Self-reporting has a high potential risk for bias by under-or overestimation of exposure, which could distort results, conclusions, and recommendations, especially in medical research. Although such biases can affect results, they are still often ignored in practice [1], raising the important issue of identifying and addressing them. In the present study, data from two population-based studies were used to investigate whether self-reported caffeine consumption, which is widely used in studies, correlated with plasma levels of caffeine and its two main metabolites. Our null hypothesis was that participants self-reporting no caffeine consumption would exhibit undetectable plasma levels of caffeine and its metabolites (i.e., plasma levels less or equal to 5ng/ml for each metabolite). Thus, an intercept indicating a mean log plasma level less than 2.71 (plasma levels less than 15ng/ml) was expected in the multivariable analyses. Subsequently, a positive association was expected between the number of self-reported cups of coffee/caffeinated beverage consumed daily and plasma levels of caffeine and its two main metabolites, which was rejected by the results showing that self-reported caffeine consumption could be a source of bias.

In both SKIPOGH and Colaus|PsyColaus, for subjects reporting no caffeine consumption (representing between 7 and 12% of both cohorts), plasma levels of caffeine and its two metabolites were over 652ng/ml. Thus, such participants likely underestimated their caffeine consumption, intentionally or not. In addition, some CoLaus|PsyCoLaus participants, representing less than 1% of the whole cohort, had undetectable plasma levels of caffeine and its two metabolites despite reporting daily caffeine consumption, resulting in an overestimation of their caffeine intake.

Furthermore, in both studies, the associations between increasing log plasma levels of caffeine and its two metabolites and the number of reported coffee/caffeinated beverages cups were not linear. Indeed, in SKIPOGH 1 and considering only the reported espresso consumption, for up to three espresso cups consumed per day, the increase in the number of cups was in agreement with the increase in plasma levels of caffeine and its two metabolites. However, above three espresso cups per day, the associations were no longer significant. Considering all caffeine sources listed in the SKIPOGH questionnaire, the change of slope was pushed from 3 to 5 cups of 60 ml espresso, however, the same results were found in the presence of a change of slope. In SKIPOGH2, the change of slope was not statistically significant, implying a positive association between plasma levels of caffeine and its two metabolites and reported espresso consumption, with a trend toward a change in the magnitude of the associations beyond a reported consumption of 4 cups of espresso. The change of slope and the change in the associations’ magnitude may be due to an overestimation of caffeine consumption as well as to the low number of individuals reporting consumption of more than 3–5 cups per day. On the other hand, although both caffeine and two main metabolites were measured, it cannot be excluded that individuals with high CYP1A2 activity metabolized and eliminated both caffeine and its metabolites more rapidly and therefore consumed more coffee. Therefore, despite high caffeine consumption (3–5 cups of espresso per day), this was no longer significantly associated with increased plasma levels of caffeine and metabolites. Of note, excessive caffeine consumption has been associated with CYP1A2 metabolic enzyme capacity saturation [26], which can lead to an accumulation of caffeine metabolites, specifically theophylline. Furthermore, several clinical and genetic factors are known for inducing or inhibiting CYP1A2 activity (e.g., oral contraceptives, smoking, some genetic variants located in the CYP1A2 gene or in other genes/regulatory regions) [27,28,29]. Therefore, CYP1A2 inducer users and individuals with high CYP1A2 activity should have elevated plasma levels of paraxanthine and theophylline and reduced levels of caffeine, and vice versa. However, this should have no effect on our results since we modeled the sum of the three components.

Predictions of plasma levels of caffeine and its two metabolites when increasing the number of reported daily consumption of espresso cups showed that up to about 3 cups (considering espresso consumption only) and 5 cups (considering all caffeinated beverages), positive associations were observed between the two variables. Beyond these thresholds, controversial associations were noticed, with even a slight decrease in plasma levels in SKIPOGH 1 while a slight increase was observed in SKIPOGH2. Of note, in the first wave of the SKIPOGH study, participants were not specifically informed that questions about caffeine consumption would be asked, allowing little time to adequately estimate it. As more than 87% of the participants took part in both study waves, participants in SKIPOGH2 could have better estimated their caffeine consumption the second time, remembering that such questions would be asked. Therefore, better estimation and self-reporting of caffeine consumption by SKIPOGH2 participants may explain the slight increase in plasma levels in SKIPOGH2 compared with SKIPOGH1.

The estimation of caffeine consumption by CoLaus|PsyCoLaus participants was not detailed, assessing only coffee and caffeinated beverage consumption without indicating the type of coffee/beverage or volume. Nevertheless, similar results to those of the SKIPOGH study were found. In fact, considering coffee consumption, plasma levels of caffeine and its two metabolites were high in participants reporting no coffee consumption. However, because the frequency of consumption was not detailed beyond two cups per day, a change of slope was not observed, which was the case when the consumption of all caffeinated beverages was considered. Indeed, no significant difference in plasma levels of caffeine and its two metabolites was observed between participants consuming 4 to 6 cups per day and those with more than 6 cups per day, indicating a likely change of slope at 3 cups per day.

Misreporting (voluntary or not), and/or ignorance of caffeine consumption in some foods/beverages, and/or not indicating caffeine-containing drugs in the questionnaire could explain the mismatch between plasma levels of caffeine and its metabolites and the reported caffeine consumption. Thus, the similar results of two large population-based studies with very different questionnaires on caffeine consumption highlighted the importance of objectively assessing exposure to caffeine.

The present study has several limitations. The quantification of plasma levels of caffeine and metabolites has some limitations in itself. First, the time interval between the last caffeine intake and blood sampling may have been misreported (SKIPOGH) or was unknown (CoLaus|PsyCoLaus). However, summing the plasma levels of caffeine and its two main metabolites (paraxanthine and theophylline) should reduce the influence of variable time intervals, a short interval of time leading to higher plasma levels of caffeine and lower plasma levels of paraxanthine and theophylline, the reverse being true for longer periods of time. In addition, summing the concentrations of 3 substances should allow a better estimation of the pharmacologically active molecules. It should also be mentioned that theobromine, another caffeine metabolite, was not taken into account because its main source is chocolate while the present study focused on caffeine consumption [30]. However, considering also theobromine levels did not influence our results (data not shown). Secondly, while the present study focused on two main metabolites of caffeine, 70 to 80% and 7 to 8% of caffeine being metabolized to paraxanthine and theophylline, respectively [31], other metabolites than theobromine have been described and were not quantified in the present study. Third, not all sources of caffeine or its metabolites (e.g., theophylline) were covered by the SKIPOGH questionnaire, namely some foods (e.g., cakes, pastries, breakfast cereals), over-the-counter drugs (e.g., some analgesics and bronchodilators), and some dietary supplements. Nevertheless, these foods contain negligible amounts of caffeine, and less than 3% of SKIPOGH participants reported using drugs that contained theophylline or caffeine. In addition, by examining solely self-reported espresso consumption as opposed to all sources of caffeine addressed by the questionnaire (i.e., estimated 60 ml espresso consumption), the results remained consistent. Thus, our results should not be significantly altered by incorporating additional caffeine sources. Fourth, the SKIPOGH questionnaire has not been validated, which is also the case for the vast majority of studies about caffeine consumption.