All general practices in Nottingham, north Lincolnshire, and southern Derbyshire were invited to take part in the study. Practices in the north of Leicestershire and in the south of north Nottinghamshire were also invited to take part. In total 73 practices took part, of which 52 recruited participants to the trial. The study population comprised patients with low back pain consulting doctors in participating practices between November 1995 and January 1999.
Identification of patients—Patients with low back pain were identified either by searches of computerised medical records based on the Read code used by each practice for low back pain or, in practices not recording all consultations on computer, by the doctor flagging the notes of patients seen with low back pain. The computerised searches were conducted by research nurses.
Inclusion and exclusion criteria—
Patients were included if they had low back pain on the day of randomisation and for at least the preceding six weeks for the first episode of low back pain. Patients with recurrent low back pain were included if they had pain on the day of randomisation and for at least six weeks in the preceding six months. Patients were excluded if they were outside the age range specified for simple backache in the guidelines of the Clinical Standards Advisory Group and the Royal College of General Practitioners (under 20 or over 55), if they had chronic back pain (persistent pain for more than six months), if they had had radiography of the lumbar spine within the preceding year, had unexplained weight loss or fever, were taking oral steroids, had a history of malignancy, tuberculosis, injecting drug use, or a positive result on a HIV test, had symptoms or signs of a cauda equina lesion, or were pregnant or planning a pregnancy.11
Patients were also excluded if the doctor considered they were unable to give informed written consent—for example, patients with a learning disability.
Ascertaining eligibility—Patients were invited to participate by letter from the general practitioner. Patients responding to the letter were interviewed on the telephone by the research nurses to ascertain eligibility criteria. Patients who seemed eligible were visited at home where the baseline structured interview and physical examination were undertaken by the research nurse. Eligible patients were then asked to give informed consent before randomisation.
Assignment to treatment group—Randomisation was by individual participant. At the baseline interview the research nurse opened a sealed envelope containing the treatment group allocation. Block randomisation (using blocks of 20) was used to ensure equality of numbers between the two groups. A member of the research team (KF) who was not involved in assigning the participants to treatment group generated the allocation schedule. Participants and research nurses were not blinded to treatment group. In addition the study included a preference arm for participants in which those who did not consent to randomisation could choose whether to have radiography or not.
Intervention—In addition to receiving the usual care provided by the practice for patients with low back pain, patients in the intervention group were given a card to attend for a radiograph of the lumbar spine at their local hospital. They were asked to contact their doctor for the result of the radiography either by telephone or by consulting the doctor, depending on the usual procedure for each participating practice. Participants in the control group received the usual care from their doctor. The doctor was able to request radiography if they considered it clinically necessary at any time.
Primary and secondary outcome measures—
The primary outcome measure was difference in the mean Roland score (an adaptation of the sickness impact profile).14
Secondary outcome measures included a visual analogue scale for pain, EuroQol, including the health status scale,15
duration of low back pain, duration of certificated sick leave, use of health and other services, and drug use. The research nurse measured primary and secondary outcomes before randomisation and at three and nine months after randomisation by structured face to face interviews. Interviews were conducted by telephone if the participant was not able to be interviewed face to face.
Sample size—The sample size calculation indicated that 388 patients in total in both arms of the study would allow a change in mean Roland score of 1.5 to be detected with 90% power at the 5% significance level, based on a baseline mean Roland score of 10.1 (SD 4.5). This was obtained from the first 88 patients recruited to the study. The sample size was based on showing a difference between the two groups that we judged would not be clinically important rather than equivalence, as showing equivalence would have required a much larger sample size.
Data analysis—The data were double entered into an Access 97 database and analysed using SPSS for Windows version 8.0. We undertook all analyses on an intention to treat basis. We compared non-normally distributed continuous variables with Mann-Whitney U tests, and we compared categorical variables with χ2 tests (with Yates correction and Fisher's exact test where appropriate). We calculated relative risks with 95% confidence intervals.
Ethics committee approval—Ethical approval was obtained from the Queens Medical Centre, Nottingham, southern Derbyshire's ethics committee, north Lincolnshire's research ethics committee, north Nottinghamshire health authority, and Leicestershire health authority.