Utilizing the DRUID® Impairment Assessment App to Predict Senior Adults’ Driving Performance

MassAITC study shows participants with higher DRUID impairment scores were less steady in their driving performance

CAMBRIDGE, MA -- Impairment Science, Inc. (ISI) has completed a pilot project that validates using DRUID® to predict the driving performance of senior adults, ages 64 to 85 years, a result that could open new commercial markets for the company.

As they age, seniors vary day to day in their cognitive and motor functioning. Having a reliable, valid, and easy-to-use tool to assess their functioning will make significant contributions to the health, safety, and quality of life of this growing population. DRUID is the most promising tool available.

The pilot project compared the performance of 40 seniors on a) Realtime Technology’s full-scale cab driving simulator, housed at the Human Performance Lab (HPL), Riccio College of Engineering, University of Massachusetts-Amherst (UMass-Amherst), and b) their pre-simulator performance on the 3-minute DRUID Benchmark app.

DRUID Benchmark Procedure. DRUID has users perform four tasks that measure cognitive and motor performance by assessing balance, decision-making accuracy, reaction time, hand-eye coordination, and time estimation under conditions of divided attention. The app collects and integrates hundreds of measurements to produce an impairment score that ranges from 25 to 75, with a higher score indicating greater impairment.

After introducing DRUID Benchmark, we asked the participants to watch a short video that presented an age peer who demonstrated the app’s four tasks. They completed three DRUID tests in a row and then up to two more tests with the goal of having two out of three sequenced tests produce impairment scores < 3.0 points apart. For research projects, we set the participants’ baseline score as the lowest impairment score they can produce when first using the app.

Driving Simulator: Scenarios. Working with HPL researchers, we created a series of short, self-contained mini scenarios rather than a single extended drive as is typically used in HPL’s driving simulator experiments. Each mini scenario presented a single, well-defined driving event under controlled conditions, after which the simulator environment was reset before proceeding to the next scenario. This approach minimized cumulative exposure to the driving simulator, important because older drivers often experience simulator sickness with longer drives.

The participants completed a total of eleven mini scenarios. The first, a two-minute introductory drive on a four-lane divided highway, served as a warm-up exercise which imposed minimal demands other than maintaining lane position and speed while traffic flowed in the opposite lane. For the 10 experimental mini scenarios, we sought to introduce a range of traffic conditions to test the participants’ driving abilities. Below is a list and brief description of the ten experimental drives we developed and analyzed.

HPL Driving Simulator

In developing the experimental scenarios, we sought to balance the need to a) avoid features that might induce simulator sickness and b) present enough challenges to create variable driving performances. Their duration ranged from 1 to 2.5 minutes, for a total of 18 minutes (20 minutes, including the two-minute warm-up drive).

• Scenario 1: Merging onto Highway. The driver is instructed to proceed down an on-ramp to enter a highway.

• Scenario 2: Tailgating. On a rural road, the driver approaches a vehicle traveling 5-10 mph below the speed limit, which they are expected to follow at a safe distance without tailgating. Note: Unfortunately, for technical reasons, we were unable to download data for this scenario.

• Scenario 3: Being Overtaken. On a divided four-lane highway, a large truck overtakes the driver’s vehicle on the left and remains adjacent for several seconds. The participants are expected to maintain stable lane position and, if needed, make speed adjustments.

Mini Scenarios: Varying Traffic Conditions

• Scenario 4: Occluding Truck. On a residential street, the driver is instructed to pass a delivery truck parked on the right side of the road that partially obscures the view of oncoming traffic, pedestrians, and potential hazards.

• Scenario 5: Hidden Driveway. On a residential street, a vehicle emerges unexpectedly from a driveway obscured by vegetation, prompting the driver to brake or otherwise respond safely.

• Scenario 6: Hedge at 3-Way. On a residential street, the driver proceeds to a three-way intersection with hedges obstructing the view on the right side, which requires them to approach cautiously, stop as indicated by traffic control devices, and visually check for hazards before making a right turn.

• Scenario 7: Sudden Yellow. On an urban street, a traffic signal at a four-way intersection changes to yellow just seconds before the driver reaches the stop line.

• Scenario 8: Straight Crossing Path. On an urban street, the driver proceeds toward an intersection where a car emerged from a street on the right side without stopping and turned left. They are expected to approach the intersection cautiously to check for other vehicles that might turn without stopping or yielding.

• Scenario 9: Left Turn at Path (1.5 minutes). On an urban street, the driver approaches a signalized three-way intersection and are instructed to turn left across oncoming traffic when safe.

• Scenario 10: Gap Acceptance (2 minutes). The driver is instructed to make a left turn at a signalized four-way intersection on urban street, with oncoming traffic spaced at increasingly long intervals.

Driving Simulator: Data Sources. The driving sessions generated three sources of data: the driving simulator, video recordings, and an integrated eye-tracking system.

Kinematic Data from the Simulator. The driving simulator records real-time kinematic metrics that indicate vehicle dynamics and positional changes during the driving scenarios. Key variables include vehicle speed, acceleration, deceleration, heading, lane position, steering angle, and coordinates within the simulated environment. This data reveals how participants control the vehicle under various roadway conditions. For instance, lane position data can be analyzed to assess how well drivers maintained lateral stability within the traffic lanes, while speed and acceleration profiles reveal their ability to manage longitudinal vehicle dynamics, and metrics such as steering inputs and braking forces provide insights into their motor control.

Behavioral Data from Video Recordings. Simultaneous video recordings of the participants’ hand, foot, and body movements can be analyzed to quantify their reactions to events (e.g., responding to a traffic signal change), stopping behaviors, lane-change actions (e.g., activation of turn signals, mirror-checking patterns, and execution of lane transitions), and other behavioral responses. In addition, these recordings make it possible to identify instances of incorrect or otherwise suboptimal responses that could reflect diminished situational awareness or cognitive processing.

Eye-Tracking Data. Integrated into the driving simulator cab, the SmartEye Pro system provides high-fidelity eye-tracking data to quantify visual attention and scanning patterns. Using four infrared cameras, the system records participants’ eye and head movements, including metrics such as gaze duration, gaze location, gaze frequency, and pupil diameter. This dataset can be analyzed to show how participants allocate their visual attention across areas of interest such as roadway configurations, traffic signals, signage, other vehicles, pedestrians, potential hazards, the dashboard, and rear and sideview mirrors.

The integration of these data provided a robust dataset for examining how the study participants physically operated the vehicle and how they visually attended to and cognitively processed the driving environments presented in the mini scenarios.

Driver Simulator: Procedure. An HPL researcher provided the lab’s standard instructions for driving on the simulator: 1) Drive as you normally would. 2) The car does not move, and instead what you see on the screens will change. 3) When you brake and come to a complete stop, the car will stay in the same position, and the screen projection will stop moving. 4) You will hear recorded driving instructions through the speaker in the car. Next, the HPL researcher calibrated the cameras in the car for the eye-tracking system.

Before beginning the experimental drives, the participants completed the introductory, two-minute drive so they would become comfortable with the simulator environment, the virtual layout of the simulated roadway, and the vehicle controls, thereby minimizing potential confusion or disorientation and ensuring their focus on the experimental mini scenarios. We presented the mini-scenarios individually, with about a 20-second break after each one while the HPL researcher cued up the next drive.

Preliminary Analyses. The participants ranged in age from 65 to 84 years, with approximately two-thirds in their 70s; 92.5% were White and 57.5% were female. Ninety percent had had at least some college after high school (50.0% with a master’s degree, 10.0% with a doctoral degree). Their median household income was in the $50,000-$74,999 range.

The pilot study’s principal goal was to validate DRUID for senior adults by assessing how well the Benchmark app predicts driving simulator errors. Below, we report a set of preliminary analyses that tested whether the study participants’ DRUID baseline scores could predict any of several outcome measures derived from their performance on the driving simulator. In all cases, we report Pearson’s correlation coefficients (r) with two-tailed tests of statistical significance.

DRUID Baseline Scores. As noted, ISI determined that for our research, the DRUID baseline score should be the lowest impairment score that a study participant can produce when first using the app. In this case, participants completed between 3 and 5 tests.

Baseline DRUID Benchmark Scores (N = 40)

Two participants were unable to use DRUID effectively, with consistently high pre-session scores, well above the scores for the other participants, and high post-simulator scores. We dropped them from all subsequent analyses but will revisit that decision as we continue to work with the data.

Baseline Scores and Age. A consistent finding in ISI’s studies is that user age is correlated with baseline DRUID scores. We replicated that finding even though the participants’ ages ranged only between 65 and 84 [r = .37, p = .029, N =35].

Baseline Scores and Hours Since Last Sleep. The more hours since the participants last slept, the lower their baseline DRUID score was. This finding suggests that, for this study population, drowsiness upon waking up diminishes over time [r = -.34, p = .044, N =35].

Baseline Scores and Post-Session Alertness. Higher DRUID scores predicted the participants’ self-rated alertness following the driving simulator session (r = -.27, p < .001, N = 36). This remained the case when controlling for the sum of their four post-session ratings of how mentally and physically demanding they found DRUID and the driving simulator to be.

Driving Outcomes by Scenario. With each scenario, we examined the correlations between the participants’ baseline DRUID scores and several driving simulator outcome measures.

In all nine operative scenarios, the baseline scores were negatively correlated with the percentage of time the participants looked straight ahead during the drive, that is, the higher their baseline impairment score, the less they looked straight ahead.

As shown below, Pearson’s correlation coefficients (r) ranged between -0.32 and -0.48 and were statistically significant for six scenarios and approached significance in two others. The correlation for Scenario 3, Being Overtaken, was not statistically significant [r = -032, p < .140].

 Scenario 1, Merging onto Highway: r = -0.35, p < .044*

 Scenario 4, Occluding Truck: r = -0.38 p < .023*

 Scenario 5, Hidden Driveway: r = -0.40, p < .023*

 Scenario 6, Hedge at 3-Way: r = -0.48, p < .004*

 Scenario 7, Sudden Yellow: r = -0.44, p < .009*

 Scenario 8, Straight Crossing Path: r = -0.33, p < .054

 Scenario 9, Left Turn at Path: r = -0.33, p < .058

 Scenario 10, Gap Acceptance: r = -0.37, p < .029*

This set of analyses revealed several additional significant findings:

 In Scenario 4, Occluding Truck, higher baseline DRUID scores predicted a lower maximum braking force when stopping the car at the end of the drive (r = -.38, p = .021, N=36).

 In Scenario 6, Hidden Driveway, higher baseline DRUID scores predicted a) lower greatest rate of deceleration (r = -0.43, p = .008. N = 36); b) greater variability in the rate of acceleration (r = 0.35, p = .039, N = 36); and c) greater maximum braking pressure (r = 0.41, p = .012, N = 36).

 In Scenario 9, Left Turn at Path, higher baseline DRUID scores predicted a) a greater percentage of time above the speed limit (r = .47, p = .003, N = 37) and b) a greater number of times checking the speedometer (r = .42. p = .01, N =37).

 In Scenario 10, Gap Acceptance, higher baseline DRUID scores predicted a higher maximum rate of deceleration when stopping the car at the end of the drive (r = .42, p = .02, N = 29).

Driving Outcomes: All Scenarios Combined. There were additional statistically significant correlations between the participants’ baseline DRUID scores and various driving simulator-based outcome measures that were common to multiple scenarios. Higher baseline scores predicted:

 More changes in accelerator pressure when initiating the drive (r = 0.41, p = .011, N = 38), when approaching a stop sign (r = 0.46, p = .004, N = 38), and when making a left turn (r = 0.52, p < .001, N = 38).

 Fewer lateral lane deviations, moving to the left outside the marked lane (r = -0.33, p = .042, N = 38); and

 More speedometer checks (r = 0.34, p = .036, N = 38);

As noted previously, for other outcome measures we inspected recorded videos of the participants’ driving. Watching each video multiple times, HPL researchers documented whether the participants followed a predetermined list of safe driving practices or made any critical errors.

The HPL researchers assigned a score to each item on the list (0 = No, 1 = Yes). For an exploratory analysis, we examined the total number of “good driving points” the participants earned during Scenario 10, Gap Acceptance. The participants’ baseline DRUID scores correlated positively with the total number of good driving points (r = .31, p = .070, N = 34). Similar analyses are needed for the other experimental scenarios.

Conclusion. The DRUID app is a promising tool for objectively assessing whether aging senior adults should drive at a particular time or be formally evaluated to gauge their capacity to drive a car safely.

Our preliminary analyses showed that, on a few outcome measures, participants with higher DRUID impairment scores were less steady in their driving performance, as evidenced by more changes in accelerator pressure and greater variability in their rate of deceleration.

Other findings indicated that participants with higher DRUID impairment scores were driving more cautiously, with fewer lane deviations, more speedometer checks, lower maximum deceleration rates, and greater maximum braking pressure. A likely explanation is that, as senior adults, those who performed poorly on the DRUID app became anxious about doing well on the driving simulator and therefore became especially motivated during their 20-minute session to demonstrate satisfactory driving skills.

Future Steps. With this pilot project, we have collected a wealth of data using a novel analysis approach for driver simulator data. We have additional work to do to mine this data, refine and expand the preliminary analyses we have reported here, and update our findings.

We envision submitting a proposal to the National Institute on Aging (NIA) to build on this pilot study. If so, our study protocol could be revised in two important ways.

First, we would recruit participants who would agree to use DRUID for an extended period of time prior to their driving simulator session. As noted, we suspect that having our study participants learn DRUID immediately beforehand made the “stereotype threat” that seniors often experience when their cognitive and motor functioning is brought into question. The revised procedure would obviate this problem while also establishing more reliable baseline scores during the study session.

Second, we would develop the scenarios with our novel data analysis approach in mind to ensure that participants complete each drive and are not timed out prematurely, and to make the data extraction process less onerous. Importantly, this would give us the opportunity to reconsider how challenging the scenarios should be. As we developed the set of present mini scenarios, we were mindful of the need to avoid having the study participants experience stimulator sickness, but it may be that having only short scenarios with relatively few challenging road conditions reduced the variance in the participants’ driving performance.