The interactive effect of location, alcohol consumption and non-traffic injury

Abstract

Aim:

This study aimed to assess the extent to which the association between recent alcohol consumption and risk of non-traffic injury varies according to location at the time of the injury.

Design:

Case-crossover design.

Setting and participants:

15,625 injury patients from 49 emergency departments (EDs) in 22 countries.

Measurements:

Recent alcohol consumption and location at the time of the injury were assessed for when the injury occurred and for the same time 1 week prior to this. The confounding and interactive effects of location were examined by estimating the adjusted odds ratio (OR) of injury from alcohol consumption adjusting for location and then by examining the alcohol consumption-location interaction.

Findings:

There were significant interactive effects of location and alcohol consumption on injury risk. For example, the ORs for volume 0.1-3.0 drinks and street/public place each were 3.0 and 14.2, respectively, whereas the OR for their joint effect was 44.1, suggesting a positive additive interaction (relative excess risk due to interaction [RERI]=27.9, p<0.05) and zero multiplicative interaction (OR=1.0, p=0.895). The interactions of alcohol consumption with drinking establishment location, workplace and other locations were mostly additive and negative on the multiplicative scale (e.g. for interaction between volume 0.1-3.0 drinks and drinking establishment location: RERI=1.2, p=0.529; multiplicative interaction OR=0.54, p<0.05).

Conclusions:

Location appears to influence the relationship between alcohol consumption and risk of injury. The association between alcohol consumption and injury appears to be greater in locations such as streets and public places compared with private residences.

INTRODUCTION

Alcohol is a leading risk factor for injury mortality and morbidity worldwide (1). An elevated risk of traffic and other causes of injuries from acute alcohol consumption has consistently been found across countries and cultures using different study designs, and with a dose-response relationship observed (2–5). Also playing important roles in the association between alcohol and injury are social and contextual factors such as socio-economic status, drinking settings and the time of consumption (6–8). Among them, one important yet under-researched area is the environmental context of injury and its effect on the alcohol-injury risk estimation.

The risk of an injury is dependent upon the person(s) who contribute to the injury, the vehicle/agent which causes the injury, and the environment where the injury takes place (9). Life style and routine activities such as daily movements of individuals through time and space can escalate or diminish opportunities for accidents and violence to occur (10). Environmental hazards surrounding the injury event, or the injury context, can be characterized by the location where an injury occurs and activities of the victim at the time of the accident. Studies on risk of substance-related road accidents and traffic injuries have taken care to control for the effect of context when comparing injured drivers (cases) with control samples of drivers pulled to the roadside (11–13). When comparing pedestrian injuries with site-matched controls, Blomberg and Fell (14) found the risk of injury from alcohol intoxication was considerably lower than when comparing to random-site controls. Literature on contexts related to non-traffic injuries is limited, however, with more emphasis on drinking context rather than injury context, particularly for circumstances when drinking leads to violence (15, 16). One study on incidents of physical aggression, while investigating in detail the location of aggression, lacked appropriate controls to model risk from drinking and context (17). To our knowledge, only two studies have examined how injury context affects risk of injury from drinking. The studies, both conducted in emergency departments (EDs) on injury patients from all causes, found that, compared with unadjusted estimates, risk of injury from drinking was either moderately increased after controlling for the location and activity of injured patients as covariates (18), or barely changed after matching activity between case and control periods (19).

These studies treated context as a confounder, assuming that risk of injury from drinking is constant across various injury contexts. Conversely, one can easily hypothesize that drinking is risker when associated with certain activities/locations than others. For example, the US Surgeon General’s warning - consumption of alcoholic beverages impairs your ability to drive a car or operate machinery - indicates that drinking carries differential risk based on environmental context. However, the nature of the interactive effect from environmental contexts on substance-related injury has not been fully studied, which necessitates assessing interactions on both additive and multiplicative scales.

The difference in additive and multiplicative interaction has long been illuminated in epidemiology and it has been argued that the assessment of interaction should be mainly based on an additive scale (20, 21). However, in models commonly used such as logistic regression, assessment of the interaction effect is often performed by simply introducing the product terms, which basically examines the multiplicative interaction (22) and the test of additive interaction is not usually performed (23). The complexities surrounding different interactive effects are arguably applied to the relationship between injury, drinking and environmental context. Risk from alcohol consumption may be magnified in some contexts (e.g. walking in unsafe neighborhood after drinking), multiplying the danger of injury; while for other contexts (e.g. drinking at a friend’s house) the interactive effect of context on drinking is more likely to be additive, with risk from drinking and context independent from each other. Furthermore, identifying the nature of interaction between drinking and injury context may also help in estimating the alcohol attributable fraction (AAF) of injury more accurately (24).

Using a combined international ED injury dataset, this study explored the relationship between injury, acute alcohol consumption and the injury location as an indicator of environmental context. We examined the potential confounding and interactive effect of context by estimating the risk of injury from drinking controlling for the effect of location context and the joint effect of drinking and context for both additive and multiplicative interactions. We also evaluated how the AAF changes after adjusting for the confounding or interactive effect of context.

METHODS

Samples

Data analyzed are from 26 studies (49 ED sites) in 22 countries included in the International Collaborative Alcohol and Injury Study (ICAIS) (25). The combined dataset includes data from four international collaborative research projects: the Emergency Room Collaborative Alcohol Analysis Project (ERCAAP) (26), and three other international collaborative studies conducted by the WHO (27), the Pan American Health Organization (PAHO) (28), and the United States National Institute on Alcohol Abuse and Alcoholism. Full information regarding the included ED studies, such as year of study and the city where each study was performed, valid sample size, and references is provided in the supplemental material (Table S1). The analytical sample constituted a total 15,625 injured patients with valid data on drinking and context for both case and control periods.

In each ED study, a probability sample of injured patients 18 years and older arriving within six hours of the injury event was obtained based on consecutive arrival to each ED, with equal representation of each shift for each day of the week. Four studies (Argentina 2001, Canada 2002 and 2014, New Zealand 2000) are exceptions, oversampling some days or shifts for logistic reasons or to maximize opportunities to include drinkers and these data have been weighted to achieve representative samples. Completion rates averaged 87% across all studies (range 59-100%). Trained interviewers administered a structural instrument of about 25 minutes in length; questionnaires were similar across studies.

Measures

Alcohol consumption was assessed at both case and control periods. Patients were asked whether they had consumed any alcohol during the six hours before and up to the injury/accident and, thinking of the same time of the injury a week earlier, whether they had any alcohol during the six hours leading to that time. For those answering “yes” for either time, the beverage-specific number, size, and alcohol concentration of drinks covering all alcoholic beverage types, including those commonly consumed locally, were queried to determine the volume of consumption for alcohol exposures for both periods. Volume was then coded into no drinking, 0.1-3 drinks, 3.1-7 drinks and >7 drinks, with one standard drink defined as 16 ml of pure ethanol. The cut-points were chosen to achieve approximately equal distribution of injury patients across volume categories.

Location context of injury was measured by the place of injury. Patients were asked where they were when the injury/accident happened and where they were a week earlier at the same time of the day as when they had the accident. Two versions of location question were asked across the ED studies. The first version included the following categories: own home, other’s home, street/highway, school, pub/hotel/tavern/other drinking place, workplace, and others. The second version had the response options including: own home, other’s home, pub/hotel/tavern/other drinking place, nightclub, sports club, restaurant/café, theatre/movie, workplace, in a private vehicle, at a sporting event, at an outdoor public place such as beach and park, and others. The final location variable used had five categories for analysis (original categories shown in parenthesis): private residence (own home or other’s home), street/public places (street/highway, in a private vehicle or at outdoor public place), drinking establishment (pub, hotel, tavern, other drinking place or nightclub), workplace, and other (sports club, restaurant and café, theatre and movie, sporting event or other places).

Cause of injury was based on how the injury occurred, including response options: being in a vehicle collision as driver, passenger, or pedestrian hit by a vehicle (traffic), a fall or trip (fall), blunt force injury, stab/cut/bite, gunshot, choking/hanging, stuck against/caught between, drowning, poisoning, burn and others. Patients were also asked whether they got into a fight, were beaten, attacked or raped (violence). Cause of injury was categorized as traffic, fall, violence, and other injuries that include all injuries not belonging to the other three causes.

Analysis

Case-crossover analysis (29) was used to estimate the risk of injury associated with alcohol consumption, comparing self-reported drinking levels within 6 hours prior to the injury event (case period) with alcohol use during the same 6-hour period the previous week (control period). For case-crossover analysis, individuals serve as their own controls, therefore adjusting by design for confounding from stable factors such as socioeconomic status or risk-taking personality. The case-crossover method is a matched case-control design, modeled by conditional logistic regression (30) with time-varying risk factors (i.e. drinking and injury location) as predictors. To examine the confounding effect of location, conditional logistic regression is fit as in Formula 1

$$\mathit{logit}(P)={\beta }_{1i}{X}_i+{\beta }_{2j}{Y}_k$$ [Formula 1]

where P is the probability of injury, X_i is drinking at level i (0.1-3 drinks, 3.1-7 drinks, > 7 drinks versus no alcohol), Y_k is at location k (street/public place, drinking establishment, workplace, other places versus private residence). The odds ratio (OR) of injury from drinking is estimated by exp(β_1i), and the confounding of location is examined by comparing ORs from drinking with and without controlling for location in the model as a covariate.

Interactive effect on additive and multiplicative scales

We illustrate assessment of the statistical interaction on additive and multiplicative scales following the comprehensive review by VanderWeele and Knol (31, 32) and a recent fine application by Mehta and Preston (33). To illustrate simply, below we assume dichotomous drinking (X=1 or 0 for any or no alcohol) and location context for injury (Y=1 or 0 for public place or private residence). The model to estimate both additive and multiplicative interactions can be expressed as Formula 2, adding the product of the drinking and location indicator to Formula 1.

$$\mathit{logit}(P)={\beta }_{1}X+{\beta }_{2}Y+{\beta }_{3}X\ast Y$$ [Formula 2]

Thus exp(β₁) is the OR of injury for drinking combined with location at a private residence (denoted as OR₁₀), exp(β₂) is the OR for no drinking combined with location at a public place (denoted as OR₀₁), and exp(β₁+β₂+ β₃) is the OR for drinking combined with a public place (denoted as OR₁₁), all with no drinking combined with a private residence as reference. For additive interactions, Relative Excess Risk due to Interaction (RERI) and the Synergy Index (SI) will be estimated, based on the formulas RERI = OR₁₁ − OR₁₀ − RR₀₁ + 1, and SI = (OR₁₁ − 1)/[(OR₁₀-1)+(OR₀₁-1)]. RERI=0 or SI=1 implies a zero additive interaction (or interaction is additive), and RERI>0 or SI>1 (RERI<0 or SI<1) indicates a positive or super-additive (negative or sub-additive) interaction. Multiplicative interaction is evaluated based on Formula 2 interaction effect exp(β₃), which is equivalent to OR₁₁/(OR₁₀*OR₀₁), with the value of one implying a zero multiplicative interaction (or interaction is multiplicative) and the value >1 (<1) indicating a positive (negative) multiplicative interaction.

It is important to note that the assessment of interaction, additive or multiplicative, can be interpreted in two ways. The first answers the research question how the joint effect of two risk factors is compared to the expected (additive or multiplicative) risks from the two factors separately. Thus a positive additive interaction (RERI>0 or SI>1) suggests that injury risk exposed to the combined effect from drinking and public location together exceeds the separate risks from drinking and public place added together. A positive multiplicative interaction (exp(β₃)>1 or OR₁₁>(OR₁₀*OR₀₁)) indicates that risk from both drinking and public location is larger than the separate risks from drinking and public place multiplied together. The second research question that interaction assessment is aimed to answer is whether the effects of one risk factor (i.e. drinking on injury) vary across levels of the other risk factor (location). Here, a positive additive interaction between drinking and public place suggests that the risk difference for drinking versus no drinking is larger for public place compared to private residence (i.e. OR₁₁ − OR₀₁ > OR₁₀ − 1), while a positive multiplicative interaction indicates that the risk ratio for drinking versus no drinking is larger for public place compared to private residence (i.e. OR₁₁/ OR₀₁>OR₁₀). Whether an interaction is positive or negative may depends on the scale. Particularly, provided both exposures have an effect on the outcome (e.g. drinking and public location on injury with OR₀₁ and OR₁₀ both > 1), it has been shown there will always be an interaction on at least one scale (31) and it is likely to have a positive additive interaction but negative or null multiplicative interaction.

Injury AAF adjusting for confounding and interactive effect of context

First, AAF without adjusting for context was estimated using Formula 3. P_i is the prevalence of acute consumption at a given level among injured patients, and OR_i is the unadjusted OR of injury at that level compared to no alcohol. AAF adjusting for context as a covariate was estimated using the same formula, with ORs replaced by the adjusted ORs controlling for context as in Formula 1.

$$\mathit{AAF}=\sum P_i(1-\frac{1}{{\mathit{OR}}_i})$$ [Formula 3]

To estimate AAF adjusting for the interactive effect of context, we followed Bruzzi et al (34) as in Formula 4.

$$\mathit{AAF}=\sum P_{\mathit{ij}}(1-\frac{{\mathit{OR}}_{0j}}{{\mathit{OR}}_{\mathit{ij}}})$$ [Formula 4]

Here P_ij is the prevalence of drinking at i level in context category j, OR_ij is the OR of injury from joint exposure of drinking at i level in context j versus no drinking at a private residence, and OR_0j is OR from context j without drinking. Bootstrap confidence intervals (CIs) for AAF were estimated based on 1000 replications. All analyses were performed using STATA version 15 (35). The analysis was not pre-registered on any public available platform.

RESULTS

Table 1 shows the distribution of location context for injury and the control period, for the total injured sample and by cause of injury. Both the symmetry tests and marginal heterogeneity tests were highly significant (p<0.001) for all groups, suggesting substantial differences in locations at the time of injury occurrence and the same time one week earlier. Compared with the control period, place of injury was more likely to be in a street/public place and less likely to be at a private residence. The location of 90% of the traffic injuries were in the street/public place.

Table 1.

Place when injury happened and at the control time last week¹ (%)

	Total injuries (N=15625)		Traffic (N=3061)		Violence (N=2005)		Fall (N=4928)		Others (N=5752)
	Injury	Control	Injury	Control	Injury	Control	Injury	Control	Injury	Control
Private residence	26.5	56.2	0.9	51.5	22.2	62.4	34.2	62.2	34.8	51.4
Street/public place	37.2	6.5	90.0	11.9	38.3	5.0	32.2	6.5	13.2	4.1
Drinking establishment	4.6	2.7	0.2	2.5	18.7	6.1	3.3	2.1	3.1	2.0
Workplace	16.6	25.3	3.1	26.9	10.4	18.8	12.9	18.6	28.8	32.4
Others	15.2	9.4	5.8	7.1	10.4	7.6	17.4	10.6	20.1	10.1

¹p<0.001 for both the symmetry tests and marginal heterogeneity tests on equality of location distributions between the injury and control periods for total injured combined and by injury cause

Table 2 shows the ORs for injury from drinking at three volume levels versus no alcohol before injury and after adjusting for location as a covariate, and after matching the location between injury and control periods. Elevated risk of injury across drinking categories was observed. After controlling for and matching context, the adjusted ORs were generally reduced compared with the unadjusted ORs. Since in theory street/public place is a necessary condition for a traffic accident, controlling for context as a covariate when predicting traffic injuries may not be appropriate and the matched sample is limited in sample size. Thus risk estimates related to traffic injuries should be taken with caution. Table 2 also shows the ORs for total injuries excluding traffic injuries.

Table 2.

Odds Ratios (ORs) of injury related to drinking (ref. no alcohol) controlling for and matching injury places

	No control or matching		Controlling as covariates		Matching places at injury and control time
	OR	95% CI	OR	95% CI	OR	95% CI
Total injuries		N=15625		N=15625		N=6630
≤ 3 drinks	2.70	(2.37, 3.07)***	2.51	(2.14, 2.95)***	2.72	(2.19, 3.38)***
3.1 - 7 drinks	5.85	(4.97, 6.88)***	4.10	(3.35, 5.01)***	3.87	(2.94, 5.08)***
> 7 drinks	9.38	(7.79, 11.29)***	5.58	(4.48, 6.96)***	5.77	(4.19, 7.94)***
Traffic		N=3061		N=3061		N=464
≤ 3 drinks	2.13	(1.59, 2.86)***	3.11	(1.32, 7.36)**	2.86	(1.00, 8.13)*
3.1 - 7 drinks	4.22	(2.89, 6.15)***	2.46	(0.91, 6.69)	2.53	(0.81, 7.84)
> 7 drinks	8.05	(4.88, 13.29)***	3.16	(0.74, 13.43)	2.75	(0.40, 19.02)
Violence		N=2005		N=2005		N=657
≤ 3 drinks	5.60	(4.03, 7.78)***	3.80	(2.52, 5.73)***	3.49	(2.02, 6.02)***
3.1 - 7 drinks	12.55	(8.65, 18.22)***	6.54	(4.20, 10.20)***	6.17	(3.24, 11.76)***
> 7 drinks	21.70	(14.49, 32.49)***	10.08	(6.32, 16.08)***	10.75	(5.36, 21.57)***
Fall		N=4928		N=4928		N=2313
≤ 3 drinks	2.63	(2.08, 3.34)***	2.75	(2.06, 3.66)***	3.56	(2.37, 5.33)***
3.1 - 7 drinks	5.47	(4.08, 7.32)***	3.84	(2.72, 5.43)***	3.77	(2.41, 5.89)***
> 7 drinks	8.14	(5.83, 11.37)***	5.17	(3.52, 7.59)***	5.48	(3.22, 9.31)***
Other injury		N=5752		N=5752		N=3228
≤ 3 drinks	2.13	(1.72, 2.63)***	2.05	(1.63, 2.58)***	2.15	(1.59, 2.92)***
3.1 - 7 drinks	4.37	(3.24, 5.89)***	3.80	(2.74, 5.27)***	3.52	(2.25, 5.48)***
> 7 drinks	5.69	(4.12, 7.87)***	4.53	(3.19, 6.43)***	4.33	(2.61, 7.18)***
Total injuries (no traffic)		N=12564		N=12564		N=6166
≤ 3 drinks	2.85	(2.47, 3.28)***	2.54	(2.15, 2.99)***	2.71	(2.17, 3.38)***
3.1 - 7 drinks	6.27	(5.23, 7.52)***	4.31	(3.50, 5.30)***	3.96	(2.99, 5.24)***
> 7 drinks	9.67	(7.91, 11.81)***	5.86	(4.68, 7.33)***	5.88	(4.25, 8.14)***

^*p<0.05,

^**p<0.01,

^***p<0.001

The results for drinking and location interaction analysis on total injuries (excluding traffic) are presented in Table 3. The interaction between drinking and street/public place was not only super-additive but also multiplicative. For example, compared to no alcohol with location at a private residence, the ORs for street/public place with no alcohol and consuming 0.1-3 drinks with location at a private residence were 14.2 and 3.0, respectively. In contrast, the OR for the joint effect of 0.1-3 drinks combined with location at a street/public place was 44.1, much larger than the sum of the two separate effects and similar to the two ORs multiplied together. Positive additive interaction was also confirmed by RERI (27.9, 95% CI: 4.7-51.0, p<0.05) and SI (2.8, 95% CI: 1.6-4.9, p<0.001), while zero multiplicative interaction was shown by a multiplicative effect OR of 1.0 (95% CI: 0.6-1.8, p=0.895). Similar results were observed for volume 3.1-7.0 and >7.0 drinks.

Table 3.

Odds Ratios (ORs) and 95% confidence intervals (CIs) of total injuries (excluding traffic) from drinking, location context and their joint effects as well as the interaction between drinking and context on additive and multiplicative scale

		Conditional logistic regression estimates			Interaction on additive scale		Interaction on multiplicative scale
Context	Volume	OR context at volume = 0	OR volume at private residence1	OR for both Context & volume	RERI2	SI3	Multiplicative effect4
Street/public place	0.1-3.0	14.23 (12.27, 16.51)	2.98 (2.40, 3.71)	44.07 (25.99, 74.74)	27.86* (4.73, 50.99)	2.83*** (1.64, 4.88)	1.04 (0.59, 1.84)
	3.1-7.0		5.97 (4.47, 7.98)	60.94 (35.11, 105.78)	41.73* (8.31, 75.16)	3.29*** (1.87, 5.80)	0.72 (0.39, 1.33)
	>7.0		8.96 (6.63, 12.11)	116.10 (67.59, 199.41)	93.90** (31.44, 156.36)	5.43*** (3.13, 9.42)	0.91 (0.50, 1.67)
Drinking establishment	0.1-3.0	5.21 (3.76, 7.22)	2.98 (2.40, 3.71)	8.38 (5.59, 12.55)	1.19 (−2.51, 4.89)	1.19 (0.71, 2.01)	0.54* (0.31, 0.93)
	3.1-7.0		5.97 (4.47, 7.98)	10.74 (7.45, 15.48)	0.56 (−3.83, 4.94)	1.06 (0.67, 1.68)	0.35*** (0.20, 0.59)
	>7.0		8.96 (6.63, 12.11)	14.51 (10.29, 20.47)	1.34 (−4.09, 6.78)	1.11 (0.73, 1.68)	0.31*** (0.18, 0.52)
Workplace	0.1-3.0	1.61 (1.44, 1.81)	2.98 (2.40, 3.71)	3.99 (2.64, 6.03)	−2.93 (−7.82, 2.10)	1.15 (0.64, 2.06)	0.83 (0.53, 1.30)
	3.1-7.0		5.97 (4.47, 7.98)	11.15 (5.20, 23.89)	4.56 (−4.04, 13.17)	1.82 (0.75, 4.39)	1.16 (0.51, 2.60)
	>7.0		8.96 (6.63, 12.11)	2.68 (1.21, 5.94)	−6.89*** (−10.18, −3.60)	0.20* (0.05, 0.70)	0.19*** (0.08, 0.42)
Others	0.1-3.0	5.66 (5.00, 6.41)	2.98 (2.40, 3.71)	7.54 (5.11, 11.13)	−0.10 (−3.05, 2.85)	0.99 (0.63, 1.55)	0.45*** (0.29, 0.69)
	3.1-7.0		5.97 (4.47, 7.98)	10.02 (6.03, 16.65)	−0.61 (−5.86, 4.65)	0.94 (0.53, 1.67)	0.30*** (0.17, 0.53)
	>7.0		8.96 (6.63, 12.11)	14.62 (8.77, 24.37)	1.00 (−6.61, 8.61)	1.08 (0.61, 1.89)	0.29*** (0.16, 0.51)

^*p<0.05,

^**p<0.01,

^***p<0.001

¹Private residence was used as the reference category

²Relative Excess Risk due to Interaction (RERI): Significance test of RERI=0 with null hypothesis indicating non-significant interaction on additive scale, or the interaction is additive

³Synergy Index (SI): Significance test of SI=1 with null hypothesis indicating non-significant interaction on additive scale, or the interaction is additive

⁴Significance test of multiplicative effect=1 with null hypothesis indicating non-significant interaction on multiplicative scale, or the interaction is multiplicative

Conversely, for locations at a drinking establishment, workplace and other injury places, their interactions with drinking were mostly zero on the additive scale, with RERIs close to zero and SIs close to one. For drinking establishment and other places, negative multiplicative interactions with drinking were consistently observed. For workplace, the interactions with drinking on additive and multiplicative scales were found to be quite similar: non-significant for volume levels ≤ 7 drinks and negative for volume > 7 drinks. These consistent findings on interactions between the two scales may partly be due to the relatively low baseline risk for workplace alone (OR=1.6). The interactions between drinking and location context predicting violence, fall and other injuries, separately, were also examined. Their results were quite similar to those shown for total injuries and presented in the supplemental material (Table S2–S4).

Last, estimates of injury AAF with and without adjusting for location context are shown in Table 4, for total injuries (excluding traffic), violence, fall and other injuries. Before adjusting for context, the AAF was 15.5% for total injuries; it reduced to 14.3% controlling for injury location as a covariate, and further dropped to 13.4% after adjusting for the interactive effect of context. This pattern was observed for all three causes of injuries. The largest change was seen for violence, where AAF estimates dropped from 39.2% to 30.7% after drinking and the injury context interaction was considered.

Table 4.

Alcohol Attributable Fraction (AAF, in %) and 95% confidence intervals (CIs) of injuries with and without adjusting for location context

	Context not adjusted		Adjusting for confounding effect of context as a covariate		Adjusting for interactive effect of context
	AAF	95% CI	AAF	95% CI	AAF	95% CI
All injuries excl traffic	15.5	(14.7, 16.3)	14.3	(13.4, 15.3)	13.4	(12.2, 14.6)
Violence injuries	39.2	(36.9, 41.5)	36.3	(33.2, 39.4)	30.7	(24.5, 36.9)
Fall injuries	12.8	(11.6, 14.1)	12.2	(10.8, 13.6)	11.9	(9.8, 13.9)
Other injuries	9.1	(8.0, 10.2)	8.6	(7.4, 9.8)	7.7	(6.2, 9.2)

DISCUSSION

In this study we systematically examined the confounding and interactive effects of the environmental context of injury when estimating risk of injury related to acute drinking and injury AAF. After controlling for location context as a covariate, a moderate reduction in ORs from drinking was observed. Our main finding is the interactive effects across various locations for the alcohol and injury relationship. A positive additive interaction between drinking and street/public place indicates risk of injury from the two exposures together exceeds the sum of each considered separately, suggesting that the effect of intoxication such as impairment of vision, body balance and judgement as well as disinhibition is most magnified in high-risk locations like street/public places. Another interpretation of this finding is that the difference in injury risk between drinking versus no drinking is larger in a street/public place compared to a private residence. As a more relevant measure with public health significance (31), the super-additive interactive effect of street/public place calls for interventions targeting these locations to reduce harmful effects of drinking.

For locations such as drinking establishment, workplace and other places, zero interaction on the additive scale suggests risks from these location contexts are additive to the risk from drinking. The risk difference from drinking is similar in these contexts to that in a private residence. In addition, negative multiplicative interactions were observed, particularly for drinking establishment and other places, indicating the risk ratio from drinking when a person is in these places is smaller than when in a private residence. This may lead to the wrong conclusion that, compared to a private residence, these locations are protective when combined with drinking, and highlights the importance of assessing additive and multiplicative interactions simultaneously with appropriate interpretation. In current analysis, compared to a private residence, people in a street/public place were exposed to risk of injury roughly 14 times higher and these baseline risks for people in a drinking establishment or other places were about five times higher. Thus a constant risk ratio for drinking results in a much higher absolute risk for those in these locations than those in a private residence.

Distinction needs to be made between the location of drinking and the location of injury. Drinking in on-premise locations has routinely been found to be associated with heavy drinking (36, 37) and alcohol-related aggression (7, 38), in comparison to drinking at private settings. But in this study, location data is on the place of injury, while data on drinking concerns the six hours prior to the injury. The drinking may thus have occurred in a different location from the injury. In particular, many injuries in public places occur on the way home after a night out, often in a drinking establishment (39). Also, as a matter of routine, staff at drinking establishments will push incipient trouble, such as a noisy drunk or an intense argument or fight, out onto the street (40), motivated not only by avoiding disturbance of other customers but also by laws that an intoxicated person cannot stay in the bar, and by the threat to the establishment’s license to sell alcohol if it has a record of fighting and trouble (41, 42). Our findings suggest attention in future studies to the location of drinking as well as of the injury, in studying the influence of location on the relationship between drinking and injury. In the meantime, the strong relationship found in the present study between a street/public place location of injury and prior drinking suggests the need to reexamine laws and practices under which intoxicated patrons of drinking establishments are pushed out on the street without any provision for protection or getting them home.

This study identified a number of limitations that could be overcome in future work. A major area to improve is the accurate recording and coding of context measures. Detailed description and classification of event location and activity characteristics is undoubtedly complicated and difficult for all injuries other than those traffic-related. Moreover, individuals falling into the same context category may not be homogeneous. For example, workplace was found in this study to have the lowest risk of injury next to private residence, obscuring the high risk in some specific types of occupations such as construction work. In this study, two versions of location context measures were used across the ED studies, and misclassification error may have occurred when they were collapsed into the categories for analysis. Several other limitations need to be noted. Although case-crossover analysis controls for potential confounders which are stable over time, the results may be biased by missing time-varying covariates such as drug use. One prior ED study in Canada, however, found that the effect of alcohol was not affected after controlling for drug use (43). Further, the merged ED data provided us the needed sample size to model the full interaction between volume levels and categories of injury location, yet ignored the potential effect heterogeneity across countries. Our findings related to interaction may vary by socioeconomic status as well, with excessive risk from drinking combined with the location in a street/public place more associated with poorer individuals living in risker neighborhoods. Further analysis is needed for interaction analysis stratified by region and income. Last, the week prior to injury was chosen as the control period since drinking tends to follow a weekly pattern. There has been mixed evidence related to recall bias on alcohol consumption (44, 45) and less is known regarding the recall of context.

CONCLUSIONS

This study systematically evaluated the confounding and interactive effect of the context of injury in the relationship between drinking and injury, particularly examining both additive and multiplicative interactions. It showed that varied location contexts play different roles in the relationship between alcohol consumption and injury, and that the effect of drinking is most magnified in high-risk locations like street/public places. Findings suggest interventions targeting these locations to reduce harmful effects of drinking. The study also exemplified the importance of correctly interpreting the interaction results on additive and multiplicative scales.