Participants and setting
Our study took place between July 2001 and November 2002 in outpatient facilities of three hospitals in the United Kingdom—the Royal Victoria Infirmary, Newcastle upon Tyne; the Gartnavel General Hospital, Glasgow; and the James Cook University Hospital, Middlesbrough.
The trial protocol was shown to all clinicians who worked in the chronic pain units of the participating hospitals and clinicians who refer patients to the trial centres were told about the trial. All suitable patients (box) with neuropathic pain (such as burning, stabbing, or paraesthesia within the distribution of a peripheral nerve) and a clear clinical history of its cause were referred by the participating pain consultants for screening. One clinician (BF) screened, followed up, and monitored all participants during the trial.
Inclusion and exclusion criteria
- Sensory abnormality
- Burning pain
- Lancinating pain
- Sympathetic dysfunction
- Presence of neuropathic pain according to the diagnostic criteria
- Patient taking a stable dose of analgesic
- Age 18-90
- Liver disease
- Bipolar disorder
- Substance misuse
- Renal failure
- Adverse events to dihydrocodeine or nabilone
- Monoamine oxidase inhibitors
We screened 100 patients aged 24-84 years with chronic neuropathic pain (see box for predefined diagnostic criteria) for one to two weeks. To justify treatment and enable improvements to be seen, the patient’s mean pain score had to be greater than 40 mm on a 0-100 mm visual analogue scale. Ninety six patients were randomised to one of the two treatment sequences. We allowed participants to keep taking stable analgesics, except for dihydrocodeine, which was stopped two weeks before the start of the study drugs. All other analgesics were continued. We excluded patients taking antipsychotics, benzodiazepines (except for night sedation), and monoamine oxidase inhibitors. Patients taking any cannabinoid preparation at the time of screening were also excluded, as were those with ongoing legal action associated with their clinical condition. Patients with severe hepatic or renal disease, epilepsy, bipolar disorder, psychosis, or a history of substance misuse were not included in the trial.
After initial recruitment, patients supplied a daily pain score for one week. During this week, a urine sample was tested for cannabis and patients with a positive sample were excluded. We took blood samples to measure full blood count, electrolytes, urea, liver function, and serum glucose. Urine was tested for protein, blood, pH, and ketones using a dipstick. Patients with abnormal values were not included in the trial, except for those who had abnormal plasma glucose and known insulin dependent diabetes.
Patients underwent an abbreviated neurological examination, which included examining muscle strength in both ankles and testing for ankle reflexes (when appropriate), clinical allodynia, and hyperalgesia. Patients also had a general medical examination, which concentrated on the skin, the respiratory system, the cardiovascular system, and the gastrointestinal system.
During the screening visit, patients filled in a hospital anxiety and depression score (HAD score) and a short form 36 quality of life questionnaire (SF-36—commonly used to measure health related quality of life).10 11
All patients watched a demonstration of six psychometric tests and practised the tests on the computer until they were confident in the use of the hardware.
Our study was a randomised, double blind, controlled, crossover trial of three months’ duration. The three trial periods were—treatment period 1 (six weeks), washout period (two weeks), and treatment period 2 (six weeks). Participants made eight specified visits at weeks 0, 2, 4, 6, 8, 10, 12, and 14 (table 1).
Table 1 Timetable of visits and procedures
All patients were asked to fill in a diary recording the average daily pain score, the number of hours slept, details of interruptions to sleep (using a tick box), and the amount of study drug taken. During the washout phase the patients recorded the number of rescue tablets (paracetamol 500 mg and codeine 30 mg per tablet) taken each day.
At each visit the patients filled in a side effects assessment form. During the fourth and eighth visits they also filled in a HAD score and SF-36 form and worked through the six psychometric tests.
The pain score was the primary outcome variable. Pain scores are just one way to measure the benefit of treatment in patients with chronic pain. Improvements in mood, sleep, and quality of life are equally important for most patients. Secondary outcomes were anxiety and depression as measured by the HAD score, each of the eight domains and the “change in health” measured by the SF-36,12
and the weekly average number of hours slept each night calculated from the diaries. We calculated the scores for the psychometric tests from the absolute results of each test divided by the time taken to perform the test.
We used an eight item questionnaire that asked about the most common side effects of dihydrocodeine and nabilone. Each question could be answered by circling one of five discretional answers ranging from “Yes, very much” to “Definitely not.” Other side effects could also be recorded. Only the answers “Yes, very much” and “Yes, quite a lot” were counted as side effects for the analysis.
The trial drugs were given in an escalating manner (fig 1). If the patient developed side effects, the dosage was reduced to the previous value for the remainder of the trial period after discussion with the investigator. Patients were weaned off the drugs by halving the dose every three days, and patients did not take any trial drug for the last six days of the washout period but were allowed up to eight tablets of rescue drugs a day.
The pharmacy at St Mary’s Hospital supplied identical white capsules containing 250 µg nabilone or 30 mg dihydrocodeine. The pharmacies at the treatment centres, the patients, and all clinical personnel involved in the trial were unaware of treatment allocation at all times.
Code breaking envelopes were kept in each hospital pharmacy and each was used for only one patient. The code was disclosed only to the requesting doctor who was not involved in the study.
We randomised patients to receive nabilone first then dihydrocodeine or vice versa. Treatment was allocated by random permuted blocks of 10, stratified by centre. We used a model with fixed patient effect, period effect, and treatment effect but no term for the carryover effect of treatment to analyse the data.13
Normal errors were assumed and this was checked using normal probability plots. Calculations of sample size were complicated by the lack of information on within patient variation for the visual analogue score. A study that randomised 30 patients to each treatment sequence would have 90% power to detect a difference in mean visual analogue score between the treatments equal to 60% of the within patient standard deviation, at 5% significance. This is broadly in line with previous studies.14 15
We anticipated that as many as 30% of patients would drop out, so we aimed to recruit 100 patients.
Each treatment period lasted six weeks and we collected data at the end of each period for all variables except the pain score; this score was recorded daily and weekly means were analysed. The treatment periods were separated by a washout period of two weeks. We excluded carryover by basing the analysis on data from the last two weeks of each treatment period—weeks 5-6 and weeks 13-14. If data from one of the last two weeks were missing, we substituted data from the preceding week (week 4 or week 12). Because the use of data from earlier weeks could produce bias from carryover effects, if no data were available on a patient in the last three weeks of the period, the treatment period was excluded. The use of a fixed patient effect in the analysis meant that the patient was then excluded from the analysis.
Two analyses are presented. The available case analysis used the fullest dataset—all patients randomised who provided data in each treatment period (week 4 or beyond and week 12 or beyond). The per protocol analysis excluded patients who did not comply with the trial drugs, as assessed by their pain diary.
Data were prepared used Minitab version 13 and analyses were carried out with Stata version 7. We recoded and tabulated the side effects using SPSS version 11.