Additional analyses

Introduction

The findings from the primary trial analysis according to the prespecified SAP provide no evidence for a difference in effect between the placebo and treatment arms. Indeed, across all primary and secondary outcome variables, outcomes were remarkably similar in the trial arms at 8 weeks and do not indicate any advantage of OROS-MPH over placebo, even at trend level. Furthermore, findings from the per-protocol analyses, as laid out in the SAP, did not provide any evidence for efficacy of the OROS-MPH intervention either.

As outlined in the justification for this study, MPH has been investigated in adults with ADHD in community ADHD clinics in previous studies and was found to have an average SMD between-group effect of around 0.51.24 A 2018 comprehensive network meta-analysis32 estimated an effect size from RCTs of MPH in reduction of ADHD symptoms in adults, with a SMD of 0.49 (95% CI 0.35 to 0.64). One of the key rationales for this study was to investigate whether or not OROS-MPH had a different effect in a young adult male prison population, compared with the previous studies. It was proposed that the symptoms of inattention and hyperactivity/impulsivity could potentially have a different meaning in this population, reflecting different comorbid mental health or neurodevelopmental disorders. It was further proposed that high levels of drug and alcohol use among offenders might modify the previously reported effects. Although a change in the effect size, compared with previous studies, was envisaged, the absence of even a small effect was unexpected (the estimated between-arm standardised effect is < 0.1).

Although there appeared to be no between-group effects, there was a change over the observation period indicating an apparent improvement in symptoms in both arms. The mean CAARS-O scores dropped from 36.4 points at baseline to 28.0 points at week 8 (a difference of 8.4 points) in the OROS-MPH arm of the trial and from 37.2 points to 29.3 points (a difference of 7.9 points) in the placebo arm (see Table 8). This is not unexpected as significant change has been observed in placebo arms of most RCTs for MPH in ADHD. However, we saw a much greater change in symptoms in the pilot study for this project at HMP YOI Isis, which included 121 participants in a single unblinded group receiving OROS-MPH.29 The pilot study found the mean change over time for CAARS-O scores to be 22.6 points (decreasing from 37.6 to 15.0 points) using the primary outcome at week 12 and a last observation carried forward approach to deal with missing observations. For the week 8 secondary outcome data in the pilot using the last observation carried forward approach, the difference was even greater: 29.3 points (decreasing from 37.6 to 8.31 points). This is far larger than the change observed in either of the two treatment arms of the current trial.

For these reasons, we completed a set of further analyses to explore possible explanations for the unexpected findings in this trial. These further analyses were proposed after database lock and the review of the findings from the primary trial analyses according to the SAP. They are post hoc analyses and should be regarded as entirely exploratory in nature, with the aim of making suggestions for possible explanations of the trial findings.

Investigating differences between the open pilot study and the randomised controlled trial

We observed that the pre–post differences for both the treatment and placebo arms were smaller in this trial than in the previous open-label pilot study. To investigate the potential reasons for this we looked at differences in the conduct of the pilot study and the trial, and compared the estimated pre–post differences between the open-label pilot study (CIAO-I) and the RCT (CIAO-II) at the HMP YOI Isis site only (because the pilot was conducted at this site only).

Comparison of pre–post differences

To investigate whether or not the pre–post differences were of different sizes in the open-label pilot study (CIAO-I) and this RCT (CIAO-II), we applied similar inclusion and retention procedures across the two studies and restricted our attention to HMP YOI Isis only. Screening-level data are not available for the pilot study, but we confirmed that similar selection and inclusion criteria were applied across both studies. However, procedures for study retention differed between the two studies. A key difference is that, for the pilot study, participants were not retained in the study if they no longer wished to take the trial medication, whereas, in the current trial, an ITT approach was applied, with participants retained in the trial for outcome assessments regardless of whether or not they continued to take medication. To assess change over time under the same assumptions in both studies, we applied similar criteria for follow-up in the trial. In the pilot, participants left the study and provided no further outcome data after a period of non-compliance with trial medication. The criteria for compliance with medication and retention in the pilot were not clearly defined in the protocol and no medication log was obtained for the pilot study; however, we estimated this to be 10 days of continuously not taking the trial medication.

To illustrate this, we plot comparable data by applying similar compliance and follow-up rules to the OROS-MPH arm of the HMP YOI Isis data from this trial as those applied in the pilot study. To match the procedure followed in the pilot study as closely as possible, participant outcome data from the current trial were treated as missing following a continuous period of 10 days without taking trial medication. To match the outcome period in the current trial (baseline to week-8 assessment), we used the 8-week assessment data from the pilot study (rather than the 12-week data) and included only those participants who provided data at the 8-week time point.

A total of 58 participants were in the OROS-MPH arm and based in the HMP YOI Isis site. Of these, 41 participants were identified as having been in the OROS-MPH arm and not having a gap in taking study medication for more than 10 days at any time. The mean values for CAARS-O scores at baseline, week 5 and week 8 are presented in Table 19. The mean score change between baseline and week 8 was 11.1 points, and between baseline and week 5 it was 12.1 points. This compares with a score change of 25.0 points for the 8-week outcome and 25.1 points for the 5-week outcome in the pilot study.

In conclusion, even after ensuring that the same prison was assessed in both studies and change was estimated under the same assumptions, the observed difference over time remained much larger in the pilot study than in the trial.

Baseline measures

To identify potential differences between participants in the two studies that could account for differences in treatment effects at HMP YOI Isis, we compared baseline measures for the participants who initiated trial medication in the two studies. Age and general cognitive ability (IQ) were similar between the two studies (Table 20) (see Appendices 3 and 4).

Ethnic background was slightly different; in the pilot and current trial, the proportions of white participants were 52.9% and 34.8% respectively, and the proportions of black participants were 27.3% and 40.0%, respectively. The current trial appears to be more representative of the ethnic mix of prisoners at HMP YOI Isis (see Chapter 6, Generalisability) than the pilot study. Potentially, there could be differences between ethnic groups in the way that ADHD symptoms are recognised and reported, and the different proportions of ethnic groups in the open-label pilot study, CIAO-I, and the CIAO-II RCT might lead to minor differences in pre–post differences. However, this seems an unlikely explanation as, in the current trial, outcomes between the OROS-MPH and placebo groups were similar between HMP YOI Isis and HM YOI Polmont, where 98% of participants came from predominantly white ethnic backgrounds.

There were also differences in the highest level of education achieved, with 29% obtaining an educational qualification up to General Certificate of Secondary Education (GCSE)/General National Vocational Qualification (GNVQ) level or higher in the pilot, compared with 73% in the current trial. This difference is thought to be largely due to a change in policy in August 2015, when mandatory assessments in English and mathematics were introduced, leading to a larger number of prisoners receiving qualifications within the prison system. However, it is also notable that the reported rate of unemployment prior to entering prison was 10% in the pilot study, compared with 27% in the current trial.

Overall, there is some evidence that participants in the current trial were functioning at a higher level than participants in the pilot study, potentially indicating a less impaired group. As impairment is a core component of ADHD, this may indicate greater uncertainty over the diagnosis of ADHD. Clinical measures of ADHD, emotional dysregulation, attitudes towards violence and overall global clinical severity were comparable between the two studies

Differences in the protocols or procedures for the pilot study and current trial

To establish whether or not there were any changes in the prison service over time that might have affected the trial outcomes, we reviewed with research staff any potential changes, such as mental health or nursing staff, age of inmates and change in the prison regime or length of sentences. Overall, there were few differences, although the health-care team at HMP YOI Isis during the pilot study had been run by Health Care UK, whereas the health-care team during the trial was run by Oxleas NHS Foundation Trust. However, this was unlikely to have had an impact because levels of training and experience appeared to be similar.

In HMP YOI Isis, the age range of prisoners was lower at the start of the pilot study, when all inmates were aged 18–25 years, and the age range was then later extended up to about aged 30 years. For the CIAO-II RCT, a subgroup of older prisoners, up to around 45 years of age, was introduced into the prison. It is not clear how this would have affected the results of the RCT, but this may have increased health-care staff workload because there are more general health issues with increasing age. Oxleas NHS Foundation Trust had also increased access to psychological treatments, although few of the trial participants were engaged in such programmes, making this unlikely to have contributed to differences between the two studies.

We also reviewed the study documentation to identify any potential differences. A key difference was the introduction of placebo in the current RCT, which affected the amount of documentation required for dispensing medication for each participant, with an increased involvement of the pharmacy and the health-care staff as a result of dispensing individual trial kits in the current trial.

We also estimated how much time researchers spent with participants in the two trials, and whether or not there were any changes in the level of support provided or the level of engagement with the research team. The main difference was that a single key research assistant was engaged throughout the pilot study, whereas, in the current RCT, there was greater turnover of key staff. There was also a higher rate of recruitment and more administrative tasks for the RCT, which might have reduced the contact time between participants and members of the research team. The larger number of measures in the RCT meant that more time was spent with individual participants, but that potentially less time was spent on follow-up and general support for participants. These changes in relation to research staff could potentially have reduced the combined effect of medication and staff attention in the current trial, compared with the pilot study. In addition, there may have been stronger rater bias effects, leading to lower investigator-rated CAARS-O scores in the pilot study, particularly as this was also an open trial.

Checking data integrity

To check for the possibility of a ‘hidden’ effect in the trial data for any reason, we investigated the possibility that there are two distinct clusters of outcomes. We would expect to see a bimodal distribution if there were any drug effects, reflecting the existence of two response groups. To achieve this, we checked the distribution in the change of CAARS-O scores from baseline to week 8 to see if there is any evidence for two clusters of the primary outcome data, by plotting the distribution of change scores from baseline to week 8. As shown in Figure 5, there is no evidence for two potential clusters.

FIGURE 5 Distribution of change in CAARS-O scores from baseline to week 8. (Figure not reproduced here; see the original article.)

Investigating efficacy in specific populations (subgroup analyses for CAARS-O)

We conducted a series of subgroup analyses of the trial data to identify subgroups potentially showing a greater or smaller effect of the trial medication. These further analyses were designed to address specific questions that might explain the current findings and could form the basis for future studies. Subgroup analyses on very small subpopulations are unlikely to be informative; for this reason, we did not conduct analyses on subgroups of ≤ 30 participants.

Was the trial outcome effected by a systematic change to trial procedures?

To check whether or not a systematic change might have occurred during the trial (e.g. unidentified changes in the conduct of the trial procedures), we provide outcomes in order of randomisation to the trial. We provide outcome data by subgroups defined by order of recruitment: 1–50, 51–100, 101–150 and 151–200. As seen in Table 21, there is little evidence to suggest that CAARS-O scores (across trial arms) varied between the four groups at any assessment time point. In addition, the sizes of the estimated between-arm effects on OROS-MPH at 8 weeks did not show any OROS-MPH effects for any of the four subgroups. By definition, there are 50 participants in each subgroup, and the estimated treatment differences between the OROS-MPH and placebo arms, in groups 1–50, 51–100, 101–150 and 151–200, are as follows: 0.56 (95% CI –5.39 to 6.52), –0.001 (95% CI –6.41 to 6.41), 3.45 (95% CI –3.90 to 10.80) and –0.54 (95% CI –5.99 to 4.90), respectively.

Was the trial outcome affected by including less severe cases of attention deficit hyperactivity disorder in the trial?

To check whether or not only the most severe cases benefit from drug treatment, we assessed the treatment effects in high- and low-severity groups for ADHD symptoms at baseline. We provided a subgroup analysis for high severity based on the baseline CAARS-O score, defined as t-scores of 70 (equivalent to a raw score of ≥ 35 points). A total of 123 participants had a baseline CAARS-O score of ≥ 35 points. As shown in Table 22, this group show greater change in scores than the overall trial: 12.26 points for OROS-MPH and 9.35 points for placebo. This is an improvement in the OROS-MPH group, compared with the placebo group, of 2.81 points (95% CI –1.11 to 6.73 points; t = 1.42; p = 0.16), which is a larger difference than for the overall trial, but does not reach the effect size specified in the sample size calculation (5 points) and is not statistically significant

Was the trial outcome affected by poor diagnostic accuracy in this prison population?

To investigate the potential effect of diagnostic accuracy, we provide outcome data for a subgroup with a more clearly defined cluster of ADHD symptoms, using the DIVA diagnostic instrument data. The rationale is that, by raising the usual symptom count thresholds for the diagnosis, we will enhance the specificity of the symptom count data to the diagnosis of ADHD, thereby enhancing the certainty of the diagnosis. This was achieved by defining the subgroup as individuals meeting combined type criteria for ADHD in both childhood and adulthood from the DIVA assessment using symptom thresholds of ≥ 7 for both symptoms of inattention and hyperactivity/impulsivity, in both childhood and adulthood.

From the total sample of 200 participants, 102 met these higher symptom thresholds. In this subgroup, the estimated score difference between the OROS-MPH and placebo arms was estimated to be 1.83 (95% CI –2.34 to 6.01; t = 0.87; p = 0.39). This was a non-significant improvement in the OROS-MPH arm.

Could the trial outcome be affected by including participants with different levels of emotional dysregulation?

Emotional dysregulation is known to be associated with ADHD and might reflect an index of severity of ADHD. However, symptoms of emotional dysregulation can also reflect the severity of other mental health disorders commonly seen in the prison population. It is therefore feasible that a group with high emotional dysregulation could show a greater effect of medication (if ADHD medication has greater effects in more severe cases), or might show a smaller effect of medication (if the symptoms of ADHD are better explained by a comorbid condition). We therefore investigated the outcomes by subgroups of high and low levels of emotional dysregulation, as measured by the WRAADDS at baseline (Table 23).

The median WRAADDS score at baseline was 17 in the the CIAO-II RCT sample. The participants were roughly evenly divided between low and high levels of emotional dysregulation, with 113 participants having high levels of emotional dysregulation. Table 23 shows summaries of the primary outcome variable by high or low level of emotional dysregulation and trial arm. The estimated score difference between the OROS-MPH and placebo arms in the subgroup with high emotional dysregulation was estimated to be 2.30 (95% CI –1.81 to 6.41; t = 1.11; p = 0.27). Although this is an improvement in the OROS-MPH arm, this is still below the difference stated in the sample size and is not statistically significant. There was also little observed difference between the two arms in the low level of emotional dysregulation subgroup, with an estimated score difference between the OROS-MPH and placebo arms of –0.85 (95% CI –5.54 to 3.84; t = 0.36; p = 0.72), which was an improvement in the OROS-MPH arm.

Could the trial outcome be affected by including participants with borderline personality disorder?

Symptoms of emotional dysregulation may also reflect the severity of other mental health disorders commonly seen in the prison population, such as BPD. It can be difficult to separate ADHD from BPD on clinical grounds alone. For example, in our recent research, we found that women meeting diagnostic criteria for BPD also show high levels of inattentiveness in their daily lives, in addition to shared symptoms such as emotional dysregulation.46 To investigate the possibility that the subgroup meeting criteria for BPD might be reducing the overall effect of medication in the trial, we aimed to provide outcomes in a subgroup with and a subgroup without BPD. However, only 15 participants had a high risk of BPD, and 185 did not have BPD. These analyses were not performed as the BPD group was too small, and the small number of participants with BPD suggests that this was not an issue and was unlikely to provide an explanation.

Could the trial outcome be affected by including participants with high levels of childhood trauma?

The effect of childhood trauma has been widely discussed, but has rarely been investigated as a potential modifier of ADHD medication treatment effects. In clinical practice, some practitioners view childhood trauma as a potential cause of an ADHD-like syndrome that might not then show a typical response of ADHD symptoms to medication. To investigate the possibility that the subgroup with evidence of childhood trauma are reducing the overall effect of medication in the trial, we provide outcomes for a subgroup with and a subgroup without childhood trauma, defined as moderate to severe abuse according to the CTQ at baseline. The number of participants who reported abuse in the various categories are listed in Table 24. Out of 200 participants, 41 reported emotional abuse, 47 reported physical abuse, 12 reported sexual abuse, 167 reported emotional neglect and 197 reported physical neglect.

Emotional abuse

The estimated score difference between the OROS-MPH and placebo arms in the subgroup with high levels of emotional abuse was 0.76 (95% CI –4.67 to 6.18; t = 0.03; p = 0.78). This is a small improvement in the OROS-MPH arm. The estimated score difference between the OROS-MPH and placebo arms in the subgroup without high levels of emotional abuse was 1.20 (95% CI –2.35 to 4.75; t = 0.67; p = 0.51): a small improvement in the OROS-MPH arm.

Physical abuse

The estimated score difference between the OROS-MPH and placebo arms in the subgroup with high levels of physical abuse was 1.62 (95% CI –4.82 to 8.07; t = 0.51; p = 0.61): a small improvement in the OROS-MPH arm. The estimated score difference between the OROS-MPH and placebo arms in the subgroup without high levels of physical abuse was 0.85 (95% CI –2.58 to 4.29; t = 0.49; p = 0.62): a small improvement in the OROS-MPH arm.

Sexual abuse

Owing to the extremely small numbers, this statistical analysis was not performed.

Emotional neglect

The estimated score difference between the OROS-MPH and placebo arms in the subgroup with high levels of emotional neglect was –0.22 (95% CI –3.13 to 3.57; t = 0.13; p = 0.90). This is a small improvement in the placebo arm.

Interestingly, but not significantly, for those without high levels of emotional neglect, the estimated score difference between the OROS-MPH and placebo arms was 3.71 (95% CI –3.26 to 10.69; t = 1.09; p = 0.29), which was an improvement in the OROS-MPH arm.

Physical neglect

This statistical analysis was not performed because 98.5% of the sample reported physical neglect.

Could the trial outcome be affected by including participants with comorbid disorders?

One question raised by the trial is the possibility that comorbid disorders may reflect a source of symptoms and impairments that might mimic ADHD in the young male prison population. Including cases with clearly defined ‘comorbid’ disorders might lead to reduction in the overall medication effect in the trial. To address this question, we provide subgroup data for those with and those without evidence of another mental health disorder according to the MINI for mental health disorders.

Overall comorbidity

Thirty-six participants had no comorbidity. This subgroup was estimated to have a very small improvement in the OROS-MPH arm, relative to the placebo arm, of 0.11 (95% CI –7.23 to 7.45; t = 0.03; p = 0.98). A total of 164 participants had at least one mental health comorbidity. The estimated score difference between the OROS-MPH and placebo arms for this subgroup was 0.98 (95% CI –2.38 to 4.34; t = 0.57; p = 0.57). This is an improvement in the OROS-MPH arm, but is not statistically significant and does not come close to the 5 points specified in the sample size calculation.

Anxiety

A total of 162 participants did not have anxiety. For this subgroup, the estimated score difference between the OROS-MPH and placebo arms was 0.39 (95% CI –2.82 to 3.60; t = 0.24; p = 0.81), which is a very small improvement in the OROS-MPH arm. Thirty-eight participants had an anxiety disorder. The estimated score difference between the OROS-MPH and placebo arms for this subgroup was 2.88 (95% CI –5.68 to 11.43; t = 0.69; p = 0.50). This was an improvement in the OROS-MPH arm that, although larger than that for the group without anxiety, is still not significant statistically.

Mood disorder

A total of 137 participants did not have a mood disorder. For this subgroup, the estimated score difference between the OROS-MPH and placebo arms was 0.32 (95% CI –3.53 to 4.17; t = 0.17; p = 0.87): a very small improvement in the OROS-MPH arm. Sixty-three participants had a mood disorder. The estimated score difference between the OROS-MPH and placebo arms for this subgroup was estimated to be larger, at 1.97 (95% CI –3.14 to 7.09; t = 0.77; p = 0.44). This was a small improvement in the OROS-MPH arm, but was not statistically significant.

Post-traumatic stress disorder

Thirteen participants had PTSD, eight had depersonalisation, four had derealisation and one was not assessed for PTSD. The subgroup analysis was not conducted because of the extremely small subsamples.

Antisocial personality disorder

Fifty-one participants did not have ASPD. The estimated score difference between the OROS-MPH and placebo arms in this subgroup was –0.98 (95%: CI –7.06 to 5.09; t = –0.33; p = 0.75), which was a small improvement in the placebo arm. A total of 149 participants had ASPD. The estimated score difference between the OROS-MPH and placebo arms for this subgroup was 1.40 (95% CI –2.18 to 4.98; t = 0.77; p = 0.44), which was a small improvement in the OROS-MPH arm. This was not statistically significant.

Could the trial outcome be affected by including participants with high levels of drug and alcohol usage?

Drug and alcohol use disorders are both associated with ADHD and might have an impact on the treatment response. This could arise because of brain damage as a long-term consequence of drug and alcohol use, either generating ADHD-like symptoms or reducing the potential effects of medication for ADHD. There could also be tolerance to OROS-MPH due to previous drug abuse, which only normalises after a period of time. Furthermore, a history of drug and alcohol abuse is likely to be an indicator for risk of ongoing (unmeasured) use of drugs in the prison during the trial, which might affect ADHD symptoms. To address this question, we used the AUDIT-C data to identify participants with high risk of alcohol abuse and the data from the NIDA screening tool to identify participants with high risk of drug abuse. We aimed to complete subgroup analyses for those with high risk and those with low risk of alcohol and drug abuse, to investigate whether or not including this group reduces the overall medication effects in the trial.

High risk of problem alcohol use

A total of 89 participants had potentially problematic alcohol use (defined as a score of ≥ 16 on the AUDIT-C). The estimated score difference between the OROS-MPH and placebo arms in this subgroup was 1.00 (95% CI –3.84 to 5.85; t = 0.41; p = 0.68): a small improvement in the OROS-MPH arm. For those who did not report problematic alcohol use, the estimated score difference between the OROS-MPH and placebo arms was 1.88 (95% CI –3.65 to 7.41; t = 0.68; p = 0.50): an improvement in the OROS-MPH arm.

High risk of problem drug abuse

Using the NIDA Quick Screen V1.0 criteria, we classified individuals as at low (none or minor use), medium and high risk for substance involvement (Table 25). We initially intended to apply the NIDA screening criteria for high risk of problematic drug use, namely a NIDA score of ≥ 27. However, although rates of reported drug use were high, the numbers meeting criteria for high risk of problem use were small; for this reason, these subgroup analyses were not conducted. Using the NIDA definition for high risk use, there were only:

• Three problem users of opiates (street or prescription). The minimum, median and maximum usage scores were 0, 0 and 32, respectively.
• Twelve problem users of cannabis. The minimum, median and maximum usage scores were 0, 7 and 39, respectively.
• No problem users of ‘spice’. The minimum, median and maximum usage scores were 0, 0 and 17, respectively.
• Four problem users of cocaine and methamphetamine. The minimum, median and maximum usage scores were 0, 0 and 39, respectively.

Because of the small numbers in the high-risk groups, we decided to complete an additional analysis regarding drug use to address the question of whether or not a group free of either moderate or severe use of any drug (score of ≥ 4 on the NIDA Quick Screen) showed a difference in the CAARS-O outcome at 8 weeks between the OROS-MPH and placebo arms.

A total of 73 participants had no or very low use of any of opiates, spice, cannabis or stimulants. In this subgroup, the estimated score difference between the OROS-MPH and placebo arms was –4.26 (95% CI –9.31 to 0.78; t = –1.69; p = 0.10). The placebo arm performed better than the OROS-MPH arm. For more information regarding drug use, see Appendix 6, Tables 43 and 44).

Could the trial outcome be affected by including participants with prior experience of stimulant medication, leading to biased reporting of the attention deficit hyperactivity disorder symptom response to medication?

Prior exposure to stimulant treatment for ADHD could change the reporting of symptoms because of prior experience of effects of medication on ADHD symptoms and adverse effects, leading to biased reporting. There is also the possibility that medication was stopped in these cases because of previous non-response or significant adverse effects. To investigate the possibility that these effects might reduce the overall effect of medication in the trial, we conducted subgroup analyses for a group with and a group without previous exposure to stimulant medication for ADHD.

A total of 153 participants had no previous treatment with stimulants for ADHD; one participant was not assessed. The subgroup that had not previously been treated with stimulants for ADHD had an estimated score difference between the OROS-MPH and placebo arms of 0.63 (95% CI –2.93 to 4.19; t = –0.35; p = 0.72) and the OROS-MPH arm performed better than the placebo arm. Forty-six participants had received stimulants for ADHD in the past, with the last dose taken at least 3 months previously. The estimated score difference between the OROS-MPH and placebo arms for this subgroup was 2.01 (95% CI –4.17 to 8.19; t = 0.66; p = 0.51): a small improvement in the OROS-MPH arm, which was not statistically significant.

Could the trial outcome be affected by underdosing of participants?

The optimal dose of MPH used in the treatment of adult ADHD is highly variable; hence, we adopted a titration procedure to individually titrate each participant to the optimal dose of trial medication. However, as noted in the CIAO-I open-label pilot study, this population was concerned by minor adverse effects and were titrated to relatively low average doses of the trial medication. This suggests that, in this population, the titration protocol might lead to underdosing of participants, potentially accounting for lesser effects of the drug, compared with placebo. To investigate this possibility, we compared the trial outcome for a subgroup titrated to a low dose with that of a subgroup titrated to a high dose, defined as subgroups titrated to one or two capsules for the low dose, and those titrated to three or four capsules for the high dose.

Table 26 shows the distribution of the prescribed dose at the start of week 5 and at the start of week 6 by trial arm (see Could the trial outcome be affected by adherence to trial medication? for an explanation of week-6 prescriptions) for those who were ongoing in the trial. As noted in the primary analyses, those allocated to OROS-MPH were titrated to lower doses than those in the placebo arm. For example, at the start of week 5, 74% were prescribed the higher dose (three or four capsules) in the OROS-MPH arm, compared with 86% in the placebo arm.

At the start of week 5, 143 participants were prescribed three or four capsules. The estimated score difference between the OROS-MPH and placebo arms in this subgroup was 0.47 (95% CI –3.03 to 3.97; t = –0.27; p = 0.79), which was an improvement in the OROS-MPH arm. Fifty-seven participants were prescribed one or two capsules at the start of week 5. The estimated difference between the OROS-MPH and placebo arms in this subgroup was 0.68 (95% CI –6.15 to 7.51; t = 0.20; p = 0.84), which is a small improvement in the OROS-MPH arm, but is not statistically significant.

Could the trial outcome be affected by adherence to trial medication?

Adherence to taking the trial medication on a daily basis was highly variable; this might have affected the assessment of efficacy by ITT. The high levels of poor adherence to medication was previously documented in the CIAO-I open-label pilot study, in which 40 out of 121 (33.9%) participants dropped out of treatment before the 8-week assessment; therefore, efforts were taken in the current trial to minimise this.29 However, the per-protocol analysis, carried out according to the SAP-defined ‘compliance’ for the trial medication (i.e. having taken medication on 75% of the days during the trial period), did not show any different effect from the ITT analysis, suggesting that poor compliance is unlikely to explain the trial findings. Nevertheless, it is possible that our definition of compliance is not adequate to rule out all effects of adherence to the trial medication on the trial outcome. To address this problem, we investigated the impact of a set of alternative definitions of adherence with the trial medication on efficacy in this trial.

Summary of adherence to medication data

To provide an overview of the amount of drugs prescribed and taken, we produced descriptors of the dose prescribed and dose taken. This is already reported as part of the main results (see Chapter 4, Titration and prescribing of trial medication) for those ongoing on medication and uses the prescription data from the start of week 5 of the trial.

The protocol also allowed a final dose change at the start of week 6, prior to the final 3-week maintenance phase of the trial. Note that for weeks 1–5, the prescription was explicitly recorded in the database. For week 6, the prescription was not recorded, but is inferred from the maximum dosage taken by participants. In all cases, the nurses providing the prescriptions recorded a daily log of the dose taken. It is therefore reasonable to infer a prescription change for week 6 from the maximum dose taken on days 36–39, which is in line with the protocol for week-6 prescriptions. Using this approach, we describe the number of participants prescribed different doses of trial medication from the start of week 6 (see Table 26), which was maintained until the end of the trial. Table 27 shows the numbers recorded as prescriptions for weeks 1–5 and the inferred prescription for week 6.

To describe the overall adherence to prescribed trial medication, we describe the mean number of daily capsules taken by week and percentage of prescribed dosage taken (Table 28). Prescriptions for weeks 1–5 are based on the recorded prescription. The prescription for week 6 is inferred from the maximum number of capsules taken. The prescription for weeks 7 and 8 is identical to that of week 6. Note that, by the final week, the data are skewed; however, it is the percentage of medication taken of that prescribed (documented or inferred) that is of interest. Throughout the trial, the placebo group took a larger percentage of the capsules prescribed. By week 4, this difference had become more marked and remained more marked for the remaining weeks.

Figure 6 illustrates the mean adherence of participants by trial arm. The weekly adherence is defined as the number of days on which at least one tablet was taken, divided by seven. At the start of the trial, the mean adherence was 71.6% and 81.5% in the OROS-MPH and placebo arms, respectively. This dropped to 50.6% and 70.4% in week 5, and to 43.7% and 56.6% by week 8 for the OROS-MPH and placebo arms, respectively. The proportion of participants in the placebo group adhering was consistently larger throughout the trial, but adherence dropped off consistently throughout the trial in both arms.

FIGURE 6 Mean adherence to prescribed trial medication. (Figure not reproduced here; see the original article.)

Effects of poor adherence

Adherence to trial medication might have been higher during the titration phase of the trial (when participants were visited every week) than during the maintenance phase. To investigate this, we conducted a per-protocol analysis of the effects of OROS-MPH, compared with placebo, on the CAARS-O outcome at week 5. Good compliance during this period was defined as those who had taken prescribed trial medication on 75% of the days before the end of week 5.

According to this definition, for the 88 participants who had complied with their allocated drug by week 5 (i.e. had taken at least one tablet on 75% of the days up to and including day 35), there was a between-arm difference in the CAARS-O score of –1.32 (95% CI –6.14 to 3.50; t = –0.54; p = 0.59), which is an improvement in the placebo arm. The mean and SD by trial arm in this group is shown in Table 29.

Adherence to trial medication immediately prior to the outcome assessments might also affect the trial outcome, even if the active medication was significantly reducing ADHD symptoms on days when they did take the medication. This could occur because participants might report accurately on their level of ADHD symptoms on the day or the few days prior to the outcome assessment only. The short-term effects of OROS-MPH on reducing ADHD symptoms (for only 6–10 hours) means that participants in the active medication group might report high levels of symptoms if they had not taken medication in the period just prior to the outcome assessment, even among those who showed an effect on days when they did take the medication. To investigate the immediate effects of the trial medication, we conducted further per-protocol analyses on the week-5 and week-8 CAARS-O data. Adherence was defined as taking trial medication on the 2 days prior to the outcome assessment at week 5 and week 8, with those not taking medication on those days excluded from the analysis.

For week 5, 97 participants had taken one or more capsules on both days immediately preceding the week-5 visit. In this subgroup, there was an estimated difference between the OROS-MPH and placebo arms of 2.30 (95% CI –2.28 to 6.88; t = 1.00; p = 0.32), which was an improvement in the OROS-MPH arm. For week 8, in the subgroup of 84 participants (one of whom did not provide an outcome) who had taken at least one capsule on both of the days preceding the week-8 visit, the estimated difference between the OROS-MPH and placebo arms was –0.22 (95% CI –4.87 to 4.44; t = –0.09; p = 0.93), which was a small improvement in the placebo arm.

Could adverse effects be driving adherence to the trial medication?

To investigate the potential association of adverse effects on adherence to trial medication, we investigated the proportion of prescribed medication taken by the presence or absence of the most commonly reported adverse effects of OROS-MPH, compared with placebo. We used the AES data from across the trial. As can be seen in Table 17, the common items from the AES showing the greatest difference between the two trial arms were sweating, headache, appetite loss and dry mouth. These were reported by 16%, 18%, 16% and 30% of participants, respectively, and 50% of participants reported one or more of these items.

For this analysis, we investigated the adherence to medication for each of these four AES items, as follows: (1) we measured adherence as the proportion of days on which any dose was taken divided by number of days medication was prescribed; (2) we calculated the mean proportion for groups reporting the presence or absence of each of the four AES items, and ran t-tests to compare the present with the absent groups; and (3) we considered an overall AES item measure of having at least one of the four items.

In Table 30, the mean adherence is shown by the group for which the AES items were present and the group for which they were not present. A difference in mean proportion of adherence by symptoms was tested using a t-test. There was no evidence to suggest that the adherence varied between those who did and those who did not display the most common adverse effects.