Predictions of Longevity within the Smoking Population of the United States
12 Aug 2013
Reading time ~5 minutes
of smokers have been studied in many counties in order to understand other external influences and effects of covariates of population demographics [i][ii][iii]. An article by P. Jha et al. in the New England Journal of Medicine showed the hazards of smoking while adjusting for variables of education, alcohol consumption, and adiposity within the population of the United States. The findings in the article were fascinating because it also analyzed the mortality of individuals that have quit smoking.
We assessed the hazard model that was used in the article and found that improvements can be made in covariates of education, adiposity and levels of alcohol consumption. Since the study considers the population of those that have quit smoking, it would be appropriate to study the quality of life in terms of dependence on social assistance.
We did not use the income variable as there was a tendency for bias within the lower income brackets[iv]. Pre-analysis indicates that there are higher rates of missing income information within groups that relies on welfare support and food stamps. Thus substitution with a variable for the usage of welfare and food stamps for quality of life is sufficient. Covariates of education was reconsidered to be stratified because we do not expect linear relations with mortality. The knowledge level of the health risks of smoking will be different in elementary school, secondary School, and post-secondary education.
Modeling with adiposity variable, BMI, requires a more complicated approach to account for the non-linear relation with health[v]. Recent studies have shown that slightly overweight individuals may have an advantage over underweight individuals[vi].The relation between BMI and smoking status are also ambiguous since smoking status have been shown to influence BMI[vii].
When determining the hazard ratios for current smokers and former smokers versus non-smokers, we discovered that confidence intervals were reduced with stratification of BMI and education variables (figure 1). We saw further reduction and improvements when the variable for social status was incorporated.
A crude way of estimating the probability of survival for current smokers compared to non-smokers was used, which will helped us understand the reduction in longevity due to smoking. The results of our model gives hazards ratio and probabilities but there has to be a better way to communicate these complex statistical findings to transform probabilities into likelihood of the years reduced (Figure 2). Using our revised model, we were able to determine that in general, smoking will result in a reduction of 10 years for females and 9 years for males, compared to 11 years for females and 12 years for males in original study.
Thanks to the large data size, the results on benefits of cessation from subcategorizing the age when quitting smoking is reliable. It is evident that earlier a smoker quits, the better it is so this analysis may seem redundant. However an important factor in achieving success in life goals is whether the goal is time bounded.
Using the same technique to determine the reduction in longevity, we determine the years gained by complete cessation, with the analysis of closeness of the probability curves of former smokers compared with never smokers. By analyzing the years gained within each age bucket of time when one quits smoking, we found that quitting at the age of 40-50 has the most impact on health (figure 3). Better yet, our findings should that survival curves of those that quitting smoking before the age of 25 is close to being identical to those that have never smoked.
Although it is exciting to learn that our health is not significantly impacted if we are able to quit smoking before the age of 25, the findings in this study are based on past events. Society is constantly changing and at non-linear rates. The most that this study can offer is a historical summary of the population of United States and the figures should be used with considerations.
Data was collected from the U.S. National Health Interview Survey between the years of 1997 and 2004[viii]. In the following years up until 2006, death records of interviewed individuals were collected from the National Death Index [ix]. Data Analysis was done using SAS and Stata software with similar adjustments to the P. Jha paper.
All support and assistance from Professor Paul Grootendorst is greatly acknowledged, an additional acknowledgement goes to Professor Alison Gibbs and the SAS company for the free year subscription to the statistical program SAS.
i)Jha P, Landsman V, et al. 21st-Century Hazard of Smoking and Benefits of Cessation in the United Stata.N Engl J Med 2008; 4:341-250.
ii)Basavaraj S. Smoking and loss of Longevity in Canada. Canadian J of Public Heath-Revue 1993;3:341-345.
iii)Pirie K, Peto R, et al. The 21st-Century Hazard of Smoking and Benefits of Stopping: A prospective study of one million women in the UK.LANCET 2013; 9861:133-141.
iv)Turrell G. Income non-reporting: implications for health inequalities research. J Epidemiology and Community Health 2000;54:207-214.
v)De Gonzales AB, et al.Body-Mass Index and Mortality among 1.46 Million White Adults. N Engl J Med 2010; 363:2211-2219.
vi)Curtis JP, et al. The Obesity paradox – Body mass index and outcomes in patients with heart failure. Archives of Internal Med 2005;165:55-61.
vii)Sneve M, Jorde R. Cross-sectional study on the relationship between Body mass index and smoking, and longitudinal changes in body mass index in relation to change in smoking status: The Tromso Study. Scandinavian J of Pub Health 2008; 36:397-407.
viii)National Health Interview Survey. Hyattville, MD; National Center for Health Statistis, May 2013 (http://www.cdc.gov/nchs/nhis.htm).
ix)National Health Interview Survey (1986-2004) Linked Mortality Files, mortality follow-up through 2006: matching methodology. Hyattville, MD: National Center for Health Statistic, May 2013 (http://www.cdc.gov/nhis/data/datalinkage/matching_methdology_nhis_final.pdf).