The study was approved by Imperial College Research Ethics Committee and the Joint Research Compliance Office at Imperial College London and carried out in accordance with principles of good clinical practice. Written informed consent was obtained from all subjects. The original survey is now closed, but the revised and still active versions of related surveys are available on the PsychedelicSurvey website [42].

Data were collected as part of a larger prospective study [9], approved by the Imperial College Research Ethics Committee and Joint Research Compliance Office. Only the elements of the design and data, which are relevant to this study, are presented here. Data were collected on the internet from a convenience sample of psychedelic drug users in a noncontrolled, naturalistic, and observational manner, through the website and software PsychedelicSurvey platform [42]. The open survey study was advertised on social media platforms, and informed consent was collected through a button-click feature following information on study purpose, design, and recruitment criteria. After sign-up, participants received emails that contained links to the relevant surveys at multiple timepoints, which were implemented and hosted by the web-based service system Alchemer [43]. Data were collected through a prospective cohort design: the baseline timepoint was set at 1 week before a planned psychedelic experience (timepoint 1), preacute measures were taken 1 day before the planned psychedelic experience (timepoint 2), and postacute measures were taken 1 day after the planned psychedelic experience (timepoint 3); the first endpoint was 2 weeks after the relevant experience (timepoint 4). The subsequent key endpoint occurred 4 weeks after the planned psychedelic experience (timepoint 5).

Baseline demographic and trait variables, postacute subjective effects measures, and outcome measures at the first endpoint were used to predict attrition. The survey 4 weeks after the psychedelic experience was the key endpoint, the completion of which was used as the criterion to determine attrition vs completion of the study. Dropouts were defined as those participants who stopped responding to the study surveys and did not return to finish the key endpoint, and completers were defined as those participants who reached the key endpoint 4 weeks after the psychedelic experience, even if they missed one or more previous timepoints.

Logistic regression was used to assess the influence of the included variables on premature study termination, using the binary outcome attrition or completion as the dependent variable, where a participant was considered a completer if they had responded successfully to the key endpoint 4 weeks after the psychedelic experience. Owing to its nonreliance on the normal distribution of predictor variables, logistic regression is well-suited for the analysis of skewed data that have limited variability. Four models were fitted: 2 models containing baseline variables (1 including demographics, the other psychological measures), 1 containing postacute measures of psychedelic effects, and 1 with changes on relevant outcome variables between baseline and the 2-week endpoint. For those categorical and ordinal variables included in the demographic logit, we chose “Male” as reference category of gender owing to its greater size (n=485) compared to other levels of the variable, and the educational level “secondary education” owing to both its association with premature study termination known from the literature and its relevance in terms of sample size (n=276). Owing to a significant correlation (ρ=0.18; P<.001) and to reduce multicollinearity issues, it was decided to include only the educational level but not employment status as a predictor of attrition in the demographics logit. We also performed logistic regression analysis to investigate whether differences in psychedelic-induced changes on relevant outcome variables account for the likelihood of attrition. To compute change scores, individual baseline scores were subtracted from scores at the 2-week endpoint for each variable of interest (ie, WEMWBS, QIDS, STAI-SF, and PDI). Subsequently, a single logit was fitted, which included both change scores and absolute baseline scores, to account for the effect baseline variables. All statistical analyses were conducted using R (version 3.6.1, 2019-07-05; The R Foundation).

In total, 564, 535, 379, 315, and 212 participants were sampled at survey timepoints 1, 2, 3, 4, and 5, respectively. Since some participants did not complete the baseline survey but responded at a subsequent timepoint, the overall number of participants was >654, (n=741). Of this total of 741 participants responding at any study timepoint, 529 (71.4%) subsequently dropped out at or prior to the final survey at 4 weeks post psychedelic experience. In particular, 170 (22.9%) participants stopped responding 1 day prior to the planned psychedelic experience (preacute timepoint), 139 (18.8%) stopped 1 day post experience (postacute timepoint), 89 (12%) stopped at the 2-week endpoint, and 131 (17.7%) stopped at the second follow-up and key endpoint 4 weeks post experience (Figure 1).

The distribution of dropouts according to timepoint was significantly different (χ²₄=56.89; P<.001); more noncompleters dropped out prior to rather than after the postacute timepoint: 309 (58.4%) vs 220 (41.6%), respectively (χ²₁=29.28; P<.001). In total, 212 (28.6%) participants completed the second follow-up 4 weeks post experience and were thus classified as completers. Table 1 lists the demographics of participants at each timepoint of interest in this study.

Table 2 presents the results of logistic regression analysis. Age significantly predicted attrition (β=–0.024; P=.007); specifically, older age was associated with a reduced probability of dropping out from the study. Participants with a university degree were less likely to drop out than those with a secondary educational level (β=–0.574; P=.005), and participants with primary education were also less likely to drop out than those with secondary education (β=–0.876; P=.008). Personality trait extraversion significantly predicted attrition (β=0.082; P=.012): higher scores on extraversion were associated with an increased probability of dropping out from the study. Personality trait conscientiousness also significantly predicted attrition (β=–0.079; P=.024) in the opposite direction: higher scores on conscientiousness were associated with a reduced probability of dropping out. All other assessed variables included in logistic regression analyses were nonsignificant in predicting attrition, including any measures of the acute psychedelic state and measures potentially related to adverse events, such as suicidality, delusional thinking, physical side effects, or challenging experiences. Similarly, no psychedelic-induced changes on outcome variables significant predicted study attrition.

This study aimed to identify variables accounting for attrition in prospective web-based studies on naturalistic psychedelic use. The overall attrition rate, which increased as a function of time, was high at 71.4% (n=529 of 741 participants). Contrary to prior hypotheses, neither the intensity of challenging experiences nor the advocacy of psychedelic use measured at baseline significantly predicted study completion. Rather, demographic variables including age and education, as well as personality traits including conscientiousness and extraversion, affected the likelihood of study noncompletion.

Specifically, logistic regression analyses revealed that young age, a low educational level, and the big 5 personality traits (high) extraversion and (low) conscientiousness were predictors of study attrition. The finding of age is in line with a number of previous studies [16-19] and may act synergistically with low conscientiousness to increase the likelihood of dropout. Indeed, several cross-sectional and longitudinal studies have shown that conscientiousness tends to increase with age [57-60], and that these increases often occur only during adulthood, which is relevant to the present study population, which had an average age of 28.9 (SD 10.4) years. The demographic logit also showed that participants with secondary education were more likely to drop out than those with university degrees and those with primary education. This may in part also be related to conscientiousness, which is known to be associated with educational expectations [61]. Highly conscientious young people also perform better academically and gain more advanced educational qualifications [62]. The personality trait conscientiousness is defined as “the propensity to follow socially prescribed norms and rules, to be goal-directed, planful, able to delay gratification, and to control impulses” [63]; thus, a conscientious person may indeed be more likely to commit to any obligation that he/she undertakes, be it loyalty to one’s partner, paying taxes, or completion of a survey pertaining to psychedelic drug use. The finding that extraversion was a significant predictor of study attrition is also in line with previous studies reporting that extraversion predicts premature termination in longitudinal studies [34,64,65].

The absence of influence from any of the psychedelic-specific predictors is an important finding. Several previous studies [9-11,41] have indicated the quality of the acute psychedelic state to be a reliable predictor of longer-term psychological changes following psychedelic use. Recognizing the importance of acute subjective drug effects has been a key consideration informing the renewed interest in the therapeutic value of psychedelic compounds [41,66]. In the present study, neither the quality of the acute experience nor psychedelic advocacy or psychedelic-induced long-term psychological changes predicted study attrition. Given the accumulating number of studies reporting improvements in mental health outcomes after naturalistic psychedelic use [9,67-73], as well as impactful clinical trials involving psychedelic interventions [3,74-81], it can be considered reassuring that none of the established mediators of positive outcomes, nor outcomes themselves, seem to bias study attrition in longitudinal studies on psychedelics. Although it is, by definition, impossible to address with certainty how the noncompleters in this study sample faired in terms of their postpsychedelic mental health, the absence of a relationship between attrition and biased perspectives toward psychedelics at baseline, the nature of the participants’ acute experiences, or the reported beneficial effects (ie, psychological changes), partly ameliorates previous concerns regarding attrition bias in observational studies on psychedelics [9,10,39].

From among 529 dropouts, 309 (58.4%) stopped responding prior to the postacute timepoint, and approximately one-third did so even prior to the preacute timepoint, which could potentially be explained by impaired accessibility to and reduced desire to access the survey directly prior to and after the psychedelic experience or by the postponement or nonperformance of the experience. Although surveys could be completed through the mobile phone, it is conceivable that a portion of participants who consumed the substance under field conditions (eg, at a festival) stopped responding after the baseline assessment owing to pragmatic reasons of limited access as well as decayed motivation. As highlighted by others [29,30,82], predictors of attrition often differ for early vs late dropouts, and such differences may be important factors that allow for targeted interventions to reduce attrition; for instance, by activating reminders on mobile devices in the early phases of the prospective study. The effects of the substance use environment, associated intentions, and other contextual factors of psychedelic use on attrition should be targeted in future studies.

Together, the observed effects suggest that the principal reasons for study attrition in observational studies on psychedelics are largely similar to those evidenced by other longitudinal studies that assessed phenomena independent of and unrelated to psychedelic use, with demographics and personality traits being the most predictive. Nevertheless, the present negative results with regard to potentially problematic systematic biases should only be considered preliminary, considering the limitations associated with the prevailing study design. Most significantly, the absence of data for those individuals who dropped out before completing the relevant surveys implies that both adverse experiences may have still occurred but were merely undetected. In future studies, with prior consent, this issue could be tackled by automatically sending out very brief surveys (eg, single-item surveys) to nonresponders, to investigate their reason for nonresponse. Similarly, rare cases of extreme negative reaction driving subsequent dropout may have been missed owing to both dropout and group averaging. Future studies focusing on such negative outliers may create pre-emptive value, given the disproportionate attention that can be attracted by such cases and the damaging impact this can have on broader studies and clinical development efforts. Further limitations of our study include a self-selection recruitment bias, which, as discussed by Haijen [9], reflects in a predominantly young, male, highly educated sample displaying strong psychedelic advocacy. On a more general level, the comparability of attrition studies across scientific domains and disciplines is a nontrivial problem. For example, outcome measures will not be consistent across studies, with the exception of simple demographic factors such as age. As revealed by independent studies [83-85], those on the predictors of attrition are often inconclusive, inconsistent, lacking in generalizability, and vulnerable to design-related limitations including a lack of standardization in definitions of attrition itself. Nevertheless, some evidence is convergent and our findings did generally converge consistently.

This study sought to identify factors accounting for the high attrition rates in a prospective study on naturalistic psychedelic use. Consistent with findings from other scientific disciplines, the strongest predictors of study attrition were observed among variables including age, educational level, and personality traits. In contrast, psychedelic-specific factors were found to be poor predictors of attrition. Methods for reducing attrition, which have been validated through other fields, such as text messaging [86], gamification [87,88], monetary incentives [89,90], or the creation of web-based participant communities [91], are thus likely to be applicable also to observational studies on psychedelics, which should be investigated in future studies. While not without prevailing limitations, for which there is significant scope for improvement, these findings somewhat support the reliability and validity of large-scale prospective web-based data collection as a methodology for studying the predictors and processes of changes related to psychedelic use.