At six months, four hundred and eighty four participants completed a follow-up questionnaire, a response rate of 82%. There were 112 persistent non-responders, despite a strict follow-up protocol of postal reminders and phone calls.
Respondents and non-respondents were similar with respect to all baseline variables with the exception that non-respondents were more likely to be male, living either in a private rental or rent-free accommodation. Participants attended an average of 6 (SD 3.7) treatment sessions.
The mean Roland Morris disability score improved by 3.8 index points (95% confidence interval 3.23 to 4.32, p < 0.001) over the six month follow-up, from 11.6 at baseline assessment to 7.8 index points at six months. The distribution of change scores is illustrated in figure . The SF- 36 physical functioning scale improved by 10.7 scale points (95% confidence interval 8.4 to 13.0, p < 0.001) from 49.2 at baseline assessment to 59.8 points at six months.
Change in Roland Morris disability questionnaire score from baseline to six-month follow-up.
Persistence of symptoms at six months was predominantly associated with ethnic grouping. Participants who categorised themselves as non-white had a reduced odds ratio for recovery of 0.39 (0.20 to 0.74, p = 0.004). Those recording ethnic group as North African or Middle Eastern showed a change of less than one index point on the Roland Morris questionnaire and 2.7 points on the SF-36. There was some evidence to suggest that participants who recorded a higher frequency of exercise participation were more likely to have recovered at six month than those who rarely undertook exercise (Table ).
Ideally, where individual prognostic variables are found to be predictive of outcome, efficient clinical cut-off scores could be used to make decision rules about the need for treatment. This has theoretical and clinical relevance for binary variables like gender or previous surgery. However, in the case of BMI the main analysis used a grouping of data above and below the median. To give greater clinical relevance a further analysis was undertaken. The BMI is usually graded as underweight, optimal, overweight, obese and morbidly obese, rather than dichotomised as required for the main analysis. The mean change scores (with 95% confidence intervals) in the RMDQ for these categories were as follows: BMI less than 25 (underweight/optimal) 4.3 (3.4 to 5.2); BMI 25.01 to 30 (overweight) 4.0 (2.7 to 5.3); and BMI 30 and over (obese/morbidly obese) 4.3 (2.4 to 6.3). A correlation between the continuous variable BMI against the change score in the Roland Morris yielded a non-significant Pearson Product Moment coefficient of 0.05 supporting the results of the main analysis. The extreme categories (BMI less than 18.5 and more than 40.0) had too few participants to provide meaningful independent analysis. Prognostic indicators in these groups may warrant further study.
Although we found some evidence to suggest that those in paid employment and the self employed had a greater chance of recovery (odds ratio 2.07, 1.21 to 3.54, p = 0.008), factors including control over work, the work environment and the physical characteristics of the tasks involved were not linked to recovery (Table ). The questions addressing the physical characteristics of work were condensed to two dimensions. The question on time spent sitting was the converse of time spent walking so it was logical to reduce this to one variable. Similarly, there were few participants who recorded lifting 50 kg and those that did, also lifted 25 kg, so this was also reduced to one variable. The data provides an indication of the nature of work undertaken whether largely sedentary or involving heavy lifting.
Of the 45 participants who were in paid employment and reported being absent from work as a direct result of their back pain, only 3 (6%) reported that they were still absent at six month follow-up.
Clinical history and Presentation
The odds ratio for recovery increased in participants who had experienced less than twelve short episodes in the past twelve months compared to those who described the nature of their episodes as continuous. Participants categorised as Grade IV on the Von Korff pain scale, indicating high levels of disability and severely limiting pain, had a reduced chance of recovery (odds ratio 0.07, 0.02 to 0.23, p < 0.001) compared to those classified as Grade I; low disability and low intensity (Table ).
Clinical history and Presentation
Since completing treatment, 69% of participants reported experiencing a further spell of back pain, although only 21% felt it severe enough to see either their GP or other health practitioner. These included physiotherapists, osteopaths, chiropractors, acupuncturists, orthopaedic surgeons or rheumatologists.
Psychosocial and psychological factors
Low scores on the Zung depression inventory (odds ratio 3.43, 1.90 to 6.19, p < 0.001) and the index of somatic anxiety (odds ratio 2.36, 1.36 to 4.09, p < 0.001) were found to be linked to improvement in Roland Morris disability and SF-36 physical functioning scores, with sizable effects on both scales. However, once the individual variables were adjusted for baseline Roland Morris scores, their effect was reduced (Table ).
Psychosocial and psychological factors
The scores for the MSPQ and the Zung Depression Inventory (Table ) are comparable with other studies of similar cohorts; (Mean MSPQ: 5.6 [37
], 9.7 [38
], 6.7 [39
]); (Mean Modified Zung Depression Index: 24.9 [37
], 29.7 [38
], 23.7 [39
]). However, there are many ways of analysing and reporting data for psychological problems. Using the decision rules for the Distress and Risk Assessment method (DRAM) defined by Main [32
] and used in the UK Beam trial [13
] and other studies [33
], patients can be classified into clusters depending on their scores on the MSPQ and the Zung Depression Inventory (Table ).
DRAM classification of participants who responded at 6 months and provided both Zung and MSPQ scores (N = 471)
The psychological profile of participants in this study, categorised using the DRAM, is comparable to similar cohorts of people with back pain (N 37%, R 42%, DD 13% and DS 9% [33
]; N 24%, R 42%, DD 24% and DS 10% [40
]). This suggests that the greatest proportion of participants were in the normal or at risk categories rather than in the distressed (somatic or depressive) categories.
Multiple Regression Model
Because baseline score is a determinant of the magnitude of change, and there is likely to be co-dependency in the data, those factors found to be predictive of outcome, defined as variables with an unadjusted p-value of less than 0.1, were entered into a multiple regression (binary logistic) analysis, controlling for all other variables in the model (Table ). Adjustment was also made for age and sex.
Multiple regression analysis of predictive variables
Adjusted odds ratios associated with a reduced chance of recovery were linked to self-classification as 'non-white' as opposed to 'white' (0.41, 0.18 to 0.96, p = 0.039). The pattern of back pain over the previous twelve months had an impact on recovery, increasing in those who reported episodic rather than continuous pain (2.64, 1.25 to 5.60, p = 0.005) with greatest improvement in those with fewer, brief episodes of back pain. Change in Roland Morris disability scores for each sub-classification of the two variables with predictive value in the multiple regression model is shown in Table .
Mean change in Roland Morris for ethnic classification and episodic history (n = 472)