Conflict of interest statement: David Foreman received £20,000 from Eli Lilly to fund the evaluation of a nurse-led ADHD clinic.
The National Health Service in England has deployed guidance from the National Institute of Clinical Excellence (NICE) to assist practitioners in the diagnosis and treatment of Attention Deficit Hyperactivity Disorder (ADHD) but, though the number of prescriptions has risen since its introduction, the impact of the guidance on prescribing practice has not been studied. Clinic records of all open ADHD cases (296) in three English Child and Adolescent Mental Health Services were examined. The time from referral to either prescription or data collection was extracted for a survival analysis. It was hypothesised that NICE guidance, clinic, patient and referral characteristics would all influence the speed and likelihood of prescription. Following the introduction of NICE guidance, the median time to start prescribing medication fell from 1262 to 526 days: the minimum realistic time to complete a routine assessment was approximately 70 days. Overall, 70% were prescribed medication. Most of the wait was after face-to-face appointments at the clinic had been initiated. Waiting times differed between clinics and shorter waits were likely for older children and those referred from an educational source. While the introduction of NICE guidance has increased the rate of prescription, the time taken before prescription suggests that the tendency in England is still to postpone treatment by medication. The reasons for this require further research.
• The introduction of NICE guidance has halved the average time taken for a child with ADHD to receive a prescription for medication to around 526 days after referral, with a minimum realistic routine assessment time of 70 days
• Five hundred and twenty-six days is longer than can be accounted for either by waiting time or prior trials of psychological therapy
• Educational referral, the child being older, and greater clinic resourcing also shorten time from referral to treatment
Hyperactivity, variously known as Attention Deficit Hyperactivity Disorder (ADHD) or Hyperkinesis (HKS) is the most common neuropsychiatric disorder of childhood, affecting approximately 1-2% of children in the UK (Meltzer et al., 2003). Current UK guidelines (NICE, 2000) require diagnosis in secondary care (CAMHS, or paediatric services with appropriately skilled staff) with ADHD representing on average, 12% of CAMHS activity in the UK (Durham University, Department of Health, 2007) and accounting for approximately half of open CAMHS clinic cases (see below). There is, however, no evidence of over-diagnosis or inappropriate prescription of medication (Green et al., 2005), though this can occur when diagnosis is undertaken in primary care (Angold et al., 2000). Inappropriate prescription of stimulants continues to cause community disquiet irrespective of local relevance (Cameron, 2007; Howe, 2003; Mills, 1998) though medication remains, at least in the short term, the most cost-effective therapeutic intervention for ADHD, either alone or in combination with behaviour therapy (Foster et al., 2007). The secondary care assessment recommended by NICE is time-consuming and lengthy, requiring skilled professionals and multi-informant reports. However, while reports of increase in the worldwide use of medication since the 1990s abound (Schlander, 2007) the impact of this guidance on access to medication for children diagnosed with ADHD has never been studied. One impact of such guidance could be to increase access to treatment by increasing professional awareness and structuring clinic responses, but it is also possible that the lengthy assessment required could delay or reduce access, by imposing queues. To examine this issue, the aim of this study was to conduct a survival analysis of the time between referral to a CAMHS clinic and prescription of medication during a period when the NICE guidance was introduced. It was hypothesised that NICE guidance would decrease the time taken to receive a prescription for medication but that clinic, referral and patient variables would also affect this process.
The sample comprised all open cases of ADHD (n =296) in three community CAMHS clinics serving a population of 85,000 children in East Berkshire, UK. Paediatric services did not offer a separate ADHD service locally. Clinicians’ identification of ADHD was checked in a 2-stage design implemented as an audit of clinic accuracy against a national standard. All clinicians’ diagnoses of the three clinics’ open cases (n =500) were first compared with the diagnostic prediction of a standardised screening questionnaire for ADHD, the Strengths and Difficulties Questionnaire (SDQ) (Goodman & Scott, 1999). Cases that were discordant were further assessed using a standardised, validated assessment for common psychiatric disorders, the Development and Well-Being Assessment (DAWBA) (Goodman et al., 2000). The sample included cases concordant for both clinician and SDQ predicted diagnosis (288), those with DAWBA diagnoses without clinician concordance (6), and two cases (not included in the audit to avoid bias) that were concordant for SDQ and clinician diagnosis, but had, mistakenly, also had a DAWBA administered. ADHD thus comprised 296/500 (59%) of all open cases in the three services. Study ethics were managed under UK National Health Service audit protocols, following advice received from the Local Research Ethics Committee.
The main dependent variable was the number of days between referral and first receiving medication from the clinic, derived from clinic records. Other data abstracted from records were selected on the basis of their likely robustness for statistical hypothesis testing. These included the date and source of the initial referral; the clinic where the patient was seen; the patient’s gender and age at referral; the postcode-derived Index of Multiple Deprivation (Office of the Deputy Prime Minister, 2004) a standardised summary measure of environmental deprivation based on national census data covering income, employment, health and disability, education and skills, barriers to housing and services, living environment and crime; family composition; SDQ diagnostic predictions of the likelihood of concurrent emotional and conduct disorders; and whether referral had taken place before or after the date of publication of NICE guidance. Attempts were made to identify waiting times prior to the start of the secondary care assessment. However, it was found that the case records indicated that much assessment activity took place during the time waiting for a first clinic appointment (obtaining questionnaires, school reports or attending multi-agency meetings) and that some early appointments were for triage purposes, with the patients being returned to wait further once an immediate crisis was assessed or defused. Thus, there was no clear division between assessment activity before the first clinic appointment, and assessment activity taking place after it.
The retrospective case-note design was conceptualised as a survival analysis, with the date of referral as each case’s time of entry into the study, and the date of first prescription of medication as the event of interest. If no prescription was given at the time the case details were recorded, the case was considered censored, with the date of data collection then being the end-point. Full details of the analysis are available from the author on request. Briefly, it entailed: calculating the study’s power; using the Cox proportional hazards model to identify significant predictors; compensating for the effect of unmeasured changes which could influence the likelihood of medication by including date of referral as a frailty term; modelling the variable impact of age on likelihood of medication by using a spline transformation; removing irrelevant variables by backwards stepwise selection; finally checking the resulting model’s assumptions and predictive validity by direct testing and by graphical superimposition of the model’s predicted survival curve upon the empirical (Kaplan-Meier) one. As no variable had more than 6% missing values, median (modal for category) imputation was used, following Harrell (2001). All analyses were conducted within the R statistical environment (R Development Core Team, 2007) using the packages ‘asypow’ for the power calculations, ‘survival’ and ‘MASS’ for the Cox proportional hazards regression and Kaplan-Meier plots.
The characteristics of the sample are set out in Table 1. Though the sample is geographically localised, the national mean and standard deviation of the Index of Multiple Deprivation is 18.9, and 9.2 respectively, suggesting that the study sample is nonetheless socially typical for the country at large, including access to services.
Table 1. Demographic characteristics of the sample and descriptive statistics of variables of interest
Age at referral
Mean 6.8 years (SD 2.7)
Male 260 (88%) Female 36 (12%)
Duration in clinic
Mean 831.8 days (range 0-4299)
Proportion prescribed medication
Proportion referred before NICE guidance published
Index of Multiple Deprivation score
Mean 15.3 (SD 9.1)
B 115/296 (39%) M 47/296 (16%) S 134/296 (45%)
Both parents 112/296 (38%) Mother only 161/296 (54%) Other 23/296 (8%)
Strengths and Difficulties Questionnaire, Prediction of comorbid behaviour disorder
Unlikely 30/296 (10%) Possible 37/296 (12%) Probable 77/296 (77%)
Strengths and Difficulties Questionnaire Prediction of comorbid emotional disorder
Unlikely 152/296 (51%) Possible 129/296 (44%) Probable 15/296 (5%)
Figure 1 provides the results of the Kaplan-Meier plots and Cox predicted survival curves. Events to the right of the vertical line on the plot record prescription of medication being started after the introduction of NICE guidance, for children who were referred before it was introduced.
The two ‘out of place’ censored cases on the Kaplan-Meier curve for cases referred prior to the introduction of NICE guidelines are due to imputation: they were not removed or altered to avoid biasing the imputation process (Honaker et al., 2000). The Kaplan-Meier Log-Rank test confirmed the graphical impression that the introduction of NICE guidance had accelerated time to prescription of medication (χ2 19.7, df 1, p < .0001). However, even after the introduction of this guidance the median time to prescription was 526 days, though this was less than half the median of 1262 days before. Inspection of the plot indicates no acceleration in rate of prescribing medication following the introduction of NICE guidelines for those children who were referred before it was introduced.
Overall, the Cox Proportional Hazards model was significant (Likelihood Ratio Test 52, df 9.07, p < .0001) with R2 0.16 following stepwise selection. The proportional hazards assumption was met (test of time-dependent coefficient for model: χ2 16.96, N.S.) and Figure 1 supports with minor reservations (see discussion) the model’s predictive validity, as the predicted survival curves remain within the 95% confidence intervals of the empirical data. The coefficients and 95% confidence intervals of the variables included in the final model following stepwise selection are given in Table 2., indicating how much the base likelihood is multiplied by a 1-unit change in each variable. The frailty term was found not to be significant in the full model, and was dropped, suggesting that unmeasured time-dependent changes, such as other transient service policies, staff and resource changes in the teams, had not impacted significantly on time to prescription during the study period.
Table 2. Coefficients and 95% confidence intervals for variables included in Cox regression to prediction prescription of medication
Age at referral was highly significant (χ2 24.75, df 1, p < .0001) with a trend for non-linearity (χ2 7.23, df 3.08, p = .069) demonstrating that the largest increase in baseline likelihood occur disproportionately for the older children. The average time to prescription for cases referred from an educational source was approximately half that of cases referred either from GPs (coefficient .499) or other health sources (coefficient .54) (see Table 2). Though retained in the model, the involvement of an Educational (School) Psychologist did not separately reach the .05 standard for statistical significance (χ2 3.22, df 1, p = .073).
Medication cannot be prescribed until all assessments have been completed so any initial flattening of the survival curve can give an estimate of the minimum time for this process. Figure 2 reports Kaplan-Meier curves, stratified by clinic, for the first 526 days (the median survival of the cases referred after the introduction of NICE guidelines) together with their summary Locally Weighted Exponentially Smoothed Scatterplot (LOWESS), superimposed with estimates of the proportion of cases having received an initial face-to-face contact by 4, 13, and 26 weeks from the 2005 CAMHS mapping (Durham University, Department of Health, 2007). A spike histogram of times to prescription is included to aid interpretation.
For all three clinics and despite their significantly different overall survival times (see table 2) the rate of initiating the first prescription accelerates sharply after approximately 70 days (10 weeks). This ten week limit is equally apparent whatever stratification is used, suggesting this is the minimum realistic assessment period achieved either before or after NICE guidance, with those new prescriptions occurring during this period probably reflecting an accelerated clinic response to crisis, or patient transfers already receiving medication. Also, the bulk of the wait took place after most cases had already begun face-to-face clinic contact, with only 12% having been given a prescription by 26 weeks of waiting, when 87% had been seen.
The main finding of this study is that, overall, the introduction of NICE guidelines has more than halved the time between referral to English CAMHS clinics for ADHD and being given a prescription for medication. Nevertheless, the median waiting time remains at almost 18 months. The minimum realistic time taken to prescribe for a routine case under NICE guidelines is approximately 70 days. It is not surprising to find that NICE guidance has not altered this; the guidance merely codified what was already considered good practice, and the long duration is consistent with the complexity of the assessment NICE recommend. However, this accounts for less than one seventh of the total waiting time, which is also much greater than the time spent waiting for a first clinic appointment. All the clinics took failure of behavioural interventions in primary care as sufficient indication to prescribe medication but the typical length of such interventions, between 9 and 14 sessions (typically weekly) (Serketich & Dumas, 1996), is also insufficient to account for the measured delay even if such interventions were routinely provided in secondary care prior to prescription, which was not the case.
The finding that waiting time varies by clinic is optimistic as well as unsurprising, as this suggests that resource and process changes could significantly improve waiting time. For example, clinic S had, historically, more medically trained staff qualified to write prescriptions than clinic B, who managed more ADHD cases than clinic M. However, the 50% improvement in rate the Cox model indicates is likely from optimising clinic processes would still leave the median wait for medication at around 12 months.
The fact that referrals from an education source reduced the time before medication was prescribed, the trend for speedier prescriptions for older children and the relevance of an educational (schools) psychologist being available, together argue for an educational influence on prescribing rates. Once having been prescribed with medication to treat ADHD, in the UK the rate of receiving medication drops off sharply after 16, which is not typical of the psychopathology of the disorder but is the UK schools’ upper age limit for compulsory education. The finding that additional educational support may predict a greater likelihood of receiving a prescription has also been found in the USA (Safer & Malever, 2000) as has being older, though there the non-linearity identified suggested earlier peak prescribing, with a drop during the teenage years (Cox et al., 2003).
While the lack of impact of comorbid behaviour disorder is surprising, the unequal distribution of the measurement categories means that the impact of such disorders would have to be greater than threefold for the study to detect. Post-hoc analysis identified a ‘significant’ (p < .05) interaction: absence of the involvement of an educational (school) psychologist and conduct disorder being unlikely or only possible was associated with a lower rate of prescription. This corresponds with previous findings that high levels of symptomatology identified by parents did not lead to medication changes (Foreman, Beedie, & Kapuge, 2006). The statistical model had a bias towards overestimating the rate of prescription for those referred before the introduction of NICE guidance. Including the frailty term increased this bias, suggesting that it refers to an unmeasured, non-time dependent reluctance to prescribe medication that was diminished by the introduction of NICE guidance. Other research suggests that the use of questionnaires at initial assessment, as recommended by NICE, might contribute to this change in rate by improving awareness (Foreman et al., 2001; Sayal, Letch, & Abd, 2008) but this could not be tested here as the clinics differed in their policy of questionnaire usage. It is perhaps more surprising to find that the introduction of NICE guidance did not change the rate of prescription for those children already in the clinic. This suggests that the impact of NICE guidance is tightly focused on initial assessment rather than a more general increase in awareness of ADHD, or increasing acceptance of medication as a treatment option.
Being a retrospective study of open cases, there was no way of estimating the number of children with ADHD who left the service before data collection began. However, NICE guidance requires all those diagnosed with ADHD and prescribed medication to be retained in secondary care, resulting in an active re-engagement and monitoring policy. To be missed by the audit a child would have had to be prescribed medication but then not attend appointments for more than twelve months, while the migration rate for the clinic population is 6% per annum (National Statistics, 2006). Thus, the number prescribed medication but missing through attrition is likely to be small.
The first conclusion to draw is that there continues to be no evidence supporting the contention that there is now a rush in the UK to prescribe medication for children with ADHD. Despite the rate of prescription doubling following the introduction of NICE guidelines, the minimum likely assessment period remains 10 weeks, and median wait to prescription is nearly 18 months; longer than can be explained in terms of either waiting for a face-to-face contact with a service, or by previous trials of psychological intervention. It seems likely that schools are an important influence on the decision about whether to prescribe medication, though more research is needed to clarify how that influence is exercised. Though this study suggests that increasing resources to CAMHS ADHD clinics may improve waiting times by a third or more, it also implies that there may be an upper limit to CAMHS clinics’ ability to initiate routine prescription of medication promptly as indicated by the NICE guidance; that limit lying somewhere between 10 weeks and 12 months. So, a case can be made for the development of effective diagnostic systems in primary care for ADHD in the UK, if greater improvement is sought, or if increased funding of secondary care CAMHS is not practicable. The development of such systems could well have international value by providing improved diagnostic accuracy in primary care without imposing unacceptable costs in terms of delayed treatment.
This study was undertaken with funding of £20,000 from Berkshire Mental Health NHS Trust. Dr Foreman was also partly supported by Reading University as Health Services Research Fellow. Thanks to Ms. Suzanne Dack for providing the data set and the cooperation of the staff of Berkshire Mental Health NHS Trust.