Selective serotonin reuptake inhibitors (SSRIs) for autism spectrum disorders (ASD)

Therapies for autism spectrum disorders (ASD)

The heterogeneous nature of problems seen within the autism spectrum means that it is often difficult to be sure which individuals will benefit from the many available therapies. It is also likely that different timing of therapy in relation to age and onset of problems will change outcomes. Many therapies are invasive, time‐consuming and/or expensive and little is known about their potential to cause harm. Pharmacological treatments have been used most commonly as adjuncts to behavioural intervention to target specific symptoms and behaviours. These treatments have been associated with reductions in sleep disturbance, mood disorder, poor attention/concentration and self harm or aggression towards others (Gringras 2000; King 2006).

Selective serotonin reuptake inhibitors (SSRIs)

While antidepressants were the most commonly prescribed medications for ASD, and the fastest growing therapeutic class for this indication through the early 2000s (Aman 2005; Oswald 2007), the trend has changed with recent studies finding neuroleptics (Mandell 2008; Rosenberg 2010) and even stimulants (Rosenberg 2010) being prescribed more often. The prescribing of SSRI drugs to children and adolescents for any indication has been curtailed since the Committee on Safety in Medicines (UK) and the Food and Drug Administration (USA) released safety warnings in 2003 and 2004 respectively, concerning an increased risk of suicide‐related behaviours associated with these medications (Murray 2005; Nemeroff 2007).

Types of studies

Trials were eligible for inclusion in the review if the assignment of study participants to intervention or control group was random.

Types of participants

Inclusion was limited to individuals with a diagnosis of an ASD defined using DSM‐IV or ICD‐10 or equivalent as a Pervasive Developmental Disorder, excluding Rett syndrome and Childhood Disintegrative Disorder. Diagnosis must have been made using a standardised diagnostic instrument (Childhood Autism Rating Scale (CARS), Autism Diagnostic Interview‐Revised (ADI‐R), Autism Diagnostic Observation Schedule (ADOS), Diagnostic Interview for Social and Communication Disorders (DISCO)) or by using established diagnostic criteria (ICD‐10, DSM‐IV). No age limits were applied.

Types of interventions

Oral SSRIs, regardless of dosage used or frequency of administration. The control treatment must be a placebo.

Types of outcome measures

Electronic searches

We searched the following databases.

1. Cochrane Central Register of Controlled Trials (CENTRAL), 2013, Issue 2, part of The Cochrane Library, last searched 16 March 2013

2. Ovid MEDLINE, 1950 to March Week 1 2013, last searched 16 March 2013

3. EMBASE (Ovid), 1980 to 2013 Week 11, last searched 16 March 2013

4. CINAHL (EBSCO), 1937 to current, last searched 16 March 2013

5. PsycINFO (Ovid), 1806 to March Week 2 2013, last searched 16 March 2013

6. ERIC, 1966 to current, last searched 16 March 2013

7. Sociological Abstracts (Proquest), 1952 to current, last searched 16 March 2013

8. International Clinical Trials Registry Platform (ICTRP), last searched 19 March 2013

9. ClinicalTrials.gov (CT.gov), last searched 19 March 2013

Detailed search strategies used for this update are in Appendix 1 and the search strategies for the original review are in Appendix 2.

Searching other resources

We also searched bibliographies of articles identified through the search strategy and contacted known experts in the field.

Selection of studies

Two authors screened titles and abstracts from the searches, in 2004, 2006 and 2008 (DW, KW), December 2009 (KW, NS), one author (KW) in 2012 and two authors (AB and MR) in June 2013. We resolved disagreement by consensus and discarded articles that did not fulfil the inclusion criteria. We retrieved potentially relevant articles for full‐text assessment and data extraction.

Data extraction and management

Data were organised using Review Manager 5 software (RevMan 2012). We developed data extraction forms a priori and included information regarding methods, participant details, dose and frequency of SSRI administration, and outcomes. Three pairs of independent review authors [(i) KW and DW, (ii) KW and NS, or (iii) AB and MR] extracted data. No disagreements arose.

Assessment of risk of bias in included studies

Three pairs of authors [(i) KW and DW, (ii) KW and NS, or (iii) AB and MR] independently assessed each included study using the risk of bias criteria outlined in chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) without blinding to authorship or source. We compared the assessments for inconsistencies and resolved differences in interpretation by discussion and consensus. We assessed risk of bias according to the following five domains with ratings of low, unclear or high risk of bias.

1. Sequence generation

Was the allocation sequence adequately generated?

Low risk ‐ computer‐generated random numbers, table of random numbers, coin‐tossing or similar.
 High risk ‐ day of week, even/odd clinic record number, clinician judgment, participant preference, laboratory test result such as haemoglobin value, or similar.
 Unclear risk ‐ insufficient information about the sequence generation process to permit judgment.

2. Allocation concealment

Was allocation adequately concealed?

Low risk ‐ central independent unit, sequentially numbered drug containers or sealed envelopes of identical appearance, or similar. 
 High risk ‐ alternation or rotation, date of birth, non‐opaque envelopes, open table of random numbers or similar.
 Unclear risk ‐ randomisation stated but no information on method used is available.

3. Blinding

Was knowledge of the allocated intervention adequately prevented during the study?

We assessed blinding of participant and personnel and blinding of outcome assessors. We took into account whether the placebo and medication were similar or different and other information provided in the text, such as whether outcome assessors were aware of side effects or participants correctly identified treatment group when available, as well as whether the trialists specifically commented on blinding.

4. Incomplete outcome data

Were incomplete data dealt with adequately by the researchers?

Low risk ‐ no missing outcome data, missing outcome data balanced in numbers across intervention groups and reasons for dropouts and withdrawals described or similar.
 High risk ‐ reason for missing outcome data likely to be related to true outcome or similar.
 Unclear risk ‐ number or reasons for dropouts and withdrawals not described.

5.      Selective outcome reporting

Are reports of the study free of suggestion of selective outcome reporting?

Low risk ‐ study protocol is available, published reports include all expected outcomes or similar.
 High risk ‐ not all of the study's prespecified primary outcomes have been reported, one or more reported primary outcomes were not prespecified or similar.
 Unclear risk ‐ insufficient information to permit judgement.

In our coding approach we acknowledge that assessing 'risk of bias' for this issue is very subjective unless trials are available on trial registers. We therefore code as 'unclear' unless trials register information indicates that risk of bias is low or high.

We also explored any other potential sources of bias, such as stopping the study early or extreme baseline imbalance.

Measures of treatment effect

We present effect estimates as provided in individual studies along with one meta‐analysis.

For details of all methods planned in the protocol that were not used in this version of the review but may be used in future updates, see Table 1.

Unit of analysis issues

We did not identify any unit of analysis errors. In all included studies, randomisation, reporting and analysis were per individual participant rather than cluster‐randomised studies.

Dealing with missing data

Where possible, missing data and dropouts were assessed and reported for each included study. Reasons for missing data are provided when reported in trials. Where insufficient data were reported, we contacted the trial authors for further information, but received no replies.

Assessment of heterogeneity

We assessed consistency of results visually and by examining I² (Higgins 2002), a value which describes approximately the proportion of variation in point estimates that is due to heterogeneity rather than sampling error (chance). We will supplement this with a test of homogeneity to determine the strength of evidence that the heterogeneity is genuine.

Assessment of reporting biases

We found insufficient studies to allow for the use of funnel plots to investigate any relationship between effect size and study precision (closely related to sample size).

Data synthesis

Binary data

Since more than two studies presented outcomes from standardised instruments as proportions for adults, the relative risk with 95% confidence intervals, was calculated from meta‐analysis, using a random‐effects model. Number needed to treat was not calculated.

Subgroup analysis and investigation of heterogeneity

Subgroup analysis was not possible because of a lack of data for meta‐analysis.

Sensitivity analysis

Sensitivity analysis was not possible because there were too few data available for meta‐analysis for the same drug therapy, age group and clinical outcome.

Results of the search

The original review, based on searches till December 2009, included seven studies.

The search for this update was run in 2012. From 83 records identified, we found no new published trials. We found one new trial on ClinicalTrials.gov and the author provided methodological information and data when contacted. The most recent search was run in March 2013 from which we identified a further 33 records including one published trial. Thus, we include a total of nine randomised controlled trials (RCTs) with 320 participants in this update.

Of note, we did not search for fenfluramine directly and it is not registered for use as an antidepressant, but was identified as a selective serotonin reuptake inhibitor (SSRI) in electronic searches. We debated its inclusion and chose to retain the two trials that evaluated it to provide a complete clinical picture. The tricyclic antidepressant clomipramine was also identified using this search method and is known to have serotonin uptake inhibitor actions, but was excluded because, unlike fenfluramine, it is included in a systematic review of tricyclic antidepressants and autism spectrum disorders (ASD) (Hurwitz 2012).

Included studies

Nine studies are included in this updated review.

Excluded studies

We excluded seven studies after reading the full paper. Two studies (Gordon 1993; Remington 2001) were trials of clomipramine, a tricyclic antidepressant that has SSRI characteristics but is not classed as such. Four studies were not randomised controlled trials (McDougle 1996; Sanchez 1996; Peral 1999; Doyle 2001). One of these studies (Doyle 2001) was a cost analysis of a treatment not classed as an SSRI. A further study did not involve participants with ASD (Humble 2001).

Allocation

Blinding

Outcome assessors were reported as being blind to treatment allocation in four studies (McDougle 1996; Hollander 2005; King 2009; Hollander 2012) with a fifth study stating that "investigators" were blind (NCT00609531). In the remaining four studies it was not possible to ascertain if outcome assessors were blinded. Of these four studies, two stated that participants and treating physicians were blinded but did not report blinding of outcome assessors (Barthelemy 1989; Buchsbaum 2001) and two studies used the term "double blind" to refer to all blinding (Leventhal 1993; Sugie 2005). Given that the outcome measures used rely on subjective observation and assessment, there is potential for bias where outcome assessors were not adequately blinded to treatment allocation.

Incomplete outcome data

Three studies reported no loss to follow‐up (Barthelemy 1989; McDougle 1996; Buchsbaum 2001). One study (King 2009) reported that 13 of 76 withdrew from the placebo arm and 13 of 73 withdrew from the treatment arm of the trial. Reasons for withdrawal included adverse events (one serious in the treatment group), protocol violation and consent withdrawal. One study (Leventhal 1993) reported one withdrawal prior to the randomised phase and reported no data from this participant. One study (Sugie 2005) excluded one participant due to non‐compliance. One study (Hollander 2012) screened 48 individuals, with 37 being randomly assigned to placebo or treatment group. Thirty‐four participants with postrandomisation data were included in the analysis, including four who dropped out after randomisation but for whom postrandomisation data had been collected. Two of these participants did not comply with the study protocol, one participant discontinued due to relocation and one discontinued due to poor toleration of the treatment. One study (Hollander 2005) reported that, of 62 participants who consented, 18 were excluded for non‐eligibility or non‐compliance and 44 were randomised. Of these, 39 participants were included and their completed outcome data were published, three were excluded due to non‐compliance, one was excluded due to lack of efficacy and one was lost from analysis due to lost records. The exclusion of a participant due to lack of efficacy creates a risk of bias, as does exclusion of those who were not compliant with therapy. One study that had expected to recruit 40 participants only recruited 12, but there was no loss to follow‐up of those recruited (NCT00609531).

Four studies reported not using, or were assessed as not using, an intention‐to‐treat analysis (Leventhal 1993; Hollander 2005; Sugie 2005; Hollander 2012). Four studies reported the use of intention‐to‐treat analysis or did not require any statistical adjustments as they had no losses to follow‐up or changes in treatment allocation (Barthelemy 1989; McDougle 1996; Buchsbaum 2001; King 2009). One study provided raw data prior to publication (NCT00609531).

Selective reporting

The likelihood of selective reporting, that is reporting only those outcomes that showed evidence of treatment effect, was lowest for King 2009, which was registered at the start of the trial. One other trial was registered and provided data for listed outcomes as well as some outcomes not listed at trial registration (NCT00609531). No other trial reported being registered prior to start. Five trials reported negative outcomes (Barthelemy 1989; Leventhal 1993; Buchsbaum 2001; Hollander 2005; King 2009); one reported only positive outcomes (McDougle 1996), and one study reported on all outcome measures assessed (Hollander 2012). One study only reported effectiveness for genetic subgroups (Sugie 2005). All studies reported at least one relevant clinical outcome. We would strongly suspect selective reporting if the primary outcomes to be assessed in this review were not presented in the included trials reports.

To assess selective reporting, we assigned each included study a judgement relating to the risk of bias for that study by answering a prespecified question about the study's adequacy in relation to each of the key criteria. Judgements were 'low risk of bias', 'high risk of bias' and 'unclear risk of bias' (used if the risk of bias was unknown).

Other potential sources of bias

We are unaware of any further potential sources of bias in the included studies.

Core ASD features

Composite measures of CGI and OCB (child data only)

One study (King 2009) used a composite measure of the CGI improvement scale (CGI‐I) and the CY‐BOCS‐PDD. In this study a CGI‐I score of 1 or 2 and a 25% reduction on the CY‐BOCS‐PDD were required as evidence of improvement. The authors of this study reported that the use of the composite score was a way of "increasing the threshold for positive response". Comparative analysis showed no difference between treatment groups for the composite score at 12 weeks (20.6% for citalopram versus 13.2% for placebo; P = 0.28).

One study (Hollander 2005) used a composite score that included the CGI‐AD and a measure of change of repetitive behaviour, based on the CY‐BOCS. The authors created a composite score by creating "a change score by subtracting the pre‐test CY‐BOCS from the post‐test CY‐BOCS. Negative values on this measure indicate a reduction in repetitive behaviours at post‐test whereas positive scores indicate an increase. This raw change measure was then added to the CGI‐AD measure to augment the overall change in autism severity". Results of the mixed regression analysis indicated a trend towards reduction in this global autism composite improvement measure for participants on fluoxetine as compared to placebo (Z = 1.907, standard error (SE) = 0.703, P = 0.056).

Composite scores used were different and were presented as categorical data in one study and continuous data in the other, so that meta‐analysis was not possible.

Global assessment of health and function

Non‐core aspects of behaviour and function

Adverse effects

Quality of life

No study used any standardised measure of quality of life.

Long term outcomes

No study recorded outcome beyond the length of the trial duration, with the exception of Barthelemy 1989 who monitored weight loss (see adverse effects).