Injury, Int. J. Care Injured 46 (2015) 1497–1502
Contents lists available at ScienceDirect
Injury journal homepage: www.elsevier.com/locate/injury
Acute medical impairment among elderly patients involved in motor vehicle collisions Scott T. Youngquist a,*, Michael Liao b,c, Sydney Hartsell a,[3_TD$IF]1, Melissa Walker a, Nathan J. Kartchner a, Raminder Nirula d a
Department of Surgery, Division of Emergency Medicine, University of Utah School of Medicine, Salt Lake City, UT, United States Department of Emergency Medicine, Denver Health Medical Center, Denver, CO, United States Department of Emergency Medicine, North Suburban Medical Center, Thornton, CO, United States d Department of Surgery, Section of Acute Care Surgery, University of Utah School of Medicine, Salt Lake City, UT, United States b c
A R T I C L E I N F O
A B S T R A C T
Article history: Accepted 2 April 2015
Background: The association between acute medical illness and motor vehicle collisions (MVCs) among elderly emergency department patients is unclear. We sought to determine the prevalence of acute medical conditions that might impair driving ability among the elderly involved in MVCs and determine if there was an increased risk of the driver having an acute medical condition compared to similarly aged passengers. Methods: We reviewed charts of patients aged 65 years or older whose emergency department visit was prompted by a motor vehicle collision between 1 July 2000 and 30 June 2010 at two Level 1 trauma centres. The exposure of interest was occupancy status (driver vs. passenger), and the outcome measure was the presence of any predefined acute medical illness that might impair driving ability. Results: Final analysis included 871 drivers (cases) and 307 passengers (controls). An acute medical illness was recorded in 107 patients (9%): 97 drivers (11%) and 10 passengers (3%). Compared to passengers, drivers had significantly higher odds of presenting with acute medical illness (OR 3.7, 95% CI 1.9–7.2). After controlling for potential confounders, the adjusted odds ratio was 5.5 (95% CI 2.3–13.0). Conclusion: Acute medical conditions are a moderately common diagnosis among elderly drivers, presenting in about one in ten patients. A difference in the risk of finding an acute medical illness when comparing elderly drivers and passengers evaluated in the emergency department after a collision suggests the need for considering additional diagnostic investigation and post-discharge surveillance in this population. ß 2015 Elsevier Ltd. All rights reserved.
Keywords: Motor vehicle collision Medical impairment Elderly
Introduction The initial evaluation of elderly patients involved in motor vehicle collisions (MVCs) focuses primarily on the rapid identification and stabilisation of injuries. Secondarily, however, such encounters elicit concern over the potential for an acute medical illness acting as a causal factor in the collision. While an extensive body of literature exists regarding functional impairments in driving ability associated with chronic medical illness [1], the risk
* Corresponding author at: University of Utah Medical Center, Division of Emergency Medicine, 30 North 1900 East, Room 1C026, Salt Lake City, UT 84132, United States. Tel.: +1 801 683 9055; fax: +1 +801 581 2730. E-mail address: [email protected] (S.T. Youngquist). [4_TD$IF]1 Ms. Hartsell is now at the University of North Carolina School of Medicine, Chapel Hill, NC, United States. http://dx.doi.org/10.1016/j.injury.2015.04.012 0020–1383/ß 2015 Elsevier Ltd. All rights reserved.
of individual disease categories on collision risk [2], as well as the influence of age-related physiologic changes alone [3], little is known regarding the contribution of acute medical illness to the epidemiology of MVCs among this patient population from the immediate post-incident perspective in the emergency department. We considered a focus on the elderly driving population, frequently defined in the medical literature as individuals aged 65 or older, as important and relevant since this population represents one of the fastest growing segments of American and OECD societies. There were 40.3 million seniors living in the United States in 2010, representing 13% of the population [4]. There were 33 million licensed elderly drivers in 2009, a 23% increase from the prior decade [5]. That number is expected to grow with advancement of the baby boomer generation [6]. Since elderly drivers are at an increased risk of injury and death per vehicle-mile of travel compared to younger drivers [7], determining the
1498
S.T. Youngquist et al. / Injury, Int. J. Care Injured 46 (2015) 1497–1502
prevalence of acute medical conditions in MVCs involving the elderly would be useful for injury prevention and trauma response planning. The goals of this investigation were two-fold: (1) to estimate the prevalence of acute medical conditions that might impair driving in elderly victims of MVCs evaluated in the emergency department[5_TD$IF] (ED) and (2) to estimate through cross-sectional methods the relative odds associating occupancy status (driver vs. passenger) and an acute medical condition. The rationale for this approach is based upon the fact that if acute medical conditions had no effect on the risk of being involved in an MVC, then the likelihood of the driver having an acute medical condition should be similar to that of passengers with comparable baseline characteristics. Methods Study design and setting We performed a structured chart review of all patients seen at the University of Utah Medical Center (UUMC) in Salt Lake City, Utah and the Denver Health Medical Center (DHMC) in Denver, Colorado from 1 July 2000 through 30 June 2010. Both facilities are Level 1 academic trauma centres staffed by board-certified emergency physicians. The Institutional Review Board at each institution approved this study.
Outcomes For purposes of this cross-sectional analysis, the exposure of interest was occupancy status in the motor vehicle at the time of collision (driver vs. passenger). The outcome of interest was defined as the presence of an acute medical condition that might impair driving ability. Principal investigators selected acute medical conditions of interest a priori based on review of published literature [[7_TD$IF]9–[8_TD$IF]17] and investigators’ clinical experience. We counted the outcome as positive whenever at least one of the following were newly diagnosed or actively treated during the episode of care: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.
Transient ischaemic attack or cerebrovascular accident Seizure Syncope Hypoglycaemia (800 mg/dL) Acute coronary syndrome Cardiac arrhythmia (atrial or ventricular in origin) High-grade cardiac conduction abnormalities (Mobitz II or higher, bifascicular block, or alternating AV block) Pacemaker malfunction Aortic dissection Deep vein thrombosis or pulmonary embolism Acute exacerbation of COPD
Selection of participants Potential cases and controls were identified through electronic search of electronic ED patient tracking system, which holds initial complaint and final diagnosis and professional claims databases using ICD-9 external causes of injury codes E810-E819. Inclusion criteria Eligible charts were reviewed and included in the analysis if the patient was 65 years of age or older and the ED visit was prompted by [6_TD$IF]an MVC. After first identifying elderly patients involved in an MVC, occupancy status, among other variables, was abstracted from the chart. Results were then excluded if the visit to the ED did not occur within 24 h of presentation, the patient was transferred from an outside facility, or the patient died in the emergency department, precluding or significantly complicating a diagnostic investigation for acute medical illness. Because of variability in institutional trauma response criteria both between and within institutions over time, we elected not to restrict our search to institutional trauma databases. Methods and measurements Data abstractors were trained and proctored by principle investigators at each site (ML, STY) and were blinded to the hypothesis of interest. Closed response data collection instruments were utilised with a standardised data dictionary for both sites. We collected data on age, past medical history, circumstances surrounding the MVC, and acute medical diagnoses of interest discovered during the ED evaluation or inpatient admission (defined below). All variables were assumed negative if not listed in the chart, except when the patient’s condition precluded the ability to obtain a history, often indicated as ‘‘unknown’’ in the chart; in these cases, the variables were set to missing. Data were entered and managed using REDCap (Research Electronic Data Capture) an electronic data capture tool [8]. All variables, with the exception of age, were entered as binary or categorical responses.
Due to the potential for delayed diagnosis and the episodic nature of some acute medical conditions, we did not attempt to distinguish between those conditions that were clinically apparent in the emergency department and those diagnosed after admission. While an acute medical condition among passengers is unlikely to substantially contribute to a collision, we chose similarly aged passengers as controls because the direction of causation cannot be ascertained in many individual cases, i.e. the physiological stress of an MVC may provoke underlying coronary artery disease, causing myocardial ischaemia; or acute myocardial ischaemia may precede a collision. Thus, the ratio of odds for the prevalence of acute medical illness among drivers vs. passengers may serve as an estimate of the association between occupancy status and risk of an acute medical illness. Given the observational nature of this data, however, this is a hypothesis screening study and is not able to establish causation. Analysis Data were exported from REDCap as a spreadsheet (Excel for Mac 14.2.5, Microsoft Corp, Redmond, WA), inspected for coding errors and imported into STATA statistical software for analysis (STATA/IC 12.1, StataCorp, College Station, TX). Descriptive variables are reported as frequencies with associated percentages or medians and interquartile ranges. We performed logistic regression to estimate the association between the diagnosis of an acute medical condition and automobile occupancy status in univariate and multivariable analysis. Because select comorbidities, age, and gender differences were expected to be associated with both driving status and outcomes, we controlled for these potential confounders (age 85 years, male gender, history of diabetes, history of valvular heart disease, and history of cerebrovascular disease) based on assumptions of plausibility and whether their inclusion in the model produced a change in the effect size of the occupancy status/ outcome odds ratio of at least 10% [[9_TD$IF]18]. Age was dichotomised at 85 years based on a logit transformed LOWESS regression showing a significant rise in the trajectory of predicted outcomes after
S.T. Youngquist et al. / Injury, Int. J. Care Injured 46 (2015) 1497–1502
age 84 (results not shown). We included a variable indicating whether a single or multiple vehicles were involved in the collision as an independent predictor. We used multilevel mixed-effects logistic regression to explore the potential clustering effects of institution. We used the STATA command xtmelogit with institution as a random intercept. However, the likelihood ratio test comparing this model to fixed-effects logistic regression had an associated p-value of 1.0. In favour of simplicity, therefore, we proceeded with standard, fixed-effects logistic regression. Final model specification or goodness-of-fit were assessed satisfactorily using the STATA command linktest and the Hosmer–Lemeshow test, respectively. All p-values are two-sided. Sensitivity analysis Given the observational nature of this study, missing data, misclassification, and the potential for uncontrolled confounding are potential threats to the validity of our findings. Because logistic regression uses only complete cases, we analysed the data after multiple imputation to explore the potential influence of missing data on outcomes. We also performed a probabilistic sensitivity analysis on the adjusted odds ratio associating occupancy status and the outcome using Monte-Carlo (random number) simulations. We defined trapezoidal prior probability distributions for the sensitivity and specificity (so called bias parameters) for capturing an acute medical illness through the record review process. We also evaluated the potential effects of residual confounding with an unknown, hypothetical confounder (such as injury patterns specific to occupancy status that might be associated with outcome) by assigning a strong residual confounder-outcome relative risk with 95% confidence limits between 5 and 20 (median 10), sufficiently strong to adjust the lower confidence limit for the odds ratio of interest below 1 (95% limits of 0.88–7.63). We then recalculated the results after correction for both misclassification and residual confounding simultaneously. (For further details, see Appendix 1.) Measures of agreement We randomly selected 10% of charts for cross-abstraction to determine the proportion of abstractions in agreement regarding the exposure and outcome.
Results Agreement among abstractors for the outcome and exposure was 100% for both occupancy status and outcome among charts selected for cross-abstraction. Characteristics of study subjects
1499
Table 1 Characteristics of 1178 elderly drivers and passengers evaluated in the emergency department after involvement in motor vehicle collisions.
Age (years) Male gender Arrival via EMS Single vehicle collision Diabetes mellitus Hypertension Coronary artery disease Congestive heart failure Pacemaker Arrhythmia Valvular heart disease Stroke/TIA Active cancer Seizure Dementia a
Drivers (n = 871)a
Passengers (n = 307)a
73 491 766 159 158 432 136 36 16 53 13 38 66 6 11
75 87 275 44 68 155 46 14 7 18 3 17 21 0 8
IQR (68–79) (56%) (95%) (19%) (20%) (52%) (16%) (4%) (2%) (6%) (2%) (5%) (8%) (1%) (1%)
Percentages are based on complete cases.
Prevalence and association An acute medical condition was recorded in 107/1178 patients (9%). Among drivers an acute medical condition was recorded in 97/871 patients (11%) and among passengers in 10/307 patients (3%). Cardiac and neurologic conditions were most prominently differentially distributed among drivers and passengers (Table 2). More than one acute medical condition was diagnosed in 16 drivers. The resulting univariate odds ratio associating occupancy status and the diagnosis of an acute medical condition was 3.7 (95% CI 1.9–7.2) for drivers relative to passengers. This association was adjusted upwards when estimated using multivariable logistic regression (Table 3), with an adjusted odds ratio (aOR) of 5.5 (95% CI 2.3–13.0) but was similar to the univariate odds ratio when including records in which missing variables were imputed (aOR 3.8, 95% CI 1.9–7.8). Single vehicle collisions, male gender, age 85 years, history of diabetes, history of cerebrovascular disease, or history of valvular heart disease were all independently associated with the presence of an acute medical illness. Automated probabilistic sensitivity analysis (Table 4) to correct for misclassification of the outcome and the presence of residual confounding produced a median odds ratio of 6.4 with 95% simulation limits 1.5–30.4. Thus, the measured association would require an even larger residual confounder than the one selected to explain away the observed association while accounting for the effects of misclassification.
Table 2 Acute medical conditions diagnosed among elderly drivers and passengers involved in motor vehicle collisions. More than one acute medical condition of interest was documented in 16 drivers. Driver
Data was abstracted from 1220 charts meeting inclusion criteria. In 42 charts (3%), no information regarding vehicle occupancy was reported, leaving 1178 available for analysis. Table 1 describes characteristics of the 307 (26%) passengers and 871 (74%) drivers included in this analysis. Drivers were twice as likely to be male. Otherwise, groups were fairly balanced with respect to pre-existing comorbidities recorded in the chart. The majority of patients in both groups, 243 passengers (79%) and 728 drivers (84%), were discharged home at the end of hospitalisation. Twelve passengers (4%) and 30 drivers (3%) expired.
IQR (69–81) (28%) (96%) (16%) (24%) (54%) (16%) (5%) (2%) (6%) (1%) (6%) (7%) (0%) (3%)
TIA/CVA Seizure Syncope Hypoglycaemia Hyperglycaemia ACS Arrhythmia Conduction block Pacemaker malfunction Aortic dissection DVT/PE Acute COPD exacerbation Total
Passenger
12 6 25 16 4 10 23 7 0 2 4 5
2 0 0 1 0 1 3 2 1 0 0 0
114
10
S.T. Youngquist et al. / Injury, Int. J. Care Injured 46 (2015) 1497–1502
1500
Table 3 Results of [2_TD$IF]multivariable logistic regression describing the association between occupancy status (driver vs. passenger) and the diagnosis of an acute medical condition that might impair driving during the episode of care. aORa Standard logistic regression (1055 observations) Driver vs. passenger 5.5 Male gender 1.9 Age 85 yearsb 3.0 Single vs. multi-vehicle collision 2.3 History of valvular heart disease 7.4 History of diabetes mellitus 3.1 History of TIA or stroke 3.5
95% confidence interval
p-Value
(2.3–13.0) (1.2–3.2) (1.5–6.2) (1.4–3.8) (2.4–23.0) (1.9–5.0) (1.6–7.7)
Contents lists available at ScienceDirect
Injury journal homepage: www.elsevier.com/locate/injury
Acute medical impairment among elderly patients involved in motor vehicle collisions Scott T. Youngquist a,*, Michael Liao b,c, Sydney Hartsell a,[3_TD$IF]1, Melissa Walker a, Nathan J. Kartchner a, Raminder Nirula d a
Department of Surgery, Division of Emergency Medicine, University of Utah School of Medicine, Salt Lake City, UT, United States Department of Emergency Medicine, Denver Health Medical Center, Denver, CO, United States Department of Emergency Medicine, North Suburban Medical Center, Thornton, CO, United States d Department of Surgery, Section of Acute Care Surgery, University of Utah School of Medicine, Salt Lake City, UT, United States b c
A R T I C L E I N F O
A B S T R A C T
Article history: Accepted 2 April 2015
Background: The association between acute medical illness and motor vehicle collisions (MVCs) among elderly emergency department patients is unclear. We sought to determine the prevalence of acute medical conditions that might impair driving ability among the elderly involved in MVCs and determine if there was an increased risk of the driver having an acute medical condition compared to similarly aged passengers. Methods: We reviewed charts of patients aged 65 years or older whose emergency department visit was prompted by a motor vehicle collision between 1 July 2000 and 30 June 2010 at two Level 1 trauma centres. The exposure of interest was occupancy status (driver vs. passenger), and the outcome measure was the presence of any predefined acute medical illness that might impair driving ability. Results: Final analysis included 871 drivers (cases) and 307 passengers (controls). An acute medical illness was recorded in 107 patients (9%): 97 drivers (11%) and 10 passengers (3%). Compared to passengers, drivers had significantly higher odds of presenting with acute medical illness (OR 3.7, 95% CI 1.9–7.2). After controlling for potential confounders, the adjusted odds ratio was 5.5 (95% CI 2.3–13.0). Conclusion: Acute medical conditions are a moderately common diagnosis among elderly drivers, presenting in about one in ten patients. A difference in the risk of finding an acute medical illness when comparing elderly drivers and passengers evaluated in the emergency department after a collision suggests the need for considering additional diagnostic investigation and post-discharge surveillance in this population. ß 2015 Elsevier Ltd. All rights reserved.
Keywords: Motor vehicle collision Medical impairment Elderly
Introduction The initial evaluation of elderly patients involved in motor vehicle collisions (MVCs) focuses primarily on the rapid identification and stabilisation of injuries. Secondarily, however, such encounters elicit concern over the potential for an acute medical illness acting as a causal factor in the collision. While an extensive body of literature exists regarding functional impairments in driving ability associated with chronic medical illness [1], the risk
* Corresponding author at: University of Utah Medical Center, Division of Emergency Medicine, 30 North 1900 East, Room 1C026, Salt Lake City, UT 84132, United States. Tel.: +1 801 683 9055; fax: +1 +801 581 2730. E-mail address: [email protected] (S.T. Youngquist). [4_TD$IF]1 Ms. Hartsell is now at the University of North Carolina School of Medicine, Chapel Hill, NC, United States. http://dx.doi.org/10.1016/j.injury.2015.04.012 0020–1383/ß 2015 Elsevier Ltd. All rights reserved.
of individual disease categories on collision risk [2], as well as the influence of age-related physiologic changes alone [3], little is known regarding the contribution of acute medical illness to the epidemiology of MVCs among this patient population from the immediate post-incident perspective in the emergency department. We considered a focus on the elderly driving population, frequently defined in the medical literature as individuals aged 65 or older, as important and relevant since this population represents one of the fastest growing segments of American and OECD societies. There were 40.3 million seniors living in the United States in 2010, representing 13% of the population [4]. There were 33 million licensed elderly drivers in 2009, a 23% increase from the prior decade [5]. That number is expected to grow with advancement of the baby boomer generation [6]. Since elderly drivers are at an increased risk of injury and death per vehicle-mile of travel compared to younger drivers [7], determining the
1498
S.T. Youngquist et al. / Injury, Int. J. Care Injured 46 (2015) 1497–1502
prevalence of acute medical conditions in MVCs involving the elderly would be useful for injury prevention and trauma response planning. The goals of this investigation were two-fold: (1) to estimate the prevalence of acute medical conditions that might impair driving in elderly victims of MVCs evaluated in the emergency department[5_TD$IF] (ED) and (2) to estimate through cross-sectional methods the relative odds associating occupancy status (driver vs. passenger) and an acute medical condition. The rationale for this approach is based upon the fact that if acute medical conditions had no effect on the risk of being involved in an MVC, then the likelihood of the driver having an acute medical condition should be similar to that of passengers with comparable baseline characteristics. Methods Study design and setting We performed a structured chart review of all patients seen at the University of Utah Medical Center (UUMC) in Salt Lake City, Utah and the Denver Health Medical Center (DHMC) in Denver, Colorado from 1 July 2000 through 30 June 2010. Both facilities are Level 1 academic trauma centres staffed by board-certified emergency physicians. The Institutional Review Board at each institution approved this study.
Outcomes For purposes of this cross-sectional analysis, the exposure of interest was occupancy status in the motor vehicle at the time of collision (driver vs. passenger). The outcome of interest was defined as the presence of an acute medical condition that might impair driving ability. Principal investigators selected acute medical conditions of interest a priori based on review of published literature [[7_TD$IF]9–[8_TD$IF]17] and investigators’ clinical experience. We counted the outcome as positive whenever at least one of the following were newly diagnosed or actively treated during the episode of care: 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.
Transient ischaemic attack or cerebrovascular accident Seizure Syncope Hypoglycaemia (800 mg/dL) Acute coronary syndrome Cardiac arrhythmia (atrial or ventricular in origin) High-grade cardiac conduction abnormalities (Mobitz II or higher, bifascicular block, or alternating AV block) Pacemaker malfunction Aortic dissection Deep vein thrombosis or pulmonary embolism Acute exacerbation of COPD
Selection of participants Potential cases and controls were identified through electronic search of electronic ED patient tracking system, which holds initial complaint and final diagnosis and professional claims databases using ICD-9 external causes of injury codes E810-E819. Inclusion criteria Eligible charts were reviewed and included in the analysis if the patient was 65 years of age or older and the ED visit was prompted by [6_TD$IF]an MVC. After first identifying elderly patients involved in an MVC, occupancy status, among other variables, was abstracted from the chart. Results were then excluded if the visit to the ED did not occur within 24 h of presentation, the patient was transferred from an outside facility, or the patient died in the emergency department, precluding or significantly complicating a diagnostic investigation for acute medical illness. Because of variability in institutional trauma response criteria both between and within institutions over time, we elected not to restrict our search to institutional trauma databases. Methods and measurements Data abstractors were trained and proctored by principle investigators at each site (ML, STY) and were blinded to the hypothesis of interest. Closed response data collection instruments were utilised with a standardised data dictionary for both sites. We collected data on age, past medical history, circumstances surrounding the MVC, and acute medical diagnoses of interest discovered during the ED evaluation or inpatient admission (defined below). All variables were assumed negative if not listed in the chart, except when the patient’s condition precluded the ability to obtain a history, often indicated as ‘‘unknown’’ in the chart; in these cases, the variables were set to missing. Data were entered and managed using REDCap (Research Electronic Data Capture) an electronic data capture tool [8]. All variables, with the exception of age, were entered as binary or categorical responses.
Due to the potential for delayed diagnosis and the episodic nature of some acute medical conditions, we did not attempt to distinguish between those conditions that were clinically apparent in the emergency department and those diagnosed after admission. While an acute medical condition among passengers is unlikely to substantially contribute to a collision, we chose similarly aged passengers as controls because the direction of causation cannot be ascertained in many individual cases, i.e. the physiological stress of an MVC may provoke underlying coronary artery disease, causing myocardial ischaemia; or acute myocardial ischaemia may precede a collision. Thus, the ratio of odds for the prevalence of acute medical illness among drivers vs. passengers may serve as an estimate of the association between occupancy status and risk of an acute medical illness. Given the observational nature of this data, however, this is a hypothesis screening study and is not able to establish causation. Analysis Data were exported from REDCap as a spreadsheet (Excel for Mac 14.2.5, Microsoft Corp, Redmond, WA), inspected for coding errors and imported into STATA statistical software for analysis (STATA/IC 12.1, StataCorp, College Station, TX). Descriptive variables are reported as frequencies with associated percentages or medians and interquartile ranges. We performed logistic regression to estimate the association between the diagnosis of an acute medical condition and automobile occupancy status in univariate and multivariable analysis. Because select comorbidities, age, and gender differences were expected to be associated with both driving status and outcomes, we controlled for these potential confounders (age 85 years, male gender, history of diabetes, history of valvular heart disease, and history of cerebrovascular disease) based on assumptions of plausibility and whether their inclusion in the model produced a change in the effect size of the occupancy status/ outcome odds ratio of at least 10% [[9_TD$IF]18]. Age was dichotomised at 85 years based on a logit transformed LOWESS regression showing a significant rise in the trajectory of predicted outcomes after
S.T. Youngquist et al. / Injury, Int. J. Care Injured 46 (2015) 1497–1502
age 84 (results not shown). We included a variable indicating whether a single or multiple vehicles were involved in the collision as an independent predictor. We used multilevel mixed-effects logistic regression to explore the potential clustering effects of institution. We used the STATA command xtmelogit with institution as a random intercept. However, the likelihood ratio test comparing this model to fixed-effects logistic regression had an associated p-value of 1.0. In favour of simplicity, therefore, we proceeded with standard, fixed-effects logistic regression. Final model specification or goodness-of-fit were assessed satisfactorily using the STATA command linktest and the Hosmer–Lemeshow test, respectively. All p-values are two-sided. Sensitivity analysis Given the observational nature of this study, missing data, misclassification, and the potential for uncontrolled confounding are potential threats to the validity of our findings. Because logistic regression uses only complete cases, we analysed the data after multiple imputation to explore the potential influence of missing data on outcomes. We also performed a probabilistic sensitivity analysis on the adjusted odds ratio associating occupancy status and the outcome using Monte-Carlo (random number) simulations. We defined trapezoidal prior probability distributions for the sensitivity and specificity (so called bias parameters) for capturing an acute medical illness through the record review process. We also evaluated the potential effects of residual confounding with an unknown, hypothetical confounder (such as injury patterns specific to occupancy status that might be associated with outcome) by assigning a strong residual confounder-outcome relative risk with 95% confidence limits between 5 and 20 (median 10), sufficiently strong to adjust the lower confidence limit for the odds ratio of interest below 1 (95% limits of 0.88–7.63). We then recalculated the results after correction for both misclassification and residual confounding simultaneously. (For further details, see Appendix 1.) Measures of agreement We randomly selected 10% of charts for cross-abstraction to determine the proportion of abstractions in agreement regarding the exposure and outcome.
Results Agreement among abstractors for the outcome and exposure was 100% for both occupancy status and outcome among charts selected for cross-abstraction. Characteristics of study subjects
1499
Table 1 Characteristics of 1178 elderly drivers and passengers evaluated in the emergency department after involvement in motor vehicle collisions.
Age (years) Male gender Arrival via EMS Single vehicle collision Diabetes mellitus Hypertension Coronary artery disease Congestive heart failure Pacemaker Arrhythmia Valvular heart disease Stroke/TIA Active cancer Seizure Dementia a
Drivers (n = 871)a
Passengers (n = 307)a
73 491 766 159 158 432 136 36 16 53 13 38 66 6 11
75 87 275 44 68 155 46 14 7 18 3 17 21 0 8
IQR (68–79) (56%) (95%) (19%) (20%) (52%) (16%) (4%) (2%) (6%) (2%) (5%) (8%) (1%) (1%)
Percentages are based on complete cases.
Prevalence and association An acute medical condition was recorded in 107/1178 patients (9%). Among drivers an acute medical condition was recorded in 97/871 patients (11%) and among passengers in 10/307 patients (3%). Cardiac and neurologic conditions were most prominently differentially distributed among drivers and passengers (Table 2). More than one acute medical condition was diagnosed in 16 drivers. The resulting univariate odds ratio associating occupancy status and the diagnosis of an acute medical condition was 3.7 (95% CI 1.9–7.2) for drivers relative to passengers. This association was adjusted upwards when estimated using multivariable logistic regression (Table 3), with an adjusted odds ratio (aOR) of 5.5 (95% CI 2.3–13.0) but was similar to the univariate odds ratio when including records in which missing variables were imputed (aOR 3.8, 95% CI 1.9–7.8). Single vehicle collisions, male gender, age 85 years, history of diabetes, history of cerebrovascular disease, or history of valvular heart disease were all independently associated with the presence of an acute medical illness. Automated probabilistic sensitivity analysis (Table 4) to correct for misclassification of the outcome and the presence of residual confounding produced a median odds ratio of 6.4 with 95% simulation limits 1.5–30.4. Thus, the measured association would require an even larger residual confounder than the one selected to explain away the observed association while accounting for the effects of misclassification.
Table 2 Acute medical conditions diagnosed among elderly drivers and passengers involved in motor vehicle collisions. More than one acute medical condition of interest was documented in 16 drivers. Driver
Data was abstracted from 1220 charts meeting inclusion criteria. In 42 charts (3%), no information regarding vehicle occupancy was reported, leaving 1178 available for analysis. Table 1 describes characteristics of the 307 (26%) passengers and 871 (74%) drivers included in this analysis. Drivers were twice as likely to be male. Otherwise, groups were fairly balanced with respect to pre-existing comorbidities recorded in the chart. The majority of patients in both groups, 243 passengers (79%) and 728 drivers (84%), were discharged home at the end of hospitalisation. Twelve passengers (4%) and 30 drivers (3%) expired.
IQR (69–81) (28%) (96%) (16%) (24%) (54%) (16%) (5%) (2%) (6%) (1%) (6%) (7%) (0%) (3%)
TIA/CVA Seizure Syncope Hypoglycaemia Hyperglycaemia ACS Arrhythmia Conduction block Pacemaker malfunction Aortic dissection DVT/PE Acute COPD exacerbation Total
Passenger
12 6 25 16 4 10 23 7 0 2 4 5
2 0 0 1 0 1 3 2 1 0 0 0
114
10
S.T. Youngquist et al. / Injury, Int. J. Care Injured 46 (2015) 1497–1502
1500
Table 3 Results of [2_TD$IF]multivariable logistic regression describing the association between occupancy status (driver vs. passenger) and the diagnosis of an acute medical condition that might impair driving during the episode of care. aORa Standard logistic regression (1055 observations) Driver vs. passenger 5.5 Male gender 1.9 Age 85 yearsb 3.0 Single vs. multi-vehicle collision 2.3 History of valvular heart disease 7.4 History of diabetes mellitus 3.1 History of TIA or stroke 3.5
95% confidence interval
p-Value
(2.3–13.0) (1.2–3.2) (1.5–6.2) (1.4–3.8) (2.4–23.0) (1.9–5.0) (1.6–7.7)
