The present study sought to provide an overview of the main variables influencing antidepressant treatment response by investigating which antidepressant side effects, comorbidities, and demographic characteristics were associated with the discontinuation of antidepressant treatment. The most common side effects, weight gain and sexual dysfunction, were not associated with higher discontinuation rates. Although commonly believed to be negative [14], these side effects showed nominal evidence of being protective against discontinuation for some antidepressants. This finding is in contrast to previous studies suggesting that weight gain and sexual dysfunction are often associated with premature discontinuation [15, 16]. The nominal protective effect of the side effects identified here may indicate that antidepressants are working, which in turn may increase the tolerability for undesired side effects. Weight increases may be induced by an improvement in appetite as symptoms ease. Furthermore, sexual dysfunction, particularly in males, may be explained through the known action of antidepressants on testosterone and dopamine regulation [17, 18].

Specific side effects such as rashes, nausea, vomiting, anxiety, agitation, and suicidality were associated with an increased risk for treatment discontinuation. Our findings are partially consistent with previous observations, which reported that suicidality, mania, rashes, and headaches were associated with discontinuation from clinical trials [12]. These results suggest at least two pathways to discontinuation: an intolerance or allergic component manifested as nausea, vomiting, and rashes, and a psychopathology-related component including anxiety and suicidality. Currently, it is challenging to establish whether the latter is explained by a lack of antidepressant effectiveness as opposed to undesired adverse effects.

Studies reporting the effect of divorce on mental health suggest that married individuals are more likely to respond positively to antidepressant treatment [19]. In addition, we found that individuals with high educational attainment are less likely to discontinue antidepressant treatment, which in turn supports evidence from previous genetic studies [20]. Matching our results, discontinuation is recognized as a concern for individuals diagnosed with ADHD and PTSD [21, 22]. For instance, previous studies report low antidepressant effectiveness among patients with comorbid ADHD [23] or PTSD [24]. It is important to distinguish that these associations are more challenging to interpret due to a lack of information on age at onset. This limitation is less likely to affect associations with ADHD as it is typically an early-onset condition; however, we cannot rule out the possibility of participants starting and stopping an antidepressant regime before developing PTSD or ADHD. In addition, we note that the spectral nature of psychiatric disorders [25, 26] may influence our results. For instance, a higher risk for discontinuation among individuals with ADHD and / or PTSD may be indicative of another psychiatric comorbidity, such as personality disorder. Future studies should aim to delineate further effects and interactions between specific psychiatric disorders and treatment discontinuation.

Some limitations need to be acknowledged. We note that the generalizability of our results must be addressed with caution until confirmed in samples from other populations and ethnicities. Retrospective participant-derived reports of side effects and discontinuation may be subject to recall bias and subjective interpretations of participants. Related to this is the lack of information on comorbidity age of onset. For example, although the association between anxiety as a side effect and discontinuation was stronger compared to anxiety as a previous diagnosis, we do not have sufficient data to assess whether participants who reported anxiety as a side effect had a previous diagnosis of anxiety disorder which could continue through and after treatment. Similarly, due to the nature of the available data, we are not able to know the magnitude of certain side effects such as weight gain. For instance, it may be possible that participants in AGDS experienced weight gain that was insignificant to them and decided to continue their treatment, which is subject to personal interpretations and tolerability. The relationship between discontinuation due to adverse side effects and due to lack of efficacy is complex since these constructs are not independent. A patient could experience severe side effects but high efficacy, potentially increasing the patient’s tolerability to the side effects. However, if a patient experiences low efficacy, it is possible that the patient’s tolerability to side effects will be much lower, increasing the odds of discontinuation. Future studies should aim to further explore the intricate association between tolerability, discontinuation due to adverse side effects, and discontinuation due to lack of efficacy, which was out of the scope of this study.

We note that patients who discontinued more than one drug were considered in analyses for all the drugs they discontinued. Therefore, the effect of individual vulnerabilities predisposing participants to discontinue treatment may be exacerbated. To prevent this, we analyzed discontinuation for each drug separately, avoiding the inclusion of repeated measures in any analysis. In addition, when investigating treatment response for any given antidepressant, we are not able to account for previously taken antidepressants. Due to the nature of our data, we cannot determine if a patient who discontinued treatment was taking any other antidepressants simultaneously. Thus, we cannot account for the potential effects of taking multiple antidepressants or previously taken ones. As an alternative approach to model monotherapy, we conducted our analyses only considering the first prescribed drug for each patient. Although the sample size for this analysis was substantially reduced, which was reflected as lower statistical power, the direction and magnitude of the observed effects is largely consistent with the full results (Additional file 2: Tables S2). Finally, we currently do not have detailed data on treatment regimes such as dosages, which would enable ruling out extremely high or low starting doses as explanations for certain side effects.

In summary, we sought to shed light on which side effects, comorbidities, and demographic factors could influence antidepressant treatment discontinuation. We observed that not all side effects contribute equally to treatment discontinuation. Side effects associated with higher odds of discontinuing treatment included anxiety, agitation, suicidal thoughts, vomiting, and rashes. We showed that participants who also reported comorbid ADHD, PTSD, and a high neuroticism score were at an increased risk for treatment discontinuation. Altogether, our results elucidate which factors could potentially determine antidepressant treatment discontinuation and suggest that discontinuation due to adverse side effects can occur at early and late stages of treatment. We argue that specific drug–factor interactions should be studied in detail to develop new medication recommendations based on demographics and early side effect reports. We suggest that future studies should seek to validate further our results, evaluate discontinuation in monotherapy, estimate the effect of dosages, test for interactions between variables identified here, and investigate the relationship between discontinuation due to side effects, lack of efficacy, and tolerability.