A Motor Vehicle Crash as an Early Signal of Alzheimer's Disease and Related Dementias

ABSTRACT

Background

People with Alzheimer's disease and related dementias (ADRD) are at increased risk of motor vehicle crash (MVC) due to cognitive decline, which often begins years before diagnosis. MVC could serve as an indicator of ADRD‐related cognitive impairment and facilitate earlier diagnosis.

Methods

We conducted a matched cohort study of Medicare beneficiaries in New Jersey aged ≥ 69 years with incident ADRD diagnoses from 2010 to 2017 compared to individuals without ADRD and with four other medical conditions: acute myocardial infarction (AMI), asthma, chronic obstructive pulmonary disease (COPD), and diabetes. Comparison groups were sampled with replacement and matched on sex, age, and calendar quarter of diagnosis or claim (for the non‐ADRD group). Crash prevalence at 6 months and 1, 2, and 3 years before diagnosis was compared using chi‐square tests, and the monthly prevalence was plotted to examine longitudinal trends in crashes.

Results

We identified 43,115 eligible individuals with incident ADRD diagnosis from 2010 to 2017. MVC crash prevalence was higher in the ADRD group than in non‐ADRD and negative control groups at all measured pre‐diagnosis time points: 6 months (2.99% vs. 2.31%–2.53%), 1 year (5.26% vs. 4.18%–4.65%), 2 years (9.52% vs. 7.88%–8.67%), and 3 years (13.64% vs. 11.29%–12.37%). The ADRD group was more likely to have multiple crashes in the pre‐diagnosis period, but multiple events were rare overall (< 3%). Differences were statistically significant but represented small changes in prevalence. Crash prevalence did not reveal a clear signal of cognitive decline in the years leading up to the index diagnosis.

Conclusions

Older adults with ADRD had slightly higher prevalence of MVC in the 3 years before diagnosis than matched comparison groups, but absolute differences were small and recurrent crashes were rare. MVC alone is too rare to serve as a signal of incipient ADRD, highlighting the need for more sensitive measures.

Summary

• Key Points

  • ○
    Older adults with ADRD experienced a slightly higher prevalence of motor vehicle crashes in the 3 years prior to diagnosis compared to matched controls.
  • ○
    Monthly crash prevalence in the 3 years before diagnosis varied substantially across all groups but did not show a consistent or distinct pre‐diagnosis acceleration in the ADRD group.
  • ○
    Motor vehicle crashes are not a reliable population‐level screening tool for early ADRD; whether a crash event might serve as a useful prompt for individualized cognitive assessment remains a hypothesis to be evaluated in future research.
• Why Does This Paper Matter?

  • ○
    This study suggests that police‐reported crash history alone is too infrequent and nonspecific to function as a population‐level screening signal for early ADRD, highlighting the need for more sensitive measures.

Abbreviations

AD

Alzheimer's disease

ADRD

Alzheimer's disease and related dementias

AMI

acute myocardial infarction

CCW

chronic conditions warehouse

COPD

chronic obstructive pulmonary disease

HRS

health and retirement survey

ICD

international classification of disease

NJ‐SHO

New Jersey safety and health outcomes

1. Introduction

Individuals with Alzheimer's disease and related dementias (ADRD) are at increased risk of motor vehicle crashes due to cognitive impairments that characterize the disease [1, 2, 3]. While there is no known cure for ADRD, early detection can improve clinical outcomes by allowing individuals to appropriately treat symptoms as they arise and avoid medications that may worsen symptoms, as well as identify candidates for newly approved disease modifying treatments [4, 5, 6, 7, 8].

Driving is a cognitively complex skill, requiring visuo‐spatial awareness and executive functioning, which are compromised in people with ADRD. However, clinical studies indicate that cognitive impairment begins in a prodromal phase, when many older adults wait up to 3 years to seek medical care [9, 10]. For instance, prior research using naturalistic driving technology (data collection equipment installed in an individual's car that records features of their driving) found that unsafe driving behaviors associated with crash, such as rapid braking and lane deviations, were predictive of ADRD pathology even when individuals may still present as “cognitively normal” on neuropsychological testing [11, 12, 13].

Because cognitive changes can occur years before a dementia diagnosis, subtle changes in daily functioning other than driving may serve as indicators of prodromal ADRD [14, 15, 16]. For instance, a study linking Medicare claims to individual‐level credit data found that individuals diagnosed with Alzheimer's disease were more likely to miss credit payments up to 6 years before diagnosis [17]. Other studies have shown similar impairments in tasks, such as grocery shopping, preparing meals, and medication management [18].

To be effective, however, a driving screening tool needs to be specific enough to limit the prevalence of false positives and simple enough to use to make it accessible. Unlike other possible markers of declining driving ability, such as lane deviation, braking variability, or restricted nighttime driving, which require specific in‐vehicle technology, a police reported crash is a well‐documented and clinically significant public safety event that patients, and often their family members, can reliably recall and discuss with healthcare providers during outpatient visits. This study therefore aimed to examine the frequency of crashes in the 3 years prior to ADRD diagnosis as a potential early “signal” of cognitive decline.

2. Methods

2.1. Data Sources and Linkage

Data for this study came from the New Jersey Safety and Health Outcomes (NJ‐SHO) Data Warehouse, which contains licensing records for all New Jersey drivers between 2004 and 2019 and every police crash report filed over this same period [19]. A crash is reportable in New Jersey if it results in injury, death, or more than $500 in property damage [20]. Each individual is linked to every reported crash from January 1, 2007 through December 31, 2017. For our analyses, we limited crashes to those in which the individual was the driver (a “crash‐involved driver”) and did not include any crashes in which the individual was a passenger or pedestrian.

The NJ‐SHO data are linked to Medicare claims using a strict matching algorithm that includes last name, birthdate, sex, and zip code. Medicare data includes dates and International Classification of Disease (ICD) diagnosis codes for all inpatient, outpatient, hospice, and prescription drug claims over the study period. The study was reviewed and approved by the Brown University Institutional Review Board. Additional information about the data and methods used for this study is available on Github: https://doi.org/10.5281/zenodo.17201755.

2.2. Study Population

The study population included Medicare fee‐for‐service beneficiaries in New Jersey with an incident diagnosis of ADRD from January 1, 2010 through December 31, 2017, using the validated Bynum‐Standard algorithm [21]. We selected the Bynum‐Standard algorithm over other definitions, such as the definition provided by the Chronic Conditions Warehouse (CCW), because it provides higher specificity and a clearly defined incident diagnosis date. Individuals with ADRD were matched 1:1 to five groups: a “no ADRD” control group comprised of individuals with a healthcare visit and no prior ADRD diagnosis and four negative control groups, each with an incident diagnosis of either acute myocardial infarction (AMI), asthma, chronic obstructive pulmonary disease (COPD), or diabetes. We selected four comparison conditions (incident AMI, asthma, COPD, and diabetes) that have claims‐defined incident dates and regular healthcare contact to contextualize whether crash occurrence patterns observed prior to ADRD diagnosis were specific to ADRD versus also present around other incident diagnoses.

Matching was performed with replacement on age, sex, and calendar quarter of diagnosis or healthcare encounter for the no‐ADRD group. When multiple healthcare encounters occurred within a quarter, one was randomly selected. A single individual in a control group could serve as a match for more than one individual with ADRD.

Eligible individuals had to be 69 years or older and enrolled in Medicare due to age (i.e., not due to disability, end‐stage renal disease, or amyotrophic lateral sclerosis), continuously enrolled in Medicare fee‐for‐service parts A, B, and D in the 4 years before the ADRD or comparison diagnosis, in continuous possession of a valid New Jersey license, and a New Jersey resident for the 3 year look‐back. We limited the population to beneficiaries 69 and older to ensure that each beneficiary had at least 3 years of continuous enrollment prior to the diagnosis and then one additional year of continuous enrollment to measure clinical characteristics and medication use prior to the “follow‐back” period (i.e., the 3 years prior to the index date). We also required at least one healthcare encounter in the 14 months before the index date to ensure opportunity for ADRD detection.

2.3. Incident Diagnoses

An incident diagnosis was defined as the first diagnosis for the condition (ADRD, AMI, asthma, COPD, or diabetes) after a three‐year period with no diagnoses. The date of incident ADRD diagnosis was identified using the Bynum‐Standard algorithm with a three‐year look‐back, which uses multiple sources of Medicare claims over the specified time period to identify beneficiaries newly diagnosed with ADRD [21]. The algorithm has been validated through prior work in the Health and Retirement Survey (HRS). The agreement between the Bynum‐Standard algorithm and HRS assessments was 89%, with a 60% positive predictive value and a 92% negative predictive value [21]. The date of diagnosis for the negative control groups was obtained from the Medicare CCW Chronic Conditions Segment [22]. Individuals with prior ADRD diagnoses were excluded from control groups.

2.4. Statistical Analyses

We compared sociodemographic characteristics, pre‐existing health conditions, and prevalent medication use across the ADRD and matched control groups measured in the year prior to the three‐year “follow‐back” period. We used standardized mean differences (SMDs) to compare individual characteristics between groups and considered a difference of greater than 0.1 to indicate substantive differences between groups.

To examine longitudinal trends in the 3 years leading up to an incident diagnosis of ADRD, one of the comparison diagnoses, or the health care encounter among the non‐ADRD group, we calculated the prevalence of motor vehicle crashes in each month and constructed confidence intervals around the estimates. We then plotted longitudinal monthly crash prevalence in the 3 years before the index month to identify any trends and compared the prevalence in each month prior to the index date to the prevalence in the month of the index date using a chi‐square test for association. Last, we calculated the prevalence of motor vehicle crash in the 6 months, 1 year, 2, and 3 years before the index date and used chi‐square tests of association to compare prevalence across comparison groups. Data analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC, USA) and R 4.1.3 (R Foundation for Statistical Computing, Vienna, Austria). All analyses compared the occurrence of police‐reported motor vehicle crashes across groups without direct measurement of driving exposure, such as miles driven or driving frequency. Although possession of a valid driver's license was required, individuals may have differed substantially in how often or under what conditions they drove. Consequently, the estimates presented here reflect differences in crash occurrence rather than differences in crash prevalence per unit of driving exposure.

3. Results

We identified a total of 43,115 eligible individuals with an incident ADRD diagnosis from 2010 through 2017. After matching on age, sex, and calendar quarter of diagnosis/index date, the sociodemographic characteristics between ADRD and comparison groups were similar (Table 1). Individuals with ADRD diagnosis had a higher prevalence of comorbidities (Table 2); for example, ADRD individuals had an elevated prevalence of anxiety (23.6% vs. 14.4%–18.8%), mobility impairment (6.8% vs. 2.5%–4.6%), and schizophrenia and psychotic disorders (3.7% vs. 0.6%–1.3%).

TABLE 1.

Sociodemographic characteristics after 1:1 matching.

	ADRD	No ADRD	SMD	AMI	SM	Asthma	SMD	COPD	SMD	Diabetes	SMD
	N = 43,115	N = 43,115		N = 42,665	D	N = 42,191		N = 42,922		N = 42,731
Male Sex, n (%)	19,496 (45.2)	19,496 (45.2)	0.00	19,292 (45.2)	0.00	18,962 (44.9)	0.00	19,408 (45.2)	0.00	19,293 (45.1)	0.00
Race and ethnicity, n (%)			0.08		0.07		0.06		0.06		0.05
Asian	294 (0.7)	407 (0.9)		302 (0.7)		377 (0.9)		281 (0.7)		382 (0.9)
Black	2781 (6.5)	2123 (4.9)		2223 (5.2)		2264 (5.4)		2240 (5.2)		2509 (5.9)
Hispanic	292 (0.7)	226 (0.5)		178 (0.4)		249 (0.6)		238 (0.6)		241 (0.6)
Unknown/Other	472 (1.1)	576 (1.3)		451 (1.1)		543 (1.3)		474 (1.1)		656 (1.5)
White	39,276 (91.1)	39,783 (92.3)		39,511 (92.6)		38,758 (91.9)		39,689 (92.5)		38,943 (91.1)
Age
Mean (SD)	82.7 (6.5)	82.7 (6.5)	0.00	82.5 (6.3)	0.00	82.4 (6.3)	0.00	82.6 (6.4)	0.00	82.6 (6.3)	0.00
Categories, n (%)			0.00		0.00		0.00		0.00		0.00
69–75	5522 (12.8)	5522 (12.8)		5522 (12.9)		5522 (13.1)		5522 (12.9)		5522 (12.9)
76–80	8079 (18.7)	8079 (18.7)		8079 (18.9)		8079 (19.1)		8079 (18.8)		8079 (18.9)
81–85	11,510 (26.7)	11,510 (26.7)		11,510 (27.0)		11,510 (27.3)		11,510 (26.8)		11,510 (26.9)
86+	18,004 (41.8)	18,004 (41.8)		17,554 (41.1)		17,080 (40.5)		17,811 (41.5)		17,620 (41.2)

Note: SMD compares the ADRD group and each respective control group.

Abbreviations: ADRD, Alzheimer's disease and related dementias; AMI, acute myocardial infarction; COPD, chronic obstructive pulmonary disease.

TABLE 2.

Health characteristics after 1:1 matching.

	ADRD	No ADRD	SMD	AMI	SMD	Asthma	SMD	COPD	SMD	Diabetes	SMD
	N = 43,115	N = 43,115
				N = 42,665		N = 42,191		N = 42,922		N = 42,731
Diagnoses, n (%)
AMI	3535 (8.2)	2265 (5.3)	0.12	0 (0.0)	0.42	3710 (8.8)	0.02	3458 (8.1)	0.01	2842 (6.7)	0.06
Anxiety	10,168 (23.6)	6227 (14.4)	0.24	7488 (17.6)	0.15	7913 (18.8)	0.12	7274 (16.9)	0.17	6869 (16.1)	0.19
Arthritis	30,761 (71.3)	29,022 (67.3)	0.09	29,479 (69.1)	0.05	31,089 (73.7)	0.06	29,920 (69.7)	0.04	29,126 (68.2)	0.07
Asthma	6876 (15.9)	5791 (13.4)	0.07	7413 (17.4)	0.04	0 (0.0)	0.62	7227 (16.8)	0.02	6577 (15.4)	0.02
Cataracts	36,635 (85.0)	36,731 (85.2)	0.01	35,822 (84.0)	0.03	36,374 (86.2)	0.04	36,087 (84.1)	0.02	35,849 (83.9)	0.03
Chronic kidney disease	16,158 (37.5)	11,981 (27.8)	0.21	20,566 (48.2)	0.22	17,019 (40.3)	0.06	16,443 (38.3)	0.02	12,350 (28.9)	0.18
COPD	14,995 (34.8)	11,241 (26.1)	0.19	16,513 (38.7)	0.08	23,662 (56.1)	0.44	0 (0.0)	1.03	13,592 (31.8)	0.06
Diabetes	22,920 (53.2)	19,885 (46.1)	0.14	24,554 (57.6)	0.09	22,568 (53.5)	0.01	22,061 (51.4)	0.04	0 (0.0)	1.51
Epilepsy	1471 (3.4)	557 (1.3)	0.14	985 (2.3)	0.07	715 (1.7)	0.11	793 (1.8)	0.10	738 (1.7)	0.11
Glaucoma	16,055 (37.2)	16,255 (37.6)	0.01	15,419 (36.1)	0.02	15,797 (37.4)	0.01	15,911 (37.1)	0.00	15,215 (35.6)	0.03
Hearing impairment	9984 (23.2)	9338 (21.7)	0.04	8566 (20.1)	0.07	9869 (23.4)	0.01	9212 (21.5)	0.04	8893 (20.8)	0.05
Heart failure	20,152 (46.7)	14,769 (34.3)	0.26	27,984 (65.6)	0.39	23,137 (54.8)	0.17	22,103 (51.5)	0.10	18,033 (42.2)	0.09
Hypertension	40,478 (93.9)	39,373 (91.3)	0.10	40,658 (95.3)	0.06	39,930 (94.6)	0.04	40,166 (93.6)	0.01	39,789 (93.1)	0.03
Ischemic heart disease	32,603 (75.6)	29,273 (67.9)	0.17	41,853 (98.1)	0.71	33,330 (79.0)	0.09	32,820 (76.5)	0.02	30,830 (72.1)	0.08
Migraine	1565 (3.6)	1452 (3.4)	0.01	1297 (3.0)	0.03	1647 (3.9)	0.01	1418 (3.3)	0.02	1540 (3.6)	0.00
Mobility impairment	2929 (6.8)	1097 (2.5)	0.20	1978 (4.6)	0.09	1498 (3.6)	0.15	1617 (3.8)	0.14	1421 (3.3)	0.16
Perivascular disease	19,567 (45.4)	15,403 (35.7)	0.20	20,805 (48.8)	0.07	19,367 (45.9)	0.02	18,292 (42.6)	0.06	16,185 (37.9)	0.15
Schizophrenia and psychotic disorders	1614 (3.7)	249 (0.6)	0.22	534 (1.3)	0.16	446 (1.1)	0.18	462 (1.1)	0.18	335 (0.8)	0.20
Stroke	12,223 (28.3)	7006 (16.2)	0.29	11,007 (25.8)	0.06	9025 (21.4)	0.16	8806 (20.5)	0.18	8223 (19.2)	0.21
Visual impairment	285 (0.7)	139 (0.3)	0.05	215 (0.5)	0.02	210 (0.5)	0.02	170 (0.4)	0.04	169 (0.4)	0.04

Note: smd compares the ADRD group and each respective control group.

Abbreviations: ADRD, Alzheimer's disease and related dementias; AMI, acute myocardial infarction; COPD, chronic obstructive pulmonary disease.

Motor vehicle crash was more common among those in the ADRD group at all measured time points. Although the p value for the comparison was < 0.001 for virtually all comparisons, the magnitude of the difference was relatively small (6 months: 3.0% vs. 2.3%–2.5%; 1 year: 5.3% vs. 4.2%–4.7%; 2 years: 9.5% vs. 7.9%–8.7%; 3 years: 13.6% vs. 11.3%‐12.4%) (Table 3). Similarly, while individuals with ADRD diagnosis had a similar or greater prevalence of multiple crashes by each time point (p < 0.001 for all but two comparisons) the absolute differences were relatively small (6 months: 0.1% vs. 0.1%–0.1%; 1 year 0.4% vs. 0.2%–0.2%; 2 years: 1.1% vs. 0.6%–0.7%; 3 years 2.1% vs. 1.3%–1.5%) (Table 3). When we plotted the prevalence of a crash in the months leading up to a diagnosis, there was substantial variation in the measure across all groups although no obvious trends over time (Figure 1). When we compared the prevalence in each month prior to the index date to the prevalence in the month of the index date, we found no meaningful trend suggesting there was a point at which the prevalence significantly differed from that of the index date for any of the groups with the exception of AMI, where the prevalence was elevated for the entire “follow‐back” period (Figure 2).

TABLE 3.

Prevalence of police‐reported motor vehicle crashes prior to index month.

	ADRD	No ADRD	AMI	Asthma	COPD	Diabetes
	N = 43,115	N = 43,115	N = 42,665	N = 42,191	N = 42,922	N = 42,731
	% (95% CI)	% (95% CI)	% (95% CI)	% (95% CI)	% (95% CI)	% (95% CI)
Crash within 6 months	2.98 (2.8, 3.1)	2.29 (2.15, 2.43)	2.31 (2.17, 2.46)	2.53 (2.38, 2.68)	2.45 (2.30, 2.60)	2.44 (2.30, 2.59)
> 1 crash within 6 months	0.1 (0.1, 0.2)**	0.1 (0.00, 0.1)	0.1 (0.0, 0.1)	0.12 (0.1, 0.2)*	0.1 (0.0, 0.1)	0.0 (0.00, 0.1)
Crash within 1 year	5.3 (5.1, 5.5)	4.2 (4.0, 4.4)	4.3 (4.1, 4.5)	4.7 (4.5, 4.9)	4.5 (4.3, 4.7)	4.4 (4.2, 4.6)
> 1 crash within 1 year	0.4 (0.3, 0.5)	0.2 (0.1, 0.2)	0.2 (0.2, 0.3)	0.2 (0.2, 0.3)	0.2 (0.2, 0.3)	0.2 (0.2, 0.2)
Crash within 2 years	9.5 (9.2, 9.8)	7.9 (7.6, 8.1)	8.2 (7.9, 8.4)	8.3 (8.0, 8.5)	8.7 (8.4, 8.9)	8.2 (7.9, 8.4)
> 1 crash within 2 years	1.1 (1.0, 1.2)	0.6 (0.6, 0.7)	0.7 (0.6, 0.8)	0.7 (0.6, 0.8)	0.7 (0.6, 0.8)	0.7 (0.6, 0.8)
Crash within 3 years	13.6 (13.3, 14.0)	11.3 (11.0, 11.6)	12.1 (11.8, 12.4)	11.9 (11.6, 12.2)	12.4 (12.1, 12.7)	12.0 (11.7, 12.3)
> 1 crash within 3 years	2.1 (1.9, 2.2)	1.3 (1.2, 1.4)	1.4 (1.3, 1.5)	1.4 (1.2, 1.5)	1.5 (1.4, 1.6)	1.3 (1.2, 1.4)

Note: The p value of the comparison of prevalence in the ADRD group to each comparison group was < 0.001 for all comparisons except those noted here *p = 0.440, **p < 0.003.

Abbreviations: ADRD, Alzheimer's disease and related dementias; AMI, acute myocardial infarction; CI, confidence interval; COPD, chronic obstructive pulmonary disease.

FIGURE 1.

Monthly prevalence of motor vehicle crashes in the 3 years prior to index date, defined by either an incident diagnosis of ADRD, one of four negative control conditions or a health care visit (no ADRD group). ADRD, Alzheimer's disease and related dementias; AMI, Acute myocardial infarction; COPD, Chronic pulmonary disease.

FIGURE 2.

Percent difference in the monthly prevalence of motor vehicle crashes in the 3 years prior to index date relative to the index date, defined by either an incident diagnosis of ADRD, one of four negative control conditions or a health care visit (no ADRD group). ADRD, Alzheimer's disease and related dementias; AMI, Acute myocardial infarction; COPD, Chronic pulmonary disease.

4. Discussion

We compared the prevalence of motor vehicle crash in the 3 years prior to an ADRD diagnosis to five control groups evaluated over the same period. Monthly crash prevalence varied substantially across all groups, and comparisons of each pre‐index month to the index month did not show a consistent late pre‐diagnosis acceleration in the ADRD group. Although individuals diagnosed with ADRD were more likely than any of the control groups to be involved in a crash prior to their diagnosis, the differences were too small to serve as an early signal of preclinical or prodromal [16] disease at a population level.

People with ADRD experience symptoms of cognitive decline long before they are officially diagnosed [15, 23, 24]. However, motor vehicle crashes are a relatively rare event, and people with undiagnosed ADRD may be self‐regulating their driving in response to their declining abilities, thereby limiting their exposure. Self‐regulation of driving (e.g., driving less often, only during the day, or avoiding risky environmental conditions) is well documented in people who have been diagnosed with ADRD, as well as the prodromal [14] stage of mild cognitive impairment, and might explain the relatively low rates of crash among cognitively impaired older drivers [25, 26, 27, 28, 29, 30].

While there is no known cure for ADRD, early detection can improve clinical outcomes by allowing individuals to appropriately treat symptoms as they arise and avoid medications that may worsen symptoms, as well as identify candidates for newly approved disease modifying treatments [4, 5, 6, 7, 8]. To be effective, however, a screening tool needs to be specific enough to limit the prevalence of false positives and simple enough to use to make it accessible. Although our findings suggest that a motor vehicle crash may not be an appropriate broad‐based screening tool for this purpose, a motor vehicle crash may still serve as a useful signal that an older adult should undergo additional screening. Such considerations are consistent with existing clinical guidance for evaluating driving risk in older adults with cognitive impairment, but our findings should not be interpreted as evidence that crash history per se identifies prodromal ADRD independent of the higher burden of neurologic, psychiatric, and functional comorbidity in this group [31].

We chose to match our ADRD group to five control groups on age, sex, and calendar quarter, but not on other characteristics that may be associated with increased risk of crash. As a result, the lack of a difference in crashes could be due to differences in the characteristics of these groups aside from their sex or age. We chose not to match (or adjust) for other characteristics because our intent was to determine if a motor vehicle crash could be used as a simple screening tool in a real‐world clinical setting, where these differences naturally occur and are often not known to the clinician. Our intent was not to identify a predictive algorithm for ADRD with motor vehicle crash as an input.

A major strength of our study was the use of a novel dataset linking health records with police crash reports in a large population of older drivers. This linkage allowed us to connect the exact dates of diagnosis and crash, which few other researchers have been able to do. However, our study also includes limitations worth mentioning. First, we were only able to measure crashes that resulted in a police report. Smaller crashes, such as “fender benders,” that did not result in an injury, death, or significant property damage were not included. Thus, we may be missing a significant number of events that could serve as a signal of preclinical or prodromal ADRD. This could be particularly problematic if people with prodromal ADRD are self‐regulating their driving (i.e., choosing to drive only during the day, avoid highways, or limit other riskier environments), as smaller incidents may be more common. Second, while we used a common, validated algorithm to identify people with ADRD, the algorithm, which has an estimated PPV of 60%, may have resulted in misclassification within the ADRD group that could have diluted the signal necessary to more accurately identify a higher risk of motor vehicle in the years leading up to an ADRD diagnosis. Third, the algorithm does not distinguish the pathology of the disease. If the risk of a crash in the period prior to a diagnosis varies by type of ADRD (e.g., Alzheimer's as compared to Lewy Body Dementia), then our findings will reflect a combination of risks, with no way to distinguish the different trajectories.

The modest differences observed in police‐reported crashes likely reflect the fact that crashes are rare events and represent a late and relatively blunt marker of declining driving safety. From a clinical perspective, crash events may therefore be best viewed as a pragmatic trigger for individualized evaluation rather than a screening tool. Higher‐resolution measures of driving behavior and emerging biomarkers may ultimately complement crash history by identifying cognitive change at earlier stages when more proactive clinical approaches may actually limit the potential for an older driver to be involved in a crash. However, their value should be interpreted in light of the limited sensitivity of crash events observed in the present study.

Author Contributions

N.R.J., A.R.Z., S.B., and M.R.P. conceptualized the study; S.B. and M.R.P. conducted data management and analyses with critical input from N.R.J.; S.B., M.R.P., and N.R.J. constructed all tables and figures; N.R.J. and S.B. wrote the first draft of the manuscript with critical input from R.G., M.R.P., and A.R.Z.; A.E.C., F.B., S.A.M., and B.R.O. reviewed the manuscript and provided substantive expertise on interpretation of findings; A.R.Z. and A.E.C. provided data for the analyses.

Funding

This work is supported by the National Institute on Aging (R01AG065722, R01AG065722‐S1, R01AG079295), Joyce, Zullo, Pfeiffer, Curry, Ott, Margolis; and the Eunice Kennedy Shriver National Institute of Child Health and Human Development at the National Institutes of Health (R01HD096221), Curry, Pfeiffer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The sponsor had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.

Conflicts of Interest

A.R.Z. received grant funding from Sanofi paid directly to Brown University for research on the epidemiology of infections and vaccinations in nursing home residents and infants. B.O. served as Chair of the Data Safety and Monitoring Board of a study on older drivers that was funded by the National Institute on Aging; he serves on the Data Safety and Monitoring Boards of clinical trials funded by the National Institute on Aging, the Institute for Memory Impairments and Neurological Disorders, and Lexeo Therapeutics.