Attrition in trials evaluating complex interventions for schizophrenia: Systematic review and meta-analysis

Abstract

Essential criteria for the methodological quality and validity of randomized controlled trials are the drop-out rates from both the experimental intervention and the study as a whole. This systematic review and meta-analysis assessed these drop-out rates in non-pharmacological schizophrenia trials. A systematic literature search was used to identify relevant trials with ≥100 sample size and to extract the drop-out data. The rates of drop-out from the experimental intervention and study were calculated with meta-analysis of proportions. Meta-regression was applied to explore the association between the study and sample characteristics and the drop-out rates. 43 RCTs were found, with drop-out from intervention ranging from 0% to 63% and study drop-out ranging from 4% to 71%. Meta-analyses of proportions showed an overall drop-out rate of 14% (95% CI: 13–15%) at the experimental intervention level and 20% (95% CI: 17–24%) at the study level. Meta-regression showed that the active intervention drop-out rates were predicted by the number of intervention sessions. In non-pharmacological schizophrenia trials, drop-out rates of less than 20% can be achieved for both the study and the experimental intervention. A high heterogeneity of drop-out rates across studies shows that even lower rates are achievable.

Introduction

Two major challenges in randomized controlled trials (RCTs) include treatment noncompliance and missing outcome data. These complications are caused by participants not receiving or discontinuing the allocated intervention and loss to follow-up. Although some attrition can be expected in clinical trials, ensuring retention of participants is crucial to achieve sufficient statistical power to detect the effect of treatment. There is no standard for acceptable drop-out rates but some evidence suggests that attrition levels as low as 5% may introduce risk of bias and the rate exceeding 20% can threaten trial validity (Polit and Hungler, 1995, Schulz and Grimes, 2002). The credibility of a trial also depends on the attrition rates. A survey involving patients, caregivers, statisticians and clinicians found drop-out out rates exceeding 25–30% can affect whether pharmacological trials in schizophrenia are judged as credible (Xia et al., 2009).

Loss of data can occur at different levels within a trial. Treatment noncompliance refers to issues with following the recommendations for prescribed treatments. In non-pharmacological trials it can be the failure to attend a required number of appointments or sessions (Nosé et al., 2003). Patients can also choose to completely discontinue an intervention. Treatment drop-out been estimated to be 13% in RCTs testing psychosocial treatments for people with schizophrenia (Villeneuve et al., 2010). In contrast to pharmacological trials, discontinuing a non-pharmacological intervention usually does not automatically exclude the participant from the follow-up, so outcome data can be collected if the participant is willing to provide them. Drop-out at the study level is defined as a failure to complete follow-up assessments, usually due to withdrawal from the study, and can occur following completing an intervention. Most research investigating drop-out of people with schizophrenia at a study level have focused on pharmacological trials, reporting 35% study drop-out in a meta-analysis of RCTs of antipsychotic drugs (Leucht et al., 2013). This contrast in the drop-out rates between pharmacological and non-pharmacological trials can be expected due to the differences in procedures (e.g. single-vs. double-blinding) (Huhn et al., 2014) and non-pharmacological trials not usually prohibiting taking already prescribed medication. Receiving pharmacotherapy on top of the experimental trial intervention might reduce the likelihood of dropping out even in case of the intervention not being efficacious for the patient.

To our knowledge there have been no systematic studies establishing the scale of drop-out from RCTs evaluating non-pharmacological interventions for schizophrenia at both the experimental intervention and the study level.

Complete outcome data from all randomized participants is necessary for a full application of the intention-to-treat (ITT) approach (Gupta, 2011), which is the ‘gold standard’ for analyzing the results from trials evaluating the effectiveness of a treatment in a pragmatic setting (Altman, 1996, Begg et al., 1996). ITT analysis includes all randomized participants, regardless of whether they adhered to or received the allocated intervention. The purpose of the ITT approach is to reflect a real-life effect of an intervention in clinical practice, taking into account the deviations from protocol that would occur in routine practice. Thus, every effort should be made to obtain complete outcome data for all randomized participants, including those who did not complete the intervention but continued to complete follow-up assessments.

Developing effective and efficient retention strategies for RCTs requires an understanding of what factors affect the likelihood of premature discontinuation of intervention or loss to follow-up. The vast majority of the literature on the determinants of attrition in psychiatric treatment has focused on pharmacological trials and identified factors negatively correlated with treatment adherence, such as substance misuse (Kampman and Lethinen, 1999, Nosé et al., 2003), unemployment (Nosé et al., 2003), unpleasant side effects of medication (Kampman and Lethinen, 1999), negative attitude towards medication (Kampman and Lethinen, 1999). In one available study analyzing drop-out from psychosocial treatment for schizophrenia the following variables were found to be associated with higher drop-out rates: being male, higher age, longer illness duration, longer treatment duration (Villeneuve et al., 2010). Lower drop-out rates were associated with study quality and inpatient setting (Villeneuve et al., 2010). These findings suggest that both study and sample characteristics can affect drop-out from clinical trials. Understanding what factors influence discontinuation of intervention and loss to follow-up can guide the development of strategies to limit these. Investigating relationships between specific study and sample characteristics and drop-out rates allows for examining if the data missing from those who dropped out is missing at random or whether individuals dropping out have any characteristics in common that make them more likely to prematurely discontinue participation in trials.

The aims of this study were first: to systematically identify relevant large-scale RCTs evaluating non-pharmacological interventions for individuals with schizophrenia; second: to perform meta-analyses to establish the proportion of participants who drop-out of a) experimental intervention and b) study; and thirdly to perform a meta-regression to examine the predictors of drop-out rates.