批判的な意見を寄せたのは一人だけではなかった。
六万人を超えるデータなのに考察の対象にしている人数が極端に少ない･･･という指摘で始まる以下のレターでは、論文中の回帰分析の数値に着目している。
ポイントは、
加糖飲料をたくさん飲む人のRR(相対危険度)は、
①年齢で補正した数値
②その他の要素も加えて補正した数値
③出来るだけ多くの要素を加えて補正した数値で、それぞれ①1.49②1.32③1.30
になるのに対して、人口甘味料を加えた飲料をたくさん飲む人の場合、それぞれ①3.50②2.19③1.68という具合に、大きく異なっていることである。
このように大きく数値が異なる場合、人口甘味料を加えた飲料が糖尿病の原因になっているのではなくて、何か他の要因によって見かけ上の相関関係が現れているのではないか･･･と考えるのがセオリーであるらしい。
そして、この論者も、やはり、人口甘味料が糖尿病を引き起こすのではなくて、糖尿病になる可能性の高い人が人口甘味料を加えた飲料を選ぶようになる、という解釈のほうが妥当だと考えているようである。
Artificially and sugar-sweetened beverages and incident type 2 diabetes
The American Journal of Clinical Nutrition, Volume 98, Issue 1, July 2013, Pages 249–250, https://doi.org/10.3945/ajcn.113.062653
Dear Sir:
The French E3N cohort study included 66,118 nondiabetic women followed up between 1993 and 2007 (1). Of these, 1369 developed type 2 diabetes (T2D). After multivariate analysis, significant trends in risk of T2D were observed for consumption both of sugar-sweetened beverages (SBBs) and artificially sweetened beverages (ASBs). In the multivariate analyses, a significant association was observed only for the highest consumption categories, with an RR of 1.30 (95% CI: 1.02, 1.66) for 359 mL SSBs/wk and 1.68 (95% CI: 1.19, 2.39) for >603 mL ASBs/wk.
These consumption categories, however, were based on only 73 cases of T2D for SSBs and on 34 cases for ASBs. A key problem of these analyses, in fact, is the definition of quartiles among consumers of sweetened beverages, which, despite consumption measures defined according to single units of mL/wk, was extremely uneven, ie, the numbers of cases for ASBs were 252, 17, 20, and 34 in subsequent “quartiles.”
An additional concern with reference to ASBs—more than SSBs—is the substantial change in the RR estimates across age-adjusted and subsequent multivariate models. For SSBs, the RR in the highest consumption category was 1.49 in the model adjusted for age only, 1.32 in the simplest multivariate model, and 1.30 in the fully adjusted model (including allowance for BMI).
For ASBs, corresponding values were 3.50 in the age-adjusted model, 2.19 in the simplest multivariate model, and 1.68 in the fully adjusted model. This substantial change in RR estimates across subsequent models indicates that residual confounding (by BMI and/or other factors) may well be present in the multivariate risk estimates for ASBs. This may partly or largely explain the residual apparent association between ASBs and T2D for the 34 women at the highest exposure level.
Evidence from other studies is largely inconsistent. In the Nurses’ Health Study in women (2) and the Health Professionals Follow-Up Study in men (3), significant associations were observed for SSBs but not for ASBs. Two smaller US studies on atherosclerosis (4, 5) found apparent associations for ASBs, with RRs between 1.3 and 1.4 for the highest consumption category, and a positive trend in risk in one study only (4). In the Framingham Heart Study (6) positive associations, of borderline significance, were observed for the highest consumption levels of both SSBs and ASBs.
Even for the highest consumption levels of ASBs, however, the RRs of T2D were systematically below 2, and generally below 1.5, thus leaving open the issue of bias or residual confounding in observational studies, and particularly confounding by BMI. Reverse causation is an additional relevant issue (7, 8), because overweight subjects with possible prediabetic conditions may selectively choose to consume ASBs, thus leaving open the possibility of a false-positive finding between ASBs and T2D risk (9).
The author has received in the past unconditional grant support from the International Sweeteners Association (Brussels, Belgium) for addressing the issue of sweeteners and cancer risk.
