Methods

United States National Health Interview Survey (NHIS) adult (ages 18+) interviewees from 1990 and 1991 were followed up for vital status through 1995 using the National Death Index (NDI).⁶ The NHIS, NDI, and linkage methods have been described in previous studies.^7,8 This provided an average of five years of virtually complete⁹ follow up on a nationally representative cohort of Americans. The lack of seat belt and alcohol use data precluded use of NHIS data from years other than 1990 and 1991.

Participants were classified as never smokers (fewer than 100 lifetime cigarettes), ex-smokers, and current smokers (smokers), by baseline self report. For smokers, cigarettes per day were recorded into 1–14, 15–24, and 25+ cigarettes/day categories.

Injury deaths include International Classification of Diseases, 9th revision (ICD-9) codes from E800–E999. Injury deaths were further categorized into unintentional injury (ICD-9 E800–E949), motor vehicle crash (ICD-9 E810–E825), suicides (ICD-9 E950–E959), and homicides (ICD-9 E960–E969).

We adjusted for several injury risk factors. Demographic risk factors included age (five year age groups), race (white, black, or other), and gender. Behavioral risk factors were education (0–11, 12, 13–15, or 16+ (college graduate or more) years), seat belt use (always, sometimes, or never), marital status, body mass index, and alcohol use (0, 0.01–4.9, 5.0–14.9, 15.0–24.9, and >25 g/day of alcohol (given 13.4 g of alcohol/drink).

Stata software's Cox proportional hazards regression analysis was used to estimate relative risks of injury death among current smokers.¹⁰ NHIS final weights and primary sampling units were incorporated into the calculations. Sex was stratified in the demographic and fully adjusted calculations. The test for trend was calculated using 0 cigarettes/day (never + ex-smokers), 1–14 cigarettes/day, 15–24 cigarettes/day, and 25+ cigarettes/day.

The smoking attributable fraction of injury death equals ((RR−1)*Prevalence)/ (1+((RR−1)*Prevalence)); RR = relative risk. The estimated smoking prevalences were 70% for the homeless^11,12 and 24.7%¹³ for household members.

Results

Compared with never and ex-smokers, smokers were younger, less educated, less consistent users of seat belts, more often divorced, and consumed more alcohol (table 1). Compared with never smokers, smokers' crude RR of injury death was 1.86 (95% confidence interval (CI) 1.30 to 2.66) (table 2). After adjustment for sex, age, and race, the RR was 1.60 (CI 1.12 to 2.29). The fully adjusted smoker injury RR was 1.42 (CI 0.99 to 2.05) with a significant dose response trend (p<0.030). The above crude, demographically, and fully adjusted RR correspond to smoking attributable fractions for injury death of 18%, 13%, and 9%, respectively, for the estimated 99.8% of Americans who have homes.¹⁴ The overall (homeless + housed) American smoking attributable fractions are the same as the above housed smoking attributable fractions, unless smoker injury death RRs are higher for homeless (v housed) Americans and/or there is a 13-fold higher injury death risk averaged across the homeless compared with the housed. Assuming the homeless and housed have the same crude and adjusted injury mortality RR, the smoking attributable fractions of homeless injury deaths are 38%, 30%, and 23%, respectively.

Table 2 shows a crude increased risk of death among smokers for unintentional injury, motor vehicle crashes (crashes), and suicide. In the fully adjusted analysis smoker RRs of unintentional injury, crash, and suicide were 1.56 (CI 1.01 to 2.41), 1.83 (CI 0.97 to 3.44), and 1.36 (CI 0.60 to 3.12), respectively.

Discussion

We find that in the early 1990s, smokers in the United States had dose-response excesses of injury death, independent of smokers' excesses of divorce and heavy alcohol use and deficits of education and seat belt use. Smokers also had significantly raised unadjusted RRs of fatal unintentional injury, crash, and suicide, though the adjusted RR were not statistically significant due to low numbers of cases studied, due to lack of association, or confounding. These associations are consistent with published suggestive (p=0.10 to p<0.20) associations between smoking and injury, unintentional injury, and suicide death seen in randomized trial data.³

In this national sample, smokers' risks of injury death are similar to the relatively high RRs seen in Norwegian or Swedish general population samples.^2,15,16 Those injury risks are modestly higher than smokers' injury, suicide, or violence death RR seen in socioeconomically advantaged populations such as the United States Cancer Prevention Study-I (CPS-I) volunteers,¹⁷ British physicians, nurses in the United States, or Northern California Kaiser Permanente health examinees.² It seems possible that smoking confers a higher injury death RR in poor or addicted⁵ people included in general population samples. The NHIS may underestimate the United States smoking/injury RR since the NHIS is based on households and thus omits homeless people who may be very prone to smoking and injury.^18,19 Correctly estimating national and global RR of injury caused by smoking is important since each 0.25 added to the global smoking attributable injury RR, means an additional about 1% of the global burden of disease is due to smoking.²

This study has strengths. Inference from this study is strengthened by the multiple demographic and behavioral factors we adjusted for, and the recent data representative of households of a large and diverse nation.

This study has weaknesses. A modest number of injury deaths, and even lower numbers of suicide, unintentional, and homicide deaths were available for study. Our methods may substantially overestimate smoking's real injury risks due to our lack of adjustment for drug use, biochemically determined alcohol use, binge or similarly risky patterns of alcohol use, personality, and other possible explanations for the smoking/injury association. However previously reviewed forensic,²⁰ randomized trial, challenge-rechallenge, animal, and some cohort and cotwin control evidence suggests a smoking/injury association is likely to exist, independent of those factors.^2,3,21

Alternatively, we may have substantially underestimated United States smoking attributable injury risks. That could result from omitting the homeless from our sample, counting baseline smokers who quit during follow up as smokers; adjusting for divorce and possible effects of nicotine addiction, and ignoring smoking's roles in fires,²² massive explosions,²³ crashes,²⁴ and other threats to bystander never smokers. Given the high levels of substance abuse seen in the homeless,²⁵ their smoking injury death RR could be over 2.5, as was seen in substance addicts.⁵

Additional research is merited. The various smoking/injury associations should be assessed in other nations, especially from Asia, Latin America, and Africa, at other stages of the tobacco epidemic. The smoking/injury association may be assessed in the homeless and then included in national estimates of smoking/injury associations. The role of smoking's fires, crashes, and stressors (illnesses from secondhand smoke and smoking attributable bereavements) in injuries in never smokers may be assessed. Smoking involved crashes may be studied in the same manner as alcohol involved crashes.²⁶ The contributions of smoking or nicotine addiction to behavioral risk factors for injury such as alcohol use,²⁷ drug use,^5,28 social isolation,²⁹ and inability to advance in school³⁰ may be assessed. This would help assess whether it is appropriate to adjust the smoking/injury association for such possible effects of tobacco use.

Implications for prevention

In summary, there is a very substantial risk that smoking is contributing to many injuries and resulting costs in the United States and globally. Taxes, cigarette pack graphic warnings,³¹ and other methods of reducing smoking may be highly cost effective³² or possibly cost saving^20,33 ways of reducing injury. Further studies and warnings of smoking's injury risks are merited. Smokers, physicians, policy makers, taxpayers, and non-smoking bystanders may benefit from warnings about the risks of injury associated with smoking.