Results

Participant flow

The Consolidated Standards of Reporting Trials (CONSORT) diagram illustrates the recruitment process (Figure 2). A total of 200 participants were randomised to treatment with OROS-MPH or placebo.

FIGURE 2 The CONSORT diagram. a, Two participants in the placebo arm were transferred to an accessible prison and the outcomes were collected from those persons. Outcomes not collected on other participants labeled as transferred, deported or released. Reproduced with permission from Asherson et al.61 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure. (Figure not reproduced here; see the original article.)

During trial recruitment, all male prisoners aged 16–25 years in HMP YOI Isis (London, Oxleas NHS Foundation Trust) and HM YOI Polmont (Scotland, NHS Forth Valley) were invited to be screened for ADHD. It was estimated that 1368 prisoners aged 16–25 years were asked to consent to screening for ADHD. Of these, around 185 (13.5%) declined to take part in the screening step and did not provide consent that would have allowed us to record this information. It is unknown whether or not those declining consent for screening differ from those participants who consented to be screened.

The CONSORT diagram shows that, of the 1183 prisoners who consented to be screened, 585 screened negative on the Barkley ADHD self-rating scale and a further 52 were excluded who were identified at this stage as not meeting eligibility criteria for the trial. The exclusions included 38 prisoners who were at high risk of early transfer or release during the trial period, one who posed a serious risk of violence towards researchers, one who was not an English speaker, five who were currently prescribed ADHD medication, four who had current major depression and one who had contraindications to the use of stimulants. Another individual was excluded due to low BMI (a clinical decision) and one was mistakenly excluded who was thought to be outside of the age range for the trial, despite being aged 25 years at the time of consent. This left 546 participants who were considered for a diagnostic assessment. Of these, 28 participants were excluded because they were thought to be at high risk of transfer or early release and 518 were invited for an ADHD assessment using the DIVA 2.0. Of these, 86 did not attend the assessment and 432 completed the DIVA 2.0 assessment. Of the 432 prisoners who completed DIVA 2.0 assessments, 279 were found to meet diagnostic criteria for DSM-5 ADHD, of which six did not meet one of the other eligibility criteria and were excluded. This left 273 prisoners who were invited to sign consent for the trial, of which 54 declined their consent. This left 219 prisoners who signed consent for the trial and were formally checked for eligibility. Of the 219 prisoners, three were no longer willing to participate, two were assessed after the trial had completed recruitment and 14 failed to meet eligibility criteria, leaving 200 participants who were randomised.

Post-randomisation losses to trial

At the primary outcome time point of 8 weeks, follow-up rates were good, with 90 and 94 participants completing final outcome measures in the OROS-MPH and placebo arms, respectively. Of these, 26 (OROS-MPH arm, n = 19; placebo arm, n = 7) had stopped medication, but provided 8-week outcome data, because they refused to take further medication during the trial (n = 24) or because they moved to an accessible prison where it was not possible to provide trial medication (n = 2). The other 16 participants had withdrawn from the trial (OROS-MPH arm, n = 4; placebo arm, n = 2); withdrawn from all aspects of the trial, but allowed prison record data to be used (OROS-MPH arm, n = 1); or had been released, deported or transferred to an inaccessible prison (OROS-MPH arm, n = 6; placebo arm, n = 3).

Eligibility violations

One participant was randomised despite scoring 51 on the WASI-II assessment for IQ. However, the clinical judgement was that the participant did in fact meet the trial criterion of IQ score of > 60, although the test result did not reflect this. Nevertheless, this was recorded as an eligibility violation on the basis of the test result. The participant was included in the target population because the low score was thought to reflect inattention on the day of the WASI-II assessment; WASI-II is not validated for people with ADHD and the eligibility criterion of 60 was within the 95% CI for the test result. This event led to a change in the protocol that clarified that the IQ assessment for the purpose of eligibility would be based on clinical judgement in addition to review of the WASI-II results owing to the limitations of the WASI-II test. A list of all protocol deviations is reported in Appendix 2, Table 32.

Baseline characteristics

Tables 3–6 summarise the baseline (pre-randomisation) characteristics of this sample. Site-specific baseline data are reported in Appendix 3, Tables 33–36, for HMP YOI Isis and in Appendix 4, Tables 37–40, for HM YOI Polmont. Table 3 summarises categorical demographic variables for each trial arm and overall. Table 4 summarises continuous baseline variables, including age, IQ, height and BMI; baseline values for clinical measures; and summaries of putative baseline moderators. Table 5 summarises coexisting mental health disorders defined by the MINI 7.0.1 assessment. Table 6 summarises prison behavioural reports from prison and educational staff, and the prison records. As expected from randomisation, these tables show that the two trial arms were well balanced with regard to clinical and demographic variables.

Participants had a mean age of 20.7 years at randomisation. A total of 62.5% were of white ethnicity and 37.5% were of black and minority ethnic backgrounds, which was representative of the prison population (see Chapter 6, Generalisability). General cognitive ability was below general populations norms, with an estimated mean IQ of 89.4 (SD 13.0). Regarding education, 39.5% had no qualifications and most participants left school before the age of 16 years or refused to give an age, with 39% leaving school aged ≥ 16 years. They were mostly unemployed: 66.5% unemployed, and 33.5% employed, self-employed or students. A large proportion of participants (76.5%) had not previously received ADHD medication.

Regarding mental health, participants in both trial arms were moderately ill according to the CGI severity scale, with a mean score of 3.9 (SD 1.0). Criteria for coexisting ASPD was met by 149 (74.5%) participants. Potential problem alcohol use was reported for 149 (74.5%) participants. Illicit drug use was reported by 194 (97%) of participants; however, in additional analyses, we found that around half of the sample met the criteria for high-risk problem use of alcohol, whereas few met the criteria for high risk of substance involvement, according to their self-reports (see Table 25).

Other mental health disorders recorded on the MINI were less frequent, including 38 cases with any type of anxiety disorder and 38% with any type of mood disorder. A total of 149 (74.5%) participants met diagnostic criteria for ASPD, but only 15 (7.5%) met criteria for BPD. High levels of childhood trauma were reported (see Table 24 for more details); however, only 13 (6.5%) participants met formal criteria for PTSD.

The mean baseline score for the primary outcome variable, the CAARS-O, was 36.8 (SD 9.2). Participants scored equally high in the inattentive and hyperactive symptom domains, with average symptom scores on the CAARS-O of 18.2 and 18.6, respectively.

Adherence to trial protocol and medication

Trial withdrawal

In this trial, we distinguished between formal withdrawal from the trial (withdrawal from all aspects of the trial); refusal to take further medication, but completion of outcomes; refusal to take further medication with partial completion of outcomes; transfer out of the prison with partial completion of outcomes; and transfer from the prison with no collection of outcomes. This information is summarised in the CONSORT diagram (see Figure 2).

Six participants formally withdrew (OROS-MPH arm, n = 4; placebo arm, n = 2). Four of these six participants withdrew by week 5 and two withdrew by week 8. The reason given by five of the participants was that they no longer wished to take part and one withdrew because the ‘medication didn’t work’.

Between randomisation and the start of week 5, 15 participants (OROS-MPH arm, n = 13; placebo arm, n = 2) had refused to take the trial medication, but were still providing outcome data, and four participants (OROS-MPH arm, n = 3; placebo arm, n = 1) had formally withdrawn. During this period, a further four participants (OROS-MPH arm, n = 3; placebo arm, n = 1) left the prison and were unable to provide further outcome data because they were released, transferred to an inaccessible prison or deported.

Between the start of week 5 and the end of week 8, a further nine participants (OROS-MPH arm, n = 6; placebo arm, n = 3) refused to take the trial medication, but continued to provide the primary outcome. In addition, two participants in the placebo arm had transferred to an accessible prison and were able to provide week-8 outcome data, but it was not possible to provide the trial medication once they had left the prisons taking part in the trial. Over the same period, two participants (OROS-MPH arm, n = 1; placebo arm, n = 1) formally withdrew, and one participant (in the OROS-MPH arm) allowed their prison records to be accessed, but did not provide primary outcome data. A further five participants (OROS-MPH arm, n = 3; placebo arm, n = 2) left the prison during this period because they were released, transferred to an inaccessible prison or deported.

Titration and prescribing of trial medication

All participants started on one capsule per day and were titrated weekly up to a maximum total of four capsules per day. A stable dose was prescribed for the final 3 weeks. The daily prescribed dose was recorded at the start of weeks 1, 2, 3, 4 and 5. From week 5 onwards, the dosage was stable for all participants, apart from 13 participants who had a dose increase at the start of week 6.

Participants were titrated by increasing the dose by one capsule per day up to a maximum of four capsules unless the ADHD symptoms reduced to negligible, there were limiting adverse effects or a participant objected. By the start of week 5, participants had been titrated to an average of 2.99 and 3.41 daily capsules in the OROS-MPH and placebo arms, respectively (Table 7). Concomitant medications were allowed during the trial and are listed in Appendix 6.

The dose of prescribed trial medication taken was recorded by the prison nurse on a daily basis (see Table 7). Participants received the medication directly from nursing staff each day, who then observed participants taking the medication. Although prescribed for daily use, participants did not always take the trial medication every day, which required them to attend the nursing station where the trial medication was dispensed. However, on the days when they did attend to take the trial medication, they took the full dose prescribed. By week 5, participants were taking an average of 1.60 and 2.40 capsules in the OROS-MPH and placebo arms, respectively. This dropped to 1.41 and 1.99 capsules per day for the OROS-MPH and placebo arms, respectively, by week 8. Participants in the OROS-MPH arm were less likely to take the medication each day (see Table 7) (Figure 3).

FIGURE 3 Weekly mean number of capsules taken by trial arm. (Figure not reproduced here; see the original article.)

We defined compliance with the treatment regimen as taking one or more prescribed capsules on any given day in the trial, irrespective of the dose prescribed. Full compliance would mean taking one or more capsules on all 56 days, as this was the intended treatment regimen. By the end of week 4, 177 participants (82 in the OROS-MPH arm and 95 in the placebo arm) continued on treatment; by the end of week 8, 158 participants (71 in the OROS-MPH arm and 87 in the placebo arm) continued on treatment. The status of participants (ongoing; withdrawn from treatment, but not from the trial; withdrawn from trial; and released, deported or transferred) was recorded at weeks 1–5 and 8.

Table 7 reports on the number of capsules prescribed and the number of capsules taken each day for those whose status was reported as ‘ongoing’ at each trial visit. For this report, it was assumed that all participants who were ‘ongoing’ at the end of week 5 continued to take the same dose for the remainder of the trial. Of the 56 days of the trial, the mean number of days on which participants took treatment was 31.5 (SD 16.5) and 39.5 (SD 13.3) in the OROS-MPH and placebo arms, respectively, which is 56.3% and 70.5%, respectively, of the 56 days of the trial.

The assumption that the dose prescribed at the end of week 5 was continued to the end of the trial was made because a variable with prescribing information was included in the trial database at the start of weeks 1–5, but not at the start of week 6. In Chapter 5, we report additional analyses that account for 13 participants who increased the number of capsules at the start of week 6, estimated from the reports of dose taken each day throughout the trial (see Chapter 5, Could the trial outcome be affected by adherence to trial medication?).

Treatment arm prediction

Participants and researchers were asked at week 8 to which treatment arm they believed they had been assigned. In the OROS-MPH arm, 89 researchers and 88 participants reported their guess; in the placebo arm, 93 researchers and 92 participants reported their guess. In the OROS-MPH arm, 44 participants (50%) guessed correctly, and 47 researchers (52.8%) guessed correctly. In the placebo arm, 72 participants (78.3%) guessed correctly, and 68 researchers (73.1%) guessed correctly. Using a chi-squared test to compare the proportion of participants guessing that they were in the OROS-MPH arm between the trial arms, we found that the probability of guessing that a participant was assigned to the OROS-MPH arm was greater when a participant was randomised to OROS-MPH than when they were randomised to control for both participants’ (p = 0.001) and researchers’ guesses (p = 0.001). This may suggest that blinding was not adequately maintained.

Analysis of primary and secondary outcomes

Handling of missing data

Follow-up rates were high, so results should not be heavily affected by assumptions regarding the process of generating missing data. At week 8, 184 out of 200 participants completed the primary outcome, leaving 16 missing values. We found that withdrawal from treatment (p < 0.001) and employment status (p = 0.08) predicted missing primary outcome at 8 weeks. Thus, these two variables were included as predictors of missing values in the imputation step of our MI procedure (see Chapter 2, Need for multiple imputation). All results reported here are based on MI with k = 100 imputations to avoid missing data biases.

Outcome measures

Outcome measures are summarised by assessment time point and trial arm in Tables 8 and 9. Results of formal trial arm comparisons are provided in Tables 10–12. Figure 4 provides a graphical display of mean CAARS-O scores by trial arm. Table 8 and Figure 4 show that improvements over time were observed for the primary outcome, CAARS-O score, in both the OROS-MPH and placebo trial arms.

FIGURE 4 The CAARS-O mean scores and 95% CIs by trial arm. (Figure not reproduced here; see the original article.)

Primary outcome measure: CAARS-O score

The estimated score difference between the OROS-MPH and placebo arms for CAARS-O at 8 weeks was a reduction of 0.57 points (95% CI –2.41 to 3.56 points). This was an improvement in the OROS-MPH arm, but the effect on the CAARS-O score when standardised was very small, at 0.06 points. The difference was not statistically significant (see Table 10) and was smaller than the difference that the trial was powered to detect (minimum clinically important difference = 5 points, as per our sample size calculation).

Table 11 reports the change in CAARS-O scores relative to baseline for each subscale and the responder rate. To investigate the responder rate, we applied the operational definition of a responder applied by NICE14 of a 20% reduction in the baseline CAARS-O score. The percentage of responders was 48.3% for the OROS-MPH arm and 47.9% for the placebo arm.

Continuous secondary outcome measures

Small improvements between the active and placebo arms were seen for WRAADDS, MEWS, MVQ, BSI and CGI (therapeutic effects), but deterioration was seen for ARI-S and CORE-OM. However, none of the secondary outcomes showed statistically significant differences between the OROS-MPH and placebo arms (see Table 10). The continuous secondary outcomes had no defined clinically important score differences; however, in all cases, the standardised difference was very small (SMD < 0.2).

Education

At baseline, 187 out of 200 participants had some form of education scheduled. Only seven participants in the OROS-MPH arm and six participants in the placebo arm were not in any kind of education. The median number of sessions scheduled was 21 in the OROS-MPH arm and 25 in the placebo arm, and the median number of sessions attended was 15 and 19, respectively. The mean proportion of sessions attended in the 8 weeks prior to randomisation was 0.78 (SD 0.31) and 0.84 (SD 0.26) in the OROS-MPH and placebo arms, respectively. However, reports from education staff about behaviour in education sessions were difficult to obtain, with only 83 MOAS-E and 67 BRC-E forms completed at baseline.

The picture was similar at 8 weeks. Only seven participants in the OROS-MPH arm and two participants in the placebo arm were not in any kind of education, and the median number of sessions scheduled across the 8 weeks of the trial was 34 and 32 in the OROS-MPH and placebo arms, respectively. Of the sessions scheduled, the median number of sessions attended was 23 in the OROS-MPH arm and 22 in the placebo arm, giving mean proportions attended of 0.80 (SD 0.28) and 0.82 (SD 0.26), respectively.

Owing to the extreme lack of participants without any sessions scheduled, it was not possible to analyse education scheduled using logistic regression. As it was recorded as a count of sessions with noticeable skew, it was analysed as a negative binomial. The estimated incident rate ratio between the OROS-MPH and placebo arms was 0.98 (95% CI 0.84 to 1.14; p = 0.78), which is a very small and non-significant decrease in sessions scheduled in the active arm (see Table 12).

At baseline, the median MOAS-E score was 0 in both trial arms, with maximum scores of 12 and 10 in the OROS-MPH and placebo arms, respectively. At week 8, the median MOAS-E score was still 0 in both trial arms, with maximum scores of 1 and 2 in the OROS-MPH and placebo arms, respectively. The BRC-E median scores were 18 and 18 at baseline and 17 and 18 at week 8 in OROS-MPH and placebo arms, respectively.

Behavioural reports from prison officers

Reports of behaviour by prison officers are reported in Table 9. BRC-P and MOAS-P are prison officer reports of behaviour completed at baseline. Of 200 participants, 50 (25%) had the minimum score of 6 for the BRC-P at baseline and 143 out of 200 (71%) participants had a score of 0 for the MOAS-P at baseline. This means that they had zero incidents of aggression reported in the week before baseline.

For the BRC-P, the median score at baseline was 8 in both arms. At week 8, the median BRC-P score was 9 for the OROS-MPH arm and 8 for the placebo arm. The BRC-P was formally analysed using a negative binomial model, which estimated the incident rate ratio between the OROS-MPH and placebo arms as 0.95 (95% CI 0.85 to 1.06; p = 0.33) (see Table 12).

The median MOAS-P score at baseline was 0 in both arms, with 75 (74.3%) of the OROS-MPH arm participants and 68 (68.7%) of the placebo arm participants having zero incidents. At week 8, the median MOAS-P score was still 0 in both arms, with the majority (123/179) of participants having a score of 0, with a slightly larger percentage (68.7%) in the placebo arm than in the OROS-MPH arm (64.8%). Owing to the small size of the subpopulation with recorded numbers of events, this outcome variable was dichotomised (1 = yes, any aggressive events seen; 0 = no, no aggressive events) and analysed using logistic regression. At 8 weeks, the odds ratio between trial arms of any aggressive events as measured using MOAS-P was 0.57 (95% CI 0.28 to 1.15; p = 0.12), implying a non-significant improvement in the placebo arm (see Table 12). This surprisingly low odds ratio estimate is due to adjusting for the baseline level of events in the analysis model, which appears to be the driver of the aggressive events (MOAS-P) at week 8.

Critical incidents were prison records of negative behaviours noted in the 8 weeks immediately prior to baseline and before the final time point of 8 weeks. These records were partial at baseline for 11 participants who had not been in prison for 56 days prior to randomisation, and at week 8 for 16 participants who left the prison prior to the week-8 assessment. At baseline, this outcome was observed in all 200 participants; 125 (62.5%) participants had no negative behaviours reported. The number of reports at baseline of negative behaviours ranged from 0–10, with a median of 0 (see Table 9). Trial arms were very similar at baseline, with a median of 0 reported for both the OROS-MPH and placebo arms. By week 8, the trial arms were still very similar, with a median of 0 for both arms. The estimated incident rate ratio in critical incidents between the OROS-MPH and placebo arms at 8 weeks was 0.75 (95% CI –0.45 to 1.25; p = 0.28) (see Table 12).

Sensitivity analyses

Four sets of sensitivity analyses were conducted to check whether or not the results on the primary outcome were sensitive to collection of outcomes within the specified window, starting treatment within the specified time, sufficiently high IQ confirmed by a standardised test, or sufficient level of compliance with the treatment. In each case, the primary analysis, estimating the difference between trial arms in CAARS-O scores at 8 weeks, was repeated. The primary analysis was robust to all assumptions tested in these sensitivity analyses.

Sensitivity analysis 1: outcome window

As per protocol, the primary outcome time was at 8 weeks or 56 days, and not > 2 days before or after the expected date. Six participants had the primary outcome completed at > 58 days and 18 participants had the primary outcome completed before 54 days. Of 176 participants who completed the final outcomes on time, the estimated difference was 0.73 (95% CI –2.35 to 3.81; p = 0.64), which does not differ appreciably from that of the primary analysis.

Sensitivity analysis 2: acceptable treatment start time

Participants, once randomised, should have been prescribed a first dose and initiated treatment within no more than 7 days of randomisation. Fourteen of the 200 participants did not start treatment within the defined time. Of 186 participants who started treatment on time, the estimated difference was 0.65 (95% CI –2.47 to 3.78; p = 0.68), which does not differ appreciably from that of the primary analysis.

Sensitivity analysis 3: intelligence quotient

Attention deficit hyperactivity disorder symptoms can make IQ tests less reliable. Clinical judgement was used to assess IQ in addition to the use of the WASI-II IQ test. One participant was measured as having an IQ of < 60 on the test. This sensitivity analysis confirms that the result is valid if the sample is restricted to those who met the eligibility criterion relating to IQ test. The estimated difference was 0.51 (95% CI –2.51 to 3.53; p = 0.74).

Sensitivity analysis 4: compliance with treatment

A total of 83 participants complied with trial medication according to our definition of compliance of taking some or all of their prescribed trial medication on at least 75% of the days for which it was prescribed. In the OROS-MPH arm, 34 participants complied, compared with 49 in the placebo arm. The estimated trial arm difference in the compliers (–0.29) was a small difference in the opposite direction (a greater effect in the placebo arm), but was not statistically significant (95% CI –5.09 to 4.52; p = 0.91).

Moderator analyses

The following baseline variables were tested as putative moderators of the treatment effect: ZAN-BPD, CTQ, RPQ-Reactive and RPQ-Proactive. A term representing the interaction between trial arm and each potential moderator was included in the primary analysis model in addition to the moderator. No moderator had any significant effect either on its own or through interaction with trial arm (Table 13).

Mediation analyses

We analysed the individual mediating effects of CAARS-O hyperactivity/impulsivity subscores, CAARS-O inattention subscores and WRAADDS emotional dysregulation, measured at 5 weeks, on BRC-P scores and the number of critical incidents measured at 8 weeks. The natural direct effect, natural indirect effect and total effect are shown in Table 14. The mediation observed was negligible (around zero) and no mediation test was statistically significant. All mediator effects were in the same direction as the total effect. The CIs shown are derived by bootstrapping (1000 replicates), with zero included in the respective interval, implying that an effect was not statistically significant.

Serious adverse events

One SAE took place during the trial. The SAE was classified as an IME after a participant tied a cord around his neck as a protest to a prison officer’s treatment. It was concluded that the participant did not intend to do any harm to himself, he did not require any treatment after the event, and it was not thought to be related to the trial medication. The blind was kept and the prisoner continued in the trial.

Adverse events

The numbers of AEs in a category are reported by person, rather than by number of events (Table 15). Of 336 AEs reported, 184 occurred in the OROS-MPH arm and 152 in the placebo arm. The numbers of participants reporting AEs were broadly similar across the two trial arms, with the exception of the High-Level Group Term categories for ‘appetite and general nutritional disorders’, ‘depressed mood disorders and disturbances’ and ‘dizziness’. This reflected appetite loss (13 participants in the OROS-MPH arm vs. 2 in the placebo arm), depressive symptoms (12 in the OROS-MPH arm vs. 4 in the placebo arm) and reports of dizziness (6 in the OROS-MPH arm vs. 0 in the placebo arm).

Expected adverse effects were also followed-up systematically at each visit using the medication AES (Tables 16 and 17). Question 23 in the AES, ‘Feeling worse or different when the medication wears off (rebound)’, was not applicable, so we report on only 22 items. Each item on this scale is rated as follows: 0 = not at all, 1 = sometimes, 2 = often and 3 = all the time. The scale was dichotomised 0 (not at all or sometimes) and 1 (often or all the time) to reflect the absence or presence of the AE, respectively. The mean and SD of the total AES scores show that there was only a minor difference in total burden of AEs at 8 weeks in the OROS-MPH arm [mean 7.4 (SD 7.5)], compared with the placebo arm [mean 5.5 (SD 5.4)]. At baseline, these values were slightly larger [mean 10.3 (SD 7.5) for the OROS-MPH arm; mean 9.8 (SD 7.3) for the placebo arm], indicating that some reported AEs may reflect phenomena of the clinical condition that improved during the 8-week period of the trial.

The numbers and percentages of participants reporting each individual symptom are reported at baseline (see Table 16) and across the entire post-randomisation period between randomisation and week 8 (see Table 17). This shows that the most common adverse effects in the post-randomisation period that were related to the use of OROS-MPH, compared with the use of placebo, were headache (17.8% vs. 10.1%, respectively), dry mouth (19.8% vs. 10.1%, respectively), sweating (19.8% vs. 8.1%, respectively) and appetite reduction (34.7% vs. 19.2%, respectively).

Finally, blood pressure and heart rate were recorded at baseline, at each titration point during weeks 1–5 and at week 8. BMI (weight/height) was recorded at baseline, week 5 and week 8. The values overall and by trial arm are shown in Table 18. There was no noticeable difference between trial arms at any point in the trial.

For a list of all concomitant medication, see Appendix 5, Tables 41 and 42.