nearly every aspect of driving ability including braking, SDLP, incidence of collisions, driver behavior, tracking, steering, car following, and reaction time, compared to placebo conditions.^90-105

Flunitrazepam, a benzodiazepine, has been studied in one simulation¹⁰⁶ and one driving assessment.¹⁰⁷ In the simulation study, 16 healthy subjects underwent a double blind cross-over study where they took Flunitrazepam the night before than completed the simulation the next morning to assess residual effects of the drug. Participants tended to speed when taking Flunitrazepam.¹⁰⁶ In the driving assessment, 32 outpatients with sleep disorders received either 2 mg of Flunitrazepam or 20 mg of Temazepam daily for seven days and partook in a driving test on day one and day seven.¹⁰⁷ Steering, lateral acceleration, and velocity were used to assess driving ability. Those taking Flunitrazepam experienced greater decreases in steering ability compared to Temazepam.¹⁰⁷

Flurazepam, a benzodiazepine derivative, has been investigated in two driving assessments^108,109 and two simulations.^105,110 In a blinded, cross-over trial, 12 female volunteers received placebo or Flurazepam the night before and then drove a course the following morning. Under Flurazepam, participants incorrectly performed passable gaps.¹⁰⁸ In another double-blind, cross-over study, 16 patients treated for insomnia performed a 75km driving assessment. When treated with Flurazepam, SDLP was decreased, but was more pronounced in female subjects during morning assessments.¹⁰⁹ In a double-blind, cross-over simulation using 12 healthy volunteers, general reaction time to signals was impaired under Flurazepam compared to control periods.¹¹⁰ In another placebo-controlled, double-blind simulation using 54 healthy volunteers, executed tasks while driving were decreased in those taking Flurazepam compared to placebo periods.¹⁰⁵

Lorazepam, a benzodiazepine, has been evaluated in seven driving assessments^94,111-116 and one simulation.¹¹⁷ All of the studies were randomized, double-blind, and cross-over by design. Four of the driving assessments were limited to males,^111,114-116 two to females,^112,113 and one to a group of clinically anxious and non-anxious participants of mixed gender.⁹⁴ The simulation study encompassed healthy volunteers.¹¹⁷ Driving ability (i.e. SDLP, inappropriate line crossings, car handling, lane-keeping, speed, following distance, and tracking) was affected in all driving assessments and simulations.^94,111-117

Temazepam, a benzodiazepine prescribed to treat short-term insomnia, has been evaluated in five driving assessments^{107,108,118-120} and one simulation study.¹²¹ All of the driving assessments and simulations were randomized, double-blind, and cross-over by design. The driving assessments used female volunteers in two studies,^108,120 one used rotating shift workers,¹¹⁹ one used healthy volunteers,¹¹⁸ and one used insomniacs.¹⁰⁷ The driving simulation used only female participants.¹²¹ Temazepam affected driving ability (i.e. participants hit more bollards) in only one study, but none of the others.¹⁰⁸

Triazolam, a sedating benzodiazepine used to treat severe insomnia, has been studied in three driving assessments^119,122,123 and three driving simulations.^15,124,125 All three driving assessments were randomized, double-blind, and cross-over by design.^119,122,123 Two of the studies used rotating shift workers^119,123 while the third used healthy volunteers.¹²² In the two driving assessments using shift workers, Triazolam severely affected driving ability;^119,123 driving performance was not as affected among the healthy volunteers.¹²² In the simulation studies, all three were similar in design and used commercial bus drivers as the study population.^15,124,125 Driving ability (i.e. lane keeping, driving path, steering, braking, and SDLP) was affected in all of the simulation studies simulations.^15,124,125

Zolpidem and Zopiclone, two sleep-promoting medications, were evaluated simultaneously in two driving assessments^126,127 and four driving simulations.^106,128-130 The two driving assessments were randomized, double-blind, and cross-over by design that used healthy participants of all age ranges.^126,127 In both studies, Zopiclone and Zolpidem affected driving ability. In the simulation studies, all four were randomized, double-blind, and cross-over by design; three studies used participants 55 years of age and older, ^106,128,130 while the mean age in the fourth study was 38 years.¹²⁹ Zopiclone affected driving ability in all four simulations; ^106,128-130 Zolpidem affected driving ability in three simulations, ^106,129,130 but not in the fourth.¹²⁸ Zolpidem and Zopiclone were studied individually in three additional studies each. Zolpidem affected driving ability in all three studies, ^121,131,132 while Zopiclone affected driving ability in two of the three studies.^118,133,134

In addition to the eight medications described, four of the other 15 medications associated with increased risk of MVC have not been associated with affected driving ability during driving assessments or simulation studies; this includes Buprenorphine, Methadone, Levocitirizine, and Codeine. One driving simulation study investigated the effects of Buprenorphine and Methadone in a group of former heroin addicts whom were stabilized for at least three months and compared them to a group of healthy controls.¹³⁵ SDLP, speed, steering wheel angle, and reaction time were used to assess driving ability. There was no difference in driving ability between the placebo group and those treated with either Buprenorphine or Methadone.¹³⁵ Another study investigated the effects of Levocitirizine through an on-the-road assessment, which measured SDLP and speed to evaluate driving ability.¹³⁶ No difference in performance was detected among those treated with Levocitirizine and the control group.¹³⁶ Codeine has been studied in three driving simulations.^96,137,138 A randomized, double-blind, placebo-controlled, crossover study of 13 healthy volunteers were administered therapeutic levels of Codeine and assessed on SDLP and time to collision under simulation. There were no differences between exposure and control periods.¹³⁷ In another simulation study, 70, 16-22 year old professional drivers from the Finnish Army were randomized to various treatment and control groups and assessed on numerous factors during simulation.⁹⁶ The only difference among the Codeine treated group was that they caused slightly more collisions.⁹⁶ In another simulation study, which compared chronic pain users with Codeine, chronic pain users not using Codeine, and placebo group of healthy controls, participants were assessed on reaction time and missed reactions.¹³⁸ No differences were detected in the Codeine treated group.¹³⁸

While the remaining three medications, Carisoprodol, Dihydrocodeine, and Tramadol, have not been evaluated in driving assessments or driving simulations, their effects on psychomotor performance and/or cognition have been evaluated in laboratory settings. Carisoprodol, a muscle relaxant, can cause users to feel sedated, sluggish, and distracted, especially at high dosages.^139,140 At therapeutic doses, Carisoprodol users often report feeling no effects of the medication even though their psychomotor functioning is decreased; this situation may cause users of this medication to underestimate its effects and partake in behaviors, such as driving, when they may be impaired.¹⁴⁰ Dihydrocodeine, an opioid prescribed for mild to moderate pain, is known to cause dizziness and mild euphoria especially in new users,¹⁴¹ which may interfere with their ability to drive. Tramadol, an opioid for mild to moderate pain, has been shown to be fairly safe in respect to cognition and psychomotor function, though at higher doses, Tramadol has been linked to the worsening of balance.¹⁴² It should be noted that poor balance has been linked to higher incidences of MVC.¹⁴³

From these findings, it is apparent that some medications that are associated with an increased risk of MVC do not appear correlated with decreased driving ability. Conversely, while the majority of the medications included in this review were not found to be statistically associated with increased risk of MVC, the inference that these medications are innocuous on one's ability to drive is not advised. There are several plausible reasons why the association between MVC and driving ability were not congruent with all medications presented in this review. First, this incongruence may be attributed to study design. The 27 studies investigating the risk or odds of MVC were observational in nature, whereas the studies investigating driving ability were all randomized control trials (RCT) and therefore, experimental. The studies investigating driving ability were comprised of small sample sizes and were occasionally limited to certain age groups or genders. Therefore, the findings of the RCTs may not be generalizable to the entire population, which is a known limitation of this type of study design.¹⁴⁴ In addition, the observational studies may not have been adjusted for key confounders whereas this would not have affected the RCTs results if participants were randomized properly.¹⁴⁴ Also, most of the RCTs only investigated one or two aspects of driving ability. The act of driving is quite complex in nature. It is possible that other aspects of driving ability may not have been affected by the medication being investigated. It is also plausible that even if driving ability is partially affected by a medication, some drivers may compensate their driving behaviors to avoid a potential collision, which is a known phenomenon.^145-147 Second, epidemiological research investigating the associations between specific medications and MVC is lacking; this is evident by the number of studies included in this review. Some medications that have been studied via RCTs may not have been investigated observationally in larger populations. Third, many of the RCTs used healthy volunteers. This is problematic as disease-medication relationships are often difficult to distinguish. It is entirely possible that the disease in which the medication is prescribed is affecting an individual's ability to drive or their risk of MVC. It is also possible that driving ability is only affected for a short period of time and that an individual who takes a medication for an extended period of time develops a tolerance. This may not have been detected as the RCTs were often short in duration.

The findings of this review have several key clinical implications. First, the association between a specific medication and increased risk of MVC and decreased driving ability is not always clear. There are some specific medications that are most likely driver-impairing as both experimental and observational research has associated them with decreased driving ability and increased risk of MVC, respectively. These include Diazepam, Flunitrazepam, Flurazepam, Lorazepam, Temazepam, Triazolam, Zolpidem, and Zopiclone. For other medications, such as Buprenorphine, Codeine, Methadone, and Levocitirizine, there appears to be an association between them and increased risk of MVC, but not driving ability. As for Tramadol, Carisoprodol, and Dihydrocodeine, they appear to be associated with increased risk of MVC, but have not been evaluated in driving assessments or simulations. Secondly, some specific medications belonging to the same category or class may be more detrimental than others. For example, among the ‘Other Medications” included in this review, Carisoprodol had a strong association with MVC while the other medications in this category did not. Thirdly, some of these medications are widely prescribed. For example, Buprenorphine was the 39^th most sold drug in the United States during the fourth quarter of 2013.¹⁴⁸ Therefore, careful consideration of the patient and their lifestyle must be given when these medications are prescribed. Possible alternatives within a drug category or class should be considered for patients that frequently drive as certain medications may be more driver-impairing than others. Education efforts are needed to ensure that patients are aware that certain medications may increase their risk of MVC.

The strengths of this review are that over 50 years of data among fourteen different databases were searched for potential studies. This review also included relevant ‘grey literature’ as evidenced by the inclusion of three doctoral dissertations and one conference paper. Despite its strengths, this review is not without its limitations. For example, publication bias is a known limitation of systematic reviews.¹⁴⁴ It is also possible that some studies may have been missed despite two authors independently reviewing studies for inclusion. As much of the DUID research arises from Europe, ⁵⁹ it is likely that several studies were not included or missed as they were not available in the English language, a requirement of the a priori inclusion criteria. In some instances, only abstracts and not complete studies could be retrieved, especially with reports or conference papers that were dated. Also, some studies were adjusted for potential confounders, while others were crude estimates. The findings of this review also indicate that future research opportunities in this area of study are plentiful. Future research could entail, 1) RCTs of medications found to be associated with increased risk of MVC that were never evaluated in a driving assessment or simulation, 2) epidemiological studies investigating the effects of other medications or new medications that may be driver-impairing, 3) investigating different methods of measuring driving ability in simulations and actual assessments, and 4) pursuing patient or clinician educational efforts to raise awareness about the possible effects of specific medications on MVC or affected driving ability.

This systematic review sought to determine which specific medications were associated with increased risk of motor vehicle collision among licensed drivers. While the majority of medications investigated were found to not be significantly associated with increased MVC, several medications were associated with such risks. It was also determined that some medications may be more driver-impairing than others. These findings pose numerous clinical implications. Due to the increasing public health and traffic safety concerns regarding driving under the influence of drugs, future research in this area of study is worthy of investigation as the associations between specific medication use and the increased risk of motor vehicle collision or affected driving ability are complex and not always lucid.