Use of alcohol as a risk factor for bicycling injury

Skip below if you’ve visited this page specifically to see the Johns Hopkins’ study.

FARS Alcohol Results

The FARS data has a number of alcohol (and drug) fields — the fields ATST_TYP, ALC_RES relate actual test type, and results. To simplify things, I’ve added a derived field sALC_RES to breaks down test results into: negative, .01 through .07, and .08+, or no results. Most fatally injured drivers, cyclists, and pedestrians do get tested. (for those that are not, there are imputed results available from a separate data file, see below).

Note that the ALC_RES field, the numerical result, has changed over the years, before 2015, it was listed as the number of hundredths of a percent BAC, e.g. 0.12 was coded as 12. In 2015 and later, it is coded as thousandths of a percent BAC, so the same result would be listed at 120.  The logic for this is encapsulated in the file 20xx_person.sql in the synthetic value sALC_RES: intox / not intox.

FARS and Drug Testing

The coding for drug results in FARS is similar to the alcohol scheme, except there are no quantitative results, only positive/negative. Also there is no equivalent to the imputation of results for drugs.

FARS coding: positive results for drugs shows up in the field DSTATUS=2 (i.e. “test given”) and DRUGRES1, 2, or 3 have a number up to 999; all in the person table. 0 meand test not given; 1 means No Drugs Reported/Negative. Potentially illicits are in groups generally in hundreds, e.g. 100-295 are narcotics, 300’s are depressants, 600’s cannaboids.  Anything 996 or above are various meanings for unknown.

Examples: Zolpidem (Ambien) is 375. See pages 579-594 of the FARS Coding and Validation Manual.

FARS and Imputation of Alcohol Results

Driving while intoxicated has been recognized as a significant serious safety factor for decades; at the same time, it’s long been recognized that many involved in fatal traffic collisions (mostly drivers but sometimes peds and bicyclists) do not have recorded alcohol test results. This nhtsa report published in 2002 explains most of the deep background and terminology on the scheme to “fill in” missing results:

DOT HS 809 403 Transitioning to Multiple Imputation – A New Method to Impute Missing Blood Alcohol Concentration (BAC) values in FARS

  • In all cases “Estimates are generated only for drivers and nonoccupants (pedestrians, pedalcyclists) for whom alcohol test results were not reported”.
  • The method established in 1986 is referred to as the “linear discriminant model that estimates the probability that a driver or nonoccupant has a BAC in grams per deciliter (g/dl) of 0, .01 to .09 or .10 and greater”; that is to say: the result is one of three discrete choices. (note in that era, legally intoxicated was .10+). Since it produced one result, it was sometimes called single imputation.
  • The new (and still current, as of 2015) method as of 2000 is referred to as multiple imputation. It does not use the arbitrary discrete groups of the discriminant model; rather it is a set of ten imputed BAC values that vary all across the “plausible range” of BACs.
  • BAC values in FARS (and the imputed values) are truncated (not rounded) to two decimal places; e.g. a 0.009 is simply zero, not rounded to 0.01. There are three groups of BAC:
  • 0 (i.e. zero) Sometimes referred to as negative for alcohol
  • .01+ Sometimes referred to as positive for alcohol. If any driver or nonoccupant involved in the collision, it is said to be alcohol related.
  • .08+ (or .10+ depending on timeframe) Usually referred to as legally intoxicated. Note this is a subset of .01+. This is used for breakdowns such as Table 2 which lists percentages of those killed by intoxication of any driver and not-intoxicated drivers, passengers, intoxicated nonoccupants, and not-intoxicated nonoccupants.
  • It helps to remember that in nhtsa-speak, the term nonoccupants means anyone not in a vehicle; i.e. Pedestrians, and Pedalcyclists (bicyclists). I’ve been meaning to check: what are occupants of a non-MV classified as, e.g. a horse-drawn wagon? It also seems to me sometimes motorcyclists are not counted as occupants; but other times are.

Eventually I am going to elaborate here on the fields in FARS data related to impairment (e.g. DR_DRINKING, and DRINKING; which heavily “undercount” drinking because of incomplete drug/alcohol testing). And eventually hope to understand how the multiple imputation files, which are in addition to the normal FARS tables, fit in.

This very handy (and very huge, it’s over 500 pages) 2004 report explains a lot of the FARS alcohol results and reporting procedures. For specific states’ laws, refer to the second document because it is regularly updated with current laws. These documents have almost no discussion of nonoccupants — there are a few references to pedestrians, and none for bicyclist (or pedalcyclist). The

DOT HS 809 756
State Laws and Practices for BAC Testing and Reporting Drivers Involved in Fatal Crashes

Digest of State Alcohol Highway Safety-Related Legislation

In Arizona, §28-668 requires testing and reporting of deceased drivers if probable cause exists that they were DUI. This particular law has no effect on surviving drivers — though other procedures apply to them, such as implied consent §28-673 which also has specific provisions for required testing of surviving drivers in a fatal or serious injury collision, where they are believed to have caused the crash. There is no sort of mandatory testing on passengers, pedestrian, or bicyclists [implied consent applies specifically to motor vehicle drivers].

 

The Johns Hopkins study

This now somewhat dated study was done by researchers from Johns Hopkins University and the Univ of Maryland Medical Center based on bicyclist injuries and fatalities occurring in Maryland from 1985-1997 (thirteen years). This has come up from time to time and I wanted to get the details down on how the study was performed and how it plays with FARS; for the purposes of this article, only the fatal cases with be examined.

JAMA. 2001 Feb 21;285(7):893-6.
Use of alcohol as a risk factor for bicycling injury.
Li G1, Baker SP, Smialek JE, Soderstrom CA.

The design is known as a matched case-control study. And while interesting, for my purposes I was interested in the raw data, the “cases”, and only the fatal cases.

It wasn’t completely clear to me what the universe of cases was (was it all fatalities in Maryland in that time? It may well be, buzzing through some mid-90’s TSF there were indeed ~ 10 per year then. OCME is the Office of the Chief Medical Examiner of Maryland), they state:

During 1985 through 1997, the OCME recorded a total of 133 bicyclist fatalities. Excluded from the study were 40 bicyclists who died at age 14 or younger, 42 who were injured at night, and 12 for whom valid estimated BAC information was unavailable… At OCME, alcohol testing is routinely performed using the head-space gas chromatography method.

So 133 – 40 – 42 – 12 = 39. Furthermore they ultimately excluded 5 more becasue they could for whatever reason not get matching controls. So 34 cases; which were all >14 years old, had labratory measured BAC results, all of whom were killed during the “day” (which they defined as from 5am to 9pm).

The result is “estimated BAC (>=0.02 g/dL) was detected in 23.5% of the 34 fatally injured“. I.e. they made the arbitrary cutoff of testing positive for alcohol at 0.02 (NHTSA tends to put it at 0.01). And by the way, it’s probably not the case that there were a lot of relatively low BACs; “for those who tested positive for alcohol, the mean estimated BAC was 0.18”.

In other words, 21.5% (8 of the 34) had at least some alcohol. They didn’t publish how many were above any other particular threshold; i.e. how many were legally intoxicated. This result is broadly consistent with national results published in the Multiple Imputation whitepaper, referenced above, see table 12 — the percent of pedalcyclists .10+ from the time period 1985-2000 (the last year published in that paper) varied from 11% at the beginning of the period to 23%. I say it’s broadly consistent because Table 12 lists fatally injured pedalcyclists of all ages, and the number of children, who are highly unlikely to be intoxicated, fatally injured had dropped considerably over that time period.

A note on the phrase “estimated BAC”, they say that few times — I don’t know why they phrase it like that; they were very clear that these were in fact measured; i.e. they are not imputed or unknown.

The Controls

The way they did the controls was sort of interesting / sort of kooky; they had research assistants return to the specific spot where the injury/fatality occurred on the same day of week, same month, and same time of day. (Thus the reason for excluding “night” hours was they felt it wouldn’t be safe for the field assistants).

The research assistant would then flag down any bicyclist and ask them if they would volunteer to answer questions, and submit to a breathalyzer (anonymously). They claim 97% of flagged down cyclist did so(!).

5 thoughts on “Use of alcohol as a risk factor for bicycling injury”

  1. This report, DOT HS 809 756:
    State Laws and Practices for BAC Testing and Reporting Drivers Involved in Fatal Crashes
    http://www.nhtsa.gov/people/injury/alcohol/BAC-Testing/
    Has some interesting stuff; discusses pedestrians somewhat but doesn’t even mention bicyclist (or pedalcyclist).

    The pro-drink-driving (yes there are such groups) tend to see imputation as just another trick to make DUI fatalities stats sound larger than they really are. see, e.g. http://www.drunkard.com/issues/08_02/08_02_fighting_madd.htm

    In FARS Coding Manual, Person Level P16, 17, 18 and Nonmotorist Level NM15, 16, 17 are the various police-reported alcohol stuff; there are also variable for drugs, and reporting for up to 3 sets of drug results.

    In the “accident” (incident) data set: DRUNK_DR 00-99 Number of Drunk Drivers Involved in the Fatal Crash. (DR_DRINK is similar in the individual vehicle table; and DRINKING in the person table).
    “This is a derived variable. Data from the Vehicle file are analyzed and if there is sufficient information to conclude that a driver was drunk, i.e., if the blood alcohol concentration (BAC) is positive, or if the police reported alcohol involvement, then the driver is counted as a drunk driver. A driver being charged with an alcohol violation by itself does not have the driver counted as a drunk driver. Note that alcohol data is often missing. For that reason this variable may undercount the actual number of drunk drivers. For detailed analysis of alcohol involvement, the alcohol files should be used”
    Alcohol files are, presumably, the MI files that reside in a separate download and contain three tables: miacc (i.e. one row per incident), midracc (not sure how this varies from miacc) and miper (one row per person; or at least one row per vehicle… it has a field for veh_no and per_no; but per_no is almost always 1?? i.e. do nonmotorists get in here at all? or maybe that’s because they don’t estimate for people who have actual test results.).

  2. most deceased cyclists (26 of 30 in AZ 2013) do get alc tested, and typically few are intoxicated, e.g. 1 (of the 26 tested) was 0.28 — the only other positive result was one at 0.03. Quick check of recent prior years 2010-2012 show 3 intoxicated in each of those years.
    Made the synthetic value sALC_RES — can do a straightforward (no joins) like so:

    SELECT sALC_RES,ALC_RES,count(*) FROM 2013_person WHERE eINJ_SEV LIKE ('Fatal%') AND ePER_TYP IN ('Bicyclist', 'Other Cyclist') AND STATE=4 GROUP BY 1,2;

    Here’s a more involved query that also joins in the driver’s results and also adds in the YEAR join so can use the ‘ignore’ option to search across mulitple years (since fars does not use unique case numbers from year-to-year):

    SELECT 
    i.ST_CASE, i.MONTH month, i.DAY day, p_bike.eATST_TYP bTst_Typ, p_bike.sALC_RES bAlc_Res, p_bike.AGE bAge, 
    p_car.eATST_TYP cTst_Typ, p_car.sALC_RES cAlc_Res, p_car.AGE cAge
    FROM ((( 2013_incident as i 
    JOIN 2013_person AS p_bike ON (i.ST_CASE = p_bike.ST_CASE AND i.YEAR = p_bike.YEAR AND p_bike.eINJ_SEV LIKE ('Fatal%') AND p_bike.ePER_TYP IN ('Bicyclist', 'Other Cyclist') )) 
    JOIN 2013_vehicle as vehicle ON (i.ST_CASE = vehicle.ST_CASE AND i.YEAR = vehicle.YEAR AND vehicle.VEH_NO=1 ) ) 
    JOIN 2013_person as p_car ON (i.ST_CASE = p_car.ST_CASE AND i.YEAR = p_car.YEAR AND p_car.ePER_TYP LIKE ('Driver%') AND p_car.VEH_NO=1 ) )
    WHERE i.STATE=4;

    Similar to above, joins the driver and also joins PBType table (valid 2010,11, and 13 only), also notably, it add the YEAR condition to the joins so can use the “ignore year” option:

    SELECT 
    i.ST_CASE, i.YEAR, i.MONTH month, i.DAY day, eBIKECTYPE, eBIKEDIR, eBIKEPOS, eBIKELOC, eBIKECGP, p_bike.eATST_TYP bTst_Typ, p_bike.sALC_RES bAlc_Res, p_bike.AGE bAge, 
    p_car.eATST_TYP cTst_Typ, p_car.sALC_RES cAlc_Res, p_car.AGE cAge
    FROM (((( 2013_incident as i 
    JOIN 2013_person AS p_bike ON (i.ST_CASE = p_bike.ST_CASE AND i.YEAR = p_bike.YEAR AND p_bike.eINJ_SEV LIKE ('Fatal%') AND p_bike.ePER_TYP IN ('Bicyclist', 'Other Cyclist') )) 
    JOIN 2013_vehicle as vehicle ON (i.ST_CASE = vehicle.ST_CASE AND i.YEAR = vehicle.YEAR AND vehicle.VEH_NO=1 ) ) 
    JOIN 2013_person as p_car ON (i.ST_CASE = p_car.ST_CASE AND i.YEAR = p_car.YEAR AND p_car.ePER_TYP LIKE ('Driver%') AND p_car.VEH_NO=1 ) )
    JOIN 2013_PBType as pbcat ON (i.ST_CASE = pbcat.ST_CASE AND i.YEAR = pbcat.YEAR ) ) 
    WHERE i.STATE=4 GROUP BY 1 ORDER BY i.YEAR, i.MONTH, i.DAY ;

    Made the synthetic value sALC_RES — can do a straightforward (no joins) like so:

    SELECT sALC_RES,ALC_RES,count(*) FROM 2013_person WHERE eINJ_SEV LIKE ('Fatal%') AND ePER_TYP IN ('Bicyclist', 'Other Cyclist') AND STATE=4 GROUP BY 1,2;
  3. here’s a quickie query to list out the drug results (STATUS of 2 means tested / result of 1 means negative).
    Similar to alchol query above but removed the join to PBCAT so would work to ignore year from 2010-2013:

    SELECT 
    i.ST_CASE, i.YEAR, i.MONTH month, i.DAY day, p_bike.eATST_TYP bTst_Typ, p_bike.sALC_RES bAlc_Res, 
    p_bike.DSTATUS , p_bike.DRUGRES1, p_bike.DRUGRES2,p_bike.DRUGRES3,
    p_bike.AGE bAge, 
    p_car.eATST_TYP cTst_Typ, p_car.sALC_RES cAlc_Res, 
    p_car.DSTATUS , p_car.DRUGRES1, p_car.DRUGRES2,p_car.DRUGRES3, 
    p_car.AGE cAge
    FROM ((( 2013_incident as i 
    JOIN 2013_person AS p_bike ON (i.ST_CASE = p_bike.ST_CASE AND i.YEAR = p_bike.YEAR AND p_bike.eINJ_SEV LIKE ('Fatal%') AND p_bike.ePER_TYP IN ('Bicyclist', 'Other Cyclist') )) 
    JOIN 2013_vehicle as vehicle ON (i.ST_CASE = vehicle.ST_CASE AND i.YEAR = vehicle.YEAR AND vehicle.VEH_NO=1 ) ) 
    JOIN 2013_person as p_car ON (i.ST_CASE = p_car.ST_CASE AND i.YEAR = p_car.YEAR AND p_car.ePER_TYP LIKE ('Driver%') AND p_car.VEH_NO=1 ) )
    WHERE i.STATE=4 GROUP BY 1 ORDER BY i.YEAR, i.MONTH, i.DAY ;

Comments are closed.