Although not highlighted in their paper, the results of Chen et al ^{Reference Chen, Yang, Tseng, Sun, Wang and Wang1} are fundamentally different from the preponderance of evidence that suggests a lack of association, as summarised in recent systematic reviews and a large case-control study. ^{Reference Seitz, Reimer and Siddiqui2-Reference Sprung, Jankowski, Roberts, Weingarten, Aguilar and Runkle4} This is an important topic that is difficult to study in humans, and all observational studies in this area have substantial limitations. The authors point out some; we wish to amplify their identified limitation and to raise others.

The primary concern is that their matched cohort design, based on exposure history, is highly susceptible to confounds. Simply put, individuals who have conditions requiring surgery are fundamentally different from those who do not - as demonstrated by the finding that they were sicker and older. The ability to adjust analytically for such differences, even if the right covariates are chosen, is limited. Case-control methodologies, such as we utilised in a recent longitudinal population-based analysis using standardised diagnostic criteria which did not find an association between exposure to anaesthesia and Alzheimer’s dementia, ^{Reference Sprung, Jankowski, Roberts, Weingarten, Aguilar and Runkle4} are less susceptible to such confounding (although such designs have their own problems). The use of large administrative data-sets can be very valuable given the ability to examine entire populations, but the lack of any consistent diagnostic criteria and the potential for incomplete or inaccurate coding makes it difficult to estimate the true population incidence of a condition such as Alzheimer’s dementia. For example, there is the potential for substantial ascertainment bias, as those patients who need surgery likely have greater contact with the healthcare system, and are more likely to have the opportunity for an Alzheimer’s dementia diagnosis.

Another design question is raised by the stated inclusion criteria for controls. For those who received anaesthesia, the date of their first anaesthesia exposure after 2004 is used as the index date. The authors do not indicate what index date was used for controls, only that controls were selected from patients who did not receive anaesthesia during the study period. It is not clear whether individuals were excluded from the pool of potential controls if they had a diagnosis of dementia at any time prior to the end of the study period (31 December 2007) or whether these exclusion criteria were applied separately for each potential exposed/unexposed matched set based on an assigned index date. The results would be substantially biased if those with diagnoses of dementia at any time prior to 31 December 2007 were excluded from being potential controls.

Concerns regarding the impact of these and other methodological issues are heightened by the fact that if, as they speculate, anaesthesia and surgery is causative of Alzheimer’s dementia, Chen et al’s results in many instances are simply not plausible. They report a uniformly positive association between anaesthesia and Alzheimer’s dementia, regardless of the type, number of anaesthetics and cumulative anaesthesia duration. Those patients who received ‘intravenous/intramuscular’ anaesthesia, presumably mainly representing monitored anaesthesia care that most anaesthesiologists would not characterise as constituting a ‘general anaesthetic’, were just as likely to develop Alzheimer’s dementia as those who were undergoing major surgical procedures. Those receiving regional anaesthesia, who in many cases have little central exposure to anaesthetic drugs, were even more likely to develop Alzheimer’s dementia. It could be argued that the association is caused by the surgery, not the anaesthesia, but there are similar problems with the results here. More minor procedures, such as ophthalmological and dermatological surgery, were associated with risk, but major surgery such as cardiac and respiratory procedures were not. The most likely explanation for these implausible findings is a significant contribution of confounding factors.

More data on this important question are always welcome, and there are significant limitations of all human observational studies in this area. However, the introduction of Chen et al’s paper states that the purpose of the study is ‘to determine whether the risk of neurodegenerative dementia increases after anaesthesia and surgery’ (which to us implies causation), and we suggest that this purpose has not been fulfilled.