Robin Hanson has questions
We usually see strong correlations between death and smoking, and we see those same correlations within each random arm (i.e., group) of a randomized trial. Nevertheless, we see no significant net death differences between control arms and arms induced to smoke less.
So we don’t have clear evidence that smoking kills on net; it could be that most or all of the death-smoking correlation is due to selection effects, and not smoking causing death
Digging through the literature this question is shockingly open, but I did manage to track down at least one study that shows a reduction in all cause mortality from smoking cessation.
Background: Randomized clinical trials have not yet demonstrated the mortality benefit of smoking cessation. (KS: This study was completed in 2005!)
Objective: To assess the long-term effect on mortality of a randomly applied smoking cessation program.
Design: The Lung Health Study was a randomized clinical trial of smoking cessation. Special intervention participants received the smoking intervention program and were compared with usual care participants. Vital status was followed up to 14.5 years.
Setting: 10 clinical centers in the United States and Canada.
Patients: 5887 middle-aged volunteers with asymptomatic airway obstruction.
Measurements: All-cause mortality and mortality due to cardiovascular disease, lung cancer, and other respiratory disease.
Intervention: The intervention was a 10-week smoking cessation program that included a strong physician message and 12 group sessions using behavior modification and nicotine gum, plus either ipratropium or a placebo inhaler.
Results: At 5 years, 21.7% of special intervention participants had stopped smoking since study entry compared with 5.4% of usual care participants. After up to 14.5 years of follow-up, 731 patients died: 33% of lung cancer, 22% of cardiovascular disease, 7.8% of respiratory disease other than cancer, and 2.3% of unknown causes. All-cause mortality was significantly lower in the special intervention group than in the usual care group (8.83 per 1000 person-years vs. 10.38 per 1000 person-years; P = 0.03). The hazard ratio for mortality in the usual care group compared with the special intervention group was 1.18 (95% CI, 1.02 to 1.37). Differences in death rates for both lung cancer and cardiovascular disease were greater when death rates were analyzed by smoking habit.
So it does seem that smoking in and of itself kills. What’s truly scary to me is how seemingly difficult it is to get these studies done. The public good here is clear, knowledge about what does and does not cause cancer, heart disease, stroke, is a non-rivalous and non-excludable input into health.