Cummings P, Quan L. Trends in Unintentional DrowningThe Role of Alcohol and Medical Care. JAMA. 1999;281(23):2198-2202. doi:10.1001/jama.281.23.2198
Author Affiliations: Harborview Injury Prevention and Research Center (Drs Cummings and Quan), Department of Epidemiology, School of Public Health and Community Medicine (Dr Cummings), and Department of Pediatrics, School of Medicine (Dr Quan), University of Washington, Seattle; and Children's Hospital and Medical Center, Seattle, Wash (Drs Cummings and Quan).
Context During the last few decades, mortality from drowning
has decreased in the United States for unknown reasons. It has been
hypothesized that this decline may be due to decreased use of alcohol
in and around water or improved medical treatment after a submersion.
Objectives To estimate changes in unintentional mortality due to
submersion, estimate trends in drownings attributable to alcohol use,
and assess the role of medical care in these mortality trends.
Design A 21-year longitudinal study of case findings, from January
1, 1975, through December 31, 1995.
Setting and Participants All residents of King County,
Washington, who died unintentionally from submersion and 284 persons
hospitalized for submersion who survived.
Main Outcome Measures Changes in submersion-related
mortality incidence over time, proportion of this mortality that could
be attributed to alcohol use, changes over time in the case-fatality
rate of treated patients, and estimate of deaths prevented in 1995
compared with projected estimates had there been no change in incidence
Results There were 539 deaths due to drowning in King County
during 21 years. Mortality rates during this period declined by 59%
(95% confidence interval [CI], −70% to −46%). The incidence of
death attributable to alcohol use decreased by 81% (95% CI, −91% to
−57%); this could account for 51% of deaths prevented in 1995. Among
249 comatose patients who received prehospital care, 205 died; the odds
of survival decreased 40% over 21 years
(P=.40). Among 101 comatose patients who were
hospitalized, 63 died; the odds of survival decreased 29%
(P=.75). The incidence of survival of comatose
hospital patients decreased by 29% from 1975 to 1995 (95% CI, −78%
to +125%). We found no evidence that trends in medical treatment
prevented any deaths due to drowning in 1995.
Conclusions Drowning incidence in King County, Washington,
declined because of a decrease in severe submersion episodes rather
than an increase in success of medical interventions. Our data support
the theory that less use of alcohol around water prevents some deaths.
About half of the decrease was unexplained.
1995, 4350 deaths in the United States were attributed to unintentional
drowning: 1.7 per 100,000 person-years.1 Drowning
was the fourth most common mechanism of unintentional injury death in
1995.2 In the United States, mortality due to drowning has
been declining,2- 6 but the reasons for this decline are not
known. From 1971 through 1988, unintentional mortality due to
non–boat-related submersion fell by about 5.5% per year among
children aged 10 through 19 years.5 The authors who
reported these data speculated that declining mortality among older
children might be related to reduced use of alcohol while in or around
water or to reduced exposure to water due to migration of the
population to more urban settings. On the other hand, Healthy
People 2000, a report on US health objectives, raised the
possibility that while submersion-related case-fatality rates may have
decreased, submersion-related neurological damage may have become more
common.7 Other authors have suggested that aggressive care
might increase the incidence of neurologically injured
We collected information regarding submersion episodes over a 21-year
period in part of Washington State. We analyzed these data to address 3
questions: (1) What were the trends in mortality? (2) What was the
trend in deaths attributable to alcohol use? and (3) What role did
medical treatment play in mortality trends?
We sought data regarding submersion episodes in King County,
Washington, from January 1, 1975, through December 31, 1995. King
County, which includes the city of Seattle, had 1,507,319
residents in 1990.10 A data collection form with 346 fields
was used to abstract information from the records of all 19 acute care
hospitals in King County, the medical examiner, and the 2 emergency
medical services agencies. Case finding was supplemented by using the
records of all 13 acute care hospitals and the medical examiners in 2
adjacent counties, the computerized files of Washington State death
all study years, and computerized files of state
civilian hospital discharges from 1987 through 1996.
Analyses, except as noted, were limited to submersions that
occurred within King County. To estimate population-based rates, we
excluded submersions of persons who were visiting King County or
patients referred to the county for hospital care. Denominators for
rates were yearly estimates of county population by sex and age in
5-year groups (Office of Financial Management, Olympia, Wash,
unpublished data [electronic files of Washington State population,
1980-1995]).11 We used Stata statistical software for all analyses.12
We estimated the overall linear trend in mortality for the
21-year period, 1975-1995, using negative binomial
regression.13- 16 Interactions were tested using the
likelihood ratio test.16 We used predicted counts from the
regression models to estimate the number of deaths in 1995, given the
observed data, and compared this with the expected number for 1995 had
there been no change in incidence since 1975. Activity prior to
drowning was grouped into 5 categories: (1) boating, including fishing
from a boat; (2) swimming or diving; (3) occupant of a car that went
into water; (4) fell into deep water from a dock or from shore or while
wading; (5) bathing.
If within-county drownings declined, this might have occurred because
residents went outside King County to engage in water-related
activities. We therefore assessed the overall trend in submersion
deaths of King County residents in areas outside King County.
A case-control study estimated that the relative risk (RR) of drowning
was 31.8 (95% confidence interval [CI], 5.8-176) among persons with
a blood alcohol level of 21.7 mmol/L or greater, and 4.6 (95% CI,
1.6-13.1) among those with a blood alcohol level greater than 0 but
less than 21.7 mmol/L, compared with persons who had no alcohol in
their blood.17 We used these RRs and the standard formula
for attributable fraction ([RR−1]/RR),18
to estimate the proportion of deaths among the exposed that could be
attributed to alcohol: 97% of deaths were attributed to alcohol when
the blood level was 21.7 mmol/L or greater, and 78% of deaths were
attributed to alcohol when the level was between 0 and 21.7 mmol/L.
Measures of alcohol in the blood of drowning victims were obtained from
emergency department, hospital, and medical examiner records. Blood
alcohol levels may rise after death due to putrefaction, so we
classified the level as unknown if the body was not recovered within 24
hours.19 We categorized known blood alcohol levels into 3
groups: 0, greater than 0 but less than 21.7 mmol/L, and 21.7 mmol/L or
Because blood alcohol information was often missing, we used a
multiple-imputation method.20,21 Deaths were classified by
year and victim's age (15-49 years, ≥50 years), and within each
stratum of year and age the missing values for blood alcohol category
were imputed 20 times using an approximate Bayesian bootstrap
Two methods were used to examine the role of alcohol. First, we
multiplied each death by its attributable fraction to generate a count
of deaths attributable to alcohol in each year. Then we used negative
binomial regression to estimate the trend in mortality attributable to
alcohol. Second, we used logistic regression to estimate any change in
the proportion of deaths that could be attributed to alcohol. Deaths
that involved alcohol were assigned to alcohol or not according to the
attributable fractions; for example, if the victim's blood alcohol
level was 21.7 mmol/L or greater, we counted this as 0.97 deaths due to
alcohol and 0.03 deaths unrelated to alcohol.
We conducted these regression analyses in each imputed data set and
averaged the log of the 20 rate ratios or odds ratios to calculate a
point estimate. Variances, confidence limits, and significance levels
were calculated using methods described by Rubin.22
For this part of the analysis, we included all persons who died, plus
284 additional persons who survived after prehospital, emergency
department, or hospital care. We excluded 2 persons who were released
to home after prehospital care and 60 who were released to home from an
emergency department. None of these 62 persons were later hospitalized
and none died.
Using published definitions,23 we classified patient
outcomes as death, vegetative state, severe disability, moderate
disability, mild disability, or normal. Presubmersion function was
classified using the same scale; survival with worse function after
submersion was defined as a new neurological injury. For our analyses,
we used 3 outcomes: death, survival with new neurological injury, and
survival with no new neurological injury.
Medical care could improve survival in 2 ways. First, if medical
treatment improved, outcomes might improve among those who received
care. We estimated the linear change in the odds of survival, compared
with death, using logistic regression. Previous studies have reported
that after submersion, death is likely only among comatose
victims.24- 26 Therefore, to control for the severity of
treated cases, we restricted this analysis to comatose victims.
Potential confounders examined were age (categorized as <5 years,
5-14 years, 15-49 years, or ≥50 years), sex, category of submersion
site (open water, pool, or bathtub), pupillary response (none,
reactive), and whether or not cardiopulmonary resuscitation was
Second, even if treatment outcomes did not improve, overall mortality
might decrease if, over time, comatose submersion victims were more
likely to receive medical care.
Therefore, we estimated trends in the incidence rate of comatose
patients who were treated. To assess the combined effects of any change
in access to care
and any change in outcome among those who
received care, we estimated the change in the incidence rate of
patients who were initially comatose and survived. We used 2
definitions of medical care: (1) prehospital care and (2) hospital
admission. Mental status was classified when the patient was first seen
by ambulance personnel, or when first admitted, respectively.
In addition, we estimated the change in incidence of survival with new
We identified 539 deaths (1.86/100,000
person-years) of King County residents due to unintentional submersion
in King County. Most deaths occurred at the scene, without any medical
treatment; only 12% occurred during hospitalization.
Mortality due to submersion declined by 59% during the
21-year period (Table 1). In 1995, an estimated 27.8
deaths were prevented by this decline in mortality, compared with what
would have been expected had mortality remained unchanged since 1975.
The incidence of death at the scene, without any medical care, declined
by 77%; this subgroup accounted for 91% of the prevented deaths in
Categorized by age, 9% of drownings involved children
aged 0 to 4 years (rate, 2.6/100,000 person-years), 9% of
victims were aged 5 to 14 years (rate, 1.2), 24% were aged 15 to 24
years (rate, 2.9), 33% were aged 25 to 49 years (rate, 1.5), and
24% were 50 years of age or older (rate, 1.9). Mortality declined in
every age group; the changes ranged from a 47% decrease among persons
aged 50 years or older to an 84% decrease among children 5 to 14 years
old. A test that the change in incidence differed by age group was not
Mortality due to drowning in open water decreased by 65%, deaths in
pools fell by 48%, and deaths in a bathtub decreased by 9% (Table 1). Boat-related drowning declined by 74%. Drowning while swimming
decreased by nearly two-thirds; the decrease was about half for
occupants of cars and for those who fell into deep water. There was
little change in the incidence of drowning while bathing.
There were 328 additional submersion deaths of a King County resident
outside King County; 60% occurred within Washington State. The
incidence of out-of-county fatal submersion decreased by 66% (95% CI,
−78% to −46%).
Among fatal cases with blood alcohol measurements, none younger than 15
years had alcohol in their blood. Of the 440 persons aged 15 years or
older who died, 10.7% had unknown alcohol values because the blood
sample was drawn more than 24 hours after death, and 20.2% had unknown
values for other reasons. There were 304 deaths with known values;
61.5% had no alcohol, 8.6% had values between 0 and 21.7 mmol/L, and
29.9% had values of 21.7 mmol/L or greater (Table 2).
The incidence of death attributable to alcohol decreased 81% (95%
CI,−91% to−57%) for persons aged 15
years or older. About 14.2 deaths were prevented in 1995 by this
change, 51% of all prevented deaths. The proportion of deaths of
persons aged 15 years or older that were attributable to alcohol
decreased over time, from 50% in 1975 to 22% in 1995
(P=.005 for a test for trend).
A total of 468 persons received emergency prehospital care. The
proportion who died varied little over time (Table 3). Initially, 249 patients were comatose
and 205 deaths occurred in this group (initial mental status was not
known for 11 who died). The odds of survival was 40% less in 1995
compared with the odds of survival in 1975
(P=.40 for a test for trend); this estimate
was adjusted for age and sex.
Among the 346 persons who were hospitalized (Table 3), 101 patients
were comatose and 63 of these comatose patients died. After adjusting
for age, sex, and pupillary response, the odds of survival was 29%
less in 1995 compared with 1975 (P=.75 for a
test for trend). No deaths occurred among patients who were alert or
lethargic at the scene or in the hospital.
The incidence rate of prehospital care of comatose patients increased
slightly from 1975 to 1995 (12%; 95% CI, −27% to +73%). The
incidence of survival of comatose patients who received prehospital
care declined by 37% (95% CI, −76% to +68%).
The incidence of hospital admission of comatose patients decreased by
1% (95% CI, −48% to +89%). The incidence of survival of hospital
patients who were comatose decreased by 29% (95% CI, −78% to
+125%). These changes did not result in any prevented deaths in 1995.
The incidence rate of survival with new neurological injury declined
47% (95% CI, −85% to +95%).
In a county in western Washington State, mortality due to
drowning decreased by more than half from 1975 through 1995. Deaths
attributable to alcohol decreased by 81%. Decreased mortality was not
explained by any improvement in prehospital or hospital treatment.
Our estimates might be incorrect if we missed a large proportion of the
victims, especially if records were missing in a systematic way over
time. Deaths were probably ascertained in a fairly complete manner.
Although computerized death certificate files may not identify all
drowning deaths,27 we used additional sources, including
medical examiner logs, hospital records, and prehospital records. It is
possible that we failed to identify some hospitalized cases, but we
sought to minimize this problem by using multiple sources, including
computerized hospital record systems, prehospital emergency medical
services records, and computerized data that the hospitals sent to
the Washington State Department of Health. Because King County includes
referral centers for western Washington, patients whose submersion
occurred in King County were unlikely to go outside the county for
care; furthermore, we searched the hospital records in 2 adjoining
We probably missed some patients who were seen by ambulance personnel
or in an emergency department and then released to home. Of the 62
people who were known to have this disposition, none was subsequently
hospitalized and none died. None of the 539 deaths were people who were
released from care and subsequently died. Patients released to home,
without hospital admission, were so mildly injured and so unlikely to
die that they were not necessary to our analysis.
For the records that we had, information was very complete for
many items: we had information regarding age, sex, and survival for all
victims; final neurological status for 99%; initial mental status for
95%; drowning site for 96%; and predrowning activity for 93%.
However, information regarding blood alcohol levels was missing for
31% of those who died and were aged 15 years or older. We used
multiple imputation to avoid the biases associated with analyzing only
records with complete information and produce statistics that formally
accounted for the missing nature of some data.21,22
Mortality due to within-county submersions declined by 59%, similar to
the 66% decline in out-of-county submersion mortality. Therefore, the
overall trend in mortality that we found cannot be explained by a
change in where King County residents drowned. However, we could not
assess the role of alcohol or medical care in drownings outside King
Our finding that about one third of persons aged 15 years or
older who drowned had alcohol in their blood was in general agreement
with several other studies.28 Alcohol might increase the
risk of drowning through many mechanisms.29 For example,
intoxicated people might be more likely to boat or swim in dangerous
circumstances, to fall into water, or be less able to get out of water
compared with sober persons.
The hypothesis that decreased use of alcohol while in, on, or close to
water contributed to the decrease in mortality was consistent with our
findings; the incidence of deaths attributable to alcohol decreased,
and the proportion of deaths that could be attributed to alcohol
decreased. However, while some information about alcohol use among
boaters and swimmers has been reported,28- 31 we are not
aware of any studies that have measured changes in person-time spent
around water while drinking alcohol.
Our analysis regarding alcohol relied on RR estimates from 1
case-control study of alcohol use and drowning.17 The
estimates from that study were roughly similar to those from studies of
drinking and death due to a traffic crash.32- 35 However,
estimates of the association between alcohol use and
drowning would be desirable.
We found no evidence that decreased mortality could be attributed to
better treatment or improved access to care. The case-fatality rate
among comatose patients showed no improvement over time, and the
incidence of survival after treatment among persons who were comatose
did not increase. Others have expressed concern that aggressive
treatment may have resulted in more survivors with new neurological
injury.7- 9 We found that survival with neurological injury
was uncommon, and the incidence of this outcome decreased in King
Migration of the population away from bodies of water has been
suggested as a possible reason for decreasing mortality due to
drowning.5 This is not likely to explain the decrease that
we found; while some migration within the county undoubtedly occurred,
bodies of water are ubiquitous in King County and much of the
population increase has been in suburban areas near lakes and rivers.
Improvements in safety may have prevented some drownings. In 1981
and 1986, King County ordinances were passed to improve the gates and
fences around outdoor pools, and in 1981, the Seattle-King County
Department of Health expanded pool inspections.36
Lifeguards have become more common at pools and public beaches and
their training has improved.36 Observed use of life vests
has increased in King County among boaters aged 15 years or older;
prevalence was 14% in 1992, 25% in 1994, and 51% in
1997.37- 39 However, this increase in life vest use came too
late to explain much of the decrease in mortality.
Drowning incidence in King County, Washington, declined chiefly
because severe submersion episodes decreased. Some of the decrease may
be explained by less use of alcohol around water. Medical care made no
important contribution to this decrease. About half of the decrease was
unexplained. Additional hypotheses include a decrease in time spent in
water-related activities, such as wading, swimming, or boating and an
increase in safe behaviors when in or near water.