Amisulpride

We first conducted meta-analyses of the BPRS or PANSS total score at 4 weeks for the three comparisons of olanzapine vs haloperidol, amisulpride vs haloperidol and olanzapine vs placebo, using Review Manager software by the Cochrane Collaboration [21] . 4-week was chosen because all the studies reported BPRS at this point in time. Following the strict intention-to-treat principle, missing data were supplemented by the last-observation-carried-forward (LOCF) method even when a participant dropped out before the first post-baseline rating. Unless statistically significant heterogeneity was noted, we obtained the standardized mean difference (Cohen's d) based on the Mantel-Haenszel fixed effect model.
We next calculated the numbers of responders defined as 10% through 90% reduction on the BPRS or PANSS total score at 4 weeks. The percentage reduction was calculated according to the formulae: B% = (B₀−B_4LOCF) * 100/(B₀−18) for BPRS and P% = (P₀−P_4LOCF) * 100/(P₀−30) for PANSS, where B₀ and P₀ are BPRS and PANSS scores at baseline and B₄ and P₄ are respective scores at 4 weeks, because 18 and 30 are the minimum scores for BPRS and PANSS, respectively, according to the original rating system. We then ran meta-analyses of response rates defined as 10% through 90% reduction for each comparison in terms of risk difference. The pooled NNT was obtained by taking the inverse of this pooled risk difference, because the response rates for a certain cutoff did not differ substantively among the trials included in the meta-analysis, [22] .
These actual NNTs were then compared with NNTs converted from Cohen's d according to Kraemer's method and to Furukawa's method using the formulae discussed in the Introduction. The agreement between the actual and the converted was quantified by ANOVA intraclass correlation coefficient (two-way mixed effects, absolute agreement, single measure) by using SPSS Version 17.

The meta-analysis was conducted and reported according to recommendations of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) group [36] (link). A review protocol was construed following the MOOSE guidelines. This was not published but only for internal use of this study.
A PubMed and Embase search was conducted for articles on metabolic side effect profiles of antipsychotic medication. The search term used was: ((“weight gain” OR “BMI” OR “7% weight”) AND (chlorpromazine OR haloperidol OR bromperidol OR fluphenazine OR zuclopenthixol OR pentixol OR flupentixol OR levopromazine OR perphenazine OR pimozide OR penfluridol OR sulpiride OR amisulpride OR amoxapine OR asenapine OR aripiprazole OR blonanserine OR clozapine OR iloperidone OR melperone OR olanzapine OR risperidone OR paliperidone OR quetiapine OR sertindole OR lurasidone OR ziprasidone)) NOT (addition OR additive OR adjunctive OR augmentation OR lithium OR valproate OR carbamazepine OR metformin OR topiramate OR ramelteon OR rimonabant OR modafinil OR sibutramine OR genetics OR pharmacokinetics OR vomiting OR nausea OR review OR “cognitive behavioural therapy” OR “cognitive behavioral therapy” OR delirium OR steroids OR ropinirole OR sleep OR “brain volume”)
Limits Activated: Humans, Clinical Trial, Randomized Controlled Trial, Clinical Trial, Phase IV, Controlled Clinical Trial, English, German, All Adult: 18+ years, Publication Date from 1999/01/01 to 2011/12/31.

We obtained data from the Health Insurance Review and Assessment Service (HIRA) database, which contains every national healthcare service claims data from patients across South Korea. Although several other types of insurance cover medical treatments, such as industrial accident insurance, most patients with schizophrenia in South Korea are covered by the National Health Insurance plan. The HIRA database includes patient demographics, diagnosis according to the International Classification of Diseases Tenth Revision (ICD 10), procedure codes, and prescriptions. Identifying data were removed from the HIRA service records in accordance with the Act on the Protection of Personal Information Maintained by Public Agencies.
We used the HIRA database to identify patients (18–69 years old) diagnosed with schizophrenia (ICD10 code F20 Schizophrenia or F25 Schizoaffective disorder) who were admitted (index admission) and discharged from the hospital between January 2008 and December 2016. To ensure the enrollment of patients who received optimal in-hospital treatment, patients who did not receive second-generation antipsychotics (SGA; amisulpride, aripiprazole, olanzapine, paliperidone, quetiapine, risperidone, ziprasidone, and zotepine), or were hospitalized for less than 7 days or more than 120 days, were excluded from the study. Furthermore, patients readmitted within 30 days of discharge, which indicated inappropriate in-hospital treatment, and those admitted for non-medical reasons were excluded from the analysis (Fig. 1). In cases with multiple admissions, each admission was included and considered to be a unique episode. Prescription data and readmissions and deaths between January 2007 and December 2017 were assessed. The observation period for each episode started 30 days after discharge for every index admission. The end of the observation period was defined as the end of the study period (December 2017), readmission, or death, whichever occurred first. Re-hospitalization during the observation period was defined as an outcome event.Fig. 1

Flowchart of the study population

Each patient served as their own control to control for fixed confounding factors. The incidence rate (IR) was calculated as event count per 10 person-years. The IRs of outcome events for each period (LAIs, oral medication, and non-medication) were expressed as the incidence rate ratio (IRR). The operational definition of an oral medication period was the time from each first prescription date to last prescription date adding the last prescription duration × 1.25 (maximum refill date + 14 days). The follow-up time was reset to zero after each event to enable within individual comparisons for each episode. Each episode was analyzed separately according to the number of index admissions (first episode without previous SGA, first episode with previous SGA, two, three, and four or more episodes). We used readmission IRRs during the various periods to assess the effectiveness of LAIs (aripiprazole, paliperidone, and risperidone) compared with oral antipsychotics (amisulpride, aripiprazole, olanzapine, paliperidone, quetiapine, risperidone, ziprasidone, zotepine) in reducing multiple readmissions. Patients having oral medication and LAIs at the same time, were included to LAIs group.
Descriptive statistical analyses were performed using R software (ver. 3.4.0; R Development Core Team, Vienna, Austria). Continuous and categorical variables are expressed as mean ± standard deviation (SD), numbers (%), or medians (interquartile range; IQR). Continuous variables were compared using t tests, and categorical variables were compared using the Chi-square test. The outcome event rate was defined as episode counts per 10 person-years. The IRR and corresponding 95% confidence intervals (CIs) were calculated using the Poisson distribution. Two-tailed P < 0.05 were considered to indicate statistical significance.

In a double-blind, randomized, within-subject design, 56 volunteers (27 female, M_age = 23.2 years, SD_age = 3.1 years) received 400 mg amisulpride or placebo in two separate sessions (2 weeks apart) as described previously (Soutschek et al., 2017 (link)). Participants gave informed written consent before participation. The study was approved by the Cantonal ethics committee Zurich (2012-0568).

We analyzed choice data also in a model-free manner and with a hyperbolic discounting model. In the model-free analysis of the D2 antagonist study, we regressed choices of LL vs. SS options on fixed-effect predictors for Drug, Magnitude_diff, Delay_diff, and the interaction terms using Bayesian mixed models as implemented in the brms package in R (Bürkner, 2017 (link)). For the D1 agonist study, the same MGLM was used with the only difference that Drug (0, 6, 15, 30 mg) represented a between- rather than a within-subject factor. All predictors were also modeled as random slopes in addition to participant-specific random intercepts. Finally, the hyperbolic discounting model was fit using the hBayesDM toolbox (Ahn et al., 2017 (link)), using a standard hyperbolic discounting function: ${SV}_{discounted}=\frac{rewardmagnitude}{1+k\times delay}$
To translate subjective value into choices, we fitted a standard softmax function to each participant’s choices: $P\left(choiceofLLoption\right)=\frac{1}{1+e^{-{\beta }_{temp}\times \left(S{V}_{LL}-S{V}_{SS}\right)}}$
We estimated parameters capturing the strength of hyperbolic discounting (k) and choice consistency (β_temp) separately for each participant and experimental session by computing two chains of 4000 iterations (burning = 2000). We then performed a Bayesian t-test on the log-transformed individual parameter estimates under placebo vs. amisulpride using the BEST package (Kruschke, 2013 (link)).

Chi-square for categorical variables and Mann–Whitney U test for continuous variables due to non-normality were used to compare the baseline characteristics between patients with RLS and RLS-free controls. Cox proportional hazards regression models were applied to explore the association between RLS and the risk of dementia after adjusting for age, sex, income, residence, CCI, and history of other comorbidities. Among the Cox regression models, we used the Fine–Gray subdistribution hazard model with mortality as a competing risk given the old age of the study population. The proportional hazard assumption was satisfied in our Cox model (Schoenfeld individual test p-value > 0.05).
Sensitivity analyses were performed using four different models. In model 1, dementia was defined as the prescription of anti-dementia medications (donepezil, rivastigmine, galantamine, and memantine) at least twice and a diagnosis of the ICD-code of dementia. Although these medications were approved for only AD (rivastigmine additionally for Parkinson’s disease dementia), they can be used for cognitive symptoms in other types of dementia based on recommendations from multiple guidelines [31 (link)–33 (link)]. The previous study revealed that the definition of all-cause dementia by ICD-10 code plus anti-dementia medications had a positive predictive value of 94.7% when reviewing the medical records of 972 patients in two hospitals [34 (link)]. In model 2, medication history was added to the ICD code to define RLS. Patients with RLS ICD-code (G25.8) who had taken dopamine agonists (ropinirole or pramipexole) twice or more were regarded as patients with RLS (n = 1458). In this sensitivity model, we excluded patients with Parkinson’s disease because they could also take dopamine agonists. In model 3, patients taking antipsychotic agents were excluded because the antidopaminergic property of antipsychotic agents could lead to a misdiagnosis of RLS (n = 2482). The following antipsychotic agents approved in South Korea were used in this study: haloperidol, sulpiride, chlorpromazine, perphenazine, pimozide, risperidone, olanzapine, quetiapine, paliperidone, amisulpride, aripiprazole, ziprasidone, clozapine, blonanserin, and zotepine. In model 4, patients with RLS only diagnosed by psychiatrists or neurologists were included (n = 1154) to preclude the possible misdiagnosis by non-expert physicians.
To evaluate the effect of dopamine agonists (pramipexole and ropinirole) on the development of dementia, the risk of dementia was compared after dividing RLS patients by dopamine agonist use. Patients with RLS who were prescribed pramipexole or ropinirole at least once were considered dopamine agonist users. All missing data were addressed using listwise deletion. Data processing and statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA). Statistical significance was set at a two-tailed p-value of < 0.05.