Racial/Ethnic and Gender Differences in Risk of Injury and Life-Course Drinking Patterns: Data from US National Alcohol Surveys

Abstract

Aims

To estimate risk of injury associated with frequency of drinking and heavy drinking (5+ drinks on occasion) by gender and race/ethnicity in the US population.

Methods

Data were from a merged sample of two National Alcohol Surveys (telephone and web-based) (2014–2015 and 2019–2020) on 16,639 respondents, and analyzed using Cox proportional hazards models with age as the timescale in a retrospective cohort design. Life-course drinking was determined by age of onset and questions on any drinking and heavy drinking by decade of life. The outcome measure was having had an injury from a serious accident at a certainage.

Results

Frequent heavy drinking (5+ daily, weekly and monthly) was significantly predictive of injury with hazard ratios (HRs) of 2.40, 1.81 and 1.50, respectively, while frequent light drinking (alcohol at least weekly and 5+ yearly or less) was also significant for women (HR = 1.73). For White respondents, 5+ at least weekly was significant for both men (HR = 1.74) and women (HR = 2.42). Among Hispanic respondents, 5+ at least weekly and 5+ monthly were both significant for men (HR = 2.81 and 2.49, respectively) and women (HR = 2.81 and 3.48, respectively). Among Black women, risk was significant for 5+ monthly (HR = 2.90) and for any alcohol ≥ weekly (HR = 2.72), but neither frequency of any drinking or 5+ was significant for Blackmen.

Conclusions

Data suggest a greater risk of injury from a serious accident for frequent heavy drinkers among all White and Hispanic respondents, and Black women, but not for Blackmen.

INTRODUCTION

The association of alcohol and injury has been primarily studied in emergency department (ED) populations which have examined the risk of injury from drinking prior to the event (Romelsjö, 1995; Cherpitel, 2007), as well as the risk associated with usual drinking patterns (Cherpitel et al., 2018). These latter studies have found that while the amount of alcohol consumed prior to injury is an important risk factor for injury, usual drinking pattern in the past year is independently important in injury occurrence (Cherpitel et al., 2005, 2012; Gmel et al., 2006).

General population studies have also found an association of alcohol with injury, especially for heavy and dependent drinkers (Vinson et al., 2003) and for heavy spirits drinkers as well (Watt et al., 2004). These studies suggest that risk of injury increases at relatively low levels of consumption (Cherpitel et al., 1995), with risk increasing at higher levels to about 6–8 days of consuming five or more drinks per day (5+ days), although risk for women continued to increase at higher frequencies of 5+ days (Cherpitel and Ye, 2009).

Racial/ethnic differences in injury risk associated with alcohol consumption have received some attention in both ED and general population studies. Differences in injury risk from drinking may be different than that for alcohol-related social problems which may become issues at lower levels of consumption for Black and Hispanic groups among whom there is a higher level of abstention and fewer social and economic resources (Zapolski et al., 2014), while risk of injury may be more consistently defined across these subgroups. A review of ethnic/racial differences in injury mortality in the USA found alcohol-attributable deaths were overrepresented among Black and Hispanic individuals compared with White respondents, based on lower rates of alcohol use and binge drinking in these two groups (Keyes et al., 2012). An ED study of injured patients across racial/ethnic groups found Black patients were less likely to be alcohol positive or to report heavy drinking either prior to the injury event or during the last year compared with White and Hispanic patients (Cherpitel, 1998), while other studies have suggested that a higher risk of injury might be expected at lower levels of consumption for Black individuals compared with other racial/ethnic groups (Witbrodt et al., 2014; Zapolski et al., 2014). However, data from the US general population found that Black (OR = 1.54) and Hispanic respondents (OR = 1.98) were significantly more likely to report drinking prior to injury than White respondents (OR = 0.89) (Cherpitel and Ye, 2014), although this study did not examine usual drinking patterns. A US study of drinking prior to injury and usual drinking pattern found Black respondent at lower risk associated with both consumption measures than Hispanic and White respondents (Cherpitel et al., 2016). Less research has been conducted on gender-specific differences in racial/ethnic disparities and the association of alcohol and injury. Compared with White men, Black and Hispanic men have been found to report significantly higher rates of injury, however no significant differences were found among women (Witbrodt et al., 2014).

While studies of racial/ethnic differences in alcohol and injury have focused either on the association of drinking prior to the injury event or usual drinking patterns, a study using data from the 2010 National Alcohol Survey (NAS) examined racial/ethnic differences in the relationship of life-course heavy drinking (frequency of 5+ days) and ever having had an injury from a serious accident (Kerr et al., 2015), using a retrospective cohort design (Hudson et al., 2005). The study design allowed for examining the risk of injury from heavy drinking during earlier decades of life when this is most common, and taking into account less proximate influences and indirect effects of the alcohol–injury relationship. Different risk relationships were found across race/ethnicity groups. Risk of injury increased significantly with the frequency of 5+ days for white respondents, to a hazard ratio (HR) of 2.11 for daily heavy drinking. For Black respondents risk was significantly increased only for daily heavy drinkers (HR = 4.09), while for Hispanic respondents risk was significantly increased for yearly (HR = 2.71) and monthly (HR = 2.68) heavy drinkers, but not for more frequent heavy drinkers.

As the above analysis did not take into account other potentially important consumption variables such as frequency of any drinking, reported here is a replication of this life-course analysis on a larger merged sample of the two subsequent NASs (2014–2015 and 2019–2020), using more fine-grained drinking categories which now include the frequency of any drinking as well as the frequency of 5+ drinking days, and also taking into account gender along with race/ethnicity.

We hypothesize that both frequency of any drinking and frequency of 5+ drinking days will be predictive of risk of injury from a serious accident in a given year, and risk will increase with increasing frequency of both. We also hypothesize that men will be at greater risk than women and that Black and Hispanic respondents will be at greater risk across levels of any drinking and heavy drinking compared with White respondents. These data are important for a better understanding of differences in the alcohol–injury relationship across race/ethnic and gender subgroups and may help guide future intervention and prevention strategies aimed at reducing alcohol-relatedharm.

METHODS

Dataset

The analytic sample is the combined data of the two latest waves of the NAS (2014–2015 and 2019–2020). The NAS is a population-based survey of randomly selected US adults aged 18 years and over living in the 50 states and the District and has been conducted approximately every 5 years since 1979 (Kerr et al., 2018). The latest two waves were conducted between April 2014 and March 2015, and between February 2019 and April 2020, respectively. NAS14–15 survey involved computer-assisted telephone interviews (CATI), utilizing a dual-frame sampling design that included a two-stage stratified, random digit dialing (RDD) sample of adults from landline telephone households and a sample of cell phone users selected from an RDD sample of cell phone numbers. Of the completed interviews, about 60% were conducted via landline and 40% via cell phone. Major changes in sampling method and modes of administration were made for NAS19–20, however, due to growing challenges including cost and declining response rates for traditional telephone surveys. The NAS19–20 survey used two probability samples including (a) an RDD landline and cell phone telephone sample, and (b) an address-based sample (ABS). These probability samples were supplemented with (c) a nonprobability sample drawn from a pre-recruited web panel. Interviews were conducted via telephone for the RDD samples or via a web questionnaire for the ABS and panel samples. For both NAS waves, interviews were conducted in English or Spanish, by ICF, Inc., of Fairfax Virginia as the field agency. Black and Hispanic minorities were oversampled in NAS14–15 and in the ABS and web panel samples in NAS19–20.

The American Association for Public Opinion Research (AAPOR) COOP4 cooperation rate (The American Association for Public Opinion Research, 2011) for NAS14–15 was 43.4% (52.0% cell phone and 38.7% landline), with a total sample of 7071. The average interview duration was about 45 min. Overall, NAS19–20 sample included 9668 respondents, with 1572 from the RDD telephone sample, 5661 from the ABS sample and 2435 from the Web Panel. The AAPOR COOP4 cooperation rate for NAS19–20 combined RDD and ABS samples was 42.2% (33.2% for landline, 13.9% for cell phone and 92.4% ABS samples).

The average interview duration was 40 min for CATI and 36 min for web. The current analysis only included respondents aged 18–59 (N = 4268 and 6520 for NAS14–15 and NAS19–20, respectively), to match the measures of drinking pattern by decades (described below).

Both surveys are weighted to account for the disproportionate probability of selection caused by the number of telephone lines in a household, number of eligible adults in households involving landlines and from ABS sampling (for NAS19–20) and oversampling of high-density minority strata. For the non-probability web panel sample of NAS19–20, propensity score weighting was applied to combine it with the ABS sample, adjusting for socio-demographics (gender, age, race/ethnicity, education, marital status), drinking indicators and general health characteristics. Finally, both surveys were post-stratified to align the combined weighted samples with the benchmark demographic distributions for the target adult population using the iterative raking adjustment. The population targets were developed from corresponding American Community Surveys involving demographics such as gender, age, race/ethnicity, census region, education and marital status, covering both marginal distributions and some, but not full, levels of cross-tabulations (e.g. sex by race by Hispanic origin; sex by age group by education, etc.).

Measures

Lifetime drinking pattern was assessed from a series of questions on drinking during the teens, 20s, 30s, 40s (and 50s in NAS19–20). The respondents were first asked their age of drinking onset. Then for each decade period, eligible respondents were asked about their frequency of drinking five or more (5+) drinks in a day during the specific life decade (Greenfield et al., 2014). For example, for the question regarding 5+ frequency during their 30s, eligible respondents were those who were at least 30 years old at the time of the interview with age of drinking onset before 30. The response options for 5+ frequency during decades were ‘every day or nearly every day’, ‘at least once a week’, ‘at least once a month’, ‘at least once a year’, ‘less often than that’ and ‘never’. Further, for each specific decade, those who answered ‘at least once a year’ or less for the 5+ frequency question were asked ‘how often did you drink alcohol’ during the decade, with the same prior response options from ‘every day or nearly every day’ to ‘never’.

The lifetime decade measures of frequency of 5+ and frequency of any drinking described above were then cross-classified to create the lifetime drinking pattern measure separately for teens, 20s, 30s and 40s (and 50s for NAS19–20). The created pattern variable had the following categories: (1) 5+ daily, (2) 5+ at least weekly (but less than daily), (3) 5+ at least monthly (but less than weekly), (4) 5+ yearly or less and drinking at least weekly, (5) 5+ yearly or less and drinking at least monthly (but less than weekly), (6) drinking yearly or less, (7) no drinking during the decade and (8) lifetime abstainers. Since only NAS19–20 asked the decade drinking questions during their 50s, the current (last 12 months) alcohol use information was used to derive drinking pattern for the 50s in NAS14–15. Specifically, for NAS14–15 respondents in the 50–59 age group, their drinking pattern during the 50s was derived from cross classifying 12-month 5+ frequency, based on the graduated frequency series items on 12 or more, 8–11 and 5–7 drinks (Greenfield, 2000), and the usual frequency measure (frequency of any drinking during the last 12 months), in the same manner as described above for the decade series.

Lifetime injury was assessed by the question asking ‘have you ever been told by a doctor or other health professional that you had injuries from a serious accident?’ Those who answered yes were then asked ‘at what age were you first told?’ as a measure of the onset of injury.

Control variables used in survival analysis predicting lifetime are all time-invariant measured at the time of the interview, including gender, race/ethnicity, education, family income, marital status and NAS wave. Also controlled was the risk taking/impulsivity and sensation seeking scale (Cherpitel, 1993) comprised from three items ‘I like to test myself every now and then by doing something a little chancy’, ‘I like to try new things just for the excitement’ and ‘I like to experience new and different sensations’.

Analysis

Cox proportional hazards models were used to estimate the hazard of injury occurrence, predicted by time-varying lifetime drinking pattern controlling for time-invariant socio-demographic factors and the risk taking and sensation seeking items. Since the respondents were only asked about their first onset of serious injury, if any, the injury occurrence data are considered censored. Cox model estimates injury risk accounting for time-to-event, thus preferable to logistic regression which does not include a time variable. Specifically, in Cox model a relationship of h(t) = h0(t) exp(β₁X_1t + β₂X₂) is assumed, in which h0(t) is the baseline hazard function, X_1t is the time-varying drinking pattern variables at time t, X₂ is the vector of time-invariant control variables and exp(β) is the estimated HR. The dataset was structured with each respondent aligned by age, right-censored at either the age of injury occurrence or, for those who had no injury up to age 59, the age of interview. This allows for the control of age by design (Korn et al., 1997). The data were also left-censored at age 14, as injury before that is presumably highly unlikely to be related to drinking. The lifetime drinking pattern measures across decades, from teens to 50s, were combined to create the measure of annual drinking pattern status at each age, assuming a constant heavy drinking pattern for each decade. Analysis was performed for the total sample, men and women, and then separately for White, Black and Hispanic respondents. All analyses were performed in Stata 16 (StataCorp., 2019), utilizing sampling weights accounting for probability of selection, non-response and US population distributions.

One key assumption of the Cox model is the proportional hazard, i.e. HRs of predictive variables do not change with time. Violation of the assumption can lead to model misspecification and failure to uncover the true effect which may be time-varying. The proportional hazard assumption was tested using Schoenfeld residuals based on the correlation between the scaled Schoenfeld residuals and rank of survival time. The tests have both a covariate-specific form and a global form for all covariates combined (Grambsch and Therneau, 1994; Keele, 2010). Except for drinking variables, covariates highly significant in the Schoenfeld residuals tests were stratified in the final models. In stratified analysis, the assumption that each individual faces the same baseline hazard is relaxed by fitting different baseline hazards across groups. One problem with stratified analysis, however, is that coefficient estimates for the stratified variables are not available. We thus present effect estimates from both the raw model before stratification (model 1) and the final model after stratification (model 2), to show the complete analysis process.

RESULTS

Table 1 shows weighted proportions of lifetime drinking pattern for each decade from teens to 50s, for the combined NAS14–15 and NAS19–20, and separately for each wave. The drinking pattern distribution was only reported for respondents in the relevant age groups, e.g. only those aged 20 and up were included in drinking pattern during their 20s. While the drinking pattern distributions were quite consistent between NAS14–15 and NAS19–20 from teens to 40s, there was substantial difference for the 50s, due to the inconsistency in measurement across surveys (as noted above).

Table 2 shows the HRs of drinking pattern indicators and socio-demographic control variables predicting injury onset for the total sample and by gender. For each group, two models are presented. The first is the raw model before the proportional hazard assumption was tested in which HRs from all predictors are shown. The second model stratified the control variables for which the proportional hazard assumption was violated. For example, for the total sample, the P value testing the proportional hazard assumption (the global Schoenfeld residuals test) increased from <0.001 in model 1 to 0.049 in model 2. Note that the global test for the final model (model 2) was still significant at P < 0.05, due to significant variable-specific Schoenfeld residuals test on a drinking pattern category, in this case the lifetime abstainer category.

As shown in Table 2, for the total sample and both men and women, frequent heavy drinking (5+ daily, weekly and monthly) was significantly predictive of injury compared with the reference group of having no alcohol in a given year. A significantly greater hazard of injury from frequent light drinking (alcohol at least weekly and 5+ yearly or less) was also observed for women (HR = 1.73, P < 0.01, model 2), but not for men. Lifetime abstention was associated with a lower hazard of injury for men (HR = 0.63, P < 0.05), but not women. In relation to the time-invariant control variables, men were at a significantly greater hazard of injury (HR = 1.43, P < 0.001) than women. Risk taking/sensation seeking was also associated with a greater hazard of injury, but only among men (HR = 1.23, P < 0.05), while higher education (college graduate or higher) was protective for men (HR = 0.64, P < 0.05) but not women. While higher family income was also shown to be protective in raw models for both genders, it was stratified in final models due to violation of the proportional hazard assumption. Similarly, Hispanic women had a lower risk of injury than White women in the raw model, but race/ethnicity was stratified in the final model. Marital status was not found to be a significant predictor for either men or women. Respondents from NAS19–20 appeared to have a lower risk of injury compared with those from NAS14–15 in the raw model for men, but was also stratified in the final model for both men and women.

Table 3 shows the HRs for injury from drinking pattern for White, Black and Hispanic respondents, separately for the total sample and by gender, controlling for all time-invariant variables as in Table 2 (not shown). 5+ daily and weekly were combined to provide a larger number for analysis across the three race/ethnic groups. For White respondents, only 5+ at least weekly was significant for both men (HR = 1.74, P < 0.01) and women (HR = 2.42, P < 0.001), while among Hispanic respondents, 5+ at least weekly and 5+ monthly were both significantly associated with a greater hazard of injury for men (HR = 2.81, P < 0.01 and HR = 2.49, P < 0.01, respectively) and for women (HR = 2.81, P < 0.05 and HR = 3.48, P < 0.01, respectively). Less consistent results were observed for Black respondents. For Black men, the only significant drinking category was lifetime abstention which was protective for injury (HR = 0.25, P < 0.01). Among Black women, HRs were significant for 5+ monthly (but <weekly) (HR = 2.90, P < 0.05) and for any alcohol at least weekly but 5+ yearly or less (HR = 2.72, P < 0.05), while 5+ at least weekly was not significant.

DISCUSSION

In this analysis of life-course drinking and injury from a serious accident we found that frequency of both any drinking and 5+ occasions were predictive of injury, but varied by gender. Frequency of 5+ drinking was significantly predictive for both men and women, as hypothesized. Those reporting 5+ drinking days weekly were two and a half times more likely to report an injury from a serious accident compared with those reporting no drinking. Frequency of any drinking was predictive only for women, however, with those reporting drinking ≥weekly nearly twice as likely to report an injury compared with those not drinking. Men were at a significantly greater risk of injury than women, as also hypothesized. While risk taking and sensation were significantly predictive for men, lifetime abstention and higher education were both protective for injury among men but not women.

Findings by race/ethnicity were mixed in terms of the predictive value of both frequency of any drinking and of 5+ occasions for injury, and hypotheses were not supported. Both categories of 5+ drinking (≥weekly and monthly) were significant for Hispanic respondents, with those drinking at these levels almost three times more likely to report an injury compared with those not drinking, while only 5+ drinking weekly was significant for White respondents, with those drinking at this level nearly twice as likely to report an injury. Neither 5+ category was significant for Black respondents. Among Black respondents, however, frequency of any drinking ≥weekly was significantly predictive of injury, with those drinking at this level nearly twice as likely to report an injury. None of the frequency of drinking categories were predictive for either White or Hispanic respondents.

Gender-specific differences by race/ethnicity for risk from life-course drinking on injury from a serious accident were also found, primarily among Black respondents. Among Black males, the only significant drinking pattern variable was lifetime abstention, which was protective for injury, while among Black women, any drinking weekly and 5+ monthly were both significantly predictive of injury, but 5+ ≥weekly was not predictive.

Findings here are not entirely consistent with those reported previously in a similar life-course study (Kerr et al., 2015). In that study while risk of injury was found to increase with the frequency of 5+ drinking days for White respondents, for Hispanic respondents risk increased significantly for yearly and monthly 5+ drinking but not for more frequent heavy drinking, as found in the present study. For Black respondents in the previous study, unlike the present study, risk was significantly increased for daily 5+ drinking. Findings here are also not consistent with other US general population studies in which injury risk has been found to increase at relatively low levels of consumption and rise with frequency of heavy drinking (Cherpitel et al., 1995; Cherpitel and Ye, 2009). However, in another US general population study White respondents were found to be at highest risk of an alcohol-related injury based on 5+ drinking days, whereas Black respondents had the lowest risk at all levels of 5+ drinking (Cherpitel et al., 2016).

While findings for White and Hispanic respondents were not unexpected, findings for Black respondents, especially Black males, were. Neither frequency of any drinking nor frequency of higher consumption times were predictive of injury from a serious accident while lifetime abstention was protective, with abstainers having only a quarter of the risk as those who were drinkers during lifetime but not drinking at the given period. Potential explanations may be relevant to the observed reverse-disparity in risk of injury from drinking among Black males. Context of drinking among Black males may differ from that for White and Hispanic males, with Black males less likely to engage in risky injury-prone activities when drinking. They may also be more likely to underreport their drinking. Black males have been found more likely to consume beverages with a higher mean alcohol content (e.g. spirits and malt liquor) than either White or Hispanic males (Kerr et al., 2009), which would result in underreporting at a given level of consumption. Unfortunately, we were not able to adjust for drink size alcohol content in these analyses.

Elevated risk of injury for 5+ drinking ≥weekly and monthly for Hispanic men and women, on the other hand, could indicate a greater frequency of heavy occasions (such as fiesta drinking), in which a higher BAC level is reached, and in contexts encompassing greater injury risk such as outdoor drinking and activity in neighborhoods with poor maintenance of infrastructure such as sidewalks and roads, and high crime rates (Karriker-Jaffe et al., 2012; Jones-Webb and Karriker-Jaffe, 2013; Zapolski et al., 2014).

Some methodological limitations apply to this study. Data on both drinking and injury were based on self-report and retrospective recall. A test–retest analysis of the decades 5+ measure of the frequency of 5+ days averaged by the respondent over each decade of life was assessed using the 2005 NAS and a follow-up 2–3 years later. Findings indicated moderate consistency for each decade (P = 0.63–0.68), but findings for Black and Hispanic respondents were less consistent (P = 0.56) (Greenfield et al., 2014).

Only the earliest injury from a serious accident was recorded. Individuals who had multiple injuries from such accidents were removed from analysis after their first injury, potentially resulting in conservative estimates of the association of frequency of drinking and of 5+ occasions with such injury. Cause of injury was also not taken into account in these analyses, and risk from frequency of drinking and higher consumption occasions may differ by cause, For example, rates of alcohol-related motor vehicle crash fatalities have been found to be lower for Black and Hispanic compared with White drivers and passengers (Campos-Outcalt et al., 2003), an event which would require an individual to be either the driver or a passenger in a motor vehicle, putting them at risk for such an injury. Future studies of life-course drinking patterns and injury should include the cause of injury.

While models controlled for respondent’s education, family income, marital status, NAS wave and risk taking at the time of the interview, information on these variables in the year of injury was not collected and could potentially affect study findings. The time-invariant nature of control variables may also explain why the proportional-hazard assumption was violated for some of these measures when predicting injury occurrence across the lifetime. Interestingly, the effect estimates of time-varying drinking measures, the key predictors of the study, were quite stable before and after the stratified analysis, demonstrating the robustness of our study findings.

Differences in injury risks related to the frequency of any drinking and of heavy drinking episodes suggest a greater risk of injury from a serious accident for frequent heavy drinkers among White and Hispanic respondents, and among Black women, but no relationship among Black men, which should be the topic of future studies of race/ethnicity and injury. Frequency of any drinking was not found to be as significant a predictor. Findings here emphasize the importance of considering race/ethnicity and gender differences in race/ethnicity in studies of injury risks related to alcohol use and heavy drinking patterns.

Funding

This work was supported by an Alcohol Research Center grant from the US National Institute on Alcohol Abuse and Alcoholism (P50 AA005595).

Conflict of interest statement

None declared.

References