Reversibility of Antipsychotic-Induced Weight Gain: A Systematic Review and Meta-Analysis

Background and Aims Weight gain is a major adverse effect of antipsychotic medication, negatively affecting physical and mental well-being. The objective of this study was to explore if dose reduction, discontinuation, switch to a partial agonist, or switch from polypharmacy to monotherapy will lead to weight loss. Methods Controlled and uncontrolled studies reporting the effects of discontinuation, dose reduction, switch to a partial agonist, or switch from polypharmacy to monotherapy on weight were included. Primary outcome was difference in weight compared to maintenance groups based on controlled studies. Secondary outcome was change in weight from initiation of one of the included interventions until follow-up in a pre-post analysis. Results We identified 40 randomized controlled trials and 15 uncontrolled studies including 12,279 individuals. The effect of the interventions, i.e. dose reduction, drug discontinuation, or switch to a partial agonis, reduced the weight with 1.5 kg (95% CI −2.03 to −0.98; P < 0.001) compared to maintenance treatment. The weight change from pre to post was a reduction of 1.13 kg (95% CI −1.36 to −0.90; P < 0.001). Conclusion We found a significant but small reduction in weight, suggesting that antipsychotic-induced weight gain can be reversed to some degree. Only a few studies were designed to address the question as primary outcome, which limits the generalizability of our findings.


INTRODUCTION
Shared decision making is the cornerstone of modern, evidence-based medicine and requires transparency about risks and benefits associated with any treatment decisions. Second generation antipsychotic medication is the mainstay of treatment of psychotic disorders (1) and is increasingly prescribed for other indications like bipolar disorder and severe depression and off-label indications such as sleeplessness and anxiety (2,3). Weight gain is a major adverse effect of second-generation antipsychotics, affecting quality of life (4), personal recovery, and somatic morbidity and is a common reason for antipsychotic discontinuation (5). Therefore, knowledge about the reversibility of antipsychotic weight gain is important to qualify the informed decision of both initiation and discontinuation.
Shared decision-making refers to a process where expert knowledge, data from the literature, and the patient's personal values and preferences are integrated in a final decision. Information about the risk of weight gain during treatment is important, and it is well documented in the scientific literature (6). However, knowledge about the reversibility of weight gain when the antipsychotic medication is reduced or discontinued remains scarce. To fully inform decisions to start or to stop medication, information about reversibility is crucial. Many may expect adverse effects to be reversible upon discontinuation, but this is not always the case with antipsychotic medication, as exemplified by tardive dyskinesia (7). Non-pharmacological interventions to reduce weight in a general population have shown only transient effect on obesity (8). While decades have passed with different trials of behavioral interventions, it is increasingly realized that temporary lost weight is normally rapidly regained, and obesity is now defined as: "chronic, relapsing, multi-factorial, neurobehavioral disease, wherein an increase in body fat promotes adipose tissue dysfunction and abnormal fat mass physical forces, resulting in adverse metabolic, biomechanical, and psychosocial health consequences." (https:// obesitymedicine.org/obesity-algorithm/).
If antipsychotic-induced obesity is an irreversible adverse effect, it has consequences for the shared decision making and potentially also for the off-label prescribing. If there is a risk of developing life long, severe obesity, some might find the riskbenefit balance to tip against medication. On the other hand, if the patient considers discontinuing antipsychotic medication to lose weight, it is important to know if this is likely to happen, to balance against the risk of relapse.
None of the second-generation antipsychotics are weight neutral, but some are more obesogenic than others, with olanzapine and clozapine having the highest obesogenic properties. Substantial effort has been put into understanding the molecular mechanisms underlying the obesogenic properties of antipsychotic medication, without clear answers (9). While the dopamine blockade produces the antipsychotic effect, several other neurotransmitters like the muscarinic, serotonergic, and histaminergic systems account for the metabolic effects, but no definite explanations have been established (10). Recently, intense focus has been put on the gut microbiota, which is linked to antipsychotic medication, obesity, and even to clinical variables of schizophrenia (11), suggesting that the effect of antipsychotic drugs is mediated via an effect on the gut microbiome (12) and then affecting psychotic symptoms and obesity.
Peculiarly, little research has been focusing on the potential reversibility of weight gain if the antipsychotic drug is reduced, discontinued, or switched to an agent with lower obesogenic properties, but case reports (13,14) and a few studies of patients with intellectual disabilities and children with bipolar disorder (15) suggest that at least some weight can be lost after discontinuation.

How the Interventions May Work
Reducing or discontinuing second generation antipsychotic medication may cause weight loss due to negative energy balance or directly through affecting various neurotransmitters and neuroendocrine signaling. Negative energy balance could be achieved by decreased appetite and by increased physical activity due to lower level of sedation. The effect is likely mediated via neurotransmitter signaling involving histamine and serotonin (16), and some suggest that thyroid function is associated with antipsychotic medication (17). Similarly, switch to partial agonists may affect weight through lower affinity to histamine receptors and lower level of sedation.

Objective
The aim of this systematic review was to report the effect of dose reduction/discontinuation of second-generation antipsychotic, switch to a partial dopamine agonist, or switch from polypharmacy to monotherapy on weight in individuals diagnosed with severe mental illness. We hypothesized that any of the four interventions would lead to a reduction in weight, in relative values when compared to maintenance treatment and in absolute values when compared to weight before interventions.

Study Selection
Two authors (HS, CW) independently examined the remaining full list, selected relevant abstracts and examined the relevant full text determining compliance with inclusion criteria using the software from Covidence.

Data Extraction
Two authors (HS, CW) independently extracted data using Covidence software. The authors were not blinded to study results, authors, or institutions. Data extraction included data on weight and additional information regarding age, gender, number of patients, duration of treatment, type of treatment, baseline weight, type of discontinuation and duration of follow up. HS and CW independently conducted risk of bias applying Rob-2 tool (19).

Outcomes
Primary outcome was defined as mean difference in weight between any of the methods of: (discontinuation, dose reduction, switch to partial agonist, switch to monotherapy) as described in the Introduction, compared to maintenance treatment measured on a continuous scale in kg. Results for each of the four interventions will be reported both pooled and separately. Secondary outcome was pre-post change in weight from initiation of intervention to end of follow-up measured on a continuous scale in kg reported pooled and separately for each of the four groups. This analysis included non-controlled intervention studies pooled with the intervention groups from the controlled trials. Type of intervention, diagnoses and difference in duration of exposure in active versus placebo medication was tested as potential moderators of effect.

Data Synthesis
In order to include a maximum of studies, we combined end-scores and change scores for primary outcomes. Thereby, we abstain from calculating standardized measures, as the combination of dispersion of end-and change scores cannot be combined in a standardized effect size (20). If change scores as well as end scores were reported, end-scores were preferred. All results were reported with 95% confidence intervals and 95% prediction intervals (21). Missing measures of dispersion were imputed as recommended by Cochrane (20). Random effects were reported, assuming underlying heterogeneity of effects due to variations in the interventions. The degree of heterogeneity was quantified using the I 2 statistic, which can be interpreted as the percentage of variation observed between the trials attributable to between-trial differences, rather than sampling error (chance). Heterogeneity was explored by analyses of subgroups and meta-regressions. Results from randomized clinical trials and uncontrolled studies were analyzed and reported separately. Results from RCTs were pooled with pre-post studies if data for these were available. Data was analyzed using Comprehensive Meta-Analysis v. 3.11; p-values <0.05 were considered significant. Prisma reporting guidelines were followed (22).
Deviations From the Protocol 1) We changed the switch to partial agonist from only including aripiprazole to also including brexpiprazole and cariprazine. 2) We changed one moderator from study duration to difference in exposure to drug in active versus placebo drug. 3) An improved search strategy was applied.

RESULTS
The main bibliographical search was conducted on 18 th February 2021. As illustrated in Figure 1, we identified 55 studies  fulfilling the inclusion criteria and none of the exclusion criteria ( Table 1). Of these, 40 were RCTs and 15 were uncontrolled studies. Of these, 33 reported on the effect of discontinuation, seven on dose reduction, 18 on switch to a partial agonist, and no studies on switch from polypharmacy to monotherapy. A total of 12,279 (mean participants 224 per study) participants were included, with a mean age of 40 years old, and the mean baseline weight of 69.6 kg.

Risk of Bias
The risk of bias assessment is presented in Table S1 (19). We found a low risk of bias in 26 trials, and some concerns in 14 trials. All the included trials were blinded, but as weight was reported as adverse events, the methods to handle missing data were often not clearly described. Visual inspection of Funnel plots (Figures S1 and S2) did not reveal any signs of publication bias.
For the secondary outcome ( Figure 3), pre-post analyses based on 55 studies reporting on 58 groups (three studies had two intervention groups), the pooled results of interventions including discontinuation, dose reduction, and switch to partial agonist, found a weight reduction of 1.13 kg (95% CI −1.36 to −0.90; P < 0.001) at end of follow-up compared to baseline

Heterogeneity, Subgroup Analyses, and Meta-Regression
The I 2 was 90.7% for the primary outcome and 97.8% for the secondary outcome, suggesting substantial heterogeneity. This was expected, as we pooled different diagnoses, different interventions, and different designs. We explored heterogeneity by testing categorical variables (diagnoses, controlled vs. uncontrolled studies) and one continuous variable (differences in exposure between active and placebo), Table 4.  For the secondary outcome, there was no difference in effect when comparing groups from randomized trials to uncontrolled studies; −1.03 kg versus −1.39 kg (p = 0.20).

DISCUSSION
We found that the pooled effect of discontinuation, dose reduction, and switch to partial agonists reduced the weight with 1.54 kg compared to maintenance treatments with a second generation antipsychotic drug. The 95% CI interval was −2.08 to −1.2, allowing us to reject the null hypothesis, but the corresponding 95% prediction interval was −4.71 to 1.61, which does not exclude the probability of null effect in future trials, but the results could be as high as −4.7 kg. The subgroup results suggest that the effect of switch to partial agonists was larger than discontinuation and dose reduction, but this may reflect trial methodology rather than true differences, as discussed beneath. In the pooled pre-post analysis, we found a weight reduction of 1.13 (p < 0.001), with no substantial differences in subgroups.
The primary outcome, based on controlled trials, and secondary outcomes, based on uncontrolled trials, address two fundamentally different questions with different inherent methodological limitations to the answers. The primary outcome asks the clinical question: Will my weight be lower if I choose a dose reduction/discontinuation strategy, compared to if I continue the same dose? But the outcome does not inform about a weight loss compared to baseline, as the observed difference may be explained by a larger weight increase in the maintenance group. On the other hand, the secondary outcome, the change in weight form baseline to post intervention, addresses the question: Will I lose weight over time if I reduce dose/discontinue medication?  Our results suggest that a weight loss does occur, but substantial methodological considerations limit the validity of our findings. A major limitation is that most of the trials included in the primary analysis were designed to evaluate relapse prevention of antipsychotic drugs in individuals who received treatment during a short stabilization phase and were defined as responders. It is likely that individuals with severe weight gain were excluded in the stabilization phase due to adverse effects, which could explain the low mean baseline weight, which again could affect the generalizability to a real-world population. It could be argued that the short duration of placebo-treatment is insufficient to evaluate weight loss, as many patients relapse and thus have withdrawn from the study. On the other hand, high degrees of relapses are observed in real world data, and these findings therefore reflect the effectiveness rather than efficacy, as complete adherence is often not realistic. Next, the duration of   exposure to study drug, active or placebo, differed due to earlier relapse in placebo groups, leaving shorter time to weight change in placebo versus active treatment. Data on duration of exposure was available for 17 of the included trials and did not explain a significant proportion of the variance, although this could be a type II error. Weight change was reported as an adverse event for the large majority of RCTs, and it was generally difficult to assess how missing data were handled, and if the way of reporting weight was pre-registered to avoid any multiple possible methods (continuous, dichotomized, BMI, 7% change) could be affected by financial conflicts of interest, as many of the RCTs were industry sponsored. This possible bias is likely to deflate effect size, as most sponsors would be interested in reporting a lower weight gain in the maintenance groups. Analyses of publication bias did not confirm any concerns but cannot exclude selective reporting. Finally, the generalizability of the population could be affected by the short duration of exposure in the open label, stabilization phase prior to randomization, as duration of prior exposure might be important, as pointed out by Kim et al. (30). This allows only a short period of weight gain, which could deflate effect sizes compared to real-world populations. On the other hand, it may be easier to lose weight if the time being overweight is shorter, which could oppositely inflate effect size.
By comparing the results with the pre-post analysis, it is tempting to conclude that a weight loss is occurring, and that the observed effect cannot be ascribed to weight gain in the maintenance group. However, important caveats in inferring from pre-post studies should be kept in mind. Results could overestimate the effect due to regression to the mean, skewed loss to follow-up; it could also be confounded by lifestyle or other types of medication. Many of the studies reporting on the effect of switching to partial agonists were uncontrolled with metabolic disturbances being the primary outcome, which increases the risk that the participants engaged in parallel weight reducing behaviors, seen as a manualized co-treatment in one study (77), which could inflate the effect.
We did not identify any studies evaluating the effect of switching from antipsychotic polypharmacy to monotherapy on weight. This is surprising, as there has been increased focus on the lack of evidence supporting superior effect of polypharmacy compared to monopharmacy, and therefore safety in switching (78).
Even though the weight reduction is highly statistically significant, it should be discussed if this weight reduction is clinically important. For the subset of individuals who gain tens of kilograms, the moderate reduction is likely to be ignorable. On the other hand, the magnitude is similar to the magnitude of weight lost from lifestyle interventions (79,80) and by adding topiramate or metformin to current antipsychotic medication, all being recommended in clinical guidelines (81). Furthermore, the observed effects should be interpreted considering the short duration of most of the trials, with some individuals only receiving placebo treatment for a few weeks before showing signs of impending relapse and subsequently withdrew from the study. Thus, it cannot be excluded that the weight loss could continue over time. The largest weight loss was achieved by switching to partial agonists, which could be explained by longer duration of exposure to antipsychotic medication, making risk of relapse in this group lower compared to trial where an active treatment is compared to placebo.
None of the included studies were designed directly to address the reversibility of antipsychotic-induced weight gain, but some studies have been published in other populations, not included in this paper: Upadhyay et al. (15) followed a large sample of children and adolescents (N = 537) with bipolar disorder who had experienced weight gain while treated with psychotropic medication. At 12 months follow-up after discontinuation of psychotropic medication, weight loss was stabilized, but never returned to baseline, and the authors conclude that it is likely that those who gained significant weight during treatment will stay overweight or obese. Significant improvement in weight was found by De Kuijpers et al. (82) in individuals with intellectual disabilities, where a reduction of dose was associated with weight loss. An improvement in metabolic factors and with no psychiatric deterioration, was reported by Hulvershorn et al. (83) who evaluated the effect of antipsychotic discontinuation in youth with disruptive behavior. However, the majority of these were prescribed medication to treat ADHD, which, in itself will induce weight loss. Finally, case reports describing severe cases of rapid weight gain and metabolic disturbances confirm reversibility, at least to some degree (14,84).

Limitations
Adding to the limitations of using primarily relapse prevention studies, there are important limitations in the conduction of this meta-analysis that should be kept in mind. First, we did not preregister the protocol at Prospero. The protocol was circulated in the author group, and all agreed on the design before the literature search was started. Second, we a-priori decided to measure weight change as continuous variable, as dichotomizing continuous variable may inflate effect sizes and decrease power (85). When extracting data, we were surprised to identify a large number of papers only reporting >7% weight gain and not supplementing with 7% weight loss or a mean weight change. It is possible that excluding these papers has introduced a selection bias. Third, we did not assess potential adverse events associated with dose reduction, switch, or discontinuation. To fully inform a clinical decision, potential risks, such as relapses or worsening of substance abuse, are just as important. However, as we chose only to include studies where weight was reported, reporting risks of adverse events on this basis would not provide the full picture of the available literature that has been done by others (86)(87)(88)(89). A general finding was that the relative risk of relapse was doubled after discontinuation [27% vs placebo 64%; risk ratio (RR) 0.40, 95% CI 0·33-0·49] (90) on short term. However, it has been argued that relapse rates converge with time, and the chance of recovery is increased after dose reduction. Based on very few trials, switch to monotherapy does not seem to increase relapse rates (78,91). Finally, the lack of access to individual data precludes the possibility of performing subgroup analyses based on personal characteristics like magnitude of weight gain during treatment or type of antipsychotic medication. Thus, it cannot be excluded that individuals with severe gain also have an equivalent weight loss, but that these are hidden in the mean.

Implications for Research
The lack of direct studies on the reversibility of weight gain calls for future research directly focusing on this topic in rigorously designed studies. Most important, the subgroups with substantial weight gain, for whom this issue may be especially important, need to be included. Ideally, RCTs should be conducted, but the feasibility in conducting these is challenging (92), and observational designs could be an alternative (93). Close monitoring of clinical characteristics during treatment in early intervention services could provide valuable information if interpreted correctly. In order to address, specifically, the question of causality, careful considerations should be put on the potential confounders, and we suggest that transparent frameworks, such as directed acyclic graphs (DAGs) (94), could be applied to make causal inferences based on observational data.

Implications for Clinical Practice
Any health care decisions are based on information about potential risks and benefits. As weight gain and potential loss are important to many people, it should be shared that little is known about the reversibility of antipsychotic-induced weight gain. This lack of knowledge could affect the decision, especially when drugs are prescribed as off-label (3), where the potential benefits are less clear. Some may worry that patients with severe mental illness may reject the medication in fear of substantial and irreversible weight gain, and thereby put themselves at risk of relapse. However, as long as we regard the patients as having decision capacity (95) and therefore able to provide informed consent, it is our obligation to inform them about potential adverse effects, including what we currently do not know, if this could affect the decision making. In four RCTs determining the effect of dose reduction, a paradoxical weight increase was seen in the reduced group (47,56,73,96), suggesting that metabolic adverse effects may not be dose dependent, which is also found in other studies (97). This underlines the importance of reconsidering the current off-label prescribing trend, as also small doses may lead to metabolic disturbances.

CONCLUSION
The main finding of this study is a paucity of studies designed to directly evaluate the reversibility of antipsychotic-induced weight gain. This lack of knowledge is problematic, as individuals making an informed decision has the right to know if they run a risk of lifelong obesity. In spite of the limitations discussed above, our findings do suggest that at least some weight gain is reversible, even though there is doubtfully any health benefit in this small effect size when balanced against the risk of relapse. The lack of dose-related effect is important and should lead to increased awareness of off-label use of low dose second generation antipsychotic for indications like sleep and anxiety, as this might create equally serious weight problems and is increasingly being prescribed (98).

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

AUTHOR CONTRIBUTIONS
HS conceived the idea and made the analysis plan. HS and CW performed the bibliographic search, extracted data, performed the final analyses and drafted the manuscript. All authors contributed to the article and approved the submitted version.

ACKNOWLEDGMENTS
We want to thank Dr. Nina Vindegaard Sørensen for valuable input during the writing process.