Smith GS, Keyl PM, Hadley JA, Bartley CL, Foss RD, Tolbert WG, McKnight J. Drinking and Recreational Boating FatalitiesA Population-Based Case-Control Study. JAMA. 2001;286(23):2974-2980. doi:10.1001/jama.286.23.2974
Author Affiliations: Johns Hopkins Center for Injury Research and Policy, Baltimore, Md (Drs Smith, Keyl, and Hadley); Department of Emergency Medicine, Johns Hopkins School of Medicine, Baltimore (Drs Smith and Keyl); Highway Safety Research Center, University of North Carolina at Chapel Hill (Dr Foss and Messrs Bartley and Tolbert); Pacific Institute for Research and Evaluation, Rockville, Md (Dr McKnight). Mr Tolbert is now at Rho Inc, Chapel Hill, NC. Dr Smith is now also at the Center for Safety Research, Liberty Mutual Research Center for Safety and Health, Hopkinton, Mass.
Context Alcohol is increasingly recognized as a factor in many boating fatalities,
but the association between alcohol consumption and mortality among boaters
has not been well quantified.
Objectives To determine the association of alcohol use with passengers' and operators'
estimated relative risk (RR) of dying while boating.
Design, Setting, and Participants Case-control study of recreational boating deaths among persons aged
18 years or older from 1990-1998 in Maryland and North Carolina (n = 221),
compared with control interviews obtained from a multistage probability sample
of boaters in each state from 1997-1999 (n = 3943).
Main Outcome Measure Estimated RR of fatality associated with different levels of blood alcohol
concentration (BAC) among boaters.
Results Compared with the referent of a BAC of 0, the estimated RR of death
increased even with a BAC of 10 mg/dL (odds ratio [OR], 1.3; 95% confidence
interval [CI], 1.2-1.4). The OR was 52.4 (95% CI, 25.9-106.1) at a BAC of
250 mg/dL. The estimated RR associated with alcohol use was similar for passengers
and operators and did not vary by boat type or whether the boat was moving
Conclusions Drinking increases the RR of dying while boating, which becomes apparent
at low levels of BAC and increases as BAC increases. Prevention efforts targeted
only at those operating a boat are ignoring many boaters at high risk. Countermeasures
that reduce drinking by all boat occupants are therefore more likely to effectively
reduce boating fatalities.
More than 43 million people reported using a motorboat in the United
States in 1994,1 and about 800 people died
in 1998 from recreational boating.2 Alcohol
is commonly involved in drownings and other unintentional injury fatalities3- 7
and is increasingly recognized as an important factor in many boating fatalities.8,9 Data from 4 states with high testing
rates for 1980 to 1985 suggest that 51% of people involved in boating fatalities
had a blood alcohol concentration (BAC) of at least 40 mg/dL, and 30% had
a BAC higher than 100 mg/dL.4,10
Other countries such as Canada11 and Finland12 have an even higher proportion of boating fatalities
linked to alcohol use.
Alcohol use while boating affects the probability not only of ending
up in the water but also of survival once that happens. Because of this apparent
double jeopardy, alcohol use may actually be more hazardous on a boat than
in other settings, with even low BACs greatly increasing relative risk (RR).8,13,14 Although these and
suggest that alcohol increases the RR of dying while boating, this relationship
has not been well quantified.
This study sought to better define the relationship between alcohol
use and the RR of death while boating. We conducted a large population-based
case-control study of alcohol use and recreational boating fatality risk in
2 states, Maryland and North Carolina. These states include a diversity of
waterways on which recreational boating takes place. We sought to determine
the magnitude of the estimated RR of dying associated with alcohol use, adjusting
for known or potential risk factors for drowning and other boating deaths.
We also examined whether RRs were different for passengers and operators and
whether low BACs pose a significant RR.
We searched official state boating fatality records and medical examiner
files in each state to identify all recreational boating deaths classified
as "accidental" that occurred from 1990 to 1998 in Maryland and North Carolina.
Only boating deaths that occurred from April through October (n = 403 of 502
deaths) were included in the study. Boating activity declined markedly outside
these months, making control interviews prohibitively expensive and difficult.
Because of difficulty finding control subjects at night, especially in North
Carolina, boating deaths that occurred between midnight and 7:00 AM in Maryland
and between 9:00 PM and 7:00 AM in North Carolina were excluded from the study
(13.9% of eligible cases). Deaths associated with the use of sailboats, personal
watercraft (ie, jet skis), and rafts were excluded (16.1% of eligible cases).
Deaths on sailboats are rare, and personal watercraft and rafts are different
from other boat types.2,17,18
Fatality and control subjects younger than 18 years (9.7% of eligible cases)
were excluded because the parents of potential underage control subjects were
often not available to give consent. Small inland bodies of water were excluded
in Maryland, since only 3% of eligible deaths occurred in them and they were
widely dispersed. Despite the Coast Guard definition of a boating death,2 individuals who drowned while swimming from a boat
were included in our study, although some of our analyses excluded them.
Control subjects were from a stratified random sample of boats from
waterways in each state during the boating season (April through October)
from 1997 through 1999. A complex sampling design was used to ensure that
control subjects were drawn from the same locations as fatality subjects in
each state. First, the state's navigable waterways were divided according
to geographic area and type of waterway into strata that reflected cultural
and demographic differences (Table 1).
Within each stratum, areas were selected to represent locations of boating
activity. Given the large differences in the types of waterways and their
distribution in Maryland and North Carolina, sampling procedures were tailored
for each state.
North Carolina. The state was first divided into 3 geographically and culturally distinct
regions (coastal, midstate, and western) with historically different patterns
of alcohol use.
Bodies of water were categorized into ocean/bay/sound waterways, large
and medium-sized lakes, and small lakes and rivers. The ocean/bay/sound waterways
were treated as 1 stratum, large and medium-sized lakes in each region composed
3 additional strata, and all smaller lakes and rivers composed the final stratum
(a total of 5 strata). Within the ocean/bay/sound stratum,19
geographically distinct areas were identified, and 6 of these were randomly
selected. Each of the 12 largest lakes and 6 of the 10 medium-sized lakes
were randomly selected as sampling areas, with the number of medium-sized
lakes selected proportionate to the population within each region. For the
small lake and river stratum, the state was subdivided by latitude and longitude.
Of the 210 resulting subdivisions, 42 (20%) were selected, with a probability
proportionate to the population within the region. Within each selected subdivision,
2 areas, 1 small lake and 1 river, were then randomly chosen from navigable
waterways, which resulted in the selection of 30 small lakes and rivers across
the state, for a total of 54 selected areas in North Carolina.
Maryland. Recreational boating in Maryland occurs primarily in 4 bodies of water,
with the majority occurring on the Chesapeake Bay and its many river estuaries
and also on the Potomac River.
Deep Creek Lake and the Atlantic Ocean are 2 other areas where boating
is popular, but neither had any boating fatalities during the study period.
Chesapeake Bay was divided into 6 strata corresponding to the upper, middle,
and lower sections on both the western and eastern sides of the bay. The Potomac
River was divided into 3 strata, 2 below Washington, DC, and 1 nontidal part
above it. Each stratum was divided into areas that could be surveyed in 1
Teams of 2 interviewers visited designated areas (eg, a lake, river,
or region of a bay) multiple times by boat on a predetermined schedule that
included both weekdays and weekends. On each visit, interviewers moved systematically
around the water to ensure that the entire area was covered. Upon arriving
at a designated location, interviewers identified up to 6 boats nearest to
them and then used a die to randomly select 1 to interview. Only stationary
or slowly moving boats were sampled. These fell into 2 categories: those that
were anchored, moored, drifting, or berthed, and those that were arriving
at destinations in a sampling area such as a fishing area, beach, marina,
or boat ramp after being under way. In Maryland only, because of the large
size of the Chesapeake Bay, we also used shore teams to interview boaters
who were returning to a boat ramp or marina. Boats were approached in the
order in which they arrived at the shore-based sampling site.
The selected boat was approached and the operator was asked to participate
in the study. The operator was interviewed and asked to provide details on
the boat and the boat's activities in the past hour. Next, the operator and
up to 2 randomly selected passengers (≥18 years) were asked to complete
a short self-administered questionnaire that included questions on general
health and demographic characteristics. Last, the operator and the selected
passengers were asked to provide a breath sample for alcohol testing by a
handheld breathalyzer (CMI Intoxilyzer D-400R; CMI Inc, Owensboro, Ky). The
interviewer also recorded information about the boat, number of passengers,
evidence of alcohol use, apparent sobriety of the operator, and refusals.
Institutional review boards for the protection of human subjects at the Johns
Hopkins School of Public Health and the University of North Carolina School
of Public Health approved the study procedures.
When recovery of a body is delayed, decomposition can result in postmortem
alcohol production. Rather than excluding the subjects that were not recovered
within 1 or 2 days after death,19 we used a
conservative procedure based on new evidence about the time course of decomposition
to adjust those subjects' BAC levels (J.A.H. and G.S.S., unpublished data,
2001). The amount subtracted from the observed BAC started as 0 mg/dL for
cases with a submersion time of 12 hours and increased linearly to a maximum
of 40 mg/dL for bodies recovered after 96 hours in the water. Few drowning
victims produce endogenous alcohol levels as high as 40 mg/dL, even at the
longest recovery times.20,21 Cases
in which the body was recovered more than 1 week after the incident were excluded.
The increased RR of fatality associated with BAC, after adjustment for
other factors, was estimated by calculating odds ratios (ORs) using logistic
regression in Stata (StataCorp, Version 6, 2000; College Station, Tex). Effects
of the sampling design (stratification, clustering, and weighting) were accounted
for in the analysis by using the svy Stata commands.
Each area within a stratum where control boaters were sampled (eg, a lake,
a section of river, or an area of bay) was treated as a primary sampling unit
or cluster (Table 1). Because
the number of passengers in control boats ranged from 0 to 14 but a maximum
of only 2 passengers was sampled, appropriate weights were applied to provide
a valid comparison of operators and passengers. All analyses were adjusted
for the confounding effects of time of day (resulting from the sampling schedule)
by including variables for time of day in 2-hour increments. For the main
analysis, both BAC and age were treated as continuous variables. Higher-order
terms were considered for both variables. Categories of BAC were created only
to compare the results with findings from other studies.
Multiple imputations were carried out to replace missing values for
sex, race, and age (1%, 12%, and 12%, respectively, for controls and 15% for
race for subjects; Table 2). A
hot-deck procedure using the approximate Bayesian bootstrap method of Rubin
and Schenker22,23 was used. Ten
imputations were performed for each analysis. This approach assumes that within
each state (Maryland or North Carolina) and boat type, missing values for
subgroups of subjects had the same distribution as known values.
Crude analyses suggested that control operators who refused to participate
might have had higher BACs than participating operators; 5% and 2%, respectively,
were judged to be at least moderately impaired. For operators only we evaluated
the extent to which refusals to give a breath sample might have influenced
BAC RR estimates. The hot-deck method described above was used to impute the
missing BACs, assuming missing values for BACs had the same distribution as
those with known BAC within each level of the interviewer's assessment of
Of the 253 boating victims meeting inclusion criteria, 15 (6%) were
excluded from the analysis because their bodies were recovered more than 1
week after death or were recovered after an unknown length of time. Among
the 238 eligible fatality subjects, 76% were recovered within 24 hours of
death; 11%, within 25 to 48 hours; 9%, within 49 to 96 hours; and 4%, within
97 to 168 hours. Seventeen of these subjects (7.1%) were not tested for BAC.
Of the 221 subjects included in the study, 55% had a positive BAC (adjusted
for recovery time); 36% had a BAC of at least 50 mg/dL; 27%, at least 100
mg/dL; 18%, at least 150 mg/dL; 11%, at least 200 mg/dL; and 7%, at least
250 mg/dL. Most subjects had been in open motorboats at least 3 m long (69.7%),
and the largest number of them died between 6:00 and 8:00 PM (20.8%; Table 2). Subjects were predominantly male
and nonblack, less than half were operators, and most were 21 to 40 years
of age. Eligible subjects excluded because of missing BAC data did not have
different demographic factors. Eleven subjects (3.2%) died in rough water,
which precluded safely interviewing control subjects in similar conditions,
but because they had BACs similar to those of other subjects, they were kept
in the study. Passengers were more likely than operators to have a positive
BAC (68% vs 48%; P<.001) that was at least 100
mg/dL (37% vs 27%; P = .04).
The number of boats sampled from each of the 14 strata ranged from 75
to 504. Almost all (93%) of the operators of boats sampled for the control
survey agreed to participate; 87% completed the self-administered questionnaire,
and 86% provided a valid breath sample. Of those who gave a breath sample,
7.6% refused the self-administered questionnaire. The interviews yielded a
total of 4801 potential controls (2468 operators and 2333 passengers), of
whom 3943 provided a valid breath sample and were included in the analysis
(Table 2). Boating and demographic
characteristics of persons who provided a breath sample differed little from
that of those who refused, although those on open motorboats, those who were
approached earlier in the day, female subjects, and younger persons were somewhat
more likely to participate. Only 17% of participants had a positive BAC. Of
those, 7.4% had a BAC of at least 50 mg/dL; 3.4%, at least 100 mg/dL; 1.4%,
at least 150 mg/dL; 0.6%, at least 200 mg/dL; and 0.3%, at least 250 mg/dL.
These figures represent crude unweighted distributions from a stratified sample
and thus are not representative of boaters in these areas.
A greater proportion of control subjects were in motorboats at least
3 m long and were female, nonblack, and 21 to 50 years of age (Table 2). The RR of death by BAC level, compared with that of subjects
with a BAC of 0 mg/dL, was determined in analyses to be a second-order quadratic
relationship when adjusted for age, race, sex, occupant status, boat type,
location, time of day, and weekend/weekday. Age was modeled as a third-order
quadratic relationship. The ORs for dying by BAC increased most rapidly at
lower BACs, with the rate of increase leveling off at higher BACs (Figure 1). The RR of death was increased
even at a BAC of 10 mg/dL (OR = 1.3; 95% confidence interval [CI], 1.2-1.4),
increasing to an OR of 52.4 at a BAC of 250 mg/dL (95% CI, 25.9-106.1; Table 3). When only those persons meeting
the official Coast Guard definition of boating accidents were considered (ie,
when the 22 subjects [10%] who died while voluntarily swimming from a boat
and when control subjects from boats where people were swimming were excluded),
there was no significant change in the RRs of fatality (Table 3).
Additional analyses were conducted by using categories of BAC and dichotomizing
BAC at different cut points to permit comparisons with other studies (Table 4). These values have wider CIs than
estimates of RR when BAC is used as a continuous variable.
The RR associated with BAC was not significantly different between operators
and passengers, male and female subjects, black and nonblack persons, persons
of different ages, or different types of boats.
Adjusting for the potential bias resulting from control subjects who
declined to give breath samples decreased the ORs, but the differences were
not significant. Because subjective impressions of intoxication are unreliable,
we elected to present findings based on actual measurements, as has been the
practice in the few studies that have evaluated refusal bias.24- 26
The most important finding in this study is the strong positive association
of BAC with the RR of death among recreational boaters aged 18 years and older,
even at BACs less than 50 mg/dL. In addition, passenger and operator drinking
is associated with the same increased RR of death, regardless of whether the
boat is under way.
The RRs associated with alcohol use and boating fatality increase markedly
as the BAC increases, from an OR of 1.3 at a BAC of 10 mg/dL to 52 at 250
mg/dL. Our finding of increased RR at low BACs is consistent with experimental
studies that find significant impairment in many safety-related tasks at BACs
below 50 mg/dL.27- 30
Alcohol can affect boater safety in multiple ways, influencing both
the risk of ending up in the water (or crashing) and chances for survival
in the water.8,13,14,29- 31
Alcohol impairs balance and coordination, which can increase the risk of falling
overboard whether a boat is under way or not. Impaired judgment resulting
from an elevated BAC can also increase the likelihood of being in high-risk
situations, and unlike on the roadway, having a sober operator will not necessarily
protect impaired occupants. The effects of alcohol on the probability of survival
are greater than for other injury causes31
and, once a person enters the water, include an increased risk of hypothermia
and a reduced ability to keep the head above water.8,13,14,29
Thus, a simple fall overboard can prove fatal.
Although there is substantial evidence for the risk of drinking and
there is surprisingly little information about the risk of drinking and other
injuries, including those associated with boating. Besides that reported here,
the only study designed to estimate the risk of drinking for boaters was conducted
at boat ramps in California. That study had a small sample size and did not
control for several relevant factors such as region, time of day, age, sex,
or boat type.4 It found a crude 10.7-fold increased
risk of boating fatality among operators with BACs higher than 100 mg/dL,
and CIs were wide (95% CI, 4.7-68.8). In this study, we found a clear dose-response
relationship and controlled for many potential confounding variables. In addition
to elevated RR at very low BACs, we also found a much greater RR of death
at higher BACs than the California study reported. Our main analyses included
subjects swimming or diving off a boat, since swimming is a common part of
boating activities, although excluding them in accordance with Coast Guard
practice2 did not change the RR.
Alcohol use has long been a part of recreational boating; 30% to 40%
of boaters surveyed report drinking while boating.1,6,35- 38
Many of these boaters believe that they can safely drink more when at anchor
or tied up and when they are passengers rather than operators.36
Current legislation concentrates entirely on alcohol use by the boat operator
while the boat is under way, prohibiting operation of a boat while intoxicated,
as have many safety campaigns.8,9,15
Some have even promoted the use of a designated driver when boating, with
the implication that passengers can drink as much as they like as long as
the operator remains sober. Although these approaches initially appear attractive,
they ignore the reality that passengers can put themselves at risk regardless
of the operator's actions or alcohol use. Only about half the recreational
boating fatalities could be attributed to operator error.8
Most boating fatalities involve drowning; only 18% involve collisions with
other boats or objects. The majority of fatalities involve falling overboard,
and almost half (46%) of these occur when the vessel is not under way. Indeed,
our findings clearly indicate that the RR of death is similar for operators
and passengers and increases for both groups as BAC increases.
Many fatalities occur in unpowered or low-powered boats,2,8,15
and many others occur while boats are not in operation, which undermines the
assumption that boat handling by drunken operators is a primary cause of boating
fatalities. Unfortunately, since boating police rarely test surviving operators
for alcohol use, it is impossible with current data to assess the role of
impaired operators in increasing the risk of death for other boaters.
The implicit assumption of designated driver programs—that a passenger
can drink as long as the operator remains sober—is dangerous for boaters.
All persons on a boat have an increased RR of mortality if they have been
drinking, even at low BACs. These findings suggest that countermeasures directed
only at operators of moving boats are likely to have less impact on alcohol-related
boating fatalities than broader efforts to address drinking by anyone engaged
in recreational boating.
Temporal changes in drinking practice among boaters could affect alcohol
risk estimates, since fatality- and control-subject data were collected for
different years. However, throughout the study period BACs among subjects
did not change significantly over time, nor did RRs of death estimated across
cases from 1990 to 1994 and from 1995 to 1998.
Although many potentially confounding variables were taken into account,
we were unable to adjust for other variables that might affect risk, such
as the boater's swimming ability, the operator's boating skills and experience,
use of personal floatation devices, water and weather conditions, and the
condition and seaworthiness of the boat. Use of personal floatation devices
was low among control subjects (about 6.7% of adults in control boats), but
because such use was assessed only at the boat level and not for individuals,
it was impossible to include it in our analyses. However, this study was designed
to look at the total RR of death when subjects had been drinking, not to separately
examine the influence of BAC on the risk of falling in the water (or crashing)
and surviving once in the water. Personal floatation device use and swimming
ability would have a direct effect only on the latter. Finally, although we
controlled for boating exposure with the random selection of control subjects,
some groups, such as persons in boats that spent most of their time under
way, may have been underrepresented.