Background

The number of older drivers is rapidly rising due to the aging population [1,2,3,4]. It is projected that, by 2030, 20% of the population will be 65 years or older [5]. In Canada, it is expected that, by 2026, 1 driver out of 5 will be 65 years or older [6]. With increasingly active lifestyles, seniors are expected to rely even more on their vehicles, taking more trips, driving further distances, and keeping their licenses longer than prior generations [7]. In fact, it is anticipated that a large proportion of both men and women will continue driving well into their 80’s [8]. For example, a majority of Canadian seniors hold a valid driver’s license (4.7 million in 2017. representing 75% of all seniors) [9]. These trends pose significant public health concerns, as older drivers are disproportionately involved in collisions [9] causing serious injury and death, when exposure (kilometres driven) is taken into account [10].

The higher crash rates in older adults may be due to age-related medical conditions. For example, seniors may develop vision impairment [11, 12], mild cognitive impairment, early dementia, Parkinson’s disease and other neurodegenerative disorders, or may have suffered a stroke. All these conditions produce symptoms that impair the skills that are required to drive safely. Many studies show that these conditions lead to worse driving performances both on-road [13, 14] and in-simulator evaluations [15] compared to the general older adult population. Despite these numerous health conditions associated with aging that might negatively impact driving performance, the proportion of older drivers who are considered healthy still represents the largest segment of these drivers. Therefore, there is a need to assess interventions that are tailored for them.

Multiple assessment tools are available for use within clinical settings to screen for at-risk drivers. Although many assessments/tools are quick and easy to administer, a screening battery has not yet been developed [16, 17]. The potential to detect unsafe drivers versus successfully identifying safe drivers is an important consideration, particularly as removing one’s license can have negative consequences [18, 19]. Prior studies have found that driving cessation is associated with increased depression, social isolation, institutionalization and even early mortality [20, 21]. A recent survey conducted by Vrkljan et al. [22] showed inconsistency in practice among evaluations in a sample of driver assessment centres for medical fitness to drive (n = 47). Their results highlight the necessity of evidence-based guidelines for the training and assessment of at-risk drivers.

While licensing authorities must consider public safety when delivering driver’s licenses, it is important to help seniors drive for as long as possible to facilitate their autonomy and independence. This is particularly the case since there are few programs to help seniors adjust to non-driving.

Alternatively, interventions aimed at improving or maintaining driving skills offer new opportunities to help seniors drive safer, longer. Several studies have examined the impact of workbooks, seminars, and cognitive, simulator or on-road training on driving performance in older adults in general, and in those with various medical conditions. The purpose of this systematic review is to synopsize the current literature on interventions that are tailored to improve driving: Self-awareness, knowledge, behaviour, skills and/or reducing of the number of collisions in healthy older drivers.

Methods

A systematic literature review (SLR) methodology [23] was used to synopsize the current literature on interventions that are tailored to improve older individuals’ driving. This methodology is scientifically transparent, replicable, and useful to generate an in-depth analysis of the scientific literature [24]. An initial exploratory review was produced prior to conduct the full SLR [25]. This method allows to elucidate common knowledge of the topic, to identify if the proposed SLR fits the existing knowledge in the area, to determine the key concepts and to refine the research question. Also, this SLR followed a five-step approach proposed by Denyer and Tranfield [25]: 1) Question formulation; 2) Locating studies; 3) Study selection and evaluation; 4) Analysis and synthesis; and 5) Reporting and using the results. Based on witch, a review protocol was used regarding the formulation of the research question, on the selection of scientific databases and search terms, and on the inclusion and exclusion criteria for searching and analysing retrieved publications.

Step 1: question formulation

A PICO framework (Population, Intervention, Control, Outcomes) was used to generate the research question of this study (Step 1). This approach allows for a more systematic approach regarding the identification of relevant information and its understanding by using these four categories [24, 26]. Therefore, the research question formulated for this SLR was: In healthy older drivers (P), which type of intervention program (I), education, computer-based, physical training, on-road, simulator-based or mixed program (C) improved driving: Self-awareness, knowledge, behaviour, skills and/or crash rates (O)?

Step 2: locating studies

Based on the research question defined in Step 1, search strings to be used and appropriate bibliographic databases were defined in Step 2. Scopus and PubMed databases were used as they encompass a wide array of scientific areas as well as the most relevant peer-reviewed publications [27]. Articles were identified using MeSH search terms and strings (in English only): (“safety” OR “education” OR “training” OR “driving” OR “simulator” OR “program” OR “countermeasures”) AND (“older drivers” OR “senior drivers” OR “aged drivers” OR “elderly drivers”). The EndNote version X9.2 management software package was used to manage all the information.

Step 3: study selection and evaluation

To select the most relevant scientific articles to include in the review, the inclusion and exclusion criteria were defined in Step 3. The following key inclusion criteria were defined prior to the search:

  • Original articles written in English and published in peer-reviewed journals;

  • Published or in press between January 2007 and December 2017.

  • Articles were excluded if the sample presented drivers with specific health conditions (e.g. traumatic brain injury, vision impairment associated with specific pathologies, stroke or Parkinson’s Disease).

Titles and abstracts of papers were scanned independently by three of the authors to identify relevant articles to retrieve for full text analysis. In cases where the papers seemed potentially eligible, but no abstract was available, the full text of the paper was retrieved. Disagreements between authors led to a deeper joint analysis of the paper; and a decision was then made regarding its inclusion. Full texts were independently reviewed for inclusion by the same three authors.

The literature search purposely only included studies between 2007 and 2017, since previous systematic reviews on the topic had already been conducted [28, 29]. These reviews cover researches completed prior to 2008, and despite being well-conducted systematic reviews, more recent studies on different interventions to improve older drivers’ performance have been conducted but not yet been synthesized. To our knowledge, there is no more recent review in the literature, despite the need for guiding evidence-based practices.

Step 4: analysis and synthesis

Step 4 consisted in analysing, extracting and managing papers’ information to identify and highlight key components of the research conducted and its results. Primary studies meeting the inclusion criteria and reported in the included reviews were identified, and the corresponding data was extracted using a standardized data extraction form. The Quality Assessment Tool set known as “QualSyst tools” was selected as it allows appraisal of quality while assessing potential bias over a wide variety of research designs, from experimental to observational [30]. Furthermore, this set of tools has one version for quantitative studies and another one for qualitative studies, and in this review, the first one was used. The quantitative version consists in a checklist of 14 questions, with possible answers of: yes, no, partial or not applicable. The score for a “yes” answer is 2 points, for a “partial” answer 1 point, and for “no” 0 points. The sum of all answers is then calculated from the corresponding points and divided by the total of applicable responses.

The QualSyst was used by three of the authors to evaluate internal and external validity of the considered studies. The QualSyst tool was originally created as a threshold allowing a study to be included in a review through a cut-off point (0.55 to 0.75) [30]. In this review, the QualSyst cut-off score of 0.55 was chosen to capture 75% of the articles initially deemed as relevant for the review, as well as to ensure the inclusion of several descriptive articles containing valuable data [31]. More specifically, papers with a score higher than 0.8 were classified as having a strong methodology (> 0.8), between 0.79 and 0.71 as being good, and adequate if the score was between 0.7–0.55, or limited and therefore excluded if the score was lower than 0.55 [32, 33].

By using an approach adapted from Sackett et al. [24], identified papers were also categorized using a standardized value system to grade biomedical practices according to the following system:

  • Level I: Systematic reviews, meta-analyses, randomized controlled trials

  • Level II: Two groups, nonrandomized studies (e.g., cohort, case control)

  • Level III: One group, nonrandomized (e.g., before and after, pre-test and post-test)

  • Level IV: Descriptive studies including analysis of outcomes (e.g., single-subject design, case series)

  • Level V: Case reports and expert opinions including narrative literature reviews and consensus statements.

Using such an approach while conducting a review also provides a scheme of references for the clinicians interested in using such methods/approaches in their practicum. Evidence-based practices are built on the assumption that scientific evidence of the effectiveness of an intervention can be deemed more or less strong and valid according to a hierarchy of research designs, the assessment of the quality of the research, or both.

Step 5: reporting and using the results

For Step 5, the results were grouped (Tables 1, 2, 3, 4, 5, 6, 7 and 8) according to the specific type of programs (independent variables) used by Golisz [59], who considered 5 different options such as: (1) education-based training programs, (2) computer-based training, (3) physical training, (4) simulator-based training, and (5) route-based or actual driving training. Moreover, another independent variable was considered in the current study (6 - Mixed programs) since there were many investigations that used two types of interventions, therefore making it difficult to differentiate which one of the variables is responsible for the obtained results. It is noteworthy that the route-based or actual driving training (5) alone was not used in any of the studies evaluated and is therefore not presented in the tables.

Table 1 Summary of the reviewed studies
Full size table
Table 2 Summary of the dependent variables considered in the reviewed studies
Full size table
Table 3 Summary of the results
Full size table
Table 4 Synthesis of intervention studies involving education-based training programs
Full size table
Table 5 Synthesis of intervention studies involving computer-based training
Full size table
Table 6 Synthesis of intervention studies involving physical training
Full size table
Table 7 Synthesis of intervention studies involving simulator-based training
Full size table
Table 8 Synthesis of intervention studies involving mixed training
Full size table

The dependent variables were categorized according to the methods used to collect them (Table 1): Tests/questionnaires, on-road evaluations, simulator and the combination of all. Also, the dependant variables were grouped to infer on the impact of the given program: Self-awareness and/or knowledge, behaviour, skills and crash rates (Tables 2, 3, 4, 5, 6, 7 and 8).

Self-awareness/Knowledge: Self-awareness of one’s ability to drive and of their capacities and/or limitations to do so safely is mainly evaluated by conducting interviews or via questionnaire. As for knowledge, it is often associated with traffic regulations and laws, as well as the effect of health and other factors on driving.

Behaviour: The drivers’ behaviour is documented as moment of the day, roads used, driven speed, or what has been described by Michon [60] in his model of driving as strategic and tactical levels.

Skills: Described as the operational level of Michon’s model [60], skills are linked with direct control of the vehicle as well as with visual searches surrounding a manoeuvre.

Crash rates (or collisions/accidents): Either collected as self-reported value by participants or by cross-referencing available databases, crash rates (ex. Collisions/accidents per distance driven or per year) are used as a predictor of an intervention’s effectiveness.

Results

Figure 1 shows the results of the search strategy using PRISMA. An initial number of 1510 papers was identified through search of databases (SCOPUS: 934 and Pubmed: 576), from which 484 duplicates were removed. After screening the remaining 1026 title, abstract and keywords of each article, 36 papers were identified as being potentially relevant. Following a complete review of the corresponding full-texts, 29 papers were then selected based on the previously mentioned inclusion criteria. Seven (7) papers were not considered due to different situations [61,62,63,64,65,66,67], for example: the objective of the paper by Joanisse et al. [61] was to report the findings from an evaluability assessment of the 55 alive mature driver-refresher course offered by the Canada Safety Council. Another example is the study by Musselwhite [63] where different issues were addressed through an expert group opinion identifying age related physiological and cognitive changes that may be involved in collisions. Finally, after applying the QualSyst [30], 4 papers were removed due to the methodology quality. Thus, 25 papers were included in the final review.

Fig. 1
figure 1

Flow diagram of paper selection process based on PRISMA

Full size image

Table 1 shows the summary of studies reviewed. It should be noted that none of the reviewed studies considered route-based or actual driving training as a pure independent variable. Also, it can be seen that 8 of the 25 studies evaluated a combination of more than one type of intervention (Mixed approach), followed by computer-based and education interventions with 7 and 6 studies, respectively. On the other hand, interventions with less studies in this review are those based on physical training and simulator-based training, with 4 and 2 studies each. For more specific information on studies identified in this literature review, detailed descriptions of protocols can be found in Tables 4, 5, 6, 7 and 8.

Tests/questionnaires were used more frequently to evaluate programs, secondly, on-road evaluations and the combination of these two approaches and thirdly, in-simulator alone or combined with on-road evaluation (Table 1).

Table 2 presents the principal dependant variables used to infer on the impact of the given program: Self-Awareness/Knowledge (SAK), Behaviour (B), Skills (S) and Crash Rates (CR). Although 25 studies were reviewed, the number of dependent variables was 28 as three of the studies presented more than 1 dependent variable [35, 52, 53]. The most studied dependent variable was Skills, tested 12 times, followed by studies that considered Behaviour, tested 10 times.

Before presenting the summary of the results (Table 3) and the synthesis of reviewed studies (Tables 4, 5, 6, 7 and 8), it is important to highlight that the effect of the independent variable was classified as (+) when the effect resulted in a significant improvement in the dependent variable, (−) when the effect was significantly negative or no change was observed in the dependent variable and (+/−) when the obtained results were not clear (i.e. non-significant effect or a combination of significantly positive and negative effects on driving).

Overall results show that 60% of reviewed studies presented positive (+) results, 24% presented negative (−), and the remaining 16% of the studies showed unclear results (+/−). For example, from Table 4, the study by Coxon et al. [34] presented unclear overall results (+/−) since self-regulatory driving practices generally showed positive results, but a negative result in the distance driven per week, the restriction of driving space, the use of alternate transportation and community participation 12-months + post-intervention.

Regarding independent variables, the highest overall positive results are for physical training (Table 6) and mixed programs (Table 8) with values of 100 and 88%, respectively. Finally, the lowest overall positive results can be observed when the reviewed studies considered education (43%) and computer-based programs (33%).

In all reviewed studies, primary research approaches were randomized controlled trial (RCT), observed in 17 studies, followed by a non-randomized controlled study (NRCT) and a cohort study used 4 and 3 times respectively. Also, only 1 study presented a pre-test and post-test approach. Finally, regarding sample size, studies evaluated ranged from 11 to 4880 drivers.

Education-based training programs

Education-based training programs were quite variable in terms of their duration (Table 4), some of them lasting from 1 day up to a full month, the number of classes ranging between 1 and 4. Programs were developed mainly in classroom format. The follow-up evaluation from the intervention was also very variable, going from immediately to 2, 3, 6, and up to 12 months post-intervention. Four (4) of the studies were conducted using a RCT (Level 1) while the remaining 2 used a retro-cohort design (Level 3).

Regarding the dependent variables, 5 of the reviewed studies used questionnaires/tests to evaluate the programs with self-reported driving knowledge, driving behaviours, and driving habits, among others. Drivers reported changing their driving habits following the program and add increased knowledge of road safety facts, but the impact of the intervention faded over time [39]. Only one study used on-road data, measured through collisions and violations of traffic regulations. The remaining study used a mix of questionnaires and monitoring of driving. For the study that evaluated implications in collision [36], drivers who participated in the program add 1.5-time greater odds of being involved in a crash than their matched controls.

Computer-based training

Table 5 presents reviewed papers regarding computer-based training. Four (4) of the studies were conducted using a RCT (Level 1), 1 used a non-RCT 2-group approach (Level 2), 1 a pre/post-test intervention (Level 3) and 1 a cohort design (Level 3). The most studied dependent variable was assessed through questionnaires/tests (3 of 7 studies). It is also important to acknowledge that 3 of the studies additionally used the on-road assessment or simulator driving for evaluation purposes. Despite differences in the form of their interventions, 4 out of the 7 reviewed studies presented computer-based training based on 10 classes. Follow-up of these studies varied between immediate evaluations, up to 1 to 5 years post-intervention. This later evaluation of the program’s impact was based on crash rates [41]. Two (2) studies assessed behaviours with mixed results on reported outcome, some factors improving such as less driving cessation over 3 years [40] while other specific manoeuvres deteriorated [43]. Speed of processing training showed a positive impact on reducing driving cessation and lowering at fault motor vehicle collisions, as well as improving reaction time [42, 45].

Physical training

Only 2 studies used physical training to improve driving skills. Both used a RCT (Level 1), 1 using an evaluation scheme similar to an instructor looking at a driver’s overall performance [47] while the second study used different types of evaluations more associated with processing and movement time, such as brake reaction time and peripheral response time tasks [48]. Both programs showed relatively similar active time for the older drivers, Marottoli et al.’s program lasting 21 h [47] and Marmeleira et al.’s program 24 h [48]. The 2 interventions showed a positive impact on driving performance by either maintaining driving capacity after 12 weeks or even improving scores of reaction time measurements. However, it is not possible to identify if there was any issue with adherence to training in these 2 studies.

Simulator-based training

Of the studies that investigated the effects of simulator-based training on driving performance, only 2 that met the inclusion criteria were reviewed (Table 7). These studies used a randomized controlled trial approach (Level 1), and the dependent variables were gathered using the same simulator. Furthermore, characteristics of the intervention are not very clear, so it is not known how long each class lasted and only one study indicates that the intervention lasted 2 days. Marchal-Crespo et al.’s [49] study showed that, despite being useful for improving driving performance for both younger and older drivers, a guidance system using haptic feedback was not sufficient to transfer learned skills for older drivers and this training was not useful at improving their skills, long-term. Once the guidance system was removed, older drivers returned to their initial driving performance.

Rogé et al.’s [50] study showed a positive effect of their training using a simulator to improve drivers’ useful visual field size. Trained participants showed an improved central and peripheral capacity to detect signals while the control group who drove the same simulator with no specific training did not improve their detection rates.

Mixed programs

The most studied independent variable in the revised papers was the mixed programs. The most common combination was done by using a classroom setting plus either an on-road intervention with a driving specialist [51,52,53] or a series of in-simulator interventions with driving-specific feedback combined with active practice in the simulator [57, 58]. Research designs were equally distributed, 4 for each, between Level 1 (randomized controlled trials) and Level 2 (2 or more groups randomly assigned to conditions but not as a RCT). When comparing the types of interventions that received mixed training, a consistent finding is that groups that only received classroom information with no specific feedback and practice of their driving (control groups) did not improve their driving when compared to other combinations of interventions [51, 57, 58]. These results are similar to those observed in the education-based training programs section above.

With the use of a specific data collection system, it is interesting to note that the approaches used by Porter [51], Romoser and Fisher [57] and Lavallière et al. [58] allowed the trainees to receive specific feedback from their own driving performance by using video collected during the initial on-road evaluation.

Discussion

The purpose of this study was to assess, by a critical review, whether interventions designed for healthy older drivers improve driving on the following components of safe driving: Self-awareness and/or knowledge (SA/K), behaviour (B), skills (S) and/or reduced crash rates (CR). Reviewing the 25 papers selected according to pre-defined criteria, 60% (15 papers) report a positive impact on different levels of driving indicating that interventions are feasible and useful at improving older drivers’ situational awareness/knowledge, behaviours, skills and/or crash rates. However, it is important to mention that some of these results are different from the findings presented by Golisz [59] in their review assessing specific interventions within the scope of occupational therapy practice and including a population with specific health concerns such as stroke survivors [68].

In this section, the primary review findings are discussed separately according to each independent variable (i.e., the effects of education-based training programs, the effects of computer-based training, etc.). The authors realized that the diverse nature of the studies and the variables used in the reviewed studies were quite different, even when testing similar variables, and that different approaches have their own specific strengths and weaknesses.

Education-based training programs

Education programs present a variety of types of interventions, from the number of classes, number of participants per class, program and class duration. In the current review, the typical class used was based on the 55 alive driving course. This was also observed in mixed training that used part of an education-based approach. On the other hand, when considering the type of class, most interventions were guided by an expert, but there were also programs developed through educational videos, more flexible programs guided by the participants, or simply reading a document. Therefore, it is difficult to propose a “standard” type of intervention or to generate a recommendation on how an intervention based on an education program should be.

Despite the positive findings related to self-regulatory driving practices found in some of the programs [34, 36, 39], the results must be considered with caution since 2 out of 3 studies with positive results related to education programs were informed by self-reports and/or questionnaires [36, 39]. This type of dependent variables could create some problems, for example, in the study developed by Selander et al. [69], all the participants self-reported as capable of driving, however, when evaluated by an objective measurement such as a test on route, 20% of them failed. In another study developed by Freund et al. [70], 38% of the participants were categorized after a simulator test as unsafe drivers, however, all of them self-reported driving performance that was equal or better than other drivers of their age group. Ross et al. [71] found that 85% of older drivers self-reported as being good or excellent drivers regardless of their previous citations or crash rates.

It is also important to highlight that 4 out of 6 of the reviewed studies, including the ones with strong methodology [34, 39], did not produce positive results supporting this type of intervention [37, 38]. Furthermore, the results obtained by Nasvadi and Vavrik [35] indicate that men and women who attended this education course had a 1.5-time greater chance of being involved in a crash than their matched controls. The previous results coincide with Janke’s study [72], which concluded that completing a course of education is not associated with a decrease in crashes after the analysis of 2 cohorts, conversely, those killed and injured in motor vehicle collisions increased. Furthermore, there is a systematic review that indicates that there is no scientific evidence to support the effectiveness of post-licensing education programs in the reduction or prevention of accidents [73]. This could translate into an increased risk of driving since the effects of the education program are not positive and participants feel more confident after participating in a program or maybe that drivers taking these types of classes have concerns about their driving and might already be at risk drivers due to declining skills and/or health conditions.

Overall, despite their widespread use among older drivers and organizations who provide these classes, their proven efficacy to increase driver’s knowledge and self-awareness are not enough to improve one’s ability to drive safely or reduce crashes. Therefore, they should not be used as a single method for an older driver who wants to continue driving. The results obtained from the current review confirm the one from Owsley et al. [74] and McKnight et al. [75]. They showed that educational interventions did not show positive results in improving driving performance and safety, even though drivers increased their knowledge on road safety. Moreover, these results have been confirmed by the mixed interventions identified in the current review showing that with a classroom intervention only, older drivers did not improve their driving performance [51, 52, 57, 58].

Computer-based training

Despite the huge variability in the methodologies used, only 2 of the 7 studies presented positive overall results on behaviour [40] and skills [45].

Considering the study by Edwards et al. [45], it can be concluded that this type of intervention could positively affect driving mobility, since the participants who completed the processing speed training were 40% less likely to cease driving during the subsequent 3 years, as compared with controls. In addition, another paper showed that the older adults with risk for mobility decline who completed the processing speed training experienced a trajectory of driving mobility similar to the subjects who were not at risk [43]. On the other hand, those who were at risk for mobility decline and did not undergo training experienced greater decrease and difficulty in mobility for the 3 subsequent years. However, those results should be analysed with caution since the dependent variable is based on self-report through the mobility questionnaire. Contrary to the positive findings mentioned above, another reviewed study showed that 5 weeks of training with 2 weekly 60-min sessions guided by a facilitator did not reveal any significant benefit associated with the intervention [46]. This could be due in part to the type of variable used, i.e. a short version of the “Hazard Perception Test”, which corresponds to a more objective tool than the application of a self-reported questionnaire, which has shown a sensitivity of 75% and a specificity of 61% in the ability to predict safe older drivers or dangerous older drivers [76].

Three (3) out of the 7 studies presented unclear results (+/−) [41, 43, 44]. Despite it being important to mention that in the study by Ball et al. [41] 2 out of 3 intervention groups (speed of processing training and reasoning training) reduced at-fault collision involvement over the subsequent 6-year period relative to controls, which would indicate that this type of computer-based program could improves driving performance. This is the only reviewed study on a computer-based program demonstrating an improvement for older drivers.

One of the advantages of computer-based programs are that they can be a great training alternative when considering costs, since today’s access to the Internet has been facilitated, and the widespread use of computers and mobile devices is not a barrier to the implementation of these interventions but rather an opportunity for people to improve their driving skills from their homes [44, 77].

Overall, computer-based interventions are an interesting opportunity for older drivers, since some of them have been shown to reduce the risk of crash involvement over time. However, more research is required to better understand how these interventions improve one’s ability to drive safely, beyond the speed of processing and reduction of reaction time. With computers and smart devices now widely available, training could be done almost anywhere for interested drivers.

Physical training

For both studies under review in this current analysis, results are positive in terms of impact on driving performance following a physical training intervention aimed at older drivers. The first study showed that physical training allowed older drivers to maintain driving performance over the course of the interventions, while the second program showed improved response time to different secondary tasks while driving. Over 2 different periods of time, 12 vs 8 weeks, they both showed that a regimen equivalent to about 21 to 24 h of exercises could be beneficial to older drivers. Further evaluation should be performed to analyse the best and most efficient modality of interventions with this clientele (ex. Short daily period of exercise or longer period spread across the week).

Moreover, since most interventions aimed at either increasing range of motion and speed of movement, little is known on the impact of cardiovascular training and its transferability to driving capacity. Positive transfer has been shown on cognitive tasks [78] and it would be of interest to evaluate this in a driving context.

Simulator-based training

Only 2 studies used simulator-based training and presented contradictory results. The negative results reported by Marchal-Crespo et al. [49] showed that older drivers did not benefit from training with haptic guidance, and long-term improvements (1 week) were only observed among younger drivers. On the other hand, Rogé et al. [50] showed that simulator training can improve visual field, allowing older adults to better identify vulnerable users on the road. Dependent variables from the 2 reviewed studies were obtained through the use of simulators, and normally indicate that performance in a driving simulator is strongly related to real driving performance and less to cognitive performance [55]. Therefore, driving tests in simulators could be used to evaluate older adults, as previously suggested by Lee et al. [79,80,81]. Despite the small number of reviewed studies using simulator-based training as the only intervention, it is interesting to report that in the mixed program, 5 out of 9 studies used simulators with another independent variable [54,55,56,57,58].

Something interesting having repercussions in future research from the study by Rogé et al. [50], is the report of simulator sickness by participants that rendered them unable to continue with the study. Studies using simulators in the mixed interventions below also reported the loss of participants due to simulator sickness. This important issue related to the use of simulator or virtual reality is at the forefront of their widespread use for clinics and programs aimed at older individuals, since they present a higher prevalence of symptoms than their younger counterparts [82]. Fortunately, interventions can be tailored to reduce the importance of such symptoms to allow the driver to accommodate to this new driving environment [83].

Mixed programs

All the Programs using a mixed approach included specific driving practice either on-road or in-simulator.

Bédard et al. [53] and Marottoli et al. [52] showed that interventions using on-road sessions with an instructor providing specific feedback improved driving scores after the intervention. Unfortunately, the use of general scores to describe a driver’s improvement does not make it possible to extract the specific effect of the intervention or the remaining generalization (ex. moving in the roadway).

Romoser and Fisher [57] compared the effectiveness of active, passive and no training on older drivers’ performance in intersections. Active training in-simulator increased a driver’s probability of looking for a hazard during a turn by nearly 100% in both post-training simulator and on-road driving sessions. Lavallière et al. [58] showed similar results in-simulator with an analysis of visual search strategies during on-road lane changes. Their results revealed that the driving-specific Feedback group increased their blind spot verifications (from 32.3 to 64.9% of the lane changes - an increase of 100%), whereas the control group did not. Porter et al. [51] also used a video intervention but the feedback was not specific to a particular set of driving skills. Porter et al. [51] used a similar paradigm utilizing video and global positioning system (GPS) data, in combination with a classroom-based education program. Their results showed a mitigated impact on driving, since only 9 out of 17 improved their driving by reducing their errors after the program. This difference between Porter et al.’s results and the 2 previous studies might be due to the number of sessions to provide feedback and practice for drivers, since Porter et al. do not report any specific practice of driving weaknesses after receiving feedback.

Only the study conducted by Romoser [56] evaluated the retention of the initial intervention and showed a positive impact 2 years after the program was completed. Only the group who received specific feedback on their driving and appropriate practice in-simulator initially continued to execute secondary looks in intersections prior to turning [57].

For all the studies using customized feedback [51, 57, 58], interventions were successful at modifying drivers’ perception of their driving abilities and positively modifying their subsequent driving skills when returning on-road. The difference between them is that, when particular feedback is aimed specifically at one driving manoeuvre (ex. visual search while turning left [57] or changing lanes [58]), one can expect that these specific manoeuvres should improve after the interventions.

Overall, the mixed approach interventions presented the highest overall score on Qualtsyst and a high frequency of randomized controlled trials (Level 1) (n = 4) or multiple groups comparisons (Level 2) (n = 4).

Limitations of this review

A probable limitation of this review includes the search process itself, which may not have allowed the identification of all studies showing the effects of the different types of programs on driving skills. The use of additional databases such as CINAHL, PsychInfo and ERIC might have led to slightly different results [27].

For the study using a simulator, either as a single intervention or with a mixed approach, simulator sickness still remains an obstacle for the large-scale use of such interventions, in particular if we cannot better understand the underlying mechanisms. The fact remains that some techniques, at best, reduce the incidence and impact of such events on subject participation, and should be used to prevent the prevalence of such impediments on individuals who follow these driving curriculums. For some of the studies, having self-reported collisions as an indicator of driving performance should be used with caution, since self-reports of collision involvement may lack validity [84, 85]. Despite this limitation, this type of reporting remains of interest, as official records might also face an issue of underreporting when addressing older drivers’ involvement for fear of losing their driver’s license [86] when such events are reported to official agencies [87].

Finally, limitation of the systematic review itself is due to lack of consistency when reporting results in training programs aimed at older individuals. The wide variety of research approaches adopted by the reviewed studies also made it difficult to summarize and obtain direct relevant findings, since not all driving parameters were assessed similarly, and the specific driving components that were evaluated were not mentioned. Some of the on-road evaluations used general score checklists similar to the one used by driving specialists [47, 52, 53] while others have assessed specific driving manoeuvres, such as visual inspection during lane change or while turning at intersections [56, 58]. In some studies, it is hard to extract proper information on the used design and method and there is limited reporting on the training program, per se. Moreover, few studies have presented a follow-up evaluation of their program to evaluate the mid or long-term retention of their interventions [56]. Despite identifying 25 interventions aimed at improving older drivers’ performance in hopes of reducing their crash risk, almost all the studies failed to show or did not address collision rate outcomes.

Conclusion

Overall, the most valuable approaches in terms of specific improvement of driving skills and performance are the ones that have put in place specific training curriculum for every single driver to tackle their specific weakness behind the wheel. This is not surprising, considering the known key-concept defined as transfer-appropriate practice [88]. One must develop his/her own capacity of error-detection in their driving if they want to be able to modify current behaviours and implement appropriate responses.

With the recent development of technology aimed at collecting driving information over longer periods of time (for example, see the SHRP 2 project: www.fhwa.dot.gov/goshrp2), evaluation of one driving’s ability could encompass more and longer driving periods, distance driven and more manoeuvres, allowing for a clearer depiction of what needs to be addressed by a driving instructor or an occupational therapist while developing specific interventions.

Moreover, the availability of computing power and artificial intelligence has brought a new set of tools that allows for automatic detection of driving errors and the possibility to provide automated feedback to the trainees [89]. However, since few of the studies have evaluated the long-term retentions of such interventions [56], future efforts should be made to include this important milestone in their projects. It is of real interest to know when a “refresher” session should be provided to prevent a decrease in performance following an improvement in their abilities [90].

Despite its complex implementation, attempts to combine the most efficient interventions presented in the current review are promising for the development of an efficient way to allow older drivers to maintain and even improve their skills, driving behaviours, and decrease their involvement in motor vehicle collisions.