JSR-01406; No of Pages 12

Journal of Safety Research xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Journal of Safety Research

journal homepage: www.elsevier.com/locate/jsr

Q10 1Q1 The state of the residential fire fatality problem in Sweden:
2 Epidemiology, risk factors, and event typologies

F
Q33Q2 Anders Jonsson, a,b,⁎ Carl Bonander, a,b Finn Nilson, a,b Fredrik Huss c,d

O
4 a
Division of Risk Management, Department of Environmental and Life Sciences, Karlstad University, Karlstad, Sweden
5 b
Centre for Public Safety, Karlstad University, Sweden
6 c
Burn Center, Department of Plastic- and Maxillofacial Surgery, Uppsala University Hospital, Uppsala, Sweden

O
7 d
Department of Surgical Sciences, Plastic Surgery, Uppsala University, Sweden
8

R
9 a r t i c l e i n f o a b s t r a c t

10 Introduction: Residential fires represent the largest category of fatal fires in Sweden. The purpose of this study was 20

P
Article history:
11 Received 14 October 2016 to describe the epidemiology of fatal residential fires in Sweden and to identify clusters of events. Method: Data 21
12 Received in revised form 25 March 2017 was collected from a database that combines information on fatal fires with data from forensic examinations and 22
13 Accepted 8 June 2017 the Swedish Cause of Death-register. Mortality rates were calculated for different strata using population statis- 23
14
19
18
17
16
15
Available online xxxx
D
tics and rescue service turnout reports. Cluster analysis was performed using multiple correspondence analysis 24
with agglomerative hierarchical clustering. Results: Male sex, old age, smoking, and alcohol were identified as 25
risk factors, and the most common primary injury diagnosis was exposure to toxic gases. Compared to non- 26
E
fatal fires, fatal residential fires more often originated in the bedroom, were more often caused by smoking, 27
and were more likely to occur at night. Six clusters were identified. The first two clusters were both smoking- 28
29
T

related, but were separated into (1) fatalities that often involved elderly people, usually female, whose clothes
were ignited (17% of the sample), (2) middle-aged (45–64 years old), (often) intoxicated men, where the fire 30
usually originated in furniture (30%). Other clusters that were identified in the analysis were related to 31
C

(3) fires caused by technical fault, started in electrical installations in single houses (13%), (4) cooking appliances 32
left on (8%), (5) events with unknown cause, room and object of origin (25%), and (6) deliberately set fires (7%). 33
Conclusions: Fatal residential fires were unevenly distributed in the Swedish population. To further reduce the 34
E

incidence of fire mortality, specialized prevention efforts that focus on the different needs of each cluster are 35
required. Practical applications: Cooperation between various societal functions, e.g. rescue services, elderly 36
R

care, psychiatric clinics and other social services, with an application of both human and technological interven- 37
tions, should reduce residential fire mortality in Sweden. 38
© 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http:// 39
R

creativecommons.org/licenses/by-nc-nd/4.0/). 40
44
42
41
O

43

45 1. Introduction Fire-related mortality has been studied in depth in developed 57

C

countries with risk factors and fatality characteristics regarding residen- 58

46 Injuries and deaths caused by fire is a major public health problem, tial fires extensively described in the existing literature (e.g. Runyan, 59
47 with more than 300,000 people being killed by fires around the world, Bangdiwala, Linzer, Sacks, and Butts (1992); Barillo and Goode (1996); 60
N

48
Q5 annually (World Health Organization, WHO et al., 2004). Most deaths Leth, Gregersen, and Sabroe (1998), Marshall et al. (1998); Brennan 61
49 occur in low-income countries where the availability of preventive (1999); Warda, Tenenbein, and Moffatt (1999); McGwin, Chapman, 62
U

50 interventions, rescue services, and advanced trauma and burn care is Rousculp, Robison, and Fine (2000); Istre, McCoy, Osborn, Barnard, 63
51 limited. However, despite a steady decrease in many countries during and Bolton (2001); Duncanson, Woodward, and Reid (2002); Istre, 64
52 the last decades, fatal fires are still a problem even in high-income McCoy, Carlin, and McClain (2002); Holborn, Nolan, and Golt (2003); 65
53 countries (U.S. Fire Administration, 2011). Residential fires are by Mulvaney et al. (2009); Turner et al. (2017)). Frequent results from 66
54 far the largest category of fatal fires in Sweden, accounting for about such studies are that the very young, elderly, males, physically disabled 67
55 three-quarters of all fire-related fatalities (Jonsson, Bergqvist, & as well as those impaired by alcohol or other substances are more likely 68
56 Andersson, 2015). to die in fires. Fatal fires are often associated with smoking materials and 69
cooking, with victims often dying as a result of smoke inhalation. In 70
terms of geographical aspects, rural communities have been shown to 71
⁎ Corresponding author at: Division of Risk Management, Department of Environmental
and Life Sciences, Karlstad University, SE-651 88 Karlstad, Sweden. have the highest fire mortality rates. The poor, as well as those with 72
E-mail address: anders.jonsson@kau.se (A. Jonsson). low educational attainment, are more likely to die in residential fires 73

http://dx.doi.org/10.1016/j.jsr.2017.06.008
0022-4375/© 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 2 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx

74 than others and the absence of a smoke alarm also appears to increase variables fire size on rescue service arrival and fire spread to supple- 137
75 the risk of death in the event of a fire. ment the event-level data. On the individual-level, the following 138
76 Recent studies add to and deepen the knowledge regarding specific variables were included: gender, age at death, primary injury diagnosis, 139
77 risk groups using sub-group analyses to study heterogeneity in risk blood alcohol concentration (BAC‰) and carboxyhemoglobin (COHb%). 140
78
Q6 factors between groups. For example, Harpur, Boyce, and McConnell All variables and categories used for this study, including coding rules 141
79 (2013), Harpur, Boyce, and McConnel (2014) investigated the circum- for the variables modified compared to the original data, are found in 142
80 stances surrounding fatal residential fires amongst the very young Tables A1 and A2 in Appendix 1. 143
81 children and the elderly population, using coronial reports. Similarly, Most of the modified variables, such as season and age group, are 144
82 Graesser, Ball, and Bruck (2009) investigated risk factors for residential self-explanatory. Others require some further explanation. For instance, 145
83 fire fatality across the lifespan, exploring differences in risk characteris- two different pre-defined municipality cluster classifications, based on 146
84 tics between age groups. Bruck, Ball, and Thomas (2011) compared the different socio-demographic variables, were used in order to study 147
85 role of alcohol in residential fire deaths for different risk factors. The aggregated socio-demographic aspects of fire mortality. The first classi- 148
86 results from these studies, which make use of detailed data on both fication, based on the municipality groups defined by the Swedish 149
87 the fire event and the characteristics of the individuals involved, show Association of Local Authorities and Regions (SKL, 2011), divides the 150

F
88 promise in guiding fire mortality prevention strategies. municipalities into ten groups based on structural parameters such 151
89 In this paper, we use a database on fatal fires compiled by Jonsson as population, commuting patterns, tourism and travel industry, and 152

O
90 et al. (2015) that derives and links data from multiple sources to de- economic structure. The second classification, which is based on a 153
91 scribe the epidemiology of residential fire deaths in Sweden in terms more recent cluster analysis by Statisticon AB (Anders Sundström, 154
92 of detailed event and fatality characteristics. By doing this, previous Statisticon AB, Stockholm, Sweden, personal communication, December 155

O
93 research is improved in terms of details as the database uses several 9, 2015), creates five clusters of municipalities with common socio- 156
94 sources of information to gain more elaborate knowledge of the etiology economic characteristics in terms of population structure, income, edu- 157
95 of residential fire mortality in Sweden. The richness of the dataset also cation, unemployment and sick leave. The classifications are detailed 158

R
96 allows us to explore and classify events that share common features in further in Appendix 2. 159
97 terms of both the characteristics of the individuals involved, as well as Based on the analyses performed by the forensic toxicology laboratory 160

P
98 the characteristics of the event itself. Thus, the purpose of this study (that belongs to the National Board of Forensic Medicine), dichotomous 161
99 is two-fold; (i) to describe the epidemiology of fatal residential fires variables were derived from the levels of alcohol and carboxyhemoglobin 162
100 in Sweden, and (ii) describe common typologies of these events by (COHb) in the blood of victims. Because of postmortem alcohol produc-
D 163
101 means of cluster analysis. tion, we used a threshold of 0.2‰ (20 mg/ml) (personal communication 164
Anita Holmgren, June 23, 2015) when classifying a crude indicator 165
102 2. Materials and methods variable for positive blood alcohol levels. Also, 0.2‰ is the current drink 166
E
driving limit in Sweden, which serves as a further justification for the 167
103 2.1. Materials use of this value as a cutoff. Concentrations of COHb ≥50% are severely 168
T

toxic and indicates carbon monoxide poisoning as the primary cause 169
104 The dataset used for this study is a subset of an already compiled of death (Widdop, 2002) and was therefore used as a threshold in the 170
C

105 database using linked individual- and event-level data from three differ- COHb dummy variable. 171
106 ent data sources covering all fatal fires in Sweden between January 1st,
107 1999 and December 31st, 2007. Included in the database are; the data- 2.2. Data analysis 172
E

108 base on fatal fires, held by MSB; the database on forensic examinations,
109 held by the National Board of Forensic Medicine; and the Cause of Death First, mortality rates for different age- and sex-specific groups were 173
R

110 Register, held by the National Board of Health and Welfare. For the calculated. The number of deaths within each group was divided by the 174
111 purpose of this study, we extracted all recorded fatal residential fires number of person-years in that group and presented as the number per 175
112 from the compiled database. Due to the definitions used in the database million population (pmp). To study between-group differences in in- 176
R

113 on fatal fires, only victims who died as a direct effect of a fire or explo- tentionality, blood alcohol levels ≥0.2‰, COHb ≥50% and primary injury 177
114 sive combustion process are included. Thus, indirect fatalities, such as diagnosis, stratum-specific shares were calculated and presented with 178
O

115 those caused by falling from heights after jumping from a burning build- 95% confidence intervals assuming a Poisson distribution. 179
116 ing are not included. For the same reason, a second restriction is that the Next, incidence rates of fatal fires were calculated by dividing the 180
117 deaths had to occur within 30 days of the fire event. For further details number of fires for the different risk factors by the number of rescue 181
C

118 regarding the data, matching methods, and compilation of the linked service-attended fires, converted to the rate of fatal fires per 1000 182
119 database, see Jonsson et al. (2015). fires. Similarly, fire rates were also calculated by dividing by days 183
In addition to this, data on all residential fires attended by rescue
N

120 when comparing different seasons, by hours when comparing different 184
121 services, for the same time period, were obtained from MSB in order time intervals and by population when comparing different municipal- 185
122 to calculate fatality rates per rescue service-attended fire, and linked ity categories. As above, 95% confidence intervals were used, assuming a 186
U

123 to the fatal fire events to gain supplementary information. MSB serves Poisson distribution. 187
124 as the national focal point for fire prevention and keeps records of Making use of both event and individual variables, we performed a 188
125 all emergency turnouts reported by local fire departments all over cluster analysis by using multiple correspondence analysis (MCA) 189
126 Sweden. The national register contains information on for example with agglomerative hierarchical clustering (AHC) to study common 190
127 incident location including coordinates, type of building/object, cause fire-related mortality typologies. In short, MCA with AHC is a generaliza- 191
128 of fire, object of origin, room of origin, fire size on arrival and fire spread. tion of standard multivariate cluster analysis techniques for categorical 192
129 Population statistics were obtained from Statistics Sweden in order datasets and presents a way to effectively summarize and present a 193
130 to calculate exposure-time in terms of person-years, which were large set of variables into a much smaller set of principal components, 194
131 estimated by aggregating the end-year population of each year during or clusters (Husson, Lê & Pagès, 2010). 195
132 the study period (1999–2007) for each age- and sex-specific group. Since the clustering method requires that all data is at the same 196
133 The final dataset contains data on two levels. At the event-level, the entity level, and one of the aims of this study was to describe common 197
134 following variables were included: date, time of day, municipality, typologies of fire events, the individual data had to, in cases with 198
135 residential category, cause of fire, object of origin and room of origin. more than one fatality (approximately7% of the fires), be converted to 199
136 Further, by linking rescue service reports, we also could add the the event level. For the age group variable this was accomplished by 200

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx 3

100 100

90 90

80 80

70 70

Mortality per million population

Number of deaths

60 60

50 50

40 40

30 30

F
20 20

O
10 10

O
0 0

R
Age group

Fig. 1. Distribution of deaths due to residential fires in Sweden from 1997 to 2007, stratified into age and sex-specific groups. The data are presented in both absolute numbers (black bars -

P
male, gray bars – female), and as mortality rates per million population (pmp) (circles – male, triangles – female).

201 letting the youngest victim represent the fire, for gender by creating
D
variables used in the cluster analysis and their categories, are detailed 219
202 three new categories; only male victims, only female victims, and a in Table A1, in Appendix 1. 220
203 mixed gender victims category, for alcohol by creating the new catego- After data processing, the cluster analysis consisted of two parts. 221
E
204 ries; all victims below 0.2‰ and at least one victim above or equal to First, the optimal amount of clusters that best summarized the data 222
205 0.2‰, and for primary injury diagnosis by creating the new categories; had to be chosen. For this, we used an automatic partitioning algorithm 223
T

206 burns (all victims), toxic effects of (all victims) and a mixed primary that first partitions the data from a hierarchical tree, selecting the opti- 224
207 injury diagnosis category. mal number of clusters based on the relative inertia gain from each new 225
C

208 In the chosen cluster analysis method, variables can be included cluster, then consolidates and improves the accuracy of the selection 226
209 either as a variable that can influence the estimation of clusters (active using a K-means algorithm (Husson, Josse, & Pagès, 2010; Husson, Lê 227
Q7
210 variables) or as a variable that merely supports the interpretation and Pagès, 2010). Since the variables entered into the analysis will clus- 228
E

211 (supplementary variables). Eight active variables were included that ter around unobserved latent components, these have to be interpreted 229
212 were deemed relevant to describe the circumstances of the event, the in order to be summarized. The second step of the cluster analysis thus 230
R

213 involved individuals as well as being relevant from a prevention per- consisted of this interpretation, which was done using multivariate 231
214 spective. These were: residential category, cause of fire, room and object significance tests of the difference between the clustered categories to 232
215 of origin, age group, gender, presence of alcohol in blood, and primary the sample average of those categories. P-values b 0.05 were considered 233
R

216 injury diagnosis. The other variables, regarding temporality, geography, as statistically significant. The cluster analysis was performed using R 234
217 the size of the fire when rescue service arrived and the fire spread were (R Core Team, 2015) and the ‘FactoMineR’ package (Husson, Josse, Le, 235
O

218 included as supplementary variables. The active and supplementary & Mazet, 2015). 236
C

t1:1 Table 1
t1:2 Residential fire fatalities in Sweden from 1999 to 2007, with shares of deaths by intentionality, toxicity metrics, and cause of death, stratified by sex and age.

t1:3 Percentage share (95% CI)

N

t1:4 No. of deaths Death rate, pmp (95% CI) Unintentional Alcohol ≥0.2 ‰ COHb ≥50% Toxic effects Burns

t1:5 Males
U

t1:6 0–19 years 20 2.0 (1.2–3.1) 90 (68–99) 10 (1–32) 65 (41–85) 75 (51–91) 25 (9–49)
t1:7 20–59 years 225 10.3 (9.0–11.8) 79 (73–84) 72 (66–78) 41 (35–48) 66 (59–72) 26 (20–32)
t1:8 60+ years 309 37.1 (33.0–41.4) 95 (92–97) 37 (31–42) 31 (26–37) 45 (40–51) 41 (36–47)
t1:9 All ages 554 13.8 (12.7–15.0) 88 (85–91) 50 (46–54) 37 (33–41) 55 (50–59) 34 (30–38)
t1:10
t1:11 Females
t1:12 0–19 years 12 1.3 (0.7–2.2) 92 (62–100) 17 (2–48) 42 (15–72) 67 (35–90) 25 (5–57)
t1:13 20–59 years 109 5.2 (4.2–6.2) 87 (79–93) 61 (51–70) 52 (43–62) 70 (60–78) 23 (15–32)
t1:14 60+ years 218 21.2 (18.5–24.2) 94 (90–97) 18 (13–24) 24 (19–31) 39 (32–46) 50 (43–56)
t1:15 All ages 339 8.3 (7.4–9.2) 92 (88–94) 32 (27–37) 34 (29–39) 50 (44–55) 40 (35–46)
t1:16
t1:17 Both genders
t1:18 0–19 years 32 1.7 (1.1–2.3) 91 (75–98) 13 (4–29) 56 (38–74) 72 (53–86) 25 (11–43)
t1:19 20–59 years 334 7.8 (7.0–8.7) 82 (77–86) 69 (63–74) 45 (39–50) 67 (62–72) 25 (20–30)
t1:20 60+ years 527 28.3 (25.9–30.8) 94 (92–96) 29 (25–33) 28 (25–33) 43 (38–47) 45 (40–49)
t1:21 All ages 893 11.0 (10.3–11.8) 90 (87–92) 43 (40–47) 36 (32–39) 53 (50–56) 37 (33–40)

t1:22 Notes: pmp = per million population COHb = carboxyhemoglobin.

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 4 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx

237 3. Results 3.3. Intentionality 263

238 3.1. Fatalities by age and sex In 90% of the fatalities the fire event was determined to be an 264
accident and thus the remaining 10% were classified as deliberately set. 265
239 Between 1999 and 2007, 893 fatalities in 830 fatal residential fires Intentionality, however, varied with regard to age- and sex-specific 266
240 occurred in Sweden with an annual average of 99 deaths and 92 fatal groups. For instance, the age group 20–59 years was more prone to 267
241 fire events. Roughly 60% of the fatalities were men and the incidence dying in an intentional fire-related incident compared to the elderly 268
242 rate increased with age. Rates were higher amongst males in all age population. This pattern was especially apparent amongst males 269
243 groups. In absolute numbers, fatalities were also more or equally (Table 1). 270
244 prevalent amongst men in all age groups apart from the age group
245 85+ (Fig. 1). 3.4. Temporality and geography 271

246 3.2. Toxicity metrics and primary injury diagnosis Fatal fires were more common during the winter months, however in 272
terms of risk per 1000 fires, no seasonal differences could be seen. In ad- 273

F
247 Approximately half (53%) of all victims died of toxic effects (whereof dition to this, more fatal fires occurred per hour during nights at week- 274
248 86% due to carbon monoxide poisoning and 14% to other toxic effects), ends, though in terms of incidence per 1000 fires, no difference could be 275

O
249 36% died from sustained burns and the remaining 11% died due to other seen compared to nightly weekday fires. Per capita, most fatal fires 276
250 primary diagnoses. These numbers do not appear to vary by sex. How- occur in the smallest municipalities, especially those that were classified 277
251 ever, after stratification into age-specific groups, we found that burns as rural, and in municipalities characterized by aging populations as 278

O
252 were more common as the primary cause of death amongst elderly well as low income and low educational levels. However, in terms of 279
253 victims (60 + years) (Table 1). The toxicity metrics showed that 36% risk per 1000 fires, no clear differences were seen with regard to munici- 280
254 of the fatalities had COHb ≥50%. The lowest proportion of COHb ≥50% pality size or category, indicating that the excess fatal events in these cat- 281

R
255 was observed in the oldest age group (60+) and no statistical signifi- egories were due to a higher concentration of residential fires (Table 2). 282
256 cant difference was seen between men and women. The toxicity metrics

P
257 also showed that having positive blood-alcohol levels was common, as 3.5. Type of housing, cause of fire, room and object of origin, fire size on 283
258 43% of the fatally injured exhibited levels that exceeded the drink driving rescue service arrival, and fire spread 284
259 limit in Sweden (≥0.2‰). However, this number varied with age and sex, D
260 as it was more than twice (69% versus 29%) as common in the age group Fatal fires occurred roughly as often in houses as in apartments. Per 285
261 20–59 years, compared to the oldest age group (60+ years). It was also fire, however, nursing homes for elderly were considerably more at risk 286
262 more common amongst men than women (Table 1). with 24 fatal fires per 1000 fires (Table 3). Smoking was the most 287
E

t2:1 Table 2
T

t2:2 Fatal residential fire event rates by rescue service-attended fires and population in Sweden from 1999 to 2007 by temporal and geographical variables.

Per 1000 fires

C

t2:3 Season Per day

t2:4 Spring (April–May) 0.25 (0.21–0.29) 15.1 (12.7–17.9)

t2:5 Summer (June–Aug) 0.15 (0.12–0.18) 11.8 (9.8–14.0)
E

t2:6 Fall (Sept–Nov) 0.25 (0.21–0.28) 16.7 (14.5–19.2

t2:7 Winter (Dec–March) 0.34 (0.30–0.37) 14.6 (13.1–16.1)
t2:8
R

t2:9 Day of week/hour Per 1000 h Per 1000 fires

t2:10 Weekdays day (06–22) 8.2 (7.3–9.2) 10.3 (9.2–11.5)

R

t2:11 Weekdays night (22–06) 10.3 (8.9–11.8) 23.8 (20.5–27.4)

t2:12 Weekends day (06–22) 11.0 (9.4–12.8) 12.1 (10.4–14.1)
t2:13 Weekends night (22–06) 15.8 (13.1–18.9) 22.7 (18.8–27.1)
O

t2:14
t2:15 Municipality classification 1 Per million pop. Per 1000 fires

t2:16 Municipalities with high income, high education 7.2 (5.5–9.2) 16.9 (12.9–21.7)
C

t2:17 Traditional middle class municipalities 9.2 (8.3–10.2) 14.3 (12.9–15.8)

t2:18 Municipalities with tourism-related economy 9.2 (7.6–10.9) 12.4 (10.3–14.8)
t2:19 Municipalities with traditional industries 12.8 (10.8–15.0) 14.6 (12.3–17.2)
N

t2:20 Small municipalities with low income, low education and negative population growth 19.8 (16.3–23.8) 17.8 (14.7–21.4)
t2:21
t2:22 Municipality classification 2 Per million pop. Per 1000 fires
U

t2:23 Metropolitan municipalities 10.3 (8.7–12.2) 15.7 (13.2–18.5)

t2:24 Suburban metropolitan municipalities 7.5 (6.1–9.2) 14.9 (12.1–18.3)
t2:25 Large cities 8.6 (7.5–9.9) 13.6 (11.8–15.6)
t2:26 Suburban large cities 11.9 (8.1–16.7) 16.5 (11.3–23.3)
t2:27 Commuter municipalities 10.9 (8.4–13.8) 14.1 (10.9–17.9)
t2:28 Tourism and travel industry municipalities 15.7 (11.2–21.3) 15.1 (10.8–20.5)
t2:29 Manufacturing municipalities 13.0 (10.5–15.8) 14.3 (11.6–17.5)
t2:30 Sparsely populated municipalities 21.6 (15.0–30.2) 17.4 (12.0–24.3)
t2:31 Municipalities in densely populated regions 10.3 (8.2–12.9) 13.7 (10.9–17.2)
t2:32 Municipalities in sparsely populated regions 13.9 (10.0–18.8) 14.9 (10.7–20.2)
t2:33
t2:34 Municipality population size Per million pop. Per 1000 fires

t2:35 0–19,999 13.2 (11.5–15.1) 14.7 (12.8–16.8)

t2:36 20,000–99,999 9.4 (8.5–10.4) 14.4 (13.0–15.9)
t2:37 100,000+ 9.6 (8.4–10.9) 14.8 (13.0–16.9)

t2:38 Notes: 95% confidence intervals assuming a Poisson distribution are presented in parentheses after the incidence rates.

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx 5

288 common fire cause, accounting for about one third of all fatal fires. Also, The fourth cluster (n = 63, 8%) consisted of kitchen fires where 325
289 of the rescue service-attended fires, those caused by smoking were sig- cooking appliances were left on. In these cases, alcohol was more likely 326
290 nificantly the most likely to result in a fatal fire. Fires that started in the to be involved than in the sample average, and the victims more often 327
291 bedroom or living room, and fires started in beds/sofas/armchairs and died of toxic effects than average. 328
292 clothing, were also more likely to cause fatalities compared to fires The fifth cluster (n = 207, 25%) was characterized by fires occurring 329
293 starting in other locations and objects. It is also clear that death was at weekday nights in houses. One defining characteristic of these fires 330
294 more common if the fire had spread from the object of origin when was the large proportion of unknown and missing data regarding the 331
295 the rescue services arrived. Likewise, if the fire engulfed an area greater cause of the fire, as well as the object and room of origin. Victims were 332
296 than the room of origin, the risk of death was considerably greater likely to belong to the age group 45–64 years and alcohol was more 333
297 (Table 3). likely to be involved than the average. Compared to the sample average, 334
the events occurred more often in small municipalities (0–19,999 335
298 3.6. Cluster analysis inhabitants) as well as in municipalities characterized as having low aver- 336
age income, low levels of education, and negative population growth. 337
299 Six clusters were identified in the cluster analysis. The characteristics The sixth and final cluster (n = 61, 7%) was characterized by delib- 338

F
300 of each cluster, with respect to the percentage distribution of the erately set fires due to ignition of flammable substances/gases. Most of 339
301 variable categories, are presented in Tables 4 and 5. Table 4 includes the victims in this cluster were men between the ages of 20–64 years. 340

O
302 the variables used in the estimation of the clusters (active variables) Compared to sample average the victims were more likely to belong 341
303 and Table 5 the variables included to support the interpretation of the to the age group 20–44 years. 342
304 clusters (supplementary variables). The most significant characteristics

O
305 are described below. 4. Discussion 343
306 The first cluster (n = 144, 17% of the sample) can be summarized as

R
307 death caused by burns after the ignition of clothing or paper materials, The results, regarding risk factors and risk groups, in this study are 344
308 often as a consequence of smoking. The fires were more often small consistent with previous studies, with men and elderly being at greater 345
309 and had occurred during daytime. The victim was often an elderly risk (Xiong, Bruck, and Ball (2015)). Smoking is the most common cause 346

P
310 woman (80+ years) living in an apartment or in a nursing home. The of fire (Barillo and Goode (1996)), a large proportion of the fatalities 347
311 victims were very rarely intoxicated, which differs greatly from the testing positive for alcohol in the blood (McGwin et al. (2000)), and 348
312 sample average. Ballard, Koepsell, and Rivara (1992) also showing that smoking and
D 349
313 The second cluster (n = 245, 30%) was characterized mostly by alcohol consumption increases the risk. A majority of all fire-related 350
314 death after smoking-related furniture fires originating in the bedroom deaths are attributed to smoke inhalation (Barillo et al. (1986); 351
315 or living room in an apartment. Alcohol was often involved, and the Stamyr, Thelander, Ernstgård, Ahlner, and Johanson (2012). We also 352
E
316 victims were more likely to die of toxic effects than burns. The victims found evidence suggesting the presence of sociodemographic inequal- 353
317 were more likely to be men and belong to the age group 45–64 years. ities, considering that the risk of fatal residential fires was higher in 354
T

318 The third cluster (n = 110, 13%) was mainly characterized by fires in municipalities with few inhabitants, municipalities in rural areas, and 355
319 houses, originating in fireplaces or due to various types of electrical fires. in municipalities characterized by an aging population, low income, 356
C

320 The fires were also more often large, having spread outside of the room and low educational attainment. 357
321 of origin. The victims more often died of toxic effects than average in Not only are the risk factors and risk groups highlighted in this study 358
322 these events and were more likely to be either young (5–19 years) or consistent with previous studies, but the results also show that the fire 359
E

323 old (80 + years). Victims were less likely to be intoxicated compared mortality risk factors have largely remained unchanged since the end of 360
324 to the sample average. the 1970’s (Berl & Halpin, 1979). As Sweden has seen a leveling off in 361
R

Q11 t3:1 Table 3

R

t3:2 Fatal residential fire event rates by rescue service-attended fires in Sweden from 1999 to 2007 by residential category, room and object of origin, fire cause, fire size on rescue service
t3:3 arrival, and fire spread.
O

t3:4 Residential category No. fires (%) Per 1000 fires Fire cause No. fires (%) Per 1000 fires

t3:5 House 376 (45%) 14.5 (13.0–16.0) Smoking 263 (32%) 119.7 (105.7–135.1)
t3:6 Apartment 363 (44%) 14.1 (12.7–15.7) Deliberately set (arson) 85 (10%) 18.8 (15.0–23.2)
C

t3:7 Summer cottage 18 (2%) 6.2 (3.7–9.8) Heat transfer 58 (7%) 14.9 (11.3–19.3)
t3:8 Nursing homes for elderly 56 (7%) 24.2 (18.3–31.4) Cooking appliance left on 58 (7%) 6.9 (5.2–8.9)
t3:9 Other (e.g. caravan, motorhome) 17 (2%) - Candle 42 (5%) 17.7 (12.7–23.9)
N

t3:10 Room of origin No. fires (%) Per 1000 fires Technical fault 31 (4%) 6.2 (4.2–8.9)
t3:11 Kitchen 165 (20%) 11.1 (9.5–12.9) Other 46 (6%) 2.5 (1.8–3.3)
t3:12 Living room 224 (27%) 46.9 (40.9–53.4) Unknown/Missing 247 (30%) 20.5 (18.1–23.3)
U

t3:13 Bedroom 194 (23%) 62.7 (54.2–72.2) Fire size on rescue service arrival No. fires (%) Per 1000 fires
t3:14 Bathroom/toilet/sauna 13 (2%) 11.2 (6.0–19.2) Fire extinguished/gone out 58 (7%) 5.0 (3.8–6.4)
t3:15 Other (e.g. hall, attic, cellar) 90 (11%) 3.1 (2.6–3.8) Only smoke 41 (5%) 2.8 (2.0–3.7)
t3:16 Unknown/missing 144 (17%) 67.6 (57.0–79.6) Fire in object of origin 90 (11%) 5.2 (4.2–6.4)
t3:17 Object of origin No. fires (%) Per 1000 fires Fire in room of origin 199 (24%) 26.4 (22.8–30.3)
t3:18 Bed 134 (16%) 359.2 (301.0–425.5) Fire in several rooms in same fire cell 294 (35%) 64.0 (57.0–71.8)
t3:19 Sofa, armchair 76 (9%) 248.4 (195.7–310.9) Fire in several fire cells 41 (5%) 54.8 (39.3–74.4)
t3:20 Other loose fittings 97 (12%) 16.0 (13.0–19.6) Unknown/missing 107 (13%) -
t3:21 Clothing 96 (12%) 359.6 (291.2–439.1) Fire spread No. fires (%) Per 1000 fires
t3:22 Paper/cardboard 10 (1%) 12.7 (6.1–23.4) Object of origin 110 (13%) 3.0 (2.5–3.6)
t3:23 Cooker 57 (7%) 6.2 (4.7–8.0) Room of origin 206 (25%) 17.4 (15.1–20.0)
t3:24 Electrical installations 50 (6%) 7.8 (5.8–10.3) Fire cell of origin 117 (14%) 44.9 (37.2–53.8)
t3:25 Flammable liquid or gas 45 (5%) 104.9 (76.5–140.4) Building of origin 276 (33%) 51.4 (45.5–57.9)
t3:26 Smoke duct, fireplace 29 (3%) 2.0 (1.3–2.8) Fire spread to other buildings 14 (2%) 34.6 (18.9–58.0)
t3:27 Other 30 (4%) 2.2 (1.5–3.1) Unknown/missing 107 (13%) -
t3:28 Unknown/missing 206 (25%) 34.7 (30.1–39.8)

t3:29 Notes: 95% confidence intervals assuming a Poisson distribution are presented in parentheses after the incidence rate.

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 6 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx

362 fire mortality in the beginning of the 21st century (Jonsson, Runefors, of smoking as a risk factor in fatal fires. The cluster analysis would 377
363 Särdqvist, & Nilson, 2016), it would seem that merely identifying risk suggest that instead of focusing on all cigarettes or all smokers, certain 378
364 factors and risk groups is insufficient in reducing mortality rates further. specific populations need to be the focus of such prevention efforts. For 379
365 Therefore, an enhanced knowledge of fire-related deaths is required. example, the first cluster with elderly women is a group of individuals 380
366 The cluster analyses performed in this study identify six very differ- that most likely have some type of contact with social services either 381
367 ent scenarios. From a prevention perspective, therefore, we hypothesize due to living in elderly care facilities or having home-help services. By fo- 382
368 that the effect of interventions will vary between clusters, as has been cusing on this group, and using both technical and behavioral solutions, 383
369 suggested previously by Runefors, Johansson, and Van Hees (2016). deaths in this group could potentially be reduced. However, the other 384
370 For example, whilst the risk factor analysis showed smoking as the smoking-related cluster might be harder to target with such solutions, 385
371 most common cause of fatal residential fires, the smoking-related fatal- and will likely require more complex prevention strategies. 386
372 ities can be grouped into two, very different types of fire scenarios. The The remaining four clusters also identify typical fire scenarios that 387
373 first mainly involves older women where the clothing is ignited, whilst clearly require different types of injury control and fire prevention 388
374 the second involves mainly middle-aged men, often intoxicated, where efforts. The cluster categorized by kitchen fires in which many victims 389
375 the bed or couch is most likely ignited by the cigarette. These two were intoxicated is also a well-defined group where technological 390

F
376 clusters comprise almost half of all fatal fires and show the importance interventions could potentially reduce the number of deaths. For 391

O
t4:1 Table 4
t4:2 Cluster characteristics, fatal residential fires in Sweden from 1999 to 2007, percentage distribution by the variables used in the estimation (active variables).

O
t4:3 Variables/categories Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 Total
(n = 144) (n = 245) (n = 110) (n = 63) (n = 207) (n = 61)

t4:4 Age group (youngest victim)

R
t4:5 0–4 years old 0% 0% − 1% 2% 3% + 2% 1%
t4:6 5–19 years old 0% − 0% − 12% + 0% 3% 0% 2%
t4:7 20–44 years old 0% − 15% 8% 17% 10% 41% + 12%

P
t4:8 45–64 years old 4% − 44% + 19% − 38% 42% + 43% 33%
t4:9 65–79 years old 31% 27% 22% 24% 24% 11% − 25%
t4:10 80+ years old 65% + 13% − 38% + 19% 18% − 3% − 27%
t4:11 D
t4:12 Gender
t4:13 Only male 41% − 65% + 66% 57% 58% 74% + 60%
t4:14 Only female 59% + 32% 28% 33% 34% 20% − 36%
E
t4:15 Both male and female 0% − 3% 5% 10% 8% + 7% 5%
t4:16
t4:17 Alcohol (‰)
t4:18 b0.2‰ (all victims) 94% + 40% − 66% + 41% − 53% 56% 57%
T

t4:19 ≥0.2‰ (at least one victim) 6% − 60% + 34% − 59% + 47% 44% 43%
t4:20
t4:21
C

Primary injury diagnosis

t4:22 Burns (all victims) 78% + 27% − 26% − 17% − 30% − 46% 37%
t4:23 Toxic effects (all victims) 6% − 61% + 71% + 63% + 53% 48% 50%
t4:24 Mixed/other/unknown 16% 13% 3% − 19% 17% 7% 13%
E

t4:25
t4:26 Residential category
t4:27 House 10% − 29% − 78% + 44% 69% + 52% 45%
R

t4:28 Apartment 53% + 70% + 16% − 49% 22% − 34% 44%

t4:29 Nursing homes for elderly 36% + 0% − 2% − 2% 0% − 0% − 7%
t4:30 Other 0% − 0% − 4% 5% 9% + 13% + 4%
R

t4:31
t4:32 Cause of fire
t4:33 Smoking 53% + 73% + 1% − 0% − 3% − 0% − 32%
O

t4:34 Deliberately set 3% − 5% − 2% − 0% − 4% − 93% + 10%

t4:35 Heat transfer 13% + 4% − 24% + 6% 0% − 0% − 7%
t4:36 Cooking appliance left on 1% − 0% − 0% − 90% + 0% − 0% − 7%
t4:37 − − −
C

Candle 13% + 5% 7% 0% 1% 0% 5%
t4:38 Technical fault 1% − 0% − 27% + 0% 0% − 0% 4%
t4:39 Other 8% 1% − 26% + 2% 0% − 2% 6%
t4:40 Unknown/missing 8% − 12% − 13% − 2% − 91% + 5% − 30%
N

t4:41
t4:42 Object of origin
t4:43 Bed/sofa/armchair 17% − 73% + 3% − 0% − 0% − 2% − 25%
U

t4:44 Other loose fittings 8% 20% + 10% 5% 7% − 11% 12%

t4:45 Clothing, paper/cardboard 69% + 2% − 0% − 0% − 1% − 2% − 13%
t4:46 Cooker 0% − 0% − 2% − 87% + 0% − 0% − 7%
t4:47 Electrical installations 1% − 0% − 44% + 2% 0% − 0% − 6%
t4:48 Flammable liquid or gas 0% − 0% − 0% − 0% − 0% − 74% + 5%
t4:49 Smoke duct, fire place 0% − 0% − 23% + 0% 2% 0% 3%
t4:50 Other 1% 2% − 17% + 2% 1% − 3% 4%
t4:51 Unknown/missing 3% − 3% − 2% − 5% − 89% + 8% − 25%
t4:52
t4:53 Room of origin
t4:54 Kitchen 28% + 2% − 24% 97% + 15% 2% − 20%
t4:55 Living room 32% 44% + 25% 3% − 15% − 15% − 27%
t4:56 Bedroom 21% 50% + 13% − 0% − 9% − 15% 23%
t4:57 Other 10% 3% − 33% + 0% − 10% 39% + 12%
t4:58 Unknown/missing 10% − 1% − 5% − 0% − 50% + 30% + 17%

t4:59 Notes: ‘-‘ indicate that the proportion is significantly (p b 0.05) lower than in the sample average, while ‘+’ indicate that the proportion is greater. Active variables are used in the
t4:60 estimation. See main text for details.

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx 7

392 example, more common use of automatic stove turn-off devices would fatalities, was characterized by the fact that a large part of the informa- 406
393 likely be an effective countermeasure to such events (Lushaka & Zalok, tion on cause of fire, or room and object of origin, is unknown. Since 407
394 2014). both these clusters are characterized by occurring in houses and being 408
395 Although only accounting for 8% of the fatal fires, deliberate, suicidal relatively large fires regarding fire size and fire spread, one hypothesis 409
396 fires in which the individual has ignited flammable liquids, is a cluster could be that they belong to the same cluster and that both clusters 410
397 that needs to be studied further. For example, this group of predomi- are grouped by the same latent factors, but are simply divided by the 411
398 nantly men may have had contact with psychiatric clinics or other fact that cause of fire, or room and object of origin in some cases was 412
399 social services, which illustrates the need to involve different societal determined, whilst in others unknown. However, the unknown cluster 413
400 functions in fire mortality prevention efforts. also shows some similarities with the kitchen fires and the male- 414
401 The remaining 38% of fatal fires were divided into two clusters, both dominated smoking cluster in terms of age group, sex and alcohol 415
402 largely consisting of fires in detached or semi-detached houses. One of consumption. This suggests that the most plausible explanation is 416
403 the clusters refers to fires started by electrical installations and heat that some of the cases in the unknown cluster probably belong to one 417
404 transfer (e.g. lamps that have fallen down in a couch/bed or clothes of the previously mentioned clusters. With further improvements 418
405 left on the electrical heater). The other cluster, consisting of 25% of all in cause determination methods, we would likely gain a better 419

F
t5:1 Table 5

O
t5:2 Cluster characteristics, fatal residential fires in Sweden from 1999 to 2007, percentage distribution by the variables used to support the interpretation (supplementary variables).

t5:3 Variables/categories Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 Total

O
(n = 144) (n = 245) (n = 110) (n = 63) (n = 207) (n = 61)

t5:4 Municipality classification 1

t5:5 Municipalities with high income, high education 8% 3% − 9% 6% 9% 16% + 7%

R
t5:6 Traditional middle class municipalities 66% + 51% 34% − 49% 35% − 43% 47%
t5:7 Municipalities with tourism-related economy 7% − 17% 21% 13% 16% 13% 15%
t5:8 Municipalities with traditional industries 13% 15% 21% 22% 19% 23% 17%

P
t5:9 Small municipalities with low income, low education and 7% − 14% 15% 10% 21% + 5% − 14%
t5:10 negative population growth
t5:11
t5:12 Municipality classification 2
t5:13 Metropolitan municipalities 31% + 21% + 9% − 11% 9% − 15% 17%
t5:14 Suburban municipalities 13% 9%
D 10% 11% 12% 16% 11%
t5:15 Large cities 28% 24% 23% 27% 23% 26% 25%
t5:16 Suburban municipalities to large cities 1% 3% 5% 6% 4% 5% 4%
E
t5:17 Commuter municipalities 4% − 8% 8% 14% 7% 13% 8%
t5:18 Tourism and travel industry municipalities 2% 5% 5% 5% 7% 3% 5%
t5:19 −
T

Manufacturing municipalities 6% 13% 15% 10% 14% 7% 11%

t5:20 Sparsely populated municipalities 1% − 3% 7% 2% 6% 7% 4%
t5:21 Municipalities in densely populated regions 8% 9% 11% 10% 12% 7% 5%
C

t5:22 Municipalities in sparsely populated regions 6% 4% 5% 5% 5% 2% 9%

t5:23
t5:24 Municipality size
t5:25 − −
E

0–19,999 inhabitants 13% 22% 35% 29% 39% + 25% 27%

t5:26 20,000–99,999 inhabitants 42% 47% 50% 46% 43% 49% 44%
t5:27 100,000+ inhabitants 46% + 32% 15% − 25% 18% − 26% 28%
R

t5:28
t5:29 Season
t5:30 Spring: Apr–May 17% 13% 18% 17% 18% 13% 16%
t5:31 Summer: Jun–Aug 11% 16% 10% 14% 15% 26% + 15%
R

t5:32 Fall: Sep–Nov 29% 25% 24% 22% 22% 21% 24%
t5:33 Winter: Dec–Mar 42% 45% 48% 46% 44% 39% 44%
t5:34
O

t5:35 Day of week/hour

t5:36 Weekdays day (06–22) 45% + 38% 43% 43% 25% − 41% 37%
t5:37 Weekdays night (22–06) 14% − 24% 25% 17% 31% + 18% 23%
− −
C

t5:38 Weekdays (unknown hour) 7% + 0% 0% 5% 4% 8% 3%

t5:39 Weekends day (06–22) 27% + 21% 19% 16% 16% 16% 20%
t5:40 Weekends night (22–06) 6% − 16% 14% 17% 18% 15% 14%
t5:41
N

Weekends (unknown hour) 1% 0% 0% 2% 2% + 0% 1%

t5:42 Unknown/missing 1% 0% 0% 0% 3% + 2% 1%
t5:43
t5:44 Fire size on rescue service arrival
U

t5:45 Fire extinguished/gone out 15% + 6% 6% 8% 3% − 7% 7%

t5:46 Only smoke 11% + 6% 3% 5% 0% − 5% 5%
t5:47 Fire in object of origin 19% + 15% + 5% − 6% 4% − 11% 11%
t5:48 Fire in room of origin 27% 32% + 20% 35% + 13% − 16% 24%
t5:49 Fire in several rooms in same fire cell 9% − 31% 49% + 29% 52% + 43% 35%
t5:50 Fire in several fire cells 0% − 4% 8% 2% 10% + 2% 5%
t5:51 Unknown/missing 19% + 6% − 8% 16% 18% + 16% 13%
t5:52
t5:53 Fire spread
t5:54 Object of origin 33% + 13% 5% − 11% 3% − 15% 13%
t5:55 Room of origin 34% + 36% + 16% − 37% + 9% − 18% 25%
t5:56 Fire cell of origin 10% 22% + 13% 6% 11% 13% 14%
t5:57 Building of origin 3% − 22% − 55% + 30% 55% + 38% 33%
t5:58 Fire spread to other buildings 0% 2% 2% 0% 4% + 0% 2%
t5:59 Unknown/missing 19% + 6% − 8% 16% 18% + 16% 13%

t5:60 Notes: ‘-‘ indicate that the proportion is significantly (p b 0.05) lower than in the sample average, while ‘+’ indicate that the proportion is greater. Supplementary variables are used in the
t5:61 interpretation of the clusters. See main text for details.

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 8 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx

420 understanding of, and the relevant interventions needed to prevent, 5. Conclusions 483
421 these events as well.
This study illustrates that the occurrence of fatal residential fires is 484
non-randomly distributed in the Swedish population, and that there 485
422 4.1. Limitations are clear clusters of events that may require specialized prevention 486
efforts. Hence, the most at-risk individuals must be identified and met 487
423 Some methodological issues are relevant to raise in relation to with actions that specifically target the risk factors with which 488
424 this paper. Firstly, a criticism could be that the data used in this paper they are associated. Through cooperation between the various munici- 489
425 is roughly a decade old. However, this is largely due to the effort in pro- pal departments and a holistic approach, this should be possible. 490
426 ducing high quality data. Previously, the external completeness has Merely focusing on classic approaches that may work for the general 491
427 been low in Sweden (Jonsson et al., 2015). As a consequence of the population (e.g. smoke detectors) seems to be ineffective in reducing 492
428
Q8 data linkage performed in Jonsson et al. (2015), Jonsson et al. (2016) the risk of fatal fires in the current at-risk population observed in this 493
429 previous work the external completeness is now assessed to be high, study. 494
430 however a drawback when linking different registers, collected with

F
431 different purposes, and different set of variables, is that the internal
Practical applications 495
Q9
432 completeness can suffer due to incomplete data, especially if a variable

O
433 exists only in one of the linked datasets. Besides missing data due to re-
The results from this study implies that a holistic approach to fire 496
434 cord linkage, the categories of some variables can be classified as un-
prevention that promotes cooperation between various societal 497
435 known because the information provider often has the opportunity to

O
functions, e.g. rescue services, elderly care, psychiatric clinics and other 498
436 code the variable in this way.
social services, may be necessary in order to decrease the fire-related 499
437 Regarding the following fire-event level variables; the date of the
mortality. This stems from an increased need for specialized interventions 500
438 fire, the residential category, and the municipality where the fire

R
for different groups of individuals, and a demand for both human and 501
439 occurred, there are no missing or unknown data. Information on the
technical solutions. 502
440 time (hour) when the fire occurred, cause of the fire, and the room

P
441 and object of origin is only collected in a structured form in the database
442 on fatal fires and in the rescue service turn-out reports and therefore has Acknowledgements 503
443 missing data due to record linkage in approximately 5% of the cases. For D
444 the time (hour) there are no unknown data. In our case, unknown The authors' gratefully thank the Swedish Civil Contingencies Agency 504
445 data is only notable for the following variables: cause of the fire (25%), for financial support (grant number 2014-5283). The funding source had 505
446 the room of origin (20%), and the object of origin (12%). The variables no role in the design of the study, the analysis and interpretation of the 506
E
447 fire size on rescue service arrival and fire spread are only present in data or the writing of, nor the decision to publish, the manuscript. This 507
448 the reports from the rescue services. There are no unknown data work was also supported by local grants from the County Council of 508
T

449 within these reports, but 13% of the values are missing due to record Uppsala (ALF). Finally, we would like to thank Colin McIntyre, Swedish 509
450 linkage. Civil Contingencies Agency for insightful suggestions for improvement 510
C

451 Regarding the individual level variables age (at death) and gender, of the manuscript and for providing language help. 511
452 there are no missing or unknown data. Primary injury diagnosis has
453 no missing values due to record linkage and the unknown values are References 512
E

454 only 2.5%. An inherent limitation that exists in the dummy variables,
Ballard, J. E., Koepsell, T. D., & Rivara, F. (1992). Association of smoking and alcohol drinking 513
455 created on the variables alcohol and COHb, is that a value of zero (beside with residential fire injuries. American Journal of Epidemiology, 135(1), 26–34. 514
R

456 that the toxic metric is zero) can also indicate an internally missing ob- Barillo, D. J., & Goode, R. (1996). Fire fatality study: Demographics of fire victims. Burns, 515
457 servation as well as externally missing values due to record linkage, 22(2), 85–88. 516
Barillo, D. J., Rush, B. F., Jr., Goode, R., Lin, R. L., Freda, A., & Anderson, E. J., Jr. (1986). Is 517
458 which means that the numbers that we assess are likely an underesti-
R

ethanol the unknown toxin in smoke inhalation injury? The American Surgeon, 518
459 mation of the true number in the population. However, as far as we 52(12), 641–645. 519
460 are aware, the proportion of internal missing values is relatively small Berl, W. G., & Halpin, B. M. (1979). Human fatalities from unwanted fires. Center for Fire 520
521
O

Research, National Bureau of Standards.

461 and the database on toxic metrics covers 90% of the sample, resulting
Brennan, P. (1999). Victims and survivors in fatal residential building fires. Fire and 522
462 in low level of missing data overall. Materials, 23(6), 305–310. 523
463 In sum, the data on which this study relies on are considered having Bruck, D., Ball, M., & Thomas, I. R. (2011). Fire fatality and alcohol intake: Analysis of key 524
C

464 very high external completeness and that the internal completeness is risk factors. Journal of Studies on Alcohol and Drugs, 72(5), 731–736. 525
Duncanson, M., Woodward, A., & Reid, P. (2002). Socioeconomic deprivation and fatal 526
465 also high for this kind of data. In our assessment, the analysis of risk Unintentional domestic fire incidents in New Zealand 1993–1998. Fire Safety Journal, 527
N

466 groups and risk factors is not influenced in any significant way by the 37(2), 165–179. 528
467 described shortcomings. However, it is clear that the unknown or Graesser, H., Ball, M., & Bruck, D. (2009). Risk factors for residential fire fatality across the 529
lifespan: Comparing coronial data for children, adults, and elders. Proceedings of the 530
468 missing data on cause of fire, room and object of origin has affected 4th international symposium on human behavior in fire (pp. 639–644). London: 531
U

469 the cluster analysis. In cases when this occur simultaneously for all Interscience Communications. 532
470 three variables (10%), these fire-events, since they have a lot in com- Harpur, A., Boyce, K., & McConnel, N. (2014). An investigation into the circumstances 533
surrounding elderly dwelling fire fatalities and the barriers to implementing fire safety 534
471 mon, can then be gathered in a separate cluster as described previously. strategies among this group. Fire Safety Science, 11, 1144–1159. 535
472 Meanwhile, as also described previously, it may be very difficult to deal Harpur, A. P., Boyce, K. E., & McConnell, N. C. (2013). An investigation into the circum- 536
473 with the problem through education and improved accident investiga- stances surrounding fatal dwelling fires involving very young children. Fire Safety 537
Journal, 61, 72–82. 538
474 tion methodology when for example fire causes are simply impossible
Holborn, P. G., Nolan, P. F., & Golt, J. (2003). An analysis of fatal unintentional dwelling 539
475 to determine. fires investigated by London fire brigade between 1996 and 2000. Fire Safety 540
476 An additional limitation is that the results in this study are specific to Journal, 38(1), 1–42. 541
477 Sweden and Swedish conditions and must therefore be viewed in that Husson, F., Josse, J., Le, S., & Mazet, J. (2015). FactoMineR: multivariate exploratory 542
data analysis and data mining. FactoMineR: multivariate exploratory data analysis 543
478 context. For example, child fire mortality is higher in countries such as and data mining. R package version 1.30 http://CRAN.R-project.org/package= 544
479 the US compared to Sweden. However, the results largely support FactoMineR. 545
480 international findings with regards to risk factors and risk groups, and Husson, F., Josse, J., & Pages, J. (2010b). Principal component methods-hierarchical 546
clustering-partitional clustering: Why would we need to choose for visualizing 547
481 therefore it is also likely that the resulting cluster analyses are general- data. Technical report – AgroCampus (Retrived from http://factominer.free.fr/docs/ 548
482 izable to many similar sociodemographic countries. HCPC_husson_josse.pdf. Accessed 10 May 2016). 549

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx 9

550 Husson, F., Lê, S., & Pagès, J. (2010a). Exploratory multivariate analysis by example using R. Runefors, M., Johansson, N., & Van Hees, P. (2016). How could the fire fatalities have been 576
551 CRC press. prevented? An analysis of 144 cases during 2011–2014 in Sweden Journal of Fire 577
552 Istre, G. R., McCoy, M., Carlin, D. K., & McClain, J. (2002). Residential fire related deaths and Sciences, 34(6), 515–527. 578
553 Injuries among children: Fireplay, smoke alarms, and prevention. Injury Prevention, Runyan, C. W., Bangdiwala, S. I., Linzer, M. A., Sacks, J. J., & Butts, J. (1992). Risk factors for 579
554 8(2), 128–132. fatal residential fires. New England Journal of Medicine, 327(12), 859–863. 580
555 Istre, G. R., McCoy, M. A., Osborn, L., Barnard, J. J., & Bolton, A. (2001). Deaths and injuries Stamyr, K., Thelander, G., Ernstgård, L., Ahlner, J., & Johanson, G. (2012). Swedish forensic 581
556 from house fires. New England Journal of Medicine, 344(25), 1911–1916. data 1992–2009 suggest hydrogen cyanide as an important cause of death in fire vic- 582
557 Jonsson, A., Bergqvist, A., & Andersson, R. (2015). Assessing the number of fire fatalities in tims. Inhalation Toxicology, 24(3), 194–199. 583
558 a defined population. Journal of Safety Research, 55, 99–103. Swedish Association of Local Authorities and Regions (SKL) (2011). Classification of Swedish 584
559 Jonsson, A., Runefors, M., Särdqvist, S., & Nilson, F. (2016). Fire-related mortality in municipalities. http://skl.se/tjanster/kommunerlandsting/faktakommunerochlandsting/ 585
560 Sweden: Temporal trends 1952 to 2013. Fire Technology, 52(6), 1697–1707. kommungruppsindelning.2051.html (Accessed 10 May 2016). 586
561 Leth, P., Gregersen, M., & Sabroe, S. (1998). Fatal residential fire accidents in the munici- Turner, S. L., Johnson, R. D., Weightman, A. L., Rodgers, S. E., Arthur, G., Bailey, R., & Lyons, 587
562 pality of Copenhagen, 1991–1996. Preventive Medicine, 27(3), 444–451. R. A. (2017). Risk factors associated with unintentional house fire incidents, injuries 588
563 Lushaka, B., & Zalok, E. (2014). Development of a sensing device to reduce the risk from and deaths in high-income countries: A systematic review. Injury Prevention 589
564 kitchen fires. Fire Technology, 50(3), 791–803. injuryprev-2016. 590
565 Marshall, S. W., Runyan, C. W., Bangdiwala, S. I., Linzer, M. A., Sacks, J. J., & Butts, J. D. (1998). U.S. Fire Administration (2011). Fire death rate trends: An international perspective, topical fire 591
566 Fatal residential fires: Who dies and who survives? JAMA, 279(20), 1633–1637. research series Vol. 12. Issue 8 Emmitsburg, MD: U.S Department of Homeland Security. 592
567 McGwin, G., Jr., Chapman, V., Rousculp, M., Robison, J., & Fine, P. (2000). The epidemiology of Warda, L., Tenenbein, M., & Moffatt, M. E. (1999). House fire injury prevention update. 593

F
568 fire-related deaths in Alabama, 1992–1997. Journal of Burn Care & Research, 21(1), 75 Part I. A review of risk factors for fatal and non-fatal house fire injury Injury 594
569 (hyhen). Prevention, 5(2), 145–150. 595
570 Mulvaney, C., Kendrick, D., Towner, E., Brussoni, M., Hayes, M., Powell, J., ... Ward, H. Widdop, B. (2002). Analysis of carbon monoxide. Annals of Clinical Biochemistry, 39(4), 596

O
571 (2009). Fatal and non-fatal fire injuries in England 1995–2004: Time trends 378–391. 597
572 and inequalities by age, sex and area deprivation. Journal of Public Health, 31(1), World Health Organization (WHO), Prentice, T., Beaglehole, R., & Irwin, A. (2004). The 598
573 154–161. world health report, 2004: Changing history. World Health Organization. 599
574 R Core Team (2015). R: A language and environment for statistical computing. Vienna, Xiong, L., Bruck, D., & Ball, M. (2015). Comparative investigation of ‘survival’ and fatality 600

O
575 Austria: R Foundation for Statistical Computing (http://www.R-project.org/). factors in accidental residential fires. Fire Safety Journal, 73, 37–47. 601

R
P
Appendix 1
D
t6:1 Table A1
t6:2 Included variables at the residential fatal fire event level, with categories and coding rules for the variables used in this study based on the database on fatal fires compiled by Jonsson et al.
E
t6:3 (2015).

t6:4 Variable Categories and coding rules Used in

T

t6:5 Seasonb Derived from date in the compiled database Table 2

Spring: Apr–May Table 4
C

Summer: Jun–Aug
Fall: Sep–Nov
Winter: Dec–Mar
t6:6 Day of week/Hourb
E

Derived from Date and Time of the day in the compiled database Table 2
Weekdays (06–22): Mon-Fri, 06:00–21:59 Table 4
Weekdays (22–06): Sun-Thu, 22:00–05:59
R

Weekdays (unknown hour): Mon-Thu, unknown hour

Weekends (06–22): Sat-Sun, 06:00–21:59
Weekends (22–06): Fri-Sat, 22:00–05:59
R

Weekends (unknown hour): Sat, unknown hour

Unknown/Not specified: Fri, Sun, unknown hour
t6:7 Municipality classification 1b (see Appendix 2) Derived from municipality in the compiled database Table 2
O

Municipalities with high income, high education Table 5

Traditional middle class municipalities
Municipalities with tourism-related economy Municipalities with traditional industries
Small municipalities with low income, low education and negative population growth
C

t6:8 Municipality classification 2b (see Appendix 2) Derived from municipality in the compiled database Table 2
Metropolitan municipalities Table 5
Suburban municipalities
N

Large cities
Suburban municipalities to large cities
Commuter municipalities
U

Tourism and travel industry municipalities

Manufacturing municipalities
Sparsely populated municipalities
Municipalities in densely populated regions
Municipalities in sparsely populated regions
t6:9 Municipality sizeb (see Appendix 2) Derived from municipality in the compiled database Table 2
0–19,999 inhabitants Table 5
20,000–99,999 inhabitants
100,000 + inhabitants
t6:10 Residential categorya (see notes) From the compiled database Table 3
House (detached, semi-detached or terraced) Table 4
Apartment
Summer cottage
Nursing homes for elderly
Other (e.g. caravan, motorhome)

(continued on next page)

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 10 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx

Table A1 (continued)

Variable Categories and coding rules Used in

t6:11 Cause of firea (see notes) From the compiled database Table 3
Smoking Table 4
Deliberately set
Heat transfer
Cooking appliance left on
Candle
Technical fault
Other
Unknown/Not specified
t6:12 Object of origina (see notes) From the compiled database Table 3
Bed Table 4
Sofa, armchair
Other loose fittings
Clothing

F
Paper/cardboard
Cooker

O
Electrical installations
Flammable liquid or gas
Smoke duct, fire place
Other

O
Unknown/not specified
t6:13 Room of origina (see notes) From the compiled database Table 3
Kitchen Table 4

R
Living room
Bedroom
Bathroom/toilet/sauna

P
Other (e.g. hall, attic, cellar)
Unknown/Not specified
t6:14 Fire size on rescue service arrivalb Linked from rescue service reports Table 3
Fire extinguished/gone out D Table 5
Only smoke
Fire in object of origin
Fire in room of origin
E
Fire in several rooms in same fire cell
Fire in several fire cells
Unknown/not specified
T

t6:15 Fire spreadb Linked from rescue service reports Table 3

Object of origin Table 5
Room of origin
C

Fire cell of origin

Building of origin
Fire spread to other buildings
E

Unknown/not specified
t6:16 Age group (oldest victim)a Derived from age in the compiled database Table 4
R

0–4 years old

5–19 years old
20–44 years old
R

45–64 years old

65–79 years old
80+ years old
t6:17 Gendera Derived from gender in the compiled database Table 4
O

Only male
Only female
Both male and female
C

t6:18 Alcohola Derived from alcohol (‰) in the compiled database Table 4
0: b0.2 ‰ or not specified (all victims)
1: ≥0.2 ‰ (at least one of the victims)
N

t6:19 Primary injury diagnosisa Derived from primary injury diagnosis in the compiled database Table 4
Burns (all victims)
Toxic effects (all victims)
U

Mixed/other/unknown

t6:20 Notes: Residential category, cause of fire, room and object of origin were derived by triangulating information from all sources. The database on fatal fires and the data on rescue service
t6:21 attended residential fires collect these variables as structured data fields. When consistent between these sources, we used the values as they were reported. In case of inconsistent data the
t6:22 values were determined by also reading the unstructured data from these sources together with unstructured data from the forensic examinations.
a
t6:23 Active variables in the cluster analysis (used in the estimation).
b
t6:24 Supplementary variables in the cluster analysis (used to support the interpretation).
t6:25
12
t7:1 Table A2
t7:2 Included variables at the individual residential fire fatality level, with categories and coding rules for the variables used in this study based on the database on fatal fires compiled by
t7:3 Jonsson et al. (2015).

t7:4 Variable Categories and coding rules Used in

t7:5 Gender From the compiled database Table 1

Male
Female

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx 11

Table A2 (continued)

Variable Categories and coding rules Used in

t7:6 Age group (5 years) Derived from Age in the compiled database Fig. 1
0–4, 5–9,…,85+
t7:7 Age group Derived from age in the compiled database Table 1
0–19 years old
20–59 years old
60+ years old
t7:8 Primary injury diagnosis From the compiled database Table 1
Burns
Toxic effects
Other/unknown
t7:9 Intention Derived from cause of fire (see notes) Table 1
1 = Unintentional
0 = Intentional
t7:10 Alcohol (dichotomous) Derived from alcohol (‰) in the compiled database Table 1

F
0: b0.2‰ or not specified
1: ≥0.2‰

O
t7:11 COHb (dichotomous) Derived from COHb (%) in the compiled database Table 1
0:b50% or not specified
1: ≥50%

O
t7:12 Notes: All fatalities in a fire classified as deliberately set were coded as intentional. Whether the death was a suicide or homicide is not always clear, but based on the underlying cause of
t7:13 death from the cause of death register and the forensic examinations, these numbers in the absolute majority will most likely reflect suicides.
t7:14

R
3

Appendix 2 4
602

P
2.1. Municipality classification 1 5
603

This classification of Swedish municipalities is made by Statisticon AB (a Swedish private company providing qualified services in statistics and
D 6
604
data management). The municipalities are clustered using a multivariate classification technique (“Random Forest”) into five categories on the 7
605
basis ten socio-economic variables (2012). These variables are: the proportion of the population that are unemployed (20–64 years), the proportion 606
8
E
of the population receiving financial support, the number of sick-days per capita, the proportion of the population in work, the median income 607
9
(20 years and above), the proportion of the population with tertiary education, the proportion of high school students eligible for college education, 608
10
609
the proportion of the population born outside of the EU/EFTA, and the number of young people (0–19 years) in proportion to the group 20–64 years. 11
T

610
The number of elderly people (65 years and above) in proportion to the group 20–64 years. 12
611
Category 1: (32 municipalities, 11% of the Swedish population). 13
C

612
In these municipalities, the inhabitants have a high level of education, a well-paid job and low levels of ill health. The municipalities include many 14
613
families with children and are within commuting distance to the three largest cities in Sweden; Stockholm, Gothenburg, and Malmö. 15
614
E

Category 2: (50 municipalities, 54% of the Swedish population). 16

615
This category includes the three largest cities, municipalities with universities, and traditionally industrial cities. These municipalities can be seen 17
616
to represent the multicultural Sweden given the breadth in socio-economic variables. Similarly to category 1, the inhabitants have a high level of ed- 18
617
R

ucation, a well-paid job and low levels of ill health. The municipalities have strong birth rates and significantly more are born outside of the EU/EFTA 618
19
area. 619
20
R

Category 3: (72 municipalities, 16% of the Swedish population). 620

21
621
This category includes many of Sweden's tourist areas. Income levels are reasonably high, unemployment is low as is the proportion of the 22
622
population receiving financial support. A crucial difference to category 1 and 2 is that municipalities in category 3 have an older population and 23
O

623
lower birth rates. 24
624
Category 4: (71 municipalities, 13% of the Swedish population). 25
625
C

In these municipalities, inhabitants have a lower educational level, higher rates of unemployment and lower levels of income. These municipal- 26
626
ities also have an older population. 27
627
Category 5: (65 municipalities, 6% of the Swedish population). 28
628
N

These municipalities are categorized by a very old population, low birth rates as well as low income and educational levels. 29
629
630
631
U

2.2. Municipality classification 2 30

632
633
This classification of Swedish municipalities is made by the Swedish Association of Local Authorities and Regions. The municipalities are divided 31
634
into ten categories based on structural parameters (2011) such as population, commuting patterns, tourism and travel industry, and economic struc- 32
635
ture. A municipality can meet the inclusion criteria for more than one category. The categories are therefore prioritized in the order they appear 636
33
below. 637
34
Metropolitan municipalities: (3 municipalities). 638
35
639
Municipalities with a population of over 200,000 inhabitants. 36
640
Suburban municipalities: (38 municipalities). 37
641
Municipalities where more than 50% of the population commute to work in another municipality. The most common commuting destination 38
642
must be one of the metropolitan municipalities. 39
643
Large cities: (31 municipalities). 40
644
Municipalities with 50,000–200,000 inhabitants where more than 70% of the population live in the urban areas. 41
645
Suburban municipalities to large cities: (22 municipalities). 42
635

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008
 12 A. Jonsson et al. / Journal of Safety Research xxx (2017) xxx–xxx

Municipalities in which more than 50% of the population commute to work in a large city. 43
636
Commuter municipalities: (51 municipalities). 637
44
Municipalities in which more than 40% of the population commute to work in another municipality. 638
45
Tourism and travel industry municipalities: (20 municipalities). 639
46
640
Municipalities where the number of guest nights in hotels, youth hostels, and camping sites is higher than 21 nights per inhabitant and the num- 47
641
ber of holiday homes is higher than 0.20 per inhabitant. 48
642
Manufacturing municipalities: (54 municipalities). 49
643
Municipalities where more than 34% of the population aged 16 to 64 is employed in manufacturing, mining, energy, environmental, and construc- 50
644
tion industries. 51
645
Sparsely populated municipalities: (20 municipalities). 52
646
Municipalities where less than 70% of the population live in urban areas and with less than eight inhabitants per km2. 53
647
Municipalities in densely populated regions: (35 municipalities). 648
54
Municipalities with more than 300,000 inhabitants within a 112.5 km radius. 649
55
Municipalities in sparsely populated regions: (16 municipalities). 650
56

F
651
Municipalities with less than 300,000 inhabitants within a 112.5 km radius. 57

O
2.3. Municipality size 652
58

This classification of Swedish municipalities is made by the authors. The municipalities are divided into three groups merely based on population 653
59

O
654
size. A yearly average of the aggregated end-year population (1999–2007) was calculated for each municipality and the municipalities were then 60
655
classified into the sizes 0–19,999, 20,000–99,999, and 100,000+. 61

R
62

P
602
656 Anders Jonsson is a PhD student in Risk Management at Karlstad University. Part-time, he Finn Nilson took his PhD in Risk Management at Karlstad University, Sweden, in 2014 on 613
664
603 is working as a statistician within the area of accident and injury analysis at the Knowledge “Fall-Related Injuries Amongst Elderly in Sweden Still an Emerging Risk?”. He currently 614
657 665
604 Development Section within the Swedish Civil Contingencies Agency. He earned a PhL in works as an associate professor and researcher at the Division of Risk Management and 615
658
605 Risk Management at Karlstad University, Sweden, in 2016 with his licentiate thesis Centre for Public Safety at Karlstad University. 666
616
659
606 “Fire-related Deaths in Sweden: Determining Current Levels and Historical Trends”. 617
607
D 618
Fredrik Huss is the director of Uppsala Burn Center, Uppsala University Hospital, Sweden 667
608 Carl Bonander earned his PhD in Risk Management at Karlstad University, Sweden, in and holds a position as associate professor in Plastic Surgery at Uppsala University, 619
660 Sweden. He finished his PhD in Medicine at Linköping University, Sweden, in 2005 on 620
668
609 2016 with his doctoral thesis titled “Assessing the effects of societal injury control inter-
E
661
610 ventions”. He currently works as a senior lecturer and researcher at the Division of Risk “In vitro and in vivo studies of tissue engineering in reconstructive plastic surgery”. 669
621
611
662 Management and Centre for Public Safety at Karlstad University. 622
670
612
663
T

623
C
E
R
R
O
C
N
U

Please cite this article as: Jonsson, A., et al., The state of the residential fire fatality problem in Sweden: Epidemiology, risk factors, and event
typologies, Journal of Safety Research (2017), http://dx.doi.org/10.1016/j.jsr.2017.06.008