Drivers are unsafe at any Blood Alcohol Concentration?

Jan 22 2014 Published by under Alcohol, Drug Abuse Science, Drug Fatality

A reasonably provocative paper which suggests that automobile drivers are impaired at a blood alcohol concentration (BAC) as low as 0.01% has recently appeared.

Phillips DP, Sousa AL, Moshfegh RT. Official blame for drivers with very low blood alcohol content: there is no safe combination of drinking and driving.Inj Prev. 2014 Jan 7. doi: 10.1136/injuryprev-2013-040925. [Epub ahead of print][Pubmed][Publisher]

When I was first told of this finding, my initial curiosity was not so much about the findings but more about the design. It is incredibly difficult to come up with ways to compare drinking driver versus non-drinking driver stats in field studies or data mining retrospectives.

The authors drew data from US traffic fatalities1 recorded in the National Center for Health Statistics database and the Fatality Analysis Reporting System database. The study sought to test the hypothesis that driver BAC would be related to the driver being determined to be solely and officially at blame for the crash. There are numerous factors that were coded for drivers including "under the influence of alcohol, drugs or medication" and "driving on wrong side of the road". The "under the influence..." factor was dropped from all analyses for the obvious reasons that it would contaminate their test of hypothesis.

This is important for the reader to grapple with his or her most obvious complaint about this design. If the police officer is determining the responsibility and can smell (or otherwise detect) alcohol on one driver, this puts a bias in the outcome measure (responsibility for the crash) that would tend to correlate with the thing being tested (BAC). So the authors focused on the factors that were seemingly unambiguous. Such as "running off road" or "driving wrong way on one-way" versus "unsafe speed for conditions" and other ambiguous factors that depend on a police judgement.

The authors calculated the Sole Official Blame (SOB) as the number of drivers officially and solely blamed for the crash divided by the number of drivers officially assigned no blame for the crash. They also calculated the percentage of drivers blamed solely and officially for the crash divided by the total number of drivers involved. Phillips14-traffic-F1Figure 1 from the paper presents the SOB by the BAC for both male and female drivers. The solid line is the "All Blame Factors" and the dotted line is for the unambiguous factors- the far better measure2, IMO. These data do make a case that fatal crash risk is an essentially linear function of BAC. Importantly, there is no inflection of the curve at either 0.08 or 0.1 BAC which have been the US legal limits during my driving lifetime. The error bars are 95% confidence intervals and they are using lack of overlap of the 95% CI as their inferential statistic indicating a significant difference (they also include a chi-square statistic for the 0.1 BAC vs sober bin). So far, so good. BAC is linearly correlated with the risk of being the sole and officially blamed driver for a fatal accident. [UPDATE: I didn't originally catch a bit of a dodge the authors are pulling here. The inferential analyses are conducted on the "all blame factors. Then they state "these patterns hold when one considered all blame factors or unambiguous factors". This "pattern" language is sometimes used to skip over the fact that the inferential analysis didn't hold up on the other variable(s). This is a big problem, given my questions about the contamination of the blame issue if the officer knows one driver had been drinking alcohol.]

An interesting side-analysis looked at the problem that BAC is not always measured which could introduce a bias. I'm assuming that the first analysis used only verified negative BAC "sober" drivers but it is hard to find this directly stated. Anyway, they looked at the correlation on a state-by-state basis between the SOB "buzzed", aka 0.1 BAC and SOB sober (which was 2.09 for the overall dataset), and the percent of unmeasured BACs in the fatal crash listing. The correlation was negative, showing that the lower the proportion of unmeasured BACs in a state, the larger the difference between sober and 0.1% BAC drivers in fatal crash responsibility. So if anything, I guess we have to assume that a lower percentage of blood testing results in an underestimate of the crash risk.

The authors next moved on to take a crack at the question of circumstances. In essence it addressed the question of whether people driving at 0.1% BAC are doing so under risky circumstances. At night, for example.
Phillips14-traffic-F3The third Figure from the paper depicts SOB ratio and the Percent Blamed for a subset of two-car crash pairings in which one driver was sober and the other was at a positive BAC. The beauty here is that nondriver circumstances are as identical as you can get for the sober and intoxicated drivers. The authors performed 16 chi-square tests but a quick multiple-comparisons adjustment to the listed p values shows they still survive as all of them being different, BAC vs sober, for SOB and P measures. Odds of being at fault are 60/40 for 0.1% BAC versus sober and about 80/20 by the time you reach two car crashes in which one driver was at 0.08% BAC. Interestingly this is the analysis that appears to show some categorical difference between BAC of 0.1-0.3% and BAC of 0.6-0.8%. They also did a cute little comparison of paired-crashes where one driver was at 0.08 and the other was 0.5-0.7 BAC. The SOB did not differ (95% CI overlap) in this analysis.

As a final note, a bunch of supplementary analyses were provided to try to rule in or out additional driver (sex, race), vehicle (speed and model year) and circumstantial (raining, time, location) factors. The relationship of SOB with BAC persisted.

Probably my largest question about traffic risks conferred by low levels of alcohol consumption is captured by the report of the relative effect size of the "most common driver factors" in Table 1. The "Driving too fast for conditions or in excess of the posted speed limit" factor is a large contributor to SOB ratio in the 0.1-0.7%BAC drivers as well as one of the larger differences from the sober drivers. This underlines a suspicion that those who are willing to drive after a low amount of alcohol consumed are perhaps innately different from those who are not. They might be somewhat more of a risk taker. Here, we'd really want to get at the population that is willing to drive after a drink or two and look at their driving crash risk when they are sober. Methodologically, this is asking for a lot, I realize.

This is only one study, of course. There may be other data out there that show a less continuous function of fatal crash risk to BAC in this range below the current US legal limit. But this is for sure an important study.

__
1authors say that there is no sufficiently detailed database for nonfatal crashes, sady.

2Still trying to wrap my head around whether these "unambiguous" factors are in fact uncontaminated by the police officer's knowledge that one of the drivers had been drinking. Presumably they write up their reports somewhat after they have investigated. Maybe I'm searching for rigor where none is needed but it still bugs me.

4 responses so far

  • chemicalbilology says:

    Isn't there a huge problem with the assumption that the dataset is representative of drivers overall? Because anyone who does not get pulled over (because they are presumably driving safely) is invisible here. There's no way to know the size of that invisible group. Am I missing something?

  • drugmonkey says:

    Well that's tricky. They make it quite clear in the paper that this is about crashes that involve a fatality. From an internal perspective, there is no assumption this is representative of drivers overall.

    From a policy perspective, sure, I bet this will lead to people acting as if this represents increased risk of traffic ...something bad... in the entire population of drivers with detectable BAC.

    Fatality may be relatively rarer but it sure does loom large as an important consequence of traffic accident, and rightly so.

    Do the individuals causing fatal accidents represent the population of those drivers at similar BAC? Yes, in a certain sense they do. And until we have plausible arguments and data on the table to show they are different, should this not be the null hypothesis?

  • drugmonkey says:

    I should point out that there are studies with a related design that may be more satisfying to you.

    The key is in trying to catch a control population. Investigators take the strategy of going to a police roadblock at the right suspect hours (usually Friday and Saturday night). They then try to recruit a representative sample of these non-offending drivers and do alcohol and drug testing. These then serve as the Controls in Case-Control designs where the cases come from the DUI or fatality records.

    There are some big gaping holes in this sort of design but at least it tries to get a peek at the drug and alcohol levels in drivers who are not in an accident or being pulled over for erratic driving.

  • Erickttr says:

    A review in J Forensic Sci (sept 2013, 58:5:1238-1250) came to a similar conclusion. The dataset was Canadian, I believe, so the legal terms and such are slightly different. But the general conclusion is that impairment can be detected in nearly any BAC level of 0.05% impairs faculties required for driving. But the determination of being in an "impaired state" is subjective to the observer and the complexity of the driving task / situation.

Leave a Reply