Background:
Proportional mortality studies using reliable death registration and smoking data can be used to assess the effects of smoking and quitting. Deaths from smoking-induced causes are treated as cases, but the selection of other causes to define controls differs. We conducted a proportional mortality study on smoking and lung cancer using four definitions of controls.

Methods:
We included 81% of all deaths aged 30+ years in 1998 in Hong Kong. Cases were lung cancer deaths (N=2909). We defined controls as deaths from non-smoking related causes according to
i) Liu (Mainland) 1998 (n₁=4898),
ii) Alam (India) 2013 (n₂=2944),
iii) Sitas (South Africa) 2013 (n₃=6076) and
iv) US Surgeon General´s Report 2014 (USSG) (n₄=4365).
Logistic regression yielded AORs of lung cancer, adjusting for sex and age at death.

Results:
In middle age (35-64 years), AOR (95% CI) for lung cancer was 2.79 (1.58-4.92) in ex-smokers (5+ years) and 2.65 (1.87-3.78) in current smokers (p_trend: 0.26) using controls based on Liu. The corresponding AORs were 1.87 (1.00-3.56) and 2.37 (1.58-3.58) (p_trend: 0.009) for Alam, 2.40 (1.42-4.04) and 4.31 (3.00-6.24) (p_trend: < 0.0001) for Sitas, and 1.99 (1.14-3.46) and 3.39 (2.31-5.01) (p_trend: < 0.0001) for USSG. In old age (65-84), the corresponding figures for Liu, Alam, Sitas and USSG were 3.31 (2.54-4.34), 5.51 (4.26-7.17), 3.27 (2.44-4.41) and 5.40 (4.04-7.24) in ex-smokers, and 3.39 (2.63-4.37), 5.38 (4.23-6.87), 3.19 (2.44-4.18) and 5.35 (4.12-6.97) in current smokers (all p_trend: < 0.0001).

Conclusions:
The effect size of smoking on lung cancer using different controls was similar in old age but varied in middle age. Benefits of quitting were observed in all except Liu's controls. For proportional mortality studies, the choice of the most appropriate causes of deaths to define controls needs to be cautious, as the effects or lack of effects of smoking on different diseases vary in different populations.