Topiramate and Metformin Are Effective Add-On Treatments in Controlling Antipsychotic-Induced Weight Gain: A Systematic Review and Network Meta-Analysis

Background: Antipsychotic drugs may lead to side effects such as obesity, diabetes, dyslipidemia, and cardiovascular disease. The current systematic review and network meta-analysis analyzes and provides an update on the clinical performance of these add-ons in comparison to placebo on body weight and body mass index (BMI) reductions. Methods: A comprehensive literature search was performed on electronic databases: PubMed (1946-), Embase (1974-), Cochrane library (1992-), and OpenGrey (2000-) until 31 July 2018. Network meta-analyses, comparing the body weight change, BMI change and withdrawn due to adverse events of different pharmacological add-ons, was performed using a multivariate meta-regression model with random-effects, adopting a frequentist approach. To rank the prognosis for all add-ons, we used surface under the cumulative ranking (SUCRA) values. Outcomes: From 614 potential studies identified, 27 eligible studies (n = 1,349 subjects) were included. All the studies demonstrated low to moderate risk of bias. For the analysis of body weight change, all add-ons except Ranitidine showed significant weight reductions comparing to placebo. The effectiveness rank based on SUCRA results from highest to lowest was Sibutramine, Topiramate, Metformin, Reboxetine, Ranitidine, and placebo. A similar pattern was seen for BMI change. The analysis of safety outcome did not detect significantly increased withdrawn number from the add-ons. Current evidence showed relatively good tolerance and safety of using the pharmacological add-ons. Interpretation: Topiramate and Metformin are effective add-on treatments in controlling antipsychotic-induced weight gain, comparing to placebo. They are well tolerated in short-term period. Although Sibutramine has the highest rank of the effectiveness, its license has been withdrawn in many countries due to its adverse effects. Hence, Sibutramine should not be adopted to treat antipsychotic-induced weight gain.


INTRODUCTION
Antipsychotic drugs (APDs) may lead to side effects such as obesity, diabetes, dyslipidemia, and cardiovascular disease. This adverse effect cluster presents an obstacle in the treatment and management of patients with schizophrenia or bipolar disorder, and limits patient adherence to medication and consequently adversely impacts treatment outcomes.
To counter the antipsychotic-induced weight gain, various pharmacological add-ons were investigated. Taking antidiabetics or antiobesity drugs as an adjuvant treatment, including metformin, orlistat, sibutramine, and naltrexone, is a popular approach for weight management and has been widely studied (Baptista et al., 2006;Henderson et al., 2007;McElroy et al., 2007;Joffe et al., 2008;Tchoukhine et al., 2011;Tek et al., 2014;Anagnostou et al., 2016;Rado and von Ammon Cavanaugh, 2016;Vishnupriya et al., 2016;Wu et al., 2016;Handen et al., 2017). Most of the studies reported significant reductions in body weight. Gastrointestinal agents, especially antacids like nizatidine, were reported may stop but not reduce the weight gain (Atmaca et al., 2003(Atmaca et al., , 2004Assuncao et al., 2006). Topiramate, a type of anticonvulsant, shows a negative association with body weight gain and has been found to control antipsychotic-induced weight gain for subjects with schizophrenia or bipolar disorder (McElroy et al., 2007;Afshar et al., 2009;Wozniak et al., 2009;Narula et al., 2010).
Until recently, no study has been published comparing various pharmacological add-ons on antipsychotic-induced weight gain, from both direct and indirect evidence. The current systematic review and network meta-analysis analyzes and provides an update on the clinical effectiveness and safety of these add-ons in comparison to placebo on body weight, body mass index (BMI) reductions and number of withdrawn due to adverse effects.

Literature Search and Eligibility Criteria
A comprehensive literature search was performed on electronic databases: PubMed (1946( -), Embase (1974, Cochrane library (1992-), and OpenGrey (2000-) until 31 July 2018. The specific concepts used in the search strategy were "antipsychotic agents" and "weight." We conducted literature search using Medical Subject Headings (MeSH) or Emtree, and free text terms. There were no restrictions on language. The bibliography listed in review papers and included publications were also checked.
Two investigators (CjZ and QZ) independently screened for eligible studies based on pre-defined eligibility criteria. Randomized controlled trials (RCTs) that examined the pharmacological interventions of weight management for antipsychotics-induced obesity were included. To avoid imprecise estimations, only those add-ons with at least two RCTs studied were included. Non-randomized or observational studies, case reports, commentaries, and letters-to-editors were excluded.

Data Extraction and Quality Assessment
The following data were extracted from the included studies: (1) study characteristics (publication year and patient population); (2) baseline characteristics (mean age, number of males, followup time, and ongoing antipsychotic treatment); and (3) outcome events (weight change [kg], BMI change [kg/m 2 ], and number of withdrawn due to adverse events).
The quality of each study was evaluated, using the Cochrane Collaboration Risk of Bias tool, by two independent investigators (CjZ and QZ). Six domains were assessed for each RCT, including random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias. Each domain would be assigned a judgment of 'Low risk' of bias, 'High risk' of bias, or 'Unclear risk' of bias. Any disagreement in quality assessment was resolved by discussion and consensus.

Statistical Analysis
A network geometry was constructed based on the included studies for each add-on treatment. Each node represented an add-on and its size was weighted by the number of subjects of each add-on. The connecting line between two nodes meant a direct comparison existed and its thickness was determined by the number of studies included.
Network meta-analysis, comparing the body weight change, BMI change, and number of patients withdrawn due to adverse events among different pharmacological add-ons, was performed using a multivariate meta-regression model with randomeffects, adopting a frequentist approach White et al., 2012). The model allows for the inclusion of potential covariates, and accounts for the correlations from multi-arm trials, and mean difference (MD) for weight and BMI change and risk ratio (RR) for number of withdrawn due to adverse events of each add-on treatment was estimated (White, 2011).
To rank the prognosis for all the add-ons, we used surface under the cumulative ranking (SUCRA) values (Salanti et al., 2011). Rank probabilities of all the add-ons were first estimated under a Bayesian framework. A step function was then applied to summarize the cumulative ranking for estimating the SUCRA values of each add-on, ranging from 0 to 1. Thus, large SUCRA values indicated a better prognosis.
The node-splitting approach and inconsistency model were used to test the consistency assumption (Dias et al., 2010). The former method involved fitting a series of node-splitting models, with one model for each add-on pairing for which there was direct and indirect evidence (Donegan et al., 2013). The latter method first fits an inconsistency model and then conduct a Wald test to check whether there is significant inconsistency among the included studies (White, 2015). Sensitivity analysis was conducted by (1) excluding studies with both "blinding of participants and personnel" and "blinding of outcome assessment" ranked as "Unclear" or "High risk, " as the outcomes (i.e., measurement of weight and BMI) were likely to be biased due to these two key components, and (2) limiting the analysis on studies with less than 12 months' followup.
The network geometry was constructed (Figure 2). Most of the studies demonstrated low to moderate risk of bias in the six domains assessed. However, due to missing information or inappropriate methods on randomization, four studies were ranked as "Unclear" or "High risk" in "Random sequence generation" (Supplementary Figure S1).  Table 2). Results from both the nodesplitting method and inconsistency model showed no evidence on the violation of consistency assumption between direct and indirect comparisons. As shown in Supplementary Figure S2, the pooled estimates were quite similar between consistency model (red diamonds) and inconsistency model (green diamonds), indicating that inconsistency covariates did not yield a significantly better fitting. The p-value = 0.166 from the Wald test further confirmed that there is no evidence on the violation of consistency assumption.

Effectiveness on Body Weight Change
To confirm the rank of effectiveness on body weight reduction, SUCRA values were calculated, and the rank from highest to lowest was Sibutramine, Topiramate, Metformin, Reboxetine, Ranitidine, and placebo (Table 3).
In the sensitivity analysis by excluding the eight studies, similar pooled estimates were obtained and the rank order remained the same. Further sensitivity analysis by excluding studies with less than 12 months' follow-up showed that metformin, sibutramine and topiramate were consistently significant with a reduction in body weight at −2.54 (95% CI: −3.29, −1.79), −2.98 (95% CI: −4.34, −1.62), and −2.95 (95% CI: −5.87, −0.03), respectively. Ranitidine did not show any significant reduction in body weight, which was consistent to the main result as well. However, Reboxetine was reported only in studies with less than 12 months' follow-up, we were unable to check its sensitivity results.

Effectiveness on BMI Change
For the BMI change outcome, 24 studies were included in the analysis. A similar pattern was seen for BMI change, where all add-ons except cardiac Ranitidine showed statistically significant BMI reductions comparing  to placebo. In particular, anticonvulsant presented the highest reduction, reaching MD = −1.59 kg/m 2 (95% CI: −2.65, −0.54) ( Table 2). SUCRA results showed a consistent order from the highest to lowest: Topiramate, Sibutramine, Metformin, Reboxetine, Ranitidine, and placebo ( Table 3).

Tolerance and Safety on Number of Withdrawn Due to Adverse Events
To assess the tolerance and safety of add-on treatments, we estimated the pooled RR for all add-ons comparing to placebo. All the pooled RR had wide 95% CI due to a relatively low prevalence of patients withdrawn due to adverse events in each included study. Among them, Topiramate had the highest but statistically insignificant RR 1.88 (95% CI: 0.44, 7.94). Therefore, current evidence did not reveal a significantly higher safety issue by using add-ons in the short-term period follow-up.

DISCUSSION
Current published systematic reviews on add-ons controlling APD-induced weight gain were either focused on one particular medication (e.g., metformin) (Housel et al., 2009;Miller, 2009;Prajapati, 2014;de Silva et al., 2016;Siskind et al., 2016), or indirectly compared different medications in a qualitative way, i.e., ranking the treatment effects simply based on the pairwise meta-analyses results (Miller, 2009;Maayan et al., 2010;De Hert et al., 2012;Kishi et al., 2014;Mizuno et al., 2014;Choi, 2015;Whitney et al., 2015;Zimbron et al., 2016). The current review aims to quantitatively synthesize the weight and BMI reduction effects through a combination of both direct and indirect evidence. A total of 27 studies were included, among which only one study reported head-to-head comparisons between any two of the add-ons (McElroy et al., 2007). Hence, by using network meta-analysis we were able to allow indirect comparisons between the add-ons as well as against placebo.
Notably, although our results showed that Sibutramine ranked first in body weight reduction and second in BMI reduction, its license has been withdrawn in many countries due to its adverse effects. Hence, Sibutramine should not be adopted to treat antipsychotic-induced weight gain. Sibutramine is an antiobesity medication affecting both serotonin and norepinephrine reuptake (Henderson et al., 2007). Although the tolerance in this current review is not significantly lower than placebo, the Sibutramine Cardiovascular Outcomes Trial confirmed that subjects with preexisting cardiovascular disease on long-term treatment with Sibutramine had a significantly increased risk for non-fatal myocardial infarction and non-fatal stroke but not cardiovascular death or all-cause mortality (Torp-Pedersen et al., 2007). Converging with previous studies, we Bottom left panel reported the pooled mean difference (MD) of body weight changes from network meta-analysis, and the name to the right of each MD was the reference add-on. Top right panel reported the pooled MD of BMI changes from network meta-analysis, and the name below each MD was the reference add-on; * statistical significant with p-value < 0.05. strongly suggest that we should not adopt sibutramine to treat antipsychotic-induced weight gain. Some longitudinal studies suggest using body composition data rather than only measurements of body weight for understanding the relationship with disability and mortality (Seidell and Visscher, 2000;Myrskylä and Chang, 2009). Current systematic reviews included BMI changes as one of the efficacy outcomes as well. Topiramate showed the best performance on controlling BMI based on our meta-analysis. It is an anticonvulsant blocking α-amino-3hydroxy-5-methylisoxazole-4-propionic (AMPA)/kainate-gated ion and sodium channels and positively modulate GABA receptors (White et al., 1997). The side effect of weight loss when using Topiramate has been documented and been applied to treatment for adiposity caused by eating disorders (Anghelescu et al., 2002;Levy et al., 2002). In our current review, it showed excellent weight and BMI reduction. The adverse effects, reported in the included studies, were mild to moderate without serious adverse effects. Paresthesia was the most common side effect observed in most of the included studies (Ko et al., 2005;Afshar et al., 2009;Narula et al., 2010).
Metformin, a major antidiabetic medication, was widely studied in the literature on controlling antipsychotic induced weight gain and was proven to be effective (de Silva et al., 2016;Siskind et al., 2016). Klein et al. reported its efficacy of reducing weight gain in children and adolescents who were treated with olanzapine as well (Klein et al., 2006). One of the reasons may be explained by its effects in enhancing the glycaemic control effects of insulin, antagonizing glucagon, and suppressing gluconeogenesis and glycogenolysis (Wiernsperger and Bailey, 1999). Similar reduction in BMI was reported in a recent pair-wise meta-analysis at −0.89 (95% CI −1.20, −0.58) (Zimbron et al., 2016). Although the rank of antidiabetics were generally lower than Topiramate and Sibutramine, Metformin showed long-term tolerance and weight loss effect for type 2 diabetes patients, even over 10 years (Ratner et al., 2006;Diabetes Prevention Program Research Group, 2012). This evidence highlights the potential long-term use of metformin for antipsychotic-induced weight gain patients as well.
Of note, apart from pharmacological add-ons, nutritional and behavioral interventions are important for weight control. A recent systematic review reported that non-pharmacological interventions, either individual or group interventions, or cognitive-behavioral therapy as well as nutritional counseling were effective in reducing or attenuating antipsychotic-induced weight gain compared with treatment as usual, with treatment effects maintained over follow-up (Alvarez-Jimenez et al., 2008). Our review did not consider including these interventions due to high heterogeneity from study to study, but the combination of both pharmacological and non-pharmacological interventions might be promising for further controlling the weight gain.
The current evidence in our study showed relatively good tolerance and safety of using these pharmacological add-ons. However, given that most of the included RCTs had relatively short period of follow-up, further evidence on potential longterm adverse effects are needed.

Study Limitations
In total, only 6 out of 27 included RCTs followed up more than half a year. The relative short period of follow-up time limited the ability of our findings to be extrapolated to longer periods, as the adherence of medication, etc., may alter the effectiveness or the rank in the long run. Therefore, primary studies on the long-term effects of pharmacological add-ons are needed. As mentioned above, nutritional and behavioral intervention, which were not included in the review, are alternatives for controlling APD-induced weight gain. We would expect primary and secondary studies on evaluating the rank of effectiveness on both intervention, and additive benefits when combining both.

CONCLUSION
Topiramate, and Metformin are effective add-on treatments in controlling antipsychotic-induced weight gain, comparing to placebo. They are well tolerated over a short-term period. More importantly, we propose to conduct a large sample long-term cohort study to explore the optimal treatment methods for treating antipsychotic-induced weight gain in long term.

ETHICS STATEMENT
Ethics committee of Tianjin Anding Hospital approved this study.