Caffeine May Raise Anxiety Scores, Review of Randomized Trials Finds

Pooling data from fifteen randomized trials involving 668 participants, the researchers found no clear evidence of a difference in depressive symptoms between caffeine and control conditions, while anxiety scores were, on average, higher after caffeine exposure. Yet the authors themselves urge strong caution: the evidence for both outcomes was rated as very low certainty, and the anxiety finding came with a prediction interval that spanned the null, meaning the true effect in individual study settings could plausibly range from a substantial increase to a decrease in anxiety.

The review, registered prospectively with PROSPERO and conducted according to the PRISMA 2020 reporting standard, searched eight English- and Chinese-language databases from inception to May 16, 2026. Eligible studies were randomized controlled trials, either parallel-group or crossover designs, that compared quantified caffeine exposure—delivered as isolated caffeine or through caffeinated foods and beverages—with placebo, decaffeinated, or caffeine-free controls, and that measured depressive or anxiety symptoms with validated psychometric instruments. From 1,139 records identified, fifteen trials published between 1994 and 2025 met the criteria. The studies came from the United States, China, Tunisia, Colombia, Brazil, the United Kingdom, Iran, and Sweden, and enrolled healthy adults, athletes, school-aged children, and people with psychiatric or other clinical conditions.

The caffeine doses tested ranged widely, from 60 milligrams per day up to 480 milligrams or weight-adjusted amounts, and exposure ranged from a single experimental session to repeated interventions lasting up to twelve weeks. Caffeine was administered as capsules, tablets, coffee, energy drinks, tea, or cola. Depressive symptoms were assessed with instruments such as the Hamilton depression scales and the Beck Depression Inventory, while anxiety was measured mainly with the state–trait anxiety inventory and the Beck Anxiety Inventory. Because these instruments capture different constructs—transient mood states versus clinical symptom severity—the authors emphasized that statistical standardization alone cannot establish clinical equivalence between them.

For depressive symptoms, the broad pooled estimate across seven independent comparisons did not show clear evidence of a difference between caffeine and control conditions. The standardized mean difference was −0.49, with a 95% confidence interval of −1.33 to 0.34, but heterogeneity was extreme, with an I-squared of 92.9%, and the 95% prediction interval stretched from −2.87 to 1.88. The authors interpreted this broad estimate as a descriptive summary of a highly diverse evidence base rather than as a common effect expected across settings. Importantly, they note that the non-significant result should not be read as proof that caffeine has no clinically relevant effect on depression, given the wide intervals and the very-low certainty of the evidence.

When the analysis was restricted to more clinically coherent subsets, the picture became more nuanced. Two acute crossover trials in non-psychiatric participants, both using the Profile of Mood States after single-session caffeine exposure, showed no clear difference from control, with a pooled standardized mean difference of −0.15 and a wide confidence interval. In psychiatric populations, the evidence was limited and summarized narratively: one acute caffeine challenge in patients with depression produced an imprecise estimate compatible with lower symptom scores, while two trials of sustained caffeine exposure alongside psychiatric care reported lower depressive symptom scores after four and twelve weeks. The authors stress that trials conducted alongside antidepressant therapy are not directly comparable with isolated caffeine exposure, and that randomized evidence therefore does not support a consistent antidepressant effect across distinct populations and exposure contexts.

Anxiety outcomes told a different statistical story. Across sixteen independent comparisons from twelve trials, anxiety scores were on average higher after caffeine exposure, with a standardized mean difference of 0.51 and a confidence interval of 0.12 to 0.89 that excluded the null. However, heterogeneity was again very high, at 90.3%, and the 95% prediction interval of −0.95 to 1.96 crossed the null, indicating that in comparable future settings the effect could differ substantially in either direction. When the analysis was restricted to eleven acute comparisons in non-psychiatric cohorts, the point estimate remained in the direction of higher anxiety but the confidence interval included the null. Evidence from psychiatric populations—patients with panic disorder and social phobia—was too limited and heterogeneous to pool, and two sustained-exposure studies in non-psychiatric populations produced strikingly divergent results, one near the null and one substantially positive.

An exploratory dose-stratified analysis divided the trials into categories of at most 200 milligrams and more than 200 milligrams of caffeine. For anxiety, the higher-dose category yielded a pooled standardized mean difference of 0.97, with a confidence interval of 0.55 to 1.39, whereas the lower-dose category yielded an estimate close to the null at −0.05. This pattern persisted in a sensitivity analysis restricted to comparisons whose dose classification was unambiguous. For depressive symptoms, there was no clear difference between the dose categories. The authors are explicit that the 200-milligram boundary was used solely as an analytical grouping criterion informed by previous literature, not as an established clinical or biological threshold, and that the categorical analysis does not establish a continuous dose–response relationship.

The methodological quality of the underlying trials added further uncertainty. For depressive outcomes, all seven independent comparisons were judged to have some concerns for risk of bias under the revised Cochrane tool, with none rated high risk; the most frequent concerns involved the randomization process. Among sixteen anxiety comparisons, three were at low overall risk and thirteen had some concerns, again mainly around randomization and outcome measurement. Using the GRADE framework, the certainty of evidence was rated very low for both outcomes, downgraded for risk of bias, substantial inconsistency, indirectness, and imprecision. For anxiety, funnel-plot asymmetry and a significant Egger regression test indicated small-study effects, raising concern about publication bias, although heterogeneity across populations, exposures, and outcome measures may also have contributed to the pattern.

The authors argue that the heterogeneous responses are biologically plausible. Caffeine acts primarily as a non-selective antagonist of central adenosine A1 and A2A receptors, influencing neuronal excitability and neurotransmitter systems involved in arousal and affect. Acute antagonism may alter dopaminergic and serotonergic neurotransmission, while adenosine receptor antagonism can increase central arousal and sympathetic activation, potentially heightening anxiety in susceptible individuals. With repeated exposure, compensatory adaptations in adenosine signaling may modify these acute effects, and interindividual differences in caffeine sensitivity, metabolism, habitual intake, and baseline anxiety vulnerability—possibly including genetic variation in pathways such as CYP1A2 and ADORA2A—may further shape responses. Such factors were seldom reported in the included trials and could not be examined.

The practical implications are deliberately restrained. The authors state that the findings do not provide a basis for defining a clinically harmful caffeine dose or for modifying existing population-level intake recommendations, and that the observed increase in anxiety scores should not be taken to indicate clinically meaningful deterioration or an increased incidence of anxiety disorders, since most trials assessed continuous symptom scores rather than diagnostic outcomes. Because much of the randomized evidence derived from acute or short-term experimental exposure, the results should not be extrapolated directly to habitual dietary caffeine consumption or long-term psychiatric outcomes. The review’s central contribution, the authors conclude, is to show that randomized evidence is more consistent with context-dependent variation in affective response than with any fixed effect of caffeine on mood—and that longer, better-designed trials with clinically meaningful outcomes are needed to settle the question.

Subject of Research: Effects of quantified caffeine intake on depressive and anxiety symptoms in randomized controlled trials

Article Title: Effects of caffeine intake on depressive and anxiety symptoms: a systematic review and meta-analysis of randomized controlled trials

Article References: Zhang, A., Ni, Y., Zhang, Z., Wang, J., Chen, W., Chang, F., Cui, H., Yuan, C., Hang, D., & Li, L. (2026). Effects of caffeine intake on depressive and anxiety symptoms: a systematic review and meta-analysis of randomized controlled trials. Frontiers in Nutrition, 13, Article 1950050. https://doi.org/10.3389/fnut.2026.1950050

Image Credits: AI Generated

DOI: 10.3389/fnut.2026.1950050

Keywords: caffeine, anxiety, depression, meta-analysis, randomized controlled trials, systematic review, mental health, nutrition, adenosine receptors, dose-response, evidence certainty, mood symptoms