Self-blinding citizen science to explore psychedelic microdosing

1There is renewed interest in the medical application of psychedelic drugs, such as lysergic acid diethylamide (LSD) and psilocybin. Contemporary research is predominantly focusing on ‘psychedelics assisted psychotherapy’, where a few (one to three) large doses of psychedelics are used as adjunct to psychotherapy. Using this paradigm, psychedelics have shown promise in the treatment of conditions such as depression, end-of-life-anxiety, addiction, and obsessive-compulsive behaviors.

2Recently, ‘microdosing’ has emerged as an alternative paradigm of psychedelic use. Due to its underground origin, microdosing does not have a universally agreed upon definition, and inconsistencies exist in substance, dose, frequency, and duration of use. However, microdosing can be broadly defined as the frequent use (one to three times per week) of low doses of psychedelics (10–20% of a typical ‘full’ dose, e.g. 10–15 μg LSD or 0.1–0.3 g of dried psilocybin containing mushrooms).

3Anecdotal evidence suggests that microdosing may improve well-being, creativity, and cognition, and recent uncontrolled, observational studies have provided some empirical support for these claims. While encouraging, these studies are vulnerable to experimental biases, including confirmation-bias and placebo effects, in particular, because microdosers are a self-selected sample with optimistic expectations about psychedelics and microdosing. This positivity bias, combined with the low dose and the subjective evaluation of effects, pave the way for a strong placebo response.

4A few recent double-blind, controlled studies have been conducted on microdosing. All studies used LSD and focused on the acute effects of a single microdose in a small number of healthy subjects. Studies have found large variability in LSD blood concentration after microdosing, along with increased BDNF blood levels. No robust evidence was found to support the positive anecdotal claims about microdosing, but some dose-related self-rated subjective effects were detected (e.g. self-ratings of ‘feel drug’, ‘feel high’, and ‘like drug’), along with concomitant changes in brain function.

5Two key issues need to be considered when assessing the scientific credibility of microdosing: the lack of placebo control in uncontrolled studies and the small sample size in controlled studies. Uncontrolled, observational studies affirm the anecdotal reports, but by design, these studies cannot provide evidence for beyond placebo benefits. Lab-based, controlled studies have small samples due to restrictive drug policies that render randomized controlled trials prohibitively expensive, and hence may be statistically underpowered.

6In the present study we conceived of a novel citizen-science initiative as a solution to this problem, exploiting modern technology and the popularity of microdosing. The key component is a self-blinding setup procedure that enabled self-experimenters, who microdose on their own initiative using their own psychedelic, to implement placebo control and randomization without clinical supervision. To investigate potential changes over the study period, participants were directed to online self-report surveys and cognitive tasks at various timepoints. The strength of this design is that it allowed us to obtain a large sample size while implementing placebo control at minimal logistic and economic costs. The primary objective of the study was to test whether psychedelics microdosing produces superior outcomes compared to placebo on psychological state and cognitive function. We hypothesized that improvements from baseline will be positively correlated with the number of microdoses taken during the dose period and that acute/post-acute outcomes will be better under/after taking a microdose.

7This study had a naturalistic design involving elements of experimental control (self-blinding), prospective data collection and online citizen-science. From baseline to the final endpoint, the study was 10 weeks long (weeks 0–9), including a core 4-week microdosing period. Primary endpoint was at week 5 and there was an optional follow-up at week 9. The self-blinding procedure randomly assigned individuals to three groups, where the groups are defined by the number of weeks taking placebos/microdoses during the dose period. The three groups were:

8Placebo (PL) group: 4 weeks of placebo,

9Half-Half (HH) group: 2 weeks of placebo and 2 weeks microdosing, and

10Microdosing (MD) group: 4 weeks of microdosing.

11Individuals took two microdoses during each microdose week, resulting in 0/4/8 total microdoses for the PL/HH/MD groups. Participants had equal probability (1/3) of being assigned to each group; Figure 1 illustrates the experimental timeline and the groups’ dose schedule.

12Outcomes can be organized into three categories capturing the effects of microdosing on different timescales.

13Accumulative: assessed monthly, first at baseline, then after the completion of the dosing regime at week 5, and finally at the optional long-term follow-up at week 9. Accumulative outcomes were: Ryff’s psychological well-being (RPWB), cognitive and affective mindfulness scale (CAMS), satisfaction with life scale (SWL), green paranoid thought scales (GPTS), big five personality traits (B5) with the addition of intellect trait and cognitive performance. To quantify cognitive performance, participants were tested in six tasks: spatial span, paired associates, rotations, odd one out, spatial planning, and feature match, see for details. Task scores were combined as the cognitive performance score (CPS) to quantify overall cognitive performance as a single outcome. Briefly, CPS is the average z-score across the six tasks after removing learning effects, see Appendix 1 for details.

14Post-acute: assessed weekly during the dose period on Sundays, when no capsule was taken. Measures were taken 48–72 hr after the last placebo/microdose capsule. Post-acute outcomes were: Warwick–Edinburgh mental well-being scale (WEMWB), Quick inventory of depressive symptomatology (QIDS), Spielberger’s state-trait anxiety inventory (STAIT), and Social connectedness scale (SCS).

15Acute: assessed weekly during the dose period on Thursdays, when either a microdose or placebo capsule was taken. The testing was carried out 2–6 hr after the ingestion of the capsule, while the potential microdose was active. Acute outcomes were positive and negative affect schedule (PANAS), visual analogue scale items (drug intensity, mood, energy, creativity, focus, and temper) and cognitive performance (see Accumulative above for details).

16An overview of the outcomes can be found in Table 1 and a description of each measure is in Appendix 1. See Figure 1 for the experimental timeline and assessment timepoints.

17A high-level overview of self-blinding is provided here; for a detailed illustration see Figure 2. First, two sets of capsules had to be prepared using non-transparent capsules: one set with microdoses inside and another set without anything inside (placebos). Next, these capsules were packaged into weekly sets, which were then placed inside envelopes together with a QR code (Figure 2A). The envelopes were grouped and shuffled. Then, using a semi-random drawing process, four of them were selected (Figure 2B) corresponding to the 4 weeks of the dose period (i.e. each envelope held capsules for 1 week of the dose period). The drawing process was constrained such that only three combinations of the envelopes were possible to draw, matching the three study groups: placebo (four placebo weeks), half-half (2–2 placebo and microdose weeks), and microdose group (four microdose weeks; Figure 2C). At this stage, participants were ready to start the experiment.

18When the dose period started, one envelope was opened per week and the capsules inside were used as scheduled (Figure 2D). Additionally, the QR code from the envelope had to be scanned, which shared a numeric code with our informatics infrastructure. The decryption key (i.e. how capsule types are encoded by the numbers) was not shared with participants, so the numeric code allowed only us to deduce which type of capsule was taken when.

19In summary, the two key elements of self-blinding are to hide the active components inside opaque capsules while preparing identical looking placebos (1) and to position non human-readable QR codes along the capsules prior to randomization (2). With the QR codes in place, it is possible for the experimenter to recover knowledge of capsule types after randomization without revealing that information to participants.

20Participants were allowed to use any psychedelic substance to microdose with. The microdose dose, which is the amount of substance to use as a microdose, was not defined for participants, rather they were instructed to use a microdose dose that they would use outside the study. The rationale for this direction was threefold. First, given that participants typically would source their substance from the black market, the precise microdose dose could not have been known even if instructions requested it. Second, based on community feedback, most experienced microdosers have a preferred dose that they would not have liked to change to participate in the study. Lastly, this study was not a clinical trial and therefore from a regulatory perspective not allowing for control over and/or directing about drug doses.

21Psychedelics users were recruited through advertisement on relevant online and offline forums. Individuals could sign up through the study’s website, https://selfblinding-microdose.org/, where they could find information about the study, including the study manual and explainer videos, the participant information’s sheet, and procedure for declaring informed consent. Once informed consent was given, individuals were able to sign up by providing their email address and planned start date. The inclusion criteria were: >18 years of age, good understanding of English, intention to microdose with psychedelics, previous experience with psychedelics (either micro- or macrodosing), no use of psychedelic drugs from a week before the start until the completion of the post-regime timepoint (other than the study’s microdoses), and willingness to follow the study protocol.

22All the questionnaires were implemented online using the SurveyGizmo platform (https://www.surveygizmo.com/). For the online assessment of cognitive performance, the Cambridge Brain Sciences (https://www.cambridgebrainsciences.com/) service was used. At each timepoint, links to each test were sent in a dedicated email via the Psychedelics Survey (https://www.psychedelicsurvey.com/) service. These links had a personal ID embedded, so each test completion could be matched to individuals.

23Participants were asked to guess which type of capsule they had taken that day during the dose period (for days when capsule was taken). This guess was a forced binary choice between microdose and placebo options. At the end of the post-acute test sessions, participants were asked separately to guess whether the current week was a microdose or a placebo week (Figure 1A). In the discussion of our results, the term ‘break blind’ indicates that the participant guessed the capsule correctly for the day (acute outcomes) or week (post-acute outcomes). No guess was collected about perceived group allocation at the end of study, because information about group structure was not shared with participants.

24Group differences in demographics, recreational drug use, and baseline scores of the accumulative outcomes were assessed with ANOVA and chi-square tests for continuous and categorical variables.

25Accumulative outcomes were analyzed with mixed-effect repeated measurement models, using the SAS PROC MIXED method with compound symmetry covariance structure. Models were constructed with change from baseline as the dependent variable, group, time and group*time interaction as factors, and individuals as experimental unit. Models were adjusted for all significant baseline covariates (the following variables were tested as potential covariates: age, sex, education, baseline score, dose, total dose, short suggestibility scale score, expectation score, number of past psychiatric diagnosis, number of current psychiatric medications, number of lifetime macrodose experiences, and number of lifetime months microdosing). To accommodate dose as a potential covariate, psilocybin mushroom mass was converted to an estimated equivalent LSD dose (0.1 g of dried mushroom ~4.6 µg LSD;; ). The following planned comparisons were made: within-group comparisons of change over time from baseline to the primary endpoint at week 5 and from baseline to the final follow-up at week 9. Additionally, between-group comparisons were made (PL vs. HH and PL vs. MD) at week 5 and week 9.

26To analyze acute and post-acute outcomes, mixed linear models were constructed. Models included score as dependent variable, subject ID as a random-effect, and condition as fixed-effect, where condition was a binary categorical variable (PL/MD). For acute outcomes, condition was PL/MD when the score was obtained under the influence of a placebo/microdose capsule, while for post-acute outcomes condition was PL/MD when the score was obtained at the end of placebo/microdose week. Planned comparisons were made between scores obtained under PL and MD conditions. Each participant contributed four scores to these models, corresponding to the four acute/post-acute assessment timepoints during the dose period. All acute/post-acute models were adjusted for all significant baseline covariates (same variables were tested for significance as in the case for the accumulative outcomes, except baseline score and total dose consumed).

27To better understand how guess influenced scores, a second set of models were constructed with the addition of guess (binary categorical variable, PL/MD) and guess*condition factors. Using these guess adjusted models, planned comparisons were made between PL and MD conditions. Finally, the two binary variables (condition and guess) divided the data into 2*2 = 4 strata, post-hoc comparisons were made between the following strata (condition/guess): PL/PL vs. MD/PL, PL/MD vs. MD/MD, PL/PL vs. PL/MD and MD/PL vs. MD/MD. This selection was made such that condition changes while guess remains fixed in the first two comparisons, and guess changes while condition remains fixed in the last two comparisons.

28The study only engaged people who planned to microdose through their own initiative with their own psychedelic substance, but who consented to incorporate placebo control to make their self-experimentation compatible with our study. Investigators did not endorse any use of psychedelics, and no financial compensation was offered to participants. Email addresses were the only personally identifiable data collected. The email address was retained after study completion if permission was given (checkbox) by the participant to receive information regarding future studies, discarded otherwise. The study was approved by Imperial College Research Ethics Committee and the Joint Research Compliance Office at Imperial College London (ICREC reference number 18IC4518).

29A total of 1630 participants signed-up, 240 started, and 191 participants completed the study. The optional follow-up at week 9 was completed by 159 individuals. No statistically significant differences were found between the groups in any demographic, recreational drug use or baseline measures, confirming efficiency of the randomization (see Supplementary file 1 for details on demographics, Supplementary file 2 for recreational drug use, and Supplementary file 3 for statistical analysis of baseline variables). Completion rate was highly similar across the three groups (χ2(12, N = 240)=0.64, p=0.99), see Figure 3.

30For the most part, the sample consisted of educated, middle-age (33.5 ± 9.4), healthy males (70% male, 19% female, 1% other) from western countries. As expected, most participants had a positive attitude toward psychedelic drugs, in particular toward medical use: 74% and 90% either agreed or strongly agreed with the statements 'I am an active advocate of psychedelic drug-use' and 'I am an active advocate of the therapeutic use of psychedelics', respectively. See Appendix for details on the sample’s expectations/attitude about microdosing and psychedelics. The sample consisted of healthy individuals for the most part: 33% of participants reported to have had at least one psychiatric diagnosis in the past, the most frequent past diagnoses were: anxiety disorder (13%), depression (13%), and PTSD (7%). Only 7% of the sample had current mental diagnosis.

31Most participants microdosed with LSD (n = 147; 61%)/LSD analogue (n = 33; 14%), followed by psilocybin containing mushrooms (n = 57; 24%) and three individuals used other psychedelics (LSA: n = 1; DOB: n = 2). The average reported dose for LSD/LSD analogues was 13 ± 5.5 µg, while for psilocybin mushroom it was 0.2 ± 0.12 g, see Appendix 1—figure 3 for further details.

32Accumulative outcomes were first collected at baseline, then at week 5 (i.e. after the completion of the 4 weeks long dose period) and at the optional long-term follow-up timepoint at week 9. The following two sets of pre-planned comparisons were made: within group comparisons of baseline vs. week 5, baseline vs. week 9 (changes over time) and between-group comparisons at the week 5 and week 9 timepoints. Sample sizes were n = 240/191/159 at baseline, week 5 and week 9, respectively. Data was also analyzed separately for LSD/LSD-analogues and psilocybin microdoses, the results from both subgroups matched the results of the combined analysis presented here.

33For the within group (change over time) comparison of baseline vs. week 5, all self-reported psychological outcomes improved significantly in the MD group: well-being (RPWB) increased with 4.2 ± 3.9 (adjusted mean estimate ±95% CI; p=0.04*), mindfulness (CAMS) increased with 2.4 ± 1.1 (p<0.001***), life satisfaction (SWL) increased with 1.2 ± 1.2 (p=0.04*), and paranoia (GPTS) decreased with −5.0 ± 1.7 (p<0.001***). Personality structure (B5) showed reduced neuroticism trait score (−1.3 ± 0.9, p<0.01**) and increased openness (0.9 ± 0.8, p=0.03*). Significant changes over the same period (from baseline to week 5) were also observed in the PL and HH groups for mindfulness (PL: 1.6 ± 1.1, p<0.01**; HH: 1.3 ± 1.2, p=0.02*) and paranoia (PL: −3.4 ± 1.7 p<0.001***; HH: −4.9 ± 1.9 p<0.001***), but not for well-being or life satisfaction. Neuroticism also decreased in the PL group (−1.0 ± 1.0, p=0.04*). Changes in mindfulness and paranoia were sustained at the week 9 follow-up timepoint for all groups, while decreased neuroticism only prolonged in the MD group, see Supplementary file 5 for details. CPS did not change in the MD group (from baseline to week 5), but significantly decreased in the HH group (−0.16 ± 0.14, p=0.03*). Among individual cognitive tests over the same period, rotations (0.34 ± 0.28, p=0.02*) and odd one out (0.52 ± 0.31, p=0.001**) increased significantly in the MD group, while spatial span (−0.49 ± 0.30, p=0.02*) and paired associates (−0.51 ± 0.30, p=0.02*) decreased in the HH group. The increased rotations score in the MD group was sustained at the follow-up (0.45 ± 0.46, p<0.01**), but not the other task scores.

34Planned comparisons revealed no significant between-group differences at either the week 5 or week 9 follow-up timepoints, including all subscales, except that in the HH group the paired associates scores decreased (PL vs HH adjusted treatment difference: −0.55 ± 0.43, p<0.01**). Time course of the adjusted mean estimates is summarized in Figure 4. See Supplementary file 4 for descriptive statistics, including subscale and individual cognitive test scores, adjusted over time and between group differences (Supplementary file 5), and model parameters (Supplementary file 6).

35As secondary analysis to further examine the role of placebo-like expectation effects in the accumulative outcomes, we performed a post-hoc adjustment by adding the ‘number of times microdose capsule was guessed’ variable as a covariate to the models (irrespective whether the guess was correct or not). This variable was significant for some models (RPWB: p<0.01**; CAMS: p=0.02*; B5 agreeableness: p=0.02*; B5 openness: p=0.03*) and further decreased the already small between-group differences on self-reported scales, while it did not affect cognitive outcomes. Specifically, the adjusted treatment difference (±95% CI) at the week 5 timepoint between PL and MD groups without/with the number of MD guesses covariate was: well-being (RPWB) 2.5 ± 5.6 (p=0.37)/0.9 ± 5.7 (p=0.76), mindfulness (CAMS): 0.8 ± 1.5 (p=0.32)/0.4 ± 1.5 (p=0.65), paranoia (GPTS): −1.6 ± 2.5 (p=0.21)/−1.2 ± 2.5 (p=0.36), life satisfaction (SWL) 0.4 ± 1.7 (p=0.67)/0.2 ± 1.8 (p=0.83), B5 intellect: −0.2 ± 1.2 (p=0.80)/−0.2 ± 1.2 (p=0.71), B5 openness: 0.3 ± 1.2 (p=0.57)/0.0 ± 1.2 (p=0.97), B5 neuroticism: −0.3 ± 1.4 (p=0.70)/−0.1 ± 1.4 (p=0.87), B5 extraversion: −0.2 ± 1.2 (p=0.81)/−0.4 ± 1.3 (p=0.52), B5 agreeableness: 0.5 ± 1.1 (p=0.37)/0.2 ± 1.1 (p=0.75), and B5 consciousness: 0.8 ± 1.3 (p=0.24)/0.5 ± 1.3 (p=0.44).

36First, outcomes are described without considering the guess component, which is discussed in the next section. Acute outcomes were measured during the dose period while the potential microdose was still active, while post-acute outcomes were measured every Sunday, when no capsule was taken, 48–72 hr after the last placebo/microdose capsule. For psychological measures the average sample size was 857 (between 849 and 884 due to partial completions; participants contributed four scores corresponding to the four acute timepoints, see Materials and methods for details), while for cognitive performance it was 684 (between 678 and 689). Data was also analyzed separately for LSD/LSD-analogues and psilocybin microdoses, and the results from both subgroups matched the results of the combined analysis presented here.

37Among acute measures, condition (PL vs. MD) was significant for acute emotional state (PANAS) (adjusted mean estimate ±95% CI: 2.2 ± 1.4, p<0.01**) and the acute drug intensity (12.5 ± 3.0, p<0.001***), mood (4.6 ± 2.9, p<0.001***), energy (5.3 ± 2.7, p<0.001***), and creativity (4.7 ± 2.6, p<0.001***) VASs, meaning that scores collected on days when a microdose was taken were significantly higher compared to scores collected on placebo days. Effect sizes, as quantified by Cohen’s d, remained small (d < 0.3) on all scales, with the exception of the drug intensity VAS (d = 0.58).

38Among post-acute measures, condition was significant only on the anxiety measure (STAIT; −1.4 ± 1.3, p=0.03*), meaning that anxiety was reduced at the end of microdose weeks compared with placebo weeks, see Table 2 for details on both acute and post-acute outcomes.

39Next, the acute and post-acute results were re-analyzed with the addition of guess into the models. Condition (PL vs. MD) was no longer significant for any scale, except for acute drug intensity VAS (adjusted mean difference ±95% CI: 3.4 ± 2.0; p<0.001***), which increased under MD (Table 2). The guess*condition interaction term was non-significant for all scales, except for drug intensity (p<0.01**).

40To better understand the role of guess, the data was further analyzed by comparing the 2*2 = 4 strata formed by the two binary variables, condition (PL/MD), and guess (PL/MD), in the models. For self-reported outcomes, no significant differences were found between microdose and placebo conditions with fixed guess (condition/guess: PL/PL vs. MD/PL and PL/MD vs. MD/MD comparisons), except for acute drug intensity visual analogue scale, which was higher when microdose was taken (adj. mean difference ±95% CI; 7.3 ± 3.1, p<0.001***). Conversely, when drug condition was fixed (condition/guess: PL/PL vs. PL/MD and MD/PL vs. MD/MD comparisons), significant differences were found in 21 of the 22 comparisons (=2*conditions*(4*post-acute+7*acute scales)), all favoring MD guess. These findings suggest that scores are significantly better when the participant believed they had taken a microdose irrespective of what was actually taken. Taking an actual microdose was only associated with a significant difference in the drug intensity scale. Figure 5 shows the stratified distribution of selected outcomes, see Supplementary file 8 for all comparisons.

41Break blind rate, defined as the proportion of correct capsule guesses (see section Blind breaking and collection of guess data for details), was 0.72 ± 0.18 (M ± SD). Specificity (true negative rate: ratio of true placebo guesses to all placebo guesses) was 0.82 ± 0.16, noticeably higher than sensitivity (true positive rate: ratio of true microdose guesses to all microdose guesses) 0.45 ± 0.30, meaning that placebo capsules were guessed correctly at a higher rate than microdoses. Based on knowledge of the ratio of PL/MD capsules (3/1) in the envelopes, which is evident to participants when they prepare the capsules, a ‘random guesser’ would have a break blind rate of 0.62 with 0.75 specificity and 0.25 sensitivity. The high sensitivity exhibited by participants (0.46 vs. the random guesser’s 0.25) suggests that the higher than random break blind rate is mostly due to superior ability to identify microdoses, see Appendix 1—table 1 for details.

42Break blind rate was positively associated with reported microdose dose (F(1, 237)=7.4, p<0.01**), meaning that the higher the dose was, the more likely participants guessed their daily condition correctly. For this analysis psilocybin mushroom doses were converted to estimated LSD dose equivalent, see Statistical analysis in Materials and methods for details. The estimated ‘detection threshold’, that is, the dose above which participants guess significantly better than random, was 12 µg.

43We employed a novel self-blinding methodology to investigate the acute, post-acute, and long-term, accumulative effects of psychedelic microdosing. To the best of our knowledge, this study is the first one to use a self-blinding methodology, the first placebo-controlled investigation of the accumulative effects of repeated microdosing, and the largest placebo-controlled psychedelic study to-date.

44When looking at changes over time from baseline to week 5 (accumulative outcomes) in the microdose group alone, results confirmed the psychological benefits reported by anecdotes and observational, uncontrolled studies: significant improvements were observed in the domains of well-being, mindfulness, life satisfaction, and paranoia. However, when looking at the between-group comparisons of the same outcomes, no significant differences were found between the placebo and microdose groups. On the cognitive tests, which are less subjective than the self-reported psychological outcomes, the microdose group did not even improve from baseline to week 5 and the between-groups comparisons were not significant either. Thus, our study validates the positive anecdotal reports about the psychological benefits of microdosing (significant improvements from baseline in a broad range of psychological measures); however, our results also suggest that these improvements are not due to the pharmacological action of microdosing, but are rather explained by the placebo effect (lack of significant between-groups differences).

45Similar conclusions can be drawn from the examination of the acute and post-acute outcomes as well. In our initial analysis without incorporation of the guess component, we detected significant effects on post-acute anxiety (STAIT), acute emotional state (PANAS), and mood, energy, creativity, and drug intensity (visual analogue scale items). Effect sizes were small on all scales (Cohen’s d < 0.3 except drug intensity); thus, the clinical and practical value of these effects is debatable. Furthermore, when the guess component was added to the models, the already small differences disappeared on all scales, except for acute drug intensity. It can be argued that the addition of the guess variable to the models may undermine the statistical significance of the condition effect due to collinearity between condition and guess. To overcome this potential issue, we conducted the stratification analysis where only one of these variables is changing, while the other remains fixed. No significant differences were observed between placebo and microdose conditions when the guess was fixed (condition/guess; PL/PL vs. MD/PL and PL/MD vs. MD/MD comparisons), except for drug intensity (MD>PL). Conversely, when condition was fixed (PL/PL vs. PL/MD and MD/PL vs. MD/MD comparisons), scores obtained under placebo and microdose guesses were significantly different in 21 out of the 22 comparisons, always favoring the microdose guess, see Figure 5 and Supplementary file 8. Importantly, neither CPS nor any cognitive subtask, the non-self-rated outcomes where beliefs and subjective feelings are likely to be less influential, were significantly different under either guess or drug conditions. In summary, these results strongly suggest that the actual content of capsules did not determine differences between the conditions, but beliefs about their content did.

46An important observation was that participants guessed their capsules correctly in 72% of the cases. This break blind rate was higher than random (random: 63% vs. participants: 72%), but not as high as reported in antidepressant studies (around 80%). It is known from a variety of clinical studies that higher break blind rate is associated with larger between-conditions effect-sizes (where placebo is the control condition). This relationship is explained by non-specific treatment factors such as expectation of a benefit and investigator alliance. The influence of such factors is likely to be large for the present study, because of highly positive expectations and favorable attitudes toward psychedelics, see attitude analysis in the Appendix. These factors together suggest that the observed ‘significant’ acute and post-acute effects may be an artifact of the combination of break blinds and expected benefits. The acute and post-acute results observed could be understood as the difference between the expected benefits when a microdose is perceived (i.e. guessed by participants) versus the absence of expected benefits when placebo is perceived. This difference in expectations could be mistaken for a ‘real’ drug effect in any study where blinding integrity is not considered during analysis. If this explanation is correct, one prediction for future microdose studies with a similarly pro-psychedelics sample is that they may observe larger effects when break blind rate is higher, or conversely, smaller effects when break blind rate is lower.

47What factors account for the blind breaking? Drug intensity was the only outcome that remained significant even after adjusting for guess (3.4 ± 2.0; p<0.001***). This observation suggests that drug intensity is a small, but true drug effect. This increased drug intensity mostly manifested as body and perceptual sensations, see Blind breaking cues in Appendix 1 for details. This finding suggests that in most cases blind breaking induced clinically irrelevant side effects, rather than deduced from improvements of outcome variables. We note that according to our data the threshold LSD dose where participants guess better than random is 12 µg, see Figure 4, which is in line with the 13 µg threshold dose estimated by a recent dose controlled study.

48It is worth noting that the current study was designed to protect blinding integrity by including placebos for the microdose group as well, administering the microdose capsules on different days of the week and by including the half-half group. The 3-arm design can be seen as a strength in this regard, adding ambiguity and thus strengthening blinding. Illustrative of the integrity of the blind, we received several emails from participants in the PL group who were in disbelief after opening their unused envelopes containing unused capsules after the conclusion of the study:

49“I counted the number of cut blotters I had in the left overs: they are 8...so you must be right... Which is incredible […] Some days during the test were really, really focused and colours more vivid. This sensation was really new to me".

50"I have just checked the remaining envelopes and it appears that I was indeed taking placebos throughout the trial. I'm quite astonished […] It seems I was able to generate a powerful 'altered consciousness' experience based only the expectation around the possibility of a microdose".

51"An empty pill with strong belief/intentions makes nearly everything. You put spirituality into an empty pill here...wow!"

52It is our view that the present part-controlled, part-observational design yields data superior to conventional observational data (inclusion of placebo control), but inferior to controlled clinical trial data (incomplete control over recruitment, screening, assessment, drug administration, etc.). This study does, however, have greater ecological validity than would a fully controlled lab study.

53A key limitation of the present study is the lack of verification of the nature, purity, and dosage of the psychedelic substance used for microdosing. Psilocybin-containing mushrooms were used by 23% of the sample, 14% used legal LSD analogues (such as 1P-LSD), whereas 62% sourced their substance from the black market, mostly LSD (61%). According to the Energy Control's drug checking service (Barcelona), LSD blotter adulteration rates were low during the period when our study was running: in both 2018 and 2019 blotters sold as LSD contained LSD only in 90% (n = 735) of tested samples [personal communication with M. Ventrua from EC, June 2020]. The exact quantity of active ingredient within a given microdose cannot be known with certainty; however, the positive relationship between dose and blind breaking (Figure 4) and that the threshold dose for psychoactivity was consistent with a recent controlled study (12 µg vs 13 µg; ) provide some reassurance. Nonetheless, our results should be not understood as clinical evidence, rather they are representative of ‘real life microdosing’.

54We could not confirm whether participants followed accurately the self-blinding procedure. Three individuals reported following an invalid sequence of weeks, but these individuals did their setups together, all committing the same mistake (1.3% error rate). Furthermore, we had no way of confirming whether the capsules were taken as instructed during the dose period. Instructions emphasized not to complete assessments planned on dosing days in case the dose schedule could not be followed for any reason, but we could not confirm whether participants adhered to this rule.

55Our stratification analysis does not allow for a strict determination of a causal relationship between guess and outcome, because guess was recorded after completion of assessments, guess was last question during test sessions. After closing the study, a survey was conducted among participants, where 86% (n = 166) responded that "I was thinking about whether I took a microdose or placebo even before I was asked to guess" (opposed to "I was not thinking about whether I took a microdose or placebo, except when I was asked to guess"), making a causal interpretation more likely. We note that the order we chose is consistent with previous work in psychiatric studies; had the guesses been requested prior to the assessments, it could have primed responses. Also, we cannot rule out that performance during the assessments influenced the guess. However, the lack of any feedback from the assessments mitigates this risk. Most participants reported to break blind due to body and perceptual sensations, rather than improved outcomes, see Blind breaking cues in the Appendix for details.

56We cannot rule out the possibility that a study in a clinical population would yield more promising results. In the present healthy sample, where well-being scores are high at baseline, there is less scope for potential improvements, which could have prevented the observation of placebo-microdose differences. Most study participants reported not to have any history of mental health problems; only 7% reported having a current psychiatric diagnosis, and 33% reported to have had a psychiatric diagnosis in the past (Supplementary file 1). We conducted two post-hoc analysis for two selective pseudo-depression subsamples: participants with the lowest 25% baseline well-being scores and those with the highest 25% baseline neuroticism scores. Results in these subsamples were entirely consistent with those from the complete sample: there were no significant differences between conditions for any of the accumulative outcomes (adjusted treatment difference ±95% CI of PL vs MD at week 5 for the lowest 25% baseline well-being subsample: well-being (RPWB) −1.6 ± 13.6 (p=0.81), mindfulness (CAMS) 0.3 ± 3.3 (p=0.85), paranoia (GPTS) −5.1 ± 6.8 (p=0.14), life satisfaction (SWL) 0.3 ± 4.5 (p=0.87), cognition (CPS) 0.1 ± 0.55 (p=0.71); same measures for the highest 25% baseline neuroticism subsample: well-being (RPWB) 4.8 ± 14.3 (p=0.50), mindfulness (CAMS) 1.3 ± 3.7 (p=0.49), paranoia (GPTS) −3.1 ± 8 (p=0.43), life satisfaction (SWL) −1.4 ± 4.6 (p=0.53), cognition (CPS) 0.04 ± 0.67 (p=0.90)). Thus, although not designed as a clinical study, data from this opportunistic naturalistic study do not provide support for clinical effects of microdosing.

57Although this was the largest placebo-controlled psychedelic research study published to-date, we note that one could argue that the study was still underpowered to detect a true effect based on the fact that the MD group did improve more than the PL group on all scales (from baseline to week 5), but just not to a statistically significant extent (Figure 4). On the well-being scale (RPWB), the adjusted PL vs. MD group difference was 2.5 ± 5.6 points. To illustrate this difference in practice, this scale consists of 42 statements that participants rate on a 6-point Likert scale (Strongly disagree - Strongly agree), thus, the full range of scores is thus 0–252, so the 2.5 point mean difference is 1% of the total scale. This difference is equivalent to scoring one item, for example ‘I like most aspects of my personality’, Strongly agree instead of Slightly agree or Slightly disagree, while responding the same to the remaining 41 items. Based on our data, we calculated that the sample size (90% power and alpha of 0.05) required to observe a true between-group difference would be: 1508 for well-being (RPWB), 1638 for mindfulness (CAMS), 4918 for life satisfaction (SWL), 1392 for paranoia (GPTS), and 366 for cognitive performance (CPS). These differences therefore are not clinically meaningful or sufficient to justify the cost of intervention.

58The successful execution of this initiative here may inspire similar initiatives throughout the world in a broad range of scientific and medical contexts. Controlling for placebo effects is important for trending phenomena, such as cannabidiol (CBD) oils, nootropics, and nutrition, where social-pressure, expectancy, positive-test strategies, and confirmation bias can lead to false-positive findings. Self-blinding citizen-science initiatives could be employed in these areas as a cost-efficient screening tool prior to conducting expensive clinical studies.

59An important feature of the self-blinding methodology is the low cost; we estimate that the current study’s costs were about 0.5–1% of an equivalent clinical study. Since the research team is not providing the study drug/placebo and on-site staffing is not required, expenses are similar to a conventional observational study, yet still with incorporation of randomization and placebo control.

60Important lessons can be taken from the current study for the design of future microdosing trials. The combination of the lack of detected efficacy in this study and an association between self-reported doses and ability to break blind (see Figure 4) suggest that selecting dosage is fraught with difficulties: if a low microdose is chosen, efficacy is unlikely if we extrapolate current results, whereas a high microdose could jeopardize the blinding. Randomization to microdose versus an active placebo conditions (e.g. niacin, which has been employed in macrodosing studies ) and careful assessment of blinding could, in principle, alleviate some of these concerns.

61The present study also has implications for full/‘macrodose’ psychedelic studies, where blinding is impossible due to the intense nature of the experience. It can be hypothesized that the intense hallucinations are essential for therapeutic outcome, questioning the suitability of placebo-controlled trials in this context. The fact that one may be unable to fully extricate belief, or ‘context’ more broadly, from the direct (e.g. pharmacological) action of a given intervention, raises interesting philosophical and ethical question with implications for drug development and regulation. One might also hypothesize that the action of microdosing and psychedelics relies on prior and continuously updating belief combining (perhaps synergistically) with a direct drug effect. Such a positive interaction could, in theory, be tested, and if endorsed, this could be interpreted as implying that belief is an active component of the psychedelic treatment model, rather than a problematic confound.

62In summary, here we created a novel, cost-effective, self-blinding, citizen-science methodology that enabled us to conduct the largest placebo-controlled study on psychedelics to-date and the first placebo-controlled examination of repeated psychedelic microdosing. Our findings confirm the anecdotal benefits of microdosing (improvements in a broad range of psychological measures); however, the results also suggest that the improvements are not due to the pharmacological action of microdosing, but are rather explained by the placebo effect (lack of significant between-groups effect).