ED data doesn't tell us much about the spatial distribution of acute alcohol-related harm

One of my main research areas in recent years has been the social affects of alcohol outlet density - essentially looking at the relationship between the number of outlets in an area and measures of alcohol related harm. So, this recent paper in the journal Addiction (sorry I don't see an ungated version anywhere) by Michelle Hobday, Tanya Chikritzhs, Wenbin Liang, and Lynn Meuleners (all Curtin University) interested me.

In the paper, the authors look at the effect of alcohol outlets, sales and trading hours on alcohol-related injuries in Perth, Australia. The idea of looking at outlet numbers (separately for on-licence and off-licence outlets), sales, and trading hours all within the same statistical framework is interesting and potentially important. However, there is a problem in that they use data from emergency department (ED) presentations (I'll explain why that's a problem shortly). Anyway, the authors find:

At postcode level, each additional on-premises outlet with extended trading hours was associated with a 4.6% increase in night injuries and a 4.9% increase in weekend night injuries. An additional on-premises outlet with standard trading hours was associated with a 0.6% increase in night injuries and 0.8% increase in weekend night injuries.

So that seems fine. However, when looking at off-licence outlets:

Conversely, counts of off-premises outlets were associated negatively with alcohol-related injury, indicating a 3.9-4.9% lower risk per additional outlet.

What the hell? So, more off-licence outlets are associated with less harm?

John Holmes and Petra Meier (both University of Sheffield) wrote a commentary on the article in the same issue of Addiction. In the commentary, they correctly note that these sort of inconsistent results are endemic in the literature on the relationship between alcohol outlets and harm. By inconsistent I mean both inconsistent between different studies (even within the same geographic area), and inconsistent with theoretical predictions.

In this case, the problem is the measure of alcohol-related harm. Hobday et al. use (alcohol-related) ED presentations, which on the surface seems like a good measure of alcohol-related harm. Person drinks too much, suffers an accident (or violent incident) and goes to the hospital. Simple enough, right? The problem lies in the address coding in the ED dataset. In health data (like ED data), patients are geographically coded to their home address. This may, or may not, coincide with the location of the harm.

For chronic harm (e.g. cirrhosis), coding to patients' home addresses makes a lot of sense. The geographic accessibility of alcohol over the long term can reasonably be measured by the extent of access to alcohol from each patient's home. However, for acute harm (e.g. injury presentations) this doesn't hold. The geographic accessibility of alcohol on the night of the incident relates to where the patient was on that night, which may or may not be their home address at all. I'd wager that a lot of alcohol-related injury presentations at night (the measure used by Hobday et al.) arise from encounters in the night-time economy away from the patient's home. Indeed, Hobday et al. recognise the problems with their data:

A limitation of using ED records is that location information is restricted to the patient's place of residence, and data on last place of drinking are not recorded.

So, there is little reason for us to believe that we would observe a positive relationship between alcohol outlet numbers (or hours or sales) and ED presentation data. In fact, my co-authors and I observed mostly statistically insignificant results when looking at similar data for Manukau City. Having acknowledged the problems with the ED data, we have avoided this approach in our subsequent work (e.g. see here or here).

Now, let's think through the unexpected results. Hobday et al. find that there are more ED presentations from people who live in areas that have fewer off-licence outlets (especially those that open later) compared with areas that have more off-licence outlets. One potential explanation is that people who live close to an off-licence outlet (especially off-licence outlets that open later) have ready access to alcohol and can easily drink at home, and have less reason to travel to entertainment precincts where they might be at higher risk of becoming a victim of violence. This might be reinforced by drinkers who don't want to go to entertainment precincts to drink, but still want to have ready access to alcohol, choosing to live in areas where an off-licence outlet is nearby. In contrast, people who don't live close to an off-licence outlet (or where such outlets close earlier) must travel further to drink, and may therefore be more likely to drink in entertainment precincts where they are at higher risk of alcohol-related harm such as violence. I'm not sure whether this explanation is the true one that underlies the results, but it might be one contributing factor.

Overall though, for the sake of credibility of results, it might be best not to use ED data as a measure of acute alcohol-related harm, unless the location data relates to the location of harm rather than the patient's residential address.