The data are from the European Schizophrenia Cohort (EuroSC), conducted in the UK, France, and Germany. A detailed description of the European Schizophrenia Cohort has been published earlier 17 . In brief, it is a naturalistic 2-year follow-up of a cohort of people suffering from schizophrenia. The principle objective of the EuroSC was to identify and describe the types of treatment and methods of care for people with schizophrenia and to correlate these with clinical outcomes, states of health, and quality of life 17 18 19 20 21 22 23 24 . In our study, we only included French samples to control for country variation in the management of schizophrenia, which can be a confounding bias. We have shown previously that service use varied considerably between the three participating countries 19 . The French health system may offer an interesting approach: universal access to care, totally free health-care, and access to the most appropriate treatment, regardless of cost. This cohort of people suffering from schizophrenia was from three catchment areas in France: northern France (Lille), central France (Lyon and Clermont-Ferrand), and southern France (Marseille and Toulon). Each of these areas covers an urban centre of approximately one million inhabitants living in a city or in medium-size towns. In each area, patients treated in the "psychiatric sector" 25 were identified according to the following criteria: diagnosis of schizophrenia according to the DSM-IV criteria 26 , aged 18 to 64 years, and French as native language. Random sampling from these patients was used to generate a representative sample.

A total of 183 patients were followed for a 2-year period from 1998 to 2000, with data collected every 6 months. If the participant withdrew consent at any time or if the participant was lost to follow-up, data collected up to this point were used in analysis. This project was conducted in accordance with the Declaration of Helsinki and French Good Clinical Practices 27 28 . The protocol of this study was approved by the Institution Review Board or the Ethic Committee responsible for the participating hospital or institution. Written informed consent was obtained from each participant after the study details had been fully explained.

The following data were collected.

1. Socio-demographic information: gender, age, and living situation.

2. Clinical characteristics: psychotic symptoms based on the Positive and Negative Syndrome Scale (PANSS), which comprises three different subscales (positive, negative and general psychopathology) 29 30 ; Functioning based on the Global Assessment of Functioning (GAF) scale 31 , the Global Assessment of Relational Functioning (GARF) 32 and the Social and Occupational Functioning Assessment Scale (SOFAS) 33 ; depression based on the Calgary Depression Scale for Schizophrenia (CDSS), specifically designed for schizophrenic patients, which evaluates depression independent of extra-pyramidal and negative symptoms 34 35 .

3. Drug information: antipsychotic medication (monotherapy vs. polypharmacy): at baseline, patients were queried about their use of medications. Monotherapy (polypharmacy) was defined as the occurrence of one (more than one) ongoing antipsychotic medication prescription on the day of the visit; sedative drugs, antidepressant and side-effects co-treatment; the Simpson and Angus Scale (SAS) 36 , the Barnes Akathisia Scale (BAS) 37 , and the Abnormal Involuntary Movement Scale (AIMS) 38 were used to assess the side-effects; the Rating of Medication Influences (ROMI) Scale was used to evaluate adherence to treatment 39 . The ROMI is a reliable and valid instrument that can be used to assess the patient's subjective reasons for medication compliance and non-compliance.

4. The number of previous hospitalisations.

5. Quality of life (QoL) questionnaire:

SF-36 is a generic, self-administered QoL questionnaire consisting of 36 items describing 8 dimensions: Physical Functioning (PF); Social Functioning (SF); Role--Physical Problems (RPP); Role--Emotional Problems (REP); Mental Health (MH); Vitality (VIT); Bodily Pain (BP); and General Health (GH). Each dimension is scored within a range from 0 (low QoL level) to 100 (high QoL level) 40 41 .

Our primary outcome was relapse on a 2-year period, defined according to a usual, clinically reproducible and validated definition 42 43 : (1) hospitalisation due to worsening of psychotic symptoms or an unequivocal worsening of psychotic symptoms of such magnitude that hospitalisation appeared imminent, or (2) a re-emerge of florid psychotic symptoms such as delusions, hallucinations, bizarre behaviour, or (3) thought disorder lasting seven days or more.

This information was collected by routine clinical interview by a psychiatrist every six months, and relapse was defined using information regarding the baseline characteristics of the patient. Additional relevant information was obtained from the medical record and also through staff interviews.

Characteristics for patients were compared using the Chi-squared or Fisher exact tests for categorical variables and the Wilcoxon rank sum test for continuous variables. We performed a propensity score analysis to adjust for imbalances in baseline characteristics between patients with monotherapy and polypharmacy. For this purpose, we first developed a nonparsimonious logistic regression model to derive a propensity score for patients receiving polypharmacy based on all the covariates 44 . This logistic regression model yielded a c-statistic of 0.80, indicating a strong ability to discriminate between monotherapy and polypharmacy. This logistic regression model was used to estimate a propensity score for each patient, corresponding to the probability of being treated using polypharmacy. Patients were stratified by quintile of increasing propensity score. To validate our propensity score adjustment, we checked for adequate overlap in propensity scores for monotherapy and polypharmacy within each quintile and for the absence of significant residual imbalances in patient characteristics after adjustment for quintile of the propensity score. A multivariable Cox proportional hazards model was used to estimate the Hazard Ratio (HR) and its corresponding 95% confidence interval (CI) of relapse associated with polypharmacy after adjusting for the strata of propensity score.

Statistical analyses were carried out using SAS 9.1. The statistical significance level was set at p < 0.05 in a two-sided test.

Characteristics for the monotherapy and polypharmacy groups are shown in Table 1. Sixty-two (62.3%) of the patients had monotherapy, and 37.7% of patients received polypharmacy. In the polypharmacy group, 54 patients received two antipsychotics (78.3%), and 15 patients received three antipsychotics (21.7%). Twenty-two patients in the polypharmacy group (31.9%) and 37 patients in the monotherapy group (32.5%) received a depot antipsychotic medication. There was no significant difference in socio-demographic and clinical characteristics between the two groups, except for the age at onset of illness, which was lower in the polypharmacy group (p = 0.03), and the general psychopathology PANSS score, which was higher in the polypharmacy group (p = 0.04). The proportion of patients receiving antidepressant and sedative drugs was significantly higher in the polypharmacy group (respectively p = 0.03 and p < 0.01). Side effects, as assessed with the AIMS, BAS, and SAS, did not differ between the two groups, whereas we identified a higher proportion of patients with drugs to correct side effects in the polypharmacy group (p < 0.01). Concerning QoL scores, we did not find any statistical difference except for the Role-Emotional limitations (problems with work or other daily activities as a result of emotional problems), which was lower in the polypharmacy group (p = 0.03).

Characteristics for the patients with relapse and no relapse are shown in Table 2. Patients with relapse were significantly younger than patients with no relapse (p = 0.04). In the same way, the age at onset of illness was lower for patients with relapse than for those without relapse (p < 0.01). Patients with relapse were also less likely to report living independently (p = 0.04). Except for these three characteristics, no significant differences were found in the univariate analysis. The proportion of polypharmacy subjects between the relapse (46.0%) and non-relapse (34.6%) groups did not differ significantly (p = 0.15).

In the propensity-adjusted analysis, antipsychotic polypharmacy was not statistically associated with an increase of relapse: HR = 1.686 (0.812; 2.505).

Our study had several limitations.

First, the treatment (monotherapy or polypharmacy) was not based on random assignment; therefore, the results may be confounded by other factors. Although the propensity score can adjust for confounding by indication and selection bias, we cannot eliminate residual confounding due to unobserved factors 57 . This limitation of the current work can be moderated by the broad spectrum of characteristics collected in our study. However, well-designed randomised controlled trials are needed to determine the impact of polypharmacy on relapse in schizophrenia.

Second, several interesting data were not analysed. Information regarding dosages of different antipsychotic was not available. Moreover, antipsychotic combinations can be considered as a heterogeneous group, and the different combinations might be associated with different risks of relapse. However, our sample was not sufficient to determine the effect of specific polytherapy combinations.

Third, our sample may not be representative of schizophrenic patients in all of France and in other countries. However, the proportion of polypharmacy and relapse that are in line with previous studies 3 5 6 8 9 14 may indicate a good representativeness.

Fourth, it is possible that our study lacked statistical power to detect a difference in the rate of relapse. However, our analytical sample comprised 183 patients, 50 of whom had a relapse. This sample size was larger than that of several published randomised controlled trials designed to test the efficacy of polypharmacy vs. monotherapy, especially when we consider the length of the follow up, which were usually shorter in these prior studies than in our study 2 .

Finally, a major methodological problem remains in the definition of relapse for schizophrenic patients and the definition of monotherapy and polypharmacy. There are no generally accepted criteria for relapse 42 . However, we have chosen the most consensual definition in the recent scientific literature 42 43 58 . Methodological issues to be addressed in future trials should include clinically relevant relapse criteria. In the same way, the definitions of monotherapy and polypharmacy are not consensual. We defined the exposure treatment groups at baseline in a standard way: receiving one antipsychotic at baseline, regardless of medication class or molecule, compared with receiving two or more antipsychotics. We assumed that exposure was relatively stable over time and that the occurrence of relapse may be related to the exposure treatment at baseline. The clinical significance of this study should be cautiously interpreted in accordance with this chosen definition.