Marshall SW, Runyan CW, Bangdiwala SI, Linzer MA, Sacks JJ, Butts JD. Fatal Residential FiresWho Dies and Who Survives?. JAMA. 1998;279(20):1633-1637. doi:10.1001/jama.279.20.1633
From the Injury Prevention Research Center (Mr Marshall, Drs Runyan and Bangdiwala, and Ms Linzer) and Departments of Epidemiology (Mr Marshall and Dr Runyan), Health Behavior and Health Education (Dr Runyan), Biostatistics (Dr Bangdiwala), and Pathology and Laboratory Medicine (Dr Butts), University of North Carolina, Chapel Hill; US Public Health Service, Centers for Disease Control and Prevention, Atlanta, Ga (Dr Sacks); and North Carolina Office of the Chief Medical Examiner, Chapel Hill (Dr Butts).
Context.— The United States has one of the highest fire fatality rates in the
developed world, and three quarters of these deaths are in residential fires.
Objective.— To compare characteristics of those who die and those who survive in
the same residential fire.
Design.— Data on fatal residential fires were collected from the medical examiner
and interviews with local fire officials.
Setting.— North Carolina.
Subjects.— Persons in residential fires with at least 1 fatality in a 1-year period.
Main Outcome Measure.— Dying vs surviving a fatal residential fire that occurred with more
than 1 person at home.
Results.— Of the 190 decedents, 124 (65%) were male, 78 (41%) were home alone,
and 69 (53%) of 130 adults who had blood alcohol measured were intoxicated
(blood alcohol content >22 mmol/L [100 mg/dL]). Of the 254 persons present
during fires in which more than 1 person was at home, 112 died. Individuals
more likely to die (high-vulnerability group) were younger than 5 years or
64 years or older, had a physical or cognitive disability, or were impaired
by alcohol or other drugs (risk of death for group, odds ratio [OR], 4.01;
95% confidence interval [CI], 2.29-7.03). The presence of an adult with no
physical or cognitive disabilities who was unimpaired by alcohol or other
drugs (a potential rescuer) reduced the risk of death in the high-vulnerability
group (OR, 0.49; 95% CI, 0.24-0.99) but not the low-vulnerability group. Overall,
a functioning smoke detector lowered the risk of death (OR, 0.39; 95% CI,
Conclusions.— Smoke detectors were equally effective in both low- and high-vulnerability
populations. The high-vulnerability group was more likely to survive if, in
addition to a smoke detector, a potential rescuer was present. Further research
should seek to identify prompts that facilitate speedy egress from a burning
structure and that can be incorporated into residential fire alarm systems.
THE UNITED STATES has one of the highest fire fatality rates in the
developed world, accounting for 2.3 deaths per 100000 population.1,2 In 1996 there were an estimated 417000
residential fires in the United States, resulting in 4035 deaths, almost 19000
injuries, and nearly $5 billion in property loss.3
There are striking regional differences in fire fatality within the United
States, with the southeastern states experiencing the highest rates.1,2,4 Most fatalities (75%)
are associated with residential fires.2 At
high risk of fire death are the very young, older adults,2,4- 6
and those impaired by alcohol.6,7
These groups are more vulnerable to fire fatality because they lack the capacity
to take "mature, independent escape action."8
We previously reported a case-control analysis of environmental and
household risk factors for fatal residential fires.9
That report compared fires with 1 or more deaths with fires with no deaths.
This study focuses solely on the fatal fires. We delineated the personal characteristics
that distinguish those who die in from those who survive fires severe enough
to result in at least 1 death.
This study included all fatal residential fires that occurred in North
Carolina between February 1, 1988, and January 31, 1989. Fatalities from residential
fires were identified at the North Carolina Office of the Chief Medical Examiner.
North Carolina law requires that all injury deaths be investigated and certified
by a county medical examiner, who must file a report with the chief medical
examiner's office. Only those deaths occurring within 30 days from injuries
received in a fire in a residential dwelling were included.
A residence was defined as a private dwelling and adjoining structures.
This included the garage, porch, and deck. Deaths of firefighters, fires determined
to be arson, fires that started when no one was home, and fires in institutions
and military housing were excluded. Fires in residential apartments and those
started in a chimney were included.
Medical examiner files include death certificates, narrative reports
from the local medical examiner, toxicologic investigation results, autopsy
reports, and newspaper clippings. The medical examiner files were abstracted
weekly, and the North Carolina chief medical examiner, a forensic pathologist,
reviewed each death and excluded any considered not due to the fire. The local
fire official in charge of responding to each fire was interviewed by telephone
as soon as possible after the fire. Information was collected on the nature
of the fire, the dwelling that burned, the fire department response, and the
characteristics of all persons present in the residence at the time of the
fire. The officials were permitted to consult standard trip reports in answering
our questions. When fire officials could not provide complete information,
we used data abstracted from the medical examiner files or from interviews
with alternate informants, such as fire survivors.
Data were obtained on all fatal residential fires during the study period
except 1 that occurred in a suspected crack house and involved 1 fatality
and 10 survivors. Fire department officials were unable to provide further
details about the occupants of this residence.
The main outcome measure was the fatality rate, defined as the number
of deaths divided by the total number of persons present in the dwelling when
the fire started. Bivariate analyses of the fatality rate were conducted and
logistic regression was used to model the odds of fatal injury while adjusting
for a range of covariates, including household, fire, and fire-response variables.
Bivariate and regression analyses were restricted to fires with multiple persons
home. SUDAAN software was used to adjust the SEs of the logistic regression
coefficients for the clustering effect due to multiple persons being in the
The initial regression model included the main study variables: age,
sex, race, physical or cognitive disability, and alcohol or other drug impairment.
The following household, fire, and fire-response variables were then added:
presence of a working smoke detector, whether the residence was a mobile home,
number of exits from the residence (≤2 vs >2), number of stories (1 vs
>1), age of residence (≥20 vs <20 years), distance to nearest neighbor
(≥ 90 vs <90 m), rental vs nonrental residence, mode of fire report
(telephone vs other), presence of a 911 system in the area, type of fire department
(all volunteer, some volunteer, or all paid), daytime fire (6:00 AM-9:59 PM)
vs nighttime fire (10:00 PM-5:59 AM), weekend (6:00 PM Friday-5:59 AM Monday)
vs weekday fire, ignition source (heating, smoking, or other), the season
in which the fire occurred, and the room of origin of the fire (living room,
Variables representing whether the subject was in the same room as the
fire when the fire started, the number of other persons vulnerable to fire
fatality present in the household at the time of the fire (defined as being
<5 or >64 years old, having a physical or cognitive disability, or being
impaired by alcohol or other drugs), and the number of other persons present
who were potential rescuers (defined as aged 18-64 years with no physical
or cognitive disability and unimpaired by alcohol or other drugs) were also
included. The final step was to remove those variables whose exclusion had
minimal effect on the regression coefficients of the main study variables.
The resulting model provided an adequate fit to the data (lack-of-fit statistic,
7.37; df, 8; P=.50).11 Interactions in the data were assessed on the basis
of departure from additivity of effects.12,13
During the 12-month study period, 155 fatal fires occurred in private
residences with 332 persons present at these residences when the fire started.
One hundred ninety (57%) of these persons died. Fire department officials
reported that 60 (18%) were injured and/or hospitalized, and 82 (25%) sustained
no reported injuries. These figures represent a rate of 6.6 fires per 100000
North Carolina households and a mortality rate of 3.0 per 100000 population.
In 50% (78/155) of the fires only 1 person was home. These fires, accounting
for 41% (78/190) of all deaths, are referred to as "one person home alone"
(OPHA) fires (Figure 1). Fires in
which more than 1 person was home at the time of the fire are referred to
as "multiple persons home" (MPH) fires. Forty-four percent (112/254) of the
subjects in the MPH fires died, and 42% (60/142) of the MPH survivors sustained
a nonfatal injury. In 8% (12/155) of residential fires, more than 1 person
was home and all occupants perished.
A smoke detector was present in 31 households (25%). Data on the presence
or absence of a smoke detector could not be obtained for 32 households. Of
the 31 households with a smoke detector installed, only 16 had detectors that
were operational (8 of these were in OPHA fires). Eleven detectors were confirmed
as nonoperational, and in 4 cases functional status was unknown.
Sixty-five percent of the 190 decedents were male and 51% were white.
The mean age was 39 years, with 17% of the decedents younger than 5 years
(n=32) and 21% older than 64 years (n=39). The OPHA and MPH decedents were
different in 2 important respects: the decedents in the OPHA fires were more
likely to be adult (99%) and male (81%) than those who died in MPH fires (58%
and 54%, respectively).
The most common mechanism of death was carbon monoxide poisoning (65%),
followed by inhalation of smoke or gases other than, or in addition to, carbon
monoxide (16%). Most deaths (93%) occurred within 24 hours of the fire. Of
the 13 deaths that occurred more than 24 hours after the fire, 11 were the
result of either burns or complications from burns.
In 153 cases, fire department officials identified what the decedent
was doing at the time of injury (81%). Of these, 47% were attempting to escape
and 31% were sleeping. All the decedents were inside the dwelling when the
fire started, and 45% were in the same room as the fire. Six decedents escaped
from the building during the fire but were fatally injured after they reentered
For those 18 years and older, the medical examiner determined the blood
alcohol content (BAC) in 92% of decedents. In more than half the cases (53%
[69/130]), the BAC exceeded 22 mmol/L (100 mg/dL). The investigating medical
examiner reported that 37 decedents had a history of alcoholism.
In the judgment of firefighters at the scene, 16% (30) of the decedents
had some form of preexisting physical or cognitive impairment, such as a hearing
or vision impairment, mental retardation, a disabling illness, or senility.
Of these, 13 were identified as bedridden or otherwise physically disabled.
Of the 48 juvenile fatalities (<18 years), 14 died in 9 fires without
adult supervision. Seven died in 4 fires in which 1 or more of the surviving
adults had been judged as being impaired by alcohol or other drugs.
Because OPHA fires, by definition, include only decedents, the comparison
of decedents and survivors was restricted to the 254 people in MPH fires for
both the bivariate and regression analyses. Persons having a physical or cognitive
disability, and those reportedly impaired by alcohol or other drugs, experienced
elevated rates of fatality (Table 1).
Age was also an important predictor, with the old (>64 years) and the young
(<5 years) being the most likely to die as a result of the fire. Based
on these and previous results,9 a "highly vulnerable"
group was defined, comprising all subjects younger than 5 years, older than
64 years, having a physical or cognitive disability, or being impaired by
alcohol or other drugs. The presence of 1 or more of these risk factors raised
the fatality rate from 31% to 54%.
Two household-level characteristics were of particular interest: the
presence of other people who might function as potential rescuers of vulnerable
persons and the presence of a smoke detector. We defined potential rescuers
as persons, other than the study subject, who were present when the fire started,
who were between the ages of 18 and 64 years, who had no physical or cognitive
disabilities, and who were not impaired by alcohol or other drugs at the time
of the fire. When a potential rescuer was present, the fatality rate dropped
from 49% to 39%. The relatively small number of MPH decedents (n=25) in fires
with a functional smoke detector experienced a lower fatality rate (36%) than
those without a smoke detector (44%).
Logistic regression was used to identify important risk factors for
death while controlling for a range of covariates. Thirty-one persons involved
in fires were not considered for the model because the severity of the fire
made it difficult for fire officials to determine whether a smoke detector
was present or, if present in the residence, whether it was functional, leaving
223 subjects for analysis. The coefficients from the logistic regression models
represent log odds ratios (ORs) for dying in vs surviving a residential fire
in which at least 1 death occurred. Odds ratios above 1 indicate that the
risk of death is increased in the presence of that factor, while ORs below
1 indicate a protective effect.
Regression results confirmed the relationships observed in the bivariate
analyses (Table 1). Following
covariate adjustment, persons meeting the "high-vulnerability" definition
were 4 times as likely to suffer a fatal injury. Those younger than 5 years
were 6 times more likely than persons between the ages of 18 and 64 years
to suffer a fatal injury.
The presence of 1 or more potential rescuers in the residence tended
to be protective of fatality, with an unadjusted OR of 0.56 (95% confidence
interval [CI], 0.32-0.97) and an adjusted OR of 0.61 (95% CI, 0.31-1.21).
The presence of a functioning smoke detector also appears to lower the risk
of death. The unadjusted OR was 0.72 (95% CI, 0.35-1.45) and the adjusted
OR was 0.39 (95% CI, 0.18-0.83). The relatively small number of subjects in
households with functional smoke detectors means that these estimates have
to be interpreted with some degree of caution.
The interaction between vulnerability and potential rescuer was investigated
(Table 2). For those classified
as less vulnerable, the presence of a potential rescuer reduced the risk of
death by a modest 19% (adjusted OR, 0.81; 95% CI, 0.28-2.31). The risk of
death for the vulnerable group was reduced by 51% (adjusted OR, 0.49 [2.36/4.86];
95% CI, 0.24-0.99).
The protective effect of having a smoke detector was similar for both
the less vulnerable (adjusted OR, 0.30; 95% CI, 0.05-1.71) and the vulnerable
groups (adjusted OR, 0.33 [1.32/3.98]; 95% CI, 0.13-0.85). There was little
or no evidence of interaction between vulnerability and smoke detector presence
Finally, in examining the 3-way interaction among smoke detector, potential
rescuer, and vulnerability, it was found that the subjects with the greatest
risk of death were those in the vulnerable group with neither a potential
rescuer nor a smoke detector (Table 3).
The subjects with the least risk, on the other hand, were those in the less
vulnerable group who were in fires with both a potential rescuer and a smoke
detector present. Though the presence of either a potential rescuer or a smoke
detector during the fire appeared to lower the risk of fatality substantially,
the presence of both a potential rescuer and a smoke detector reduced it even
This study confirmed earlier evidence that readily identifiable personal
characteristics are important determinants of survival in residential fires.
Older and younger persons,3- 5,14,15
persons with disabilities, and those impaired by alcohol or other drugs6,7,16 were observed to be
at a higher risk of death in fatal MPH fires. In addition, our findings suggest
that the increased fatality risk in these populations can, to some extent,
be mitigated through the presence of an operational smoke detector and an
able-bodied adult unimpaired by alcohol or other drugs at the time of the
This study is based on a select group of individuals—those involved
in a fire in which at least 1 person died. The analyses presented here, therefore,
enable us to make inferences about who will die and who will survive a residential
fire, given that someone will die. It does not necessarily follow that the
same relationships will apply in fires in which no fatality occurs. Nevertheless,
a complementary analysis of these data, which compared fatal with nonfatal
fires, identified a similar group of risk factors (age, disability, alcohol
or other drug impairment, and absence of a smoke detector).9
A limitation of this study is the small number of households equipped
with functioning smoke detectors. This problem resulted in some wide CIs in
the interaction analyses. Furthermore, smoke detector status was not available
for some fires, resulting in the deletion of 31 subjects from our regression
analyses. Fires with missing smoke detector data were more likely to have
resulted in a high level of destruction to the building and had a higher fatality
rate (52% vs 43% in the fires with smoke detector data present). We investigated
the potential for bias associated with missing data. If all the missing data
came from households that, in reality, had smoke detectors that operated correctly,
then the adjusted OR for smoke detectors would be 0.32 (95% CI, 0.15-0.68).
If the other extreme were true, and all the missing data represented households
without functional smoke detectors, then the adjusted OR would be 0.39 (95%
CI, 0.18-0.83). Thus, it is unlikely that a missing data bias resulted in
substantial overestimation or underestimation of the effectiveness of the
smoke detectors. Point and interval estimates for our main study variables
under each of these 2 scenarios were similar to the results presented in this
Because smoke detector status is a household-level rather than an individual-level
characteristic, a possible association between household size and smoke detector
status could have biased the results. An analysis with weights inversely proportional
to household size yielded smoke detector estimates that were similar to those
from the unweighted analysis reported above (unadjusted OR, 0.71; 95% CI,
0.33-1.50; adjusted OR, 0.33; 95% CI, 0.15-0.74), indicating that this was
an unlikely source of bias.
The vast majority of the data used in this study was obtained by interviewing
fire officials who attended the fires. For example, of the 4460 data points
entered into the regression analyses (20 variables on 223 subjects), 98% were
obtained from the fire department interview. The first priority of the fire
department, however, is the evacuation of endangered persons and containment
of the fire, not the assessment of the personal characteristics of the decedents
and survivors. The assessment of such characteristics may be problematic in
an emergency setting and consequently they could have been underreported to
our telephone interviewers, potentially resulting in some misclassification
For this reason, the effect of misclassification of vulnerability status
was investigated in a sensitivity analysis, under the working assumption that
the other variables were measured without error. If the misclassification
was nondifferential (ie, identical for decedents and survivors), and the sensitivity
and/or specificity of vulnerability status was 80% or better, then our finding
of an OR of 4.06 may have underestimated the corrected OR, which ranged up
to 13.05. If the misclassification was differential, and the sensitivity and/or
specificity was 80% or better, the corrected OR ranged from 2.79 to 26.25.
Only in cases of severe differential misclassification did the corrected OR
approach the null (eg, a specificity of 50% for decedents and 100% for survivors,
with 90% sensitivity for both, yielded a corrected OR of 1.44). This suggests
that the findings with regard to vulnerability status are reasonably robust,
and that the most probable effect of any misclassification bias would have
been to err on the conservative side by underestimating the increased risk
in the high-vulnerability group.
Our finding that smoke detectors reduce the risk of death in a fatal
fire by about 60% contributes to a mounting body of scientific evidence for
the effectiveness of these relatively inexpensive devices.2,9,17- 19
Some authors have suggested, however, that smoke detectors and other early
warning devices may not be effective in populations that have difficulty responding
to an alarm in a timely and independent manner, such as children, older adults,
persons with disabilities, or those impaired by alcohol or other drugs.6,20 The data from this study indicate
that the reduction in fatality risk associated with the presence of a smoke
detector was similar for both the vulnerable and less vulnerable groups. This
suggests that smoke detectors may be as effective in vulnerable populations
as they are in the general population. We caution that our smoke detector
findings are based on a small number of subjects. Furthermore, fatalities
in MPH fires are a highly select group within the wider population of fire
injuries. Further research is needed to investigate whether the observed relationships
hold for nonfatal injuries and OPHA fires.
Although the presence of an operational smoke detector reduced the fatality
risk, the presence of a potential rescuer in conjunction with a smoke detector
tended to lower the risk even further. This is not surprising, in that smoke
detectors are merely an early warning device, whereas potential rescuers can
be proactive in assisting with the egress of the other occupants from burning
structures. Consideration should be given to enhancing current alarm systems,
such as smoke detectors, with additional egress prompts that might mimic,
to some extent, the effect of having a potential rescuer present. For example,
the alarm could be modified to alternate the siren noise with prerecorded
vocal messages providing exit directives, or could incorporate visual cues
such as a flashing light or directional arrows. Laboratory-based testing of
various cues and prompts under simulated fire conditions is required to identify
and test an appropriate set of egress stimuli.
Smoke detectors have been widely adopted by the public and 93% of US
households have 1 or more smoke detectors installed.19
As many as 30% to 45% of these, however, are nonoperational, largely due to
nonreplacement or removal of batteries.18,19
Direct connection of the detector to the household's electrical supply is
one mechanism for ensuring an uninterrupted power source. In addition, detectors
with extended-life (up to 10 years) batteries are now available.
Sprinkler systems also afford considerable promise for the prevention
of fire fatality,20- 22
but are significantly more expensive to install and have not been well studied
in residential settings. Tests in full-scale simulated fires and surveillance
data from the National Fire Incident Reporting System indicate, however, that
they can be expected to significantly mitigate the risk of fatality in residential
This study found that smoke detectors were effective in reducing the
risk of death in fatal fires in which multiple persons were present. In addition,
they appear to perform equally well in both low- and high-vulnerability populations.
The presence of a potential rescuer further enhanced the chance of survival
for those most at risk of fire death. Further research is needed to identify
prompts and cues that facilitate a speedy egress from the burning structure
and can be incorporated into smoke detectors and other residential fire alarm