Details of study identification, data collection and event verification for studies in the APCSC are described elsewhere [12]. Briefly, studies were included if they had continued follow-up for at least 5,000 person-years and had recorded vital status at the end of follow-up. Studies were excluded from APCSC if entry to that study was dependent upon having a particular medical condition; for this current paper only, studies were additionally excluded if they did not record deaths due to UADTC. We restricted our analyses to individuals aged 20 or more years at enrolment. Mortality was classified according to the 9th Revision of the International Classification of Diseases (ICD): UADTC was selected as codes 140–149 (lips, tongue and pharynx; this includes: malignant neoplasms of lip, tongue, major salivary glands, gum, floor of mouth, other and unspecified parts of mouth, oropharynx, nasopharynx, hypopharynx, other and ill-defined sites within the lip), 150 (esophagus), and 161 (larynx). Studies were classified as Asian if their participants were recruited from mainland China, Hong Kong, Japan, Korea, Singapore, Taiwan or Thailand; and ANZ if from Australia or New Zealand.

All data on cigarette smoking and alcohol drinking were based on self-report at the time of entry into each of the included studies (‘baseline’). Data recorded on smoking included the number of cigarettes currently smoked per day, categorized here into groups of < 20 and ≥ 20 (20 cigarettes corresponds to one standard pack), and whether or not non-smokers were former smokers. Alcohol was only recorded as current drinking or not. Participants were included in this study only if they reported their current and past smoking and current alcohol drinking at baseline. In APCSC, 94% of the 600,445 participants recorded their current smoking status, 92% recorded their current drinking status and 91% reported both.

Cox proportional hazard models, stratified by study and sex, and adjusted for age, with and without additional adjustment for alcohol, were used to estimate hazard ratios (HRs) associated with smoking (current v never, and dose-response) for death due to UADTC and its subtypes. We also estimated HRs for overall UADTC by sex, geographical area (Asia/ANZ) and alcohol subgroups, and tested for effect modification by including interaction terms in the Cox models [13]. In order to avoid reporting unreliable results when the number of events was small, estimates of relative risk for UADTC subtypes are not reported for these subgroups. The effect of quitting smoking was estimated through HRs comparing ex-smokers with current smokers. All analyses used individual participant data.

The population attributable risk for mortality from UADTC associated with smoking were calculated for each country, for both males and females, using previously published recent prevalence estimates of the prevalence of smoking [10] and the formula [13]: PAR = [P (RR-1)]/ [1 + P (RR-1)]. Here, the RR was that found from APCSC.

Current smokers had more than twice the risk of mortality from UADTC, compared with never smokers: the HR was 2.36 (95% confidence interval, CI: 1.77 – 3.15), adjusted for age. The HR stayed the same after further adjustment for alcohol (Figure 1). For each subtype of UADTC, smoking was associated with a significant increase in the risk of mortality compared with never smoking. After adjusting for age and alcohol, the HR was: for lip, oral cavity or pharyngeal cancer, 2.14 (95% CI: 1.43 – 3.19); esophageal cancer, 2.84 (95% CI: 1.72 – 4.68); laryngeal cancer, 2.59 (95% CI: 1.07 – 4.68). The association between smoking and overall UADTC death was much stronger in ANZ compared with Asia, and in women compared with men (Figure 1).

The HR for current drinking, compared with not, was 1.33 (95% CI: 1.03 – 1.73). However, after adjusting for current smoking, alcohol had very little association with the risk of mortality from UADTC: 1.01 (95% CI: 0.80 – 12.7). Current smoking had a significant effect among both drinkers and non-drinkers (Figure 1). Although the effect of smoking was considerably higher among those who consumed alcohol, there was no strong evidence of an interaction (p = 0.10), perhaps due to lack of information on the level of alcohol consumption.

Overall, 48,134 individuals (62% from Asia) reported at study baseline that they had quit smoking. During follow-up, a total of 70 (40% from Asia) former smokers died of UADTC. The HR for mortality from UADTC, comparing ex-smokers with current smokers, was 0.92 (95% CI: 0.66 – 1.27), adjusted for age and alcohol drinking. This HR differed by subtypes of UADTC: for lip, oral cavity or pharyngeal cancer, 0.99 (95% CI: 0.58 – 1.69); esophageal cancer, 1.03 (95% CI: 0.66 – 1.63); and laryngeal cancer, 0.33 (95% CI; 0.11 – 0.98).

Of the 26 studies, 17 reported the number of cigarettes smoked per day at baseline, involving 231,562 study participants. There was evidence of a dose-response association between the number of cigarettes smoked per day and mortality from UADTC and its subtypes (Table 2). Individuals who smoked more than 20 cigarettes a day had approximately three times the risk of dying from UADTC compared with never smokers. However, there was evidence of a sex difference in the strength of this association: the HR for smoking ≥ 20 cigarettes per day compared with never smoking was 2.27 (95% CI: 1.77 – 2.90) in men versus 21.1 (95% CI: 9.02 – 49.4) in women, after adjusting for age and alcohol; (Figure 2A). Similarly, significant differences were found between the regions (Figure 2B); the equivalent HRs were 5.03 (95% CI: 2.70 – 9.36) in ANZ versus 2.30 (95% CI: 1.77 – 2.99) in Asia.

Figure 3 presents the population attributable fraction for males and females for UADTC caused by smoking for each country in the WHO Western Pacific and South-East Asian regions. The fraction of UADTC attributable to smoking ranged from 18–47% in males and from 1–45% in females.

Analyses from this large dataset, with almost 400 deaths, confirm that smoking is strongly associated with the risk of mortality from UADTC in both Asia and ANZ. There was a significant dose-response association between the average number of cigarettes smoked daily and the risk for all subtypes of UADTC. Smoking had a stronger association with the risk of mortality from UADTC among female, compared with male, smokers. Although not statistically significant, the current study supports previous observations that concurrent alcohol consumption may exacerbate the effect of smoking on the risk of UADTC [7,14]: amongst non-drinkers, smoking doubled the risk of UADTC, whereas amongst drinkers it trebled the risk.

The findings of this study are consistent with earlier epidemiological evidence [15–23], but the estimated hazards are smaller than in some studies [18–23], especially those conducted in Western countries [20–23]. This may be due to differences in study design and the degree to which non-fatal events are included. But it may also signify a regional difference, since the Asian studies contributed 87% of participants, and 74% of events in APCSC, whilst the ANZ hazard ratio was significantly higher in this study. The lower excess risk for UADTC associated with smoking in Asia is likely to be due to the relative immaturity of the smoking epidemic across Asian countries [24], especially in China, where the habit only reached its peak in the mid-1990’s [25–29]. Other potential contributing factors to the smaller relative risks associated with smoking observed in the Asian cohorts are the greater exposure among non-smokers to indoor air-pollution from wood-burning in homes [30] combined with extremely high levels of passive smoking, particularly in China where an estimated 450 million individuals are so exposed [31].

The APCSC estimates suggest a greater risk of mortality from UADTC among female smokers compared with male smokers, in agreement with a recent study in Japan [32], but opposite to the finding of the Swedish Smoking Survey [33], although the latter was based on few events (11 in women and 33 in men). Several epidemiological studies have found a statistically significant increase risk for lung cancer among female smokers compared with male smokers, although others have found no difference [34–39]. It is currently unclear which mechanisms mediate the reported increased susceptibility among female smokers to cancers of the lung and upper-aero digestive tract. Potential explanations include sex differences in smoking behaviours (such as the degree of inhalation, for example) or differences in the accuracy of self-reporting of smoking consumption between men and women. Alternatively, a role for female sex hormones in increasing the susceptibility of women to tobacco carcinogens has been suggested [34–38]. Clearly, further research is required to determine whether the difference in risk between men and women is real and if so, what is driving it. Answers to these two questions could have repercussions for the development of sex-specific smoking cessation programs, particularly in countries where the prevalence of smoking is on the increase in young women [40].

In contrast with an overview of studies that reported a reduction in risk of UADTC among former smokers compared with current smokers of 30% to 50% in the first five years and up to 70% after 10 years of quitting smoking [41], data from the current study failed to demonstrate a significant reduction in risk of dying from UADTC among former smokers compared with current smokers, except for laryngeal cancer. We believe that our inability to find a protective effect of quitting overall is most likely an artifact of the current situation in Asia where quitting is unusual, and those who have quit will have abstained for a relatively short time. However, differences in the distribution of subsidiary site cancers within the overall definition of UADTC could also play a role.

The main strength of the current analysis is the large number of events, which facilitated our ability to assess associations in subgroups. However, we were limited by lack of information on duration of smoking and age of uptake of smoking, which may well be important determinants of the risk of smoking cigarettes for UADTC. Our data on alcohol drinking were also limited to baseline measures of current drinking and also to whether or not the subject drank; there is evidence [7,14] of a dose-response relationship between the amount of alcohol consumed and UADTC. Thus, we may have underestimated the true effect of drinking; there may be residual confounding due to our lack of data on the frequency of alcohol use. Furthermore, we had no data on changes in smoking and drinking status during follow-up. Although our analyses were restricted to deaths, the low five-year survival from UADTC [2–4], and the lack of systematic differences in excess risk found between previous studies that did [15,16,18,21–23] and did not [17,19,20] analyse non-fatal, as well as fatal, UADTC events, suggests that this has not introduced bias.