Search strategy

The searches, conducted in July 2004, aimed to retrieve both published and unpublished papers with no language restrictions. A date restriction of 1999 onwards was placed on the methylphenidate searches to update the report produced by Lord and Paisley published in 2000. A date restriction of 1997 onwards was placed on the searches for dexamfetamine to update the AHRQ report (which included a review of this drug). Research on atomoxetine was searched for from 1981 onwards. The search strategy was based on that used in the AHRQ report.

Inclusion/exclusion criteria

To be eligible for inclusion, studies had to be randomised controlled trials (RCTs) of at least 3 weeks' duration (3 weeks per treatment arm in parallel studies and 3 weeks in overall trial length for crossover studies). In addition, systematic reviews were included to examine adverse events data. For the assessment of cost-effectiveness, a broader range of studies was considered.

The studies had to examine MPH, DEX or ATX used alone or in combination with non-drug interventions and be compared with placebo, with one another in head-to-head comparisons or with non-drug interventions. Non-drug interventions included any type of psychological and behavioural strategies (e.g. cognitive behavioural therapy, child or parent training, bibliotherapy) and/or nutritional interventions. Studies that compared MPH, DEX or ATX with other drugs (e.g. Adderall) not licensed in the UK for ADHD were included as long as there was a placebo group. This was applied to both efficacy and adverse events data.

Participants included children and adolescents under 18 years of age diagnosed with ADHD (including hyperkinetic disorder). There was no lower age limitation (although there was no lower age limitation for the report, it is noted that MPH is indicated for children older than 6 years, and ATX is indicated for children aged 6 years and over). The diagnosis must have been made in an explicit way, preferably using either the ICD-10 criteria or the DSM-IV criteria. Studies including participants with conditions other than ADHD (e.g. Tourette's syndrome) were excluded unless they reported separate analyses for patients with ADHD alone.

To be included in the review, trials had to report results on one or more of the following:

• core symptoms (including measures of inattention, hyperactivity, impulsivity)
• quality of life (QoL) (Clinical Global Impression or overall severity indices were used as a proxy of QoL)
• adverse effects (including loss of appetite, insomnia, headache, stomach ache and weight loss).

Studies that only examined tests of psychological function (e.g. the continuous performance test), measures of depression and/or anxiety, or measures of coexistent problems (including poor peer relationships, and conduct/oppositional-disorder-related outcomes) were not included in the review. Studies that presented results in figures without presenting actual numbers, or only significance values for comparisons, were excluded from the review.

Studies that met the inclusion criteria above, but were only published as abstracts or as conference presentations were not included in the review unless a full paper could be obtained that related to the abstract.

Two reviewers independently screened all titles and abstracts, including economic evaluations, identified in the updated literature search. Full paper manuscripts of any titles/abstracts that were considered relevant by either reviewer were obtained where possible. In addition, full paper copies of relevant studies presented in the NICE, AHRQ and CCOHTA reports were obtained. The full papers were then assessed against the inclusion criteria by one reviewer and checked by another. Any discrepancies were resolved by consensus and, if necessary, a third reviewer was consulted.

Data extraction and quality assessment

The quality of the clinical effectiveness studies was assessed using modified criteria based on CRD Report No. 4. Each study was assessed and data were extracted by one reviewer and independently checked for agreement with a second reviewer. Disagreements were resolved by consensus and, if necessary, a third reviewer was consulted.

Methods of analysis/synthesis

Clinical effectiveness data were reported separately for each drug and by the type of comparison. Data for MPH were also analysed separately based on whether it was administered as an immediate release or extended release formulation. For all drugs, the data were examined by dose. Data for the core outcomes of hyperactivity (using any scale), Clinical Global Impression (as a proxy of QoL) and adverse events were reported. For crossover studies, the mean and standard deviation (SD) for each outcome were data extracted for end of trial data (i.e. baseline data were not considered). Where possible, we aimed to calculate mean difference and standard errors for crossover studies in order to facilitate meta-analysis. However, owing to the lack of information needed to calculate mean differences in many of the studies, this was not possible. For parallel studies, change scores were reported where given, otherwise means and SDs were presented for end of trial data. In addition, mean differences with 95% confidence intervals were calculated for each study. For adverse events, self-ratings were reported when used, otherwise, parent reports were utilised. Percentages of participants reporting adverse events were used to calculate numbers of events in each treatment arm.

For the cost-effectiveness section of the report, details of each identified published economic evaluation, together with a critical appraisal of its quality, were presented in structured tables.

Handling company submissions

All the clinical effectiveness data included in the company submissions were assessed. Where these met the inclusion criteria they were included in the clinical effectiveness review. All economic evaluations (including accompanying models) included in the company submissions were assessed and detailed assessments of the assumptions underlying the submitted analyses were undertaken.

A new model was developed to assess the cost-effectiveness of the alternative treatments in terms of cost per quality-adjusted life-year. To achieve this, a mixed treatment comparison model was used to estimate the differential mean response rates. Monte Carlo simulation was used to reflect uncertainty in the cost-effectiveness results.

Clinical effectiveness

In the previous systematic reviews (NICE, AHRQ and CCOHTA), 65 studies were identified as potentially relevant to the current systematic review, and full paper copies were ordered. Of these, 40 met the inclusion criteria. In the updated search, a total of 2908 titles and abstracts relating to clinical effectiveness or systematic reviews of adverse events were identified and screened for relevance. Of these, 409 full paper copies were examined in detail and assessed for inclusion. Of these, 20 RCTs and one systematic review met the inclusion criteria. In addition, four commercial-in-confidence papers were included. Overall, this gives a total of 65 papers.

As reported in the previous NICE report, and in the AHRQ and CCOHTA reviews, the plethora of MPH studies suggest that MPH is effective at reducing hyperactivity and improving QoL (as determined by Clinical Global Impression) in children. It was noted, however, that the majority of studies that evaluated the effectiveness of MPH did not adequately report their study methodology. Hence, the reliability of the study results is not known. There appears to be little evidence supporting a difference in the effectiveness of immediate-release (IR) and extended-release (ER) MPH.

Similarly, DEX also appears to be effective at reducing hyperactivity and improving QoL, although this is based on a small number of studies. Only one study adequately reported the study methodology.

There was consistent evidence that ATX was superior to placebo for hyperactivity and Clinical Global Impression. Studies on ATX more often reported the study methodology well, and the results are likely to be reliable.

Very few studies made direct head-to-head comparisons between the drugs. The previous NICE report stated that there appeared to be little evidence of difference in the effectiveness of MPH and DEX. No recent studies were found in the updated search. Although the studies reported variable results, the one study that reported no statistically significant differences between the two drugs was deemed to be of good quality, whereas the quality of the others was uncertain given the poor reporting of study methodologies.

One study that compared MPH and ATX reported no differences between the drugs for hyperactivity or Clinical Global Impression. This study did not adequately report study methodology, and the results should be interpreted with caution. [Confidential information removed].

Few studies were included in the review that examined a non-drug intervention in combination with MPH, DEX or ATX. Generally, the results were variable. The studies were, however, heterogeneous regarding the type of non-drug interventions examined and the scales used to measure outcomes.

Adequate and informative data regarding the potential adverse effects of MPH, DEX and ATX are lacking. Overall, higher dosages of IR-MPH appear to be associated with the occurrence of headache, lost appetite, stomach ache and insomnia compared with placebo. ER-MPH appears to be associated with decreased appetite and increased insomnia. However, a previous systematic review highlighted the need for further research into somatic complaints, which may be associated with the disorder itself rather than methylphenidate treatment. Similarly, high doses of DEX appear to be associated with decreased appetite and increased sleeping problems. ATX of any dose may impair appetite.

Cost-effectiveness

The review highlighted a number of potential limitations in the existing literature. In particular, the review highlighted limitations in estimating treatment effectiveness and associated utility values. These limitations may stem from a lack of available data. A new economic model was developed for this report. Pooling was limited in the clinical effectiveness review, owing to heterogeneity between trials. However, some degree of pooling is necessary to proceed with an economic model. The issue of heterogeneity was overcome by basing the base case on trials that are more similar in terms of how they measure the outcome of interest. In a series of sensitivity analyses more trials were included by relaxing the criterion of similarity in outcome measurement. Data on resource use associated with ADHD in the UK were lacking, and so the model relies on estimates from experts.

Given the lack of available evidence for statistically significant differences in efficacy between the alternative drugs, the results of the economic model were largely driven by drug cost, in which there are marked differences. The economic evaluation clearly suggests an optimal treatment strategy, that is, DEX first-line, followed by IR-MPH for treatment failures, followed by ATX for repeat treatment failures. If DEX is considered not suitable as a first-line therapy, the optimal strategy is IR-MPH first-line, followed by DEX as second-line and ATX again as third-line. For patients contraindicated to stimulants, ATX is preferred to no treatment. For patients in whom a midday dose of medication is unworkable, ER-MPH is preferred to ATX, and ER-MPH12 appears more cost-effective than ER-MPH8.

The model is not without limitations. As identified in the clinical effectiveness review, the reporting of studies was poor, there are few data to discriminate between the drugs in efficacy or adverse events and there are few data on long-term efficacy and adverse events associated with medical management of ADHD. The data do not allow discrimination between patients with ADHD in terms of ADHD subtype, age, gender or previous treatment. These caveats must be borne in mind when interpreting the model results.