Dietary fructose, carbohydrates, glycemic indices and pancreatic cancer risk: a systematic review and meta-analysis of cohort studies.
Aune D, Chan DS, Vieira AR et al. Dietary fructose, carbohydrates, glycemic indices and pancreatic cancer risk: a systematic review and meta-analysis of cohort studies. Ann Oncol Apr. 26 2012.
The review and meta-analysis of Aune et al  focuses on the role of dietary fructose among other carbohydrates as a possible pancreatic cancer risk factor. They concluded their results “indicate that intake of fructose, but not total carbohydrates, sucrose, GI or GL, increases the risk of pancreatic cancer”. This study was plagued by several weaknesses that were not addressed in the discussion. While the meta-analysis is interesting, it suffers from known biases, lacks sufficient precision to draw the main conclusion, is supported by biochemical mechanisms that lean on experimental data of doubtful physiological significance and interventions that are unrepresentative of exposure to fructose among the general public. The review also under emphasises factors thought to confound epidemiological studies on fructose, and does not meet modern guidelines on the conduct and presentation of systematic reviews.
The paper identifies no study-specific protocol or registration. Authors do refer to a “predefined protocol” , which they state they followed. However, the cited item is a systematic literature review (SLR) manual and not an actual study-specific protocol. In addition, the manual used is relatively outdated (2006). Registering (e.g. PROSPERO) and reporting (e.g. PRISMA) are applicable to all systematic reviews including observational studies (3) (4) and help to predefine the study question, conduct the review, and present to scientific communities how the question was addressed and the conclusion reached. In the final literature search only one database was used. Additional published studies may therefore have been overlooked. No literature search terms list is presented for this specific study. Instead the authors cite the above SLR Manual which is too general and does not even mention fructose.
Although the food frequency questionnaires used in all studies were validated, this was not true for the carbohydrate measures (only one had been validated) or fructose measures (none were validated). It is preferable as guidelines indicate to undertake duplicate extraction of data and reporting of agreement by way of kappa or other statistics. The authors report neither duplicate extraction nor their method for reaching agreement nor kappa (or other agreement) statistics. Moreover, the supplementary file online doesn’t tabulate the agreed data for readers to assess accuracy. Error and bias in data extraction therefore cannot be excluded.
Although a heading in the Methods section refers to quality assessment, no method of study quality assessment was cited, nor was study quality assessment presented either as a whole or for individual study quality items. It is not clear, therefore, whether the studies analysed are of sufficient quality to allow conclusions about fructose to be drawn; this even if all other aspects had been satisfactory.
Too few studies were included to adequately assess symmetry of reported associations between fructose and pancreatic cancer. The Begg’s and Egger’s methods that authors used to assess study error or publication bias are based on regression or rank correlation analysis for which a minimum 10 studies would be recommended as for meta-regression. It is also possible that ‘less interesting’ neutral or even negative associations may have been omitted from other studies cited in the review (or not retrieved from the literature). In the final analysis just six prospective studies were analysed. Without undertaking further studies it cannot be excluded that a cluster of similar associations arose by chance among 3 of the 6 studies.
An overestimation of the association between pancreatic cancer and fructose consumption in the dose-response analysis (Figure 2 in the main article) is inevitable through not including the study of Patel et al 2007, who found no association (5). This would affect also the width of the confidence bounds, bringing the lower one closer to unity (no significant association). A statistically significant association would then likely be dependent upon one study, that of Jiao et al (6) . For this reason also, there are too few studies to draw the conclusion that the authors reached. An overestimation of the association between pancreatic cancer and fructose consumption in the dose-response analysis (Figure 2 in the main article) is inevitable through not including the study of Patel et al 2007, who found no association (5). This would affect also the width of the confidence bounds, bringing the lower one closer to unity (no significant association). A statistically significant association would then likely be dependent upon one study, that of Jiao et al (6) . For this reason also, there are too few studies to draw the conclusion that the authors reached.
Confidence bounds in the dose-response analysis for total carbohydrate, sucrose, and glycaemic index are very wide (Figure 3 in the supplementary file); this is not unexpected given the low incidence of pancreatic cancer. Hence, null biological responses cannot be claimed for these measures. It seems misleading, therefore, for the authors to conclude that fructose, but not these other carbohydrate measures, increase the risk of pancreatic cancer. Further, no statistical test was made of difference in risk between any of the carbohydrate measures; meanwhile, confidence bounds for fructose overlap with those for each of the other carbohydrate measures. Furthermore, the central trend for the dose-response curve for fructose when limited to the range of doses encountered falls with the reported confidence bounds for glycaemic index, sucrose, and total carbohydrates. Evidently, too few studies are available to draw any comparative conclusions.
The dose-response curves for the various carbohydrate measures (Figure 3 supplemental file) and fructose (Figure 2 main article) do not show data points. This hides the extent to which curves are drawn beyond the range of the analysed data. Particularly obvious in this respect is the figure for glycaemic index, drawn for GI=0 to 100, an exposure range not encountered among the populations studied (or any other population). Overemphasising any biological significance in this way might well be considered a reporting bias; such being emphasised for fructose in the main article, other dose response curves being less prominent in the supplementary data file online.
Mechanism the authors suggest to support a relation between dietary fructose and pancreatic cancer risk were fructose induction of transketolase activity in pancreatic cancer cells and primary cells cultures from solid tumours (7) , and elevation of insulin concentrations secondary to insulin resistance. Critical examination of recent literature suggests transketolase induction may only take place when cells are exposed to fructose for several days (continuously without overnight or ‘between meal’ rests) at concentrations excessive of those generally found in vivo (normally <0.5 mmol/l at rest). By contrast, pancreatic cell proliferation rate, nucleic acid synthesis (BrdUrd marked), and percentage of cells in G-phase and S phase (growth) in several pancreatic cancer cell types continuously exposed to excessive fructose (5 mmol/L) are reported to be no different to when continuously exposed to normal glucose concentrations (5 mmol/L) (7) .
Turning to insulin, this is often considered a growth factor for various tumours, and elevation of insulin is a possible reason for elevated tumour growth rates. However, compared with total carbohydrate, dietary fructose is well known to only weakly elevate post-prandial insulin concentrations, and although excessive fructose consumption (>150g/d) is found to cause insulin resistance in humans, addition of fructose to regular diets appears to have either no effect or may improve insulin sensitivity provided total fructose intakes does not exceed 150g/d (8, 9) . Among epidemiological studies fructose is not unique as a carbohydrate in associating with type-2 diabetes, since free glucose also does so in at least 4 prospective cohort studies (10). Moreover, medium term intervention studies in humans have not shown fructose to elevate body weight, thus not affect BMI, unless intake is provided above energy requirements—an effect not unique to fructose either since this would occur with all macronutrients (11, 12). Obesity too may not be related to fructose intake among USA populations (8, 10) .
Dietary fructose is associated with several other aspects of diet, which may confound meta-analyses of observational data (10), and which the authors do not discuss. Dietary fructose is seldom consumed as an item. By far most is consumed with similar amounts of glucose in fruits and high fructose corn syrup (in North America), and both glucose and fructose are available indirectly from sucrose via hydrolysis in the small intestine. At present it is not possible from observational studies, therefore, to focus entirely on fructose. Thus, fructose may simply be labelling a monosaccharide mixture or even glucose. For example, glucose and pancreatic cancer risk are reported to be associated, at least or more strongly, than is fructose (6) . Moreover, higher intakes of fructose in the USA and Canada (where all the reviewed studies were conducted) is likely also to be a marker of a poor diet, one plentiful in several other foods of low nutrient density, and not just soft drinks that can be a major source of dietary fructose (13) . These drinks, too, may simply mark a relatively higher energy consumption and lower nutrient density with accuracies perhaps greater than estimable from the sums for all individual dietary items.
The conduct and reporting of any systematic review and meta-analysis is a challenge. It is particularly challenging for observational studies because of high risk of bias by unmeasured confounders, and fructose has many potential confounders. The authors likely regard their paper as preliminary, in which case a more substantive assessment should be forthcoming. Meanwhile, researchers ought to be aware that more widely the systematic review here considered there is considerable controversy over the quality of research studies on fructose.