Effectiveness of the Cigarette Ignition Propensity Standard in Preventing Unintentional Residential Fires in Massachusetts

Abstract

Objectives. We evaluated the Massachusetts Fire Safe Cigarette Law’s (FSCL’s) effectiveness in preventing residential fires.

Methods. We examined unintentional residential fires reported to the Massachusetts Fire Incident Reporting System from 2004 to 2010. We analyzed FSCL effect on the likelihood of cigarette- versus noncigarette-caused fires and effect modification by fire scenario factors by using an interrupted time series regression model. We analyzed the effect of FSCL on monthly fire rates with Poisson regression.

Results. Cigarettes caused 1629 unintentional residential fires during the study period. The FSCL was associated with a 28% (95% confidence interval = 12%, 41%) reduction in the odds of cigarette- versus noncigarette-caused fires, although not in analyses restricted to casualty fires, with smaller sample size. The largest reductions were among fires in which human factors were involved; that were first ignited on furniture, bedding, or soft goods; that occurred in living areas; or that occurred in the summer or winter.

Conclusions. The FSCL appears to have decreased the likelihood of cigarette-caused residential fires, particularly in scenarios for which the ignition propensity standard was developed. Current standards should be adopted, and the need for strengthening should be considered.

Approximately 360 900 residential building fires are reported each year in the United States, resulting in an estimated 2495 deaths, 13 250 injuries, and $7 billion in property losses.¹ The number of fatalities and injuries caused by residential fires is gradually decreasing, but many residential fires caused by cigarettes remain preventable and continue to pose a significant public health problem.² Cigarettes were responsible for 90 800 fires, 610 civilian deaths, 1570 civilian injuries, and $633 million in direct property damages and other economic loss in the United States in 2010,³ and account for an estimated 10% of all fire fatalities worldwide.⁴ Furthermore, 1 in 4 deaths by cigarette-caused fires is not of the smokers themselves.⁵ Thirty-four percent are children of the smokers, 25% are neighbors or friends, 14% are spouses or partners, and 13% are the parents. Persons at increased risk include children aged 4 years and younger^6,7 and adults aged 65 years and older,^6,7 African Americans and Native Americans,^6,7 the poorest of Americans,^7,8 persons living in rural areas^7,9 or in manufactured homes or substandard housing,^10,11 and responding firefighters.²

As early as 1929, Rep Edith Nourse Rogers of Massachusetts called for the National Bureau of Standards to develop technology for a “self-snubbing” cigarette.¹² The Bureau developed such a cigarette, but stated that a manufacturer would have to adopt it.¹³ Later, in 1947, the National Fire Protection Association called on cigarette manufacturers to take some responsibility for the problem of cigarette-ignited fires¹⁴; however, tobacco manufacturers did not respond publicly to either of these appeals. Instead, they opposed legislative efforts to regulate cigarette ignition propensity, while conducting extensive cigarette ignition propensity research and development since the early 1970s.¹⁵

The Cigarette Safety Act of 1984¹⁶ required the creation of an advisory group, which issued a final report in 1987 concluding that developing a cigarette with minimum ignition propensity was technically, economically, and commercially feasible.¹⁷ Subsequently, the National Institute of Standards and Technology developed standards for cigarette ignition propensity testing^18–21 as charged by the Fire Safe Cigarette Act of 1990.²² This testing protocol, presently known as the ASTM E2187-04 Standard Test Method of Measuring the Ignition Strength of Cigarettes,²³ requires observing a lit cigarette placed on 10 layers of standard filter paper in a draft-free environment.²³ A brand is in compliance with the regulation if no more than 25% of cigarettes tested in a trial exhibit burn through their full tobacco column lengths. The State of New York in 2004^15,24,25 was the first jurisdiction to implement cigarette ignition propensity regulations using this standard, which has now been adopted by all 50 US states as well as in Canada, Australia, Finland, South Africa, and the European Union.^26–30

Massachusetts implemented its Fire Safe Cigarette Law (FSCL) on January 1, 2008. In the present study, we utilized data from the Massachusetts Fire Incident Reporting System (MFIRS), a comprehensive, standardized fire incidence reporting system maintained by the Massachusetts Department of Fire Services. The purpose was to evaluate the Massachusetts FSCL’s effectiveness in reducing cigarette-caused fires and associated casualties and to determine whether effectiveness has differed by ignition substrate, origin of fire, or human, temporal, or geographic factors.

METHODS

In 2012, the population of Massachusetts was 6 646 144, of whom 21.1% were aged younger than 18 years, 14.4% aged 65 years and older, 51.5% female, 83.7% White, 7.9% African American alone, 10.1% Hispanic or Latino, and 5.8% Asian alone.³¹ Housing units in 2011 numbered 2 818 940, 41.8% of which were in multiunit structures.³¹ In 2004, 29 462 fires (including 14 006 structure fires) were reported in Massachusetts, in which 52 civilian deaths (32 men, 16 women, and 4 children), 373 civilian injuries, 579 fire service injuries, and $186 million in property damages occurred.³²

Data Source and Management

Fire services departments throughout Massachusetts are required by law to report to the Division of Fire Safety all fires resulting in a fatal or nonfatal human casualty or a dollar loss. More than 95% of fire departments do so by participating in MFIRS. Each fire incident is assigned unique identification numbers in the system, which were used in the present study to link fire incidence and civilian and fire services casualty modules of MFIRS. The circumstances of each fire are recorded with respect to date and time, place, heat source of ignition, and other information and are compiled in a standardized structured database.

The time frame for the analyses in this study was from year 2004 through 2010, beginning with the first year after the reporting system underwent a major conversion and continuing through 3 full years following implementation of the FSCL. We restricted analyses to unintentional residential fire incidents. We excluded intentional or arson fires for having no expectation to be influenced by the FSCL.

We derived the primary outcome variable, cigarette versus other cause of fire, from the MFIRS field heat source. We coded fire incidents for which heat source was listed as “cigarette” or “heat from undetermined smoking material” for analysis as “1” and all other heat sources, including “cigarette lighter,” as “0.” Fewer than 1% of fire incidents were listed as “heat from undetermined smoking material,” and these were assumed to have been predominantly caused by cigarettes (oral communication from Derryl Dion, Massachusetts Department of Fire Services, January 3, 2012). We derived the second outcome variable, cigarette versus other cause of fire involving any casualties, from civilian and fire services casualty severity fields, each consisting of 7 categories, ranging from “report only, including exposure” to “first aid only,” “life-threatening,” and “death.”

Candidate predictor variables, including the main predictor of interest, FSCL, and covariates pertaining to the circumstances of each fire included temporal, spatial, substrate, and human factors. Temporal covariates were FSCL, a dichotomous variable coded as “1” for year of fire incident during years 2008 through 2010 and “0” for fire incident during years 2004 through 2007; chronological time, a continuous variable to control for underlying temporal trend; season; weekday versus weekend; and hour of day (day or night). Spatial factors included the area in which the fire ignited. We coded fires occurring in egress, assembly, or sales areas; function areas; structural areas; exterior balcony or unenclosed porch; and courtyard, patio, terrace, and screened-in porches as “living area” and all other locations as “non–living area.” Substrate factors consisted of the item first ignited. We coded fires ignited on finished materials, furniture, utensils, soft goods, and wearing apparel as substrate “furnishings or soft goods”; we coded all others “non–furnishing or soft goods.” Item first ignited was listed as “undetermined” for 1378 observations, and we treated these as a third category and retained them in the analyses. We created a variable representing the presence or absence of any of the following human factors: asleep, possibly impaired by alcohol or drugs, unattended or unsupervised person, possibly mentally disabled, physically disabled, multiple persons involved, or age. We considered the county of each fire for adjustment as a random effect in mixed effects models.

Statistical Analysis

We performed univariate analyses of the relationship between cigarette versus noncigarette heat source and each of the candidate fixed effects variables by using contingency tables and the Fisher’s exact test. We included all variables that were statistically significantly at P < .05 related to the outcome in the subsequent multivariate analyses. We analyzed the association between FSCL and the odds of a fire having been caused by a cigarette versus a noncigarette heat source by using an interrupted time series logistic regression model to assess the degree to which the FSCL may have affected the outcome immediately or gradually over time. We retained FSCL in all models as well as chronological month to control the underlying linear temporal trend in fire incidence. We computed stratum-specific effects of FSCL for all covariates to explore under which if any circumstances the FSCL may have been the most effective. We used a b- (parameter estimates) and P value–driven backward selection procedure to increase validity of the b_FSCL by maximum control for confounding. We removed covariates that did not change b_FSCL by 10% or more, and whose log-likelihood ratio test P value was .05 or greater from the model. We considered potential collinearity between FSCL and the other covariates by using the Spearman correlation and contingency tables analyses. We tested interactions between FSCL and all predictor variables by using a likelihood ratio test. We examined random effects with likelihood ratio tests.

We examined temporal trends in monthly rates of total, cigarette-caused, and noncigarette-caused fires and association with FSCL by using Poisson regression analyses. We performed these to discriminate between observed changes in the probability of cigarette-caused fires attributable to decreased risk of these fires as opposed to increased risk of noncigarette-caused fires.

We used the same analytic procedures and modeling approach to examine both incidence of cigarette-caused fires and cigarette-caused fires involving 1 or more casualties. We included MFIRS data indicating the reported presence or absence of a smoke detector and sprinkler system, which are believed to affect the risk of casualty fires, as covariates in the casualty–fire models. We used Stata release 11.2, MP version (StataCorp LP, College Station, TX) for all statistical analyses.

RESULTS

Of 30 767 incident fires reported in Massachusetts in 2004 through 2010, 23 413 were classified residential; 17 522 of these fires were reported unintentional, and 4543 as cause undetermined. A total of 1629 (9.30%) of the unintentional residential fires reported were caused by cigarettes or an undetermined smoking material. The mean number and range of unintentional residential fires reported per month were 209 (133–345); 19 (5–44) ignited by cigarettes and 189 (125–326) ignited by another heat source.

On univariate analysis, unintentional residential fires were more likely to have been ignited by cigarettes in the presence of human factors (17.0%); occurring in living areas (10.8%); first igniting on furnishings, bedding, or soft goods (11.0%); and occurring during the day (10.2%), on weekends (9.5%; not statistically significant), and in the spring (12.3%), summer (9.0%), and fall (9.3%; Table 1).

TABLE 1—

Factors Associated With Cigarettes as the Heat Source of Unintentional Residential Fires: Massachusetts: 2004–2010

Covariate	Noncigarette-Caused Fires, No.	Cigarette-Caused Fires, No. (%)	Unadjusted OR (95% CI)	P
FSCL
Post-FSCL	6434	657 (9.3)	1.00 (0.90, 1.11)	> .999
Pre-FSCL	9431	963 (9.3)
Area of origin
Living area	10 382	1253 (10.8)	1.74 (1.54, 1.96)	< .001
Non–living area (Ref)	5326	365 (6.4)	1.00
Undetermined	157	2 (1.3)	0.18 (0.45, 0.75)	.018
Substrate
Furniture, bedding, or soft goods	7713	956 (11.0)	1.44 (1.29, 1.61)	< .001
Undetermined	1302	76 (5.5)	0.68 (0.52, 0.87)	.002
Other (Ref)	6850	588 (7.9)	1.00
Human factors
Yes	2376	485 (17.0)	2.37 (2.11, 2.66)	< .001
No	13 156	1135 (7.9)
Day of wk
Weekend	5725	603 (9.5)	1.05 (0.94, 1.17)	.365
Weekday	10 140	1017 (9.1)
Time of d
Night	9102	854 (8.6)	0.83 (0.75, 0.92)	< .001
Day	6763	766 (10.2)
Season
Winter	4660	358 (7.1)	0.81 (0.69, 0.94)	.006
Spring	4003	564 (12.4)	1.47 (1.27, 1.69)	< .001
Summer	3613	356 (9.0)	1.03 (0.88, 1.20)	.725
Fall (Ref)	3589	342 (8.7)	1.00

Note. CI = confidence interval; FSCL = Fire Safe Cigarette Law; OR = odds ratio.

Massachusetts FSCL and Residential Fires

The monthly rates of total and noncigarette-caused unintentional residential fires displayed statistically significant, slightly decreasing, trends (incident rate ratio [IRR] = 0.998; 95% confidence interval [CI] = 0.997, 0.999) over the full study period, and no changes in association with the FSCL (total fires; P = .155; and noncigarette-caused fires; P = .628). By contrast, the monthly rate of cigarette-caused fires decreased by 26% (IRR = 0.738; 95% CI = 0.611, 0.892) following the FSCL, with control for an underlying increasing longitudinal trend (IRR = 1.005; 95% CI = 1.000, 1.009) in these fires over the full study period.

The best-fitting multivariate logistic regression model included all fixed-effect candidate predictor variables except for weekday versus weekend (P = .419) and a random effect for county level (variance = 0.045; SE = 0.24; Table 2). The FSCL was associated with a 28% (95% CI = 12%, 41%) reduction in the odds that an unintentional residential fire was caused by a cigarette versus another heat source. A sensitivity analysis showed no statistically significant difference in the result when fires listed as “heat from undetermined smoking material” were coded as noncigarette- rather than cigarette-caused fires. The interaction between FSCL and chronological month was not statistically significant (P = .582), indicating that the reduction in odds occurred abruptly and with no change in temporal trend. The odds of cigarette-caused fires were greatest in fires in which human factors were involved (odds ratio [OR] = 2.11; 95% CI = 1.87, 2.38), that occurred in living areas compared with non-living areas (OR = 1.56; 95% CI = 1.38, 1.77), or were first ignited on furniture, bedding, or soft goods as opposed to other substrates (OR = 1.47; 95% CI = 1.32, 1.64; Table 2) Cigarettes were also less likely to be the heat source of fires that occurred at night (OR = 0.87; 95% CI = 0.79, 0.97), in the winter (OR = 0.84; 95% CI = 0.72, 0.99), and that were first ignited on substrates that were undetermined as opposed to other nonfurniture, bedding, or soft-good substrates (OR = 0.72; 95% CI = 0.56, 0.93; Table 2).

TABLE 2—

Odds Ratios for Cigarette-Caused Fires After Compared With Before Massachusetts FSCL and Covariates Adjusted in Multivariate Logistic Regression: 2004–2010

Covariate	OR (95% CI)	P
FSCL, post-FSCL: pre-FSCL	0.72 (0.59, 0.88)	.001
Chronological time, per y	1.11 (1.06, 1.17)	< .001
Area of origin
Living areas: non–living areas	1.56 (1.38, 1.77)	< .001
Undetermined: non–living areas	0.22 (0.05, 0.91)	.37
Substrate
Furniture, bedding, or soft goods: other	1.47 (1.32, 1.64)	< .001
Undetermined: other	0.72 (0.56, 0.93)	.011
Human factors, yes: no	2.11 (1.87, 2.38)	< .001
Time of day,a night: d	0.87 (0.79, 0.97)	.01
Days of week,b weekend: weekday	1.05 (0.94, 1.16)	.419
Seasonc
Winter: fall	0.84 (0.72, 0.99)	.033
Spring: fall	1.57 (1.35, 1.81)	< .001
Summer: fall	1.10 (0.94, 1.29)	.241

Note. CI = confidence interval; FSCL = Fire Safe Cigarette Law; OR = odds ratio.

^aNight (6:00 PM to 5:59 AM), day (6:00 AM to 5:59 PM).

^bDays of week not included in final model.

^cWinter = December through February; spring = March through May; summer = June through August; Fall = September through November.

The greatest reductions by FSCL of the odds of cigarette-caused fires observed in exploratory analyses were fires in which human factors were involved (39%; 95% CI = 20%, 54%); first ignited on furniture, bedding, or soft goods (37%; 95% CI = 22%, 50%); occurring in the summer (34%; 95% CI = 12%, 50%) or winter (32%; 95% CI = 10%, 49%); and that occurred in living areas (29%; 95% CI = 12%, 42%; Figure 1).

FIGURE 1—

Adjusted odds ratios of cigarette-caused fires before and after Massachusetts Fire Safe Cigarette Law, by fire circumstances: 2004–2010.

Note. Whiskers indicate 95% confidence intervals.

Massachusetts FSCL and Residential Cigarette Fire Casualties

Of the 17 522 unintentional residential fires, 1782 (10.1%) involved 1 or more civilian or fire services casualties. The mean number and range of unintentional residential casualty fires per month over the full study period was 35 (10–78); 3 (0–16) caused by cigarettes and 32 (10–62) caused by another heat source. As observed for unintentional residential fires overall, fires with 1 or more casualties were more likely to have been ignited by cigarettes in the presence of human factors (21.5% vs 7.5%; P < .001); occurring in living areas (13.2% vs 5.0%; P < .001); first igniting on furnishings, bedding, or soft goods (14.6% vs 7.0% and 6.4% undetermined; P < .001); and occurring during the day (12.0% vs 9.9%; P = .062), and in the spring (14.3%) or summer (11.1%), and least likely commonly in the fall (8.4%; P = .029). The monthly rates of total unintentional residential fires with casualties over the entire study period decreased very gradually (IRR = 0.998; 95% CI = 0.997, 1.000), but neither trends in the monthly rates of cigarette-caused (P = .616) nor noncigarette-caused fires with casualties (P = .054) reached statistical significance.

The best-fitting multivariate logistic regression model of unintentional residential casualty fires showed no statistically significant association with the FSCL (OR = 0.97; 95% CI = 0.53, 1.79). The odds of cigarette-caused casualty fires were greatest in fires in which human factors were involved (OR = 2.86; 95% CI = 2.07, 3.93), that occurred in living areas (OR = 2.28; 95% CI = 1.45, 3.57), and that first ignited on furnishings, bedding, or soft goods as opposed to other substrates (OR = 2.41; 95% CI = 1.68, 3.44). We did not observe a statistically significant temporal trend in the odds of cigarette-caused fires (P = .538), and we did not observe statistically significant random effects of county. We observed 1 statistically significant stratum-specific association of FSCL with cigarette-caused casualty fires, an increase in the odds of casualty fires igniting on materials other than furnishings, bedding, or soft goods (OR = 2.23; 95% CI = 1.20, 4.13; Figure 2).

FIGURE 2—

Adjusted odds ratios of cigarette-caused casualty fires before and after Massachusetts Fire Safe Cigarette Law, by fire circumstances: 2004–2010.

Note. Whiskers indicate 95% confidence intervals.

DISCUSSION

The Massachusetts FSCL appears to have decreased the likelihood of unintentional residential fires caused by cigarettes by 28% and the monthly rate of these fires by 26%. The circumstances in which the FSCL appeared to have the greatest protective effect were the same as those most likely to be cigarette-caused: fires in which human factors were involved; that were first ignited on furniture, bedding, or soft goods; or that occurred in the living areas of the home. Similar declines were not observed in analyses restricted to cigarette-caused fires involving casualties, possibly because in part of the small sample size (< 10% of fires).

The National Institute of Standards and Technology–developed standard is not expected to differentially affect casualty versus noncasualty fires. Increased use of smoke detectors and sprinklers, which reduce the risk of injury and death from fires overall, would not be expected to differentially affect cigarette- versus noncigarette-caused fires.

Despite some limitations, many methodological challenges for assessing the effectiveness of the FSCL were effectively addressed in this study, notably the underlying trends in fire incidence and potential confounding attributable to concurrent changes among other fire risk factors. We controlled temporal, spatial, substrate, and human factor circumstances, reducing these potential sources of confounding. Relevant historical events, such as stricter upholstery regulations, are not known to have occurred in Massachusetts during the study period. Restriction of analyses to years after the statewide smoke-free law came into effect³³ minimized potential confounding from any increase in home or outdoor smoking. Increased use of flame retardants on the other hand is, however, a potential source of residual confounding.

Multiple comparisons in the stratum-specific examinations of FSCL effect may have increased type I error in these exploratory analyses. Selection bias related to determination of intentionality, which was missing in 18% of fire incidents, is a possibility. However, the most likely consequence would be to shift the result in the direction of no FSCL effect. This is because cigarette-caused fires are rarely intentional and improved documentation of intentionality of cigarette-caused fires following the FSCL is more likely than fires caused by other common heat sources such as matches and lighters with increased awareness and identification of cigarettes as a heat source.

The near-total participation rate in MFIRS among Massachusetts fire services departments ensured that the data are highly representative. No differences are theorized to suggest variation across states, which all share the same ignition propensity performance standard. Mandatory compliance by law minimizes potential exposure misclassification. Early introduction of reduced ignition propensity cigarettes, substitution by imported or Internet sales of cigarettes, or increased fire risk behaviors, would each be expected to have decreased the apparent effect. As fire services departments throughout the state receive standardized training for MFIRS reporting, any residual variation would be expected to favor the null hypothesis.

Our original assessment of the early experience with reduced ignition propensity standard in New York demonstrated the technical, economical, and commercial feasibility, contrary to previous industry claims.¹⁵ Our further research found no change in usual fire-risk ameliorating behaviors among cohorts of smokers,³⁴ and no significant changes in fire-risk behaviors, fire events, quitting intentions, or smoking topography, except that after the law was implemented smokers were less likely to report smoking more than a pack of cigarettes per day or inhaling deeply into the chest when smoking.^35,36 Smokers were also more likely to report after the law’s implementation that their cigarettes extinguished between puffs.³⁵

This study is the first comprehensive evaluation of the population effectiveness of legislated fire-safe cigarette standards. The findings confirm the National Institute of Standards and Technology’s expectations that the reduced ignition propensity standard would substantially decrease the incidence of cigarette-caused fires. Although a reduction in fire casualties has not yet been demonstrated, jurisdictions with fire-safe cigarette laws should seriously consider upgrading the standard where possible, for example by decreasing the number of filter papers serving as the test substrate or varying other laboratory conditions. Manufacturers have also developed other, potentially even more effective, technologies to reduce ignition propensity, which we have reviewed elsewhere.³⁷ On the basis of the low cost of reduced ignition propensity cigarette construction, the high costs of cigarette-caused fires, and this study’s findings, cigarette manufacturers should demonstrate responsibility for the consequences of their products and voluntarily adopt reduced ignition propensity standards globally as is required by all 50 US states, Canada, Australia, Finland, South Africa, and the European Union. Cigarettes remain an important, preventable cause of residential fires and associated morbidity and mortality. Continued surveillance and monitoring of fires and fire losses caused by cigarettes is essential to evaluate the success of existing policies and the potential need for adjusting the standards adopted.