Alcohol and pancreatic cancer: an example of how alcohol-caused deaths are greatly under-estimated globally?

Expert committees are often called upon to adjudicate which causes of death, injury or illness meet criteria for being caused by alcohol use. We have participated in these discussions, whether under the auspices of the World Health Organization (WHO) or as members of groups developing national drinking guidelines. Such expert groups face the challenge of determining whether reduced risks often observed among low-volume drinkers compared to abstainers in relation to some serious diseases are genuine or spurious.

Pancreatic cancer is one such condition. Globally, it is the 6th leading cause of cancer death, accounting for approximately 1 in 20 cancer deaths [Sung et al., 2024]. This high mortality rate reflects its particularly poor prognosis. Some individual cohort studies and systematic reviews have concluded that low-volume alcohol use may be associated with lower risk of pancreatic cancer. Pancreatic cancer is not included in the WHO list of alcohol-caused cancers [IARC, 2010], likely because of these findings, despite plausible mechanisms for causation (e.g. cell damage from acetaldehyde, alcohol’s first metabolite).

But are these apparent protective effects real or might they be the result of the well-known “sick quitter” effect? As people age, they tend to stop or cut down their alcohol use, especially if they run into health problems. The great majority of cohort studies looking at the health impacts of alcohol ignore the bias this potentially creates. The risk of later death or ill health for someone drinking at the beginning of one of these studies is contrasted with the same risk for someone who is a current abstainer, regardless of whether they drank in the past. This means that, especially in studies of older people, the comparison group of “abstainers” has filled up with less healthy individuals who are then compared with those well enough to continue drinking. This can lead to an obvious and substantial bias. Could this be the case with pancreatic cancer?

Our recent study

We recently published an updated review of 37 studies from around the world, comparing pancreatic cancer risk across different levels of alcohol consumption against so-called “abstainers”. [Zhao et al., 2026]. Only three studies met strict criteria for having excluded former drinkers from the “abstainer” reference group. In these studies, pancreatic cancer incidence or mortality risk for drinkers was 41% higher than for drinkers in the other 34 studies. Ten studies took partial steps to remove former drinkers from the abstainer reference group. Incidence or mortality risk in these studies was 19% higher than in the other 27. Furthermore, it was only in these 27 highly biased studies that significant protection against pancreatic cancer appeared – falsely, we concluded.

We also used several other approaches to address potential biases, including the sick quitter effect. Three consistent findings emerged when all 37 studies were analysed together:

1. there was no significant protection at any level of consumption
2. there was significantly increased risk at higher levels (even without addressing the sick quitter effect)
3. there was a significant dose-response relationship across all levels of consumption.

These conclusions are also consistent with one previous systematic review [Wang et al, 2016] and another recent large cohort study [Naudin et al., 2025].

What does it all mean?

Our first take-home message from this updated systematic review is, therefore, that pancreatic cancer should be carefully reconsidered by WHO’s International Agency for Research on Cancer when they update their list of cancers deemed as “definitely” caused by alcohol [IARC, 2010]. This would mean also, in turn, it could be incorporated into all estimates of alcohol’s burden of disease [global, national or regional] and in assessments of drinking guidelines.

Pancreatic cancer is not the only alcohol attributable disease for which taking account of the sick quitter effect [or “former drinker bias” as we prefer to call it] makes a night and day difference to interpretation of the alcohol-health relationship. For example, in a systematic review of alcohol use and prostate cancer, we identified a handful of studies in which former drinker bias had been avoided. These studies estimated the risk of prostate cancer for low volume drinkers to be three times higher than the estimates from studies containing former drinker bias [Zhao et al., 2016]. Prostate cancer is also not on IARC’s list of alcohol attributable cancers though it was identified as such by the US Centers for Disease Control [Esser et al., 2024]. Previously, we also demonstrated the importance of adjusting for former drinker bias in studies of alcohol use and breast cancer [Zeisser et al., 2014].

A wholesale underestimation of alcohol deaths

The implications for cancer prevention alone are immense. Not only are some alcohol attributable cancers ignored in global burden of disease estimates, even those included likely underestimate alcohol’s contribution to morbidity and mortality. This is not just an issue affecting low-volume or “moderate” drinkers, but these fundamental biases affect risk estimates across the entire spectrum of drinking levels and patterns, likely leading to a wholesale underestimation of the extent of alcohol’s contribution to cancer.

But the story does not end with underestimation of alcohol’s contribution to cancer. As we have shown in our now many systematic reviews and meta-analyses of alcohol and mortality more generally [e.g. Stockwell et al, 2016, 2024; Zhao et al, 2023], the great majority of cohort studies used to estimate alcohol’s risk to health completely ignore the possibility of former drinker bias. For example, the group estimating the Global Burden of Disease [2020 GBD Collaborators, 2022] in 2020 lost about 1 million deaths attributable to alcohol compared with 2016 [2016 GBD Collaborators, 2018], largely due to assuming that consumption of up to nine drinks per day was protective against ischemic heart disease. The 2016 GBD group had assumed protection up to a mere five drinks per day. Our own systematic review and meta-analysis [Zhao et al., 2017] estimated possible protection at one drink per day. Other examples can be cited in relation to type II diabetes and ischemic stroke.

The tip of the bias iceberg

Unfortunately, the “sick quitter effect” is only the tip of the iceberg in relation to the types of bias that come into play when health risks for drinkers are compared with even lifetime abstainers. Firstly, even from the get-go, young adults who are teetotalers have been shown to have worse health profiles, more disability and greater health risk factors [e.g. poverty] than their drinking peers [Ng Fat and Shelton, 2012]. In other words, lifetime abstainers may be biased towards ill health and likely to make people continuing to drink look good by comparison, long before any sick quitter effects come into play. It is hard to simply adjust for the effects of low income and health risk factors in cohort studies and believe it has eliminated this problem – there will always be some residual confounding.

Secondly, it is insufficient to remove former drinkers from people currently abstaining and used as the reference group in these studies. They also need to be reallocated among the current drinkers according to their previous level of drinking to avoid selection bias [e.g. Srivatsa et al, 2026]. Future studies need to not only ask current abstainers if they used to drink, they also need to ask how much, for how long, and to develop methods of including them among groups of drinkers in order to get a balanced assessment of the relative risk for drinking versus abstaining over the full life-course.

We conclude there has likely been wholesale underestimation of alcohol’s contribution to mortality and morbidity, across the full drinking spectrum and the full range of alcohol attributable diseases, not just cancer. Pancreatic cancer is only one example of a condition that is overdue for reconsideration by WHO for addition to their list of alcohol attributable diseases. It has been estimated that alcohol’s contribution to preventable deaths each year in Canada would double if risk estimates were based only on studies free from abstainer bias (Stockwell et al., 2007).  We suggest a more comprehensive and updated analysis of this question is urgently needed in relation to alcohol’s global burden.

Written by Tim Stockwell, Jinhui Zhao, James Clay, Keegan Lawrence and Tim Naimi, Canadian Institute for Substance Use Research, The University of Victoria, Canada.

All IAS Blogposts are published with the permission of the author. The views expressed are solely the author’s own and do not necessarily represent the views of the Institute of Alcohol Studies.