Generalisability
Patients were recruited between July 2009 and December 2009, and followed up until June 2010.
We considered several patient characteristics in order to evaluate the generalisability of this study: sex, age, deprivation (as measured in quintiles based on postcode), PCT and body region of MSK problem.
We compared the characteristics of patients in two groups: (a) those who were ineligible for the study on the basis of their referral letter () or because they did not confirm that they wanted physiotherapy and (b) those who were eligible.
Reasons why referrals were ineligible for invitation
Among the eligible patients, we also compared those who did or did not consent to participate and were randomised to participate in the trial.
Obvious differences between those patients who were or were not eligible, reflecting the eligibility criteria, were observed for the site of MSK problem for which patients were referred to physiotherapy. A higher proportion of patients in Bristol and a lower proportion in Somerset were deemed eligible. More women were referred than men, and slightly fewer men were eligible. Patients who were eligible were slightly older than those whose who were ineligible ().
Comparisons between those who were or were not eligible and were or were not randomised
Considering the eligible patients, those patients who were randomised were slightly younger, on average, than those who were not. Apart from this, there were few differences between those who were randomised and those who were not ().
Patients from the least deprived areas (quintile 1) were under-represented in this study in comparison with the national population. Patients who were from the most deprived areas were slightly less likely to be eligible or to consent to participate in the study and be randomised.
The recruitment rate, defined as the percentage of patients consenting to participate from those who were eligible (responding to the invitation letter confirming that they did want physiotherapy), was 2256/4523 or 50% ().
Consolidated Standards of Reporting Trials (CONSORT) diagram: flow of participants through the trial, a. Per cent is expressed as n/1506 × 100. b, Per cent is expressed as n/743 × 100.
Baseline comparability of randomised groups
No marked difference was observed between patients randomised to the two arms of the trial in terms of demographic, stratification/minimisation or outcome variables at baseline (). At baseline, 59% of participants in the trial were female, 96% were white, the majority were in employment, and patients from the least deprived areas were under-represented. Almost everyone had access to a telephone and the vast majority had access to a car ().
Baseline demographic characteristics of the patients randomised
Baseline characteristics related to the physiotherapy problem or physiotherapy service
The most common reasons for referral were lower limb problems (including hip and knee problems), lumbar problems and upper limb problems (including shoulder). Participants from each of the PCTs were reasonably well represented, with Bristol contributing the highest proportion. People were fairly evenly split in their preference for future care between usual care, PhysioDirect and having no preference (see ). There were no important differences in outcome measures assessed at baseline ().
Outcome variables at baseline for the patients randomised to the two arms of the trial
Loss to follow-up
Primary outcome data were collected for 88% (1985/2249) of patients at the 6 weeks' follow-up time point and 85% (1912/2249) of patients at the primary outcome time point of 6 months, with no difference between trial arms. Details of the relationship between loss to follow-up and patient characteristics are given in Appendix 1.
Several differences were observed between patients for whom the primary outcome measure (SF36v2 PCS at 6 months) was collected and those for whom it was missing, although these differences were similar for both trial arms (see Appendix 1, Tables 62 and 63). With regards to the demographic characteristics, patients with missing data tended to be younger, unemployed and to come from more deprived areas. There was also an association between job type and missing primary outcome data (see Appendix 1, Table 62).
As described in Chapter 1 (see Summary of changes to the project protocol), we obtained ethical permission for a protocol amendment to continue inviting patients to participate until 2000 patients had been recruited, in order to ensure that we reached our target of 1500 patients for analysis. Because of the lag time between initial approach to patients and recruitment, and because the final follow-up rate was higher than assumed, the total number of patients available for analysis (1912) was higher than originally planned (1500).
Process of care
In the context of this trial it is most helpful to describe the types of care received by participants in each arm of the trial, before describing the primary and secondary outcomes. This will aid interpretation of these outcomes.
Number of telephone and face-to-face consultations
Patients in the PhysioDirect arm were offered an initial telephone call to assess their need and priority for a face-to-face appointment. Of all patients in the PhysioDirect arm, 1281 (85%) contacted the service at least once, with 97% (1239/1281) of these having an initial telephone call and 3% (42) having an initial face-to-face appointment. Of those having an initial telephone assessment, 47% (586/1239) were managed entirely on the telephone with the remainder having at least one face-to-face consultation. Of all patients in the PhysioDirect arm, only 46% (695/1506) eventually had any face-to-face appointments ().
Types of consultations experienced by patients
By the time of the first follow-up time point at 6 weeks after randomisation, 81% of patients in the PhysioDirect arm had received some sort of assessment appointment or consultation, compared with only 50% of those in the usual-care arm.
There is strong evidence that PhysioDirect greatly increases the odds of having at least one appointment or consultation (of any type) by 6 weeks [adjusted odds ratio (OR) 4.49 (95% CI 3.68 to 5.49); p < 0.001].
The following table shows the number of consultations that patients received in each arm of the trial. Over the 6-month follow-up period, patients in the PhysioDirect arm had significantly fewer face-to-face consultations, and fewer consultations in total (including telephone and face-to-face consultations), compared with those in the usual-care arm ().
below details the distribution of the number of face-to-face consultations experienced by patients in each arm. This is further divided into all of those patients, or only those patients who had any face-to-face consultations. For those who did have any face-to-face consultations, a ‘course of treatment’ was a mean number of face-to-face consultations of 3.77 (95% CI 3.57 to 3.96) and 4.13 (95% CI 3.94 to 4.33) in the usual care and PhysioDirect arms, respectively. The number of appointments was skewed with a median number of face-to-face consultations of 3 [interquartile range (IQR) 2 to 5] and 4 (IQR 2 to 6) in the usual care and PhysioDirect arms, respectively.
Number of face-to-face consultations
Length of consultations
The mean and median lengths of the face-to-face consultations in minutes were 34.1 (SD 11.3) and 30 (IQR 30 to 40), respectively. Telephone consultations had a mean and median length of 23.9 (SD 12.2) and 23 (IQR 17 to 30) minutes, respectively.
The mean length of telephone calls in the PhysioDirect service was monitored throughout the trial and it was observed that the mean duration of calls decreased during the study. This is illustrated by . The figure shows the mean duration of telephone calls, from the first week of calls in each site. However, as the study progressed an increasing proportion of calls were follow-up calls, which tended to be shorter.
Mean duration of first telephone calls over time: all calls.
shows the same analysis restricted to first calls only, and confirms that call duration became shorter over the period of the trial.
Mean duration of first telephone calls over time: first calls only.
Did not attend rates
The rate of failed face-to-face appointments (DNA rate) at 6 months was measured in two different ways: (1) as an overall rate, namely the ratio of the number of unattended appointments over that of all appointments that had been scheduled; and (2) as an individual patient rate, with the DNA rate calculated for each patient as an incidence rate ratio (IRR) adjusted for age, sex and primary outcome at baseline and PCT.
The overall rate was substantially the same in the two trial arms, being 262/2310 (11%) for usual care and 279/2873 (10%) for PhysioDirect. The individual DNA rates had median 0 for both arms, with means of 0.12 for usual care and 0.09 for PhysioDirect. The effect estimated for PhysioDirect on the DNA rate (adjusting for age, sex, primary outcome at baseline and PCT) was 0.55 (95% CI 0.41 to 0.73) with p < 0.001. Therefore, there was strong evidence of PhysioDirect being associated with an average 45% reduction in individual patient DNA rate.
Waiting times
Waiting time to the first consultation, be it a telephone or face-to-face consultation, was calculated for each patient from both the referral date and the randomisation date (). The possible effect of PhysioDirect on each of those waiting times was estimated through Accelerated Failure Time analysis, adjusting for PCS at baseline, age, sex, main referral problem and PCT. Both in the referral and the randomisation case there was strong evidence (p < 0.001) of PhysioDirect substantially decreasing the waiting time, a reduction of up to 68%. This equates to a shorter wait of 7 days from randomisation. The median delay introduced by the need to obtain consent in the trial was 12 days (this is the difference between the date the referral was received and the date written patient consent was received).
Number of days to first consultation (telephone or face to face)
One aim of PhysioDirect is to filter out people who will not benefit from face-to-face care, so that those who do need it can be seen more quickly. There was no evidence that the waiting time to first face-to-face appointment was different between the PhysioDirect and the usual-care arms ().
Number of days to first face-to-face consultation from the randomisation date
Characteristics of patients seen in face-to-face appointments in the PhysioDirect arm
In an exploratory analysis, we compared the characteristics of patients in the PhysioDirect arm who were then seen in face-to-face appointments following initial telephone assessment with those who were managed entirely on the telephone (). There were no differences in terms of age or sex. However, there were slight differences according to site of MSK problem, and patients managed entirely on the telephone had slightly higher SF-36v2 PCS scores at baseline than those seen face to face [37.6 and 35.4, respectively, difference in means 2.36 (95% CI 1.38 to 3.33)].
Within PhysioDirect arm, characteristics of patients invited for face-to-face consultation
Primary outcome analysis
No evidence was found of a difference in the PCS of the SF-36v2 between the two trial arms at either the 6-week follow-up time point or the primary time point of 6 months, having adjusted for baseline outcome measure, sex, age, main referral problem and the patient's PCT (). The 95% confidence limits support the hypothesis of equivalence in terms of clinical outcome between the trial arms. illustrates the difference in means and confidence limits in relation to the observed variation in the primary outcome (SD) and the pre-specified equivalence limits of 2 points on the SF-36v2 PCS.
Primary outcome: SF-36v2 PCS at 6 weeks and 6 months following randomisation
Difference in means in relation to equivalence limits.
A repeated measure analysis was conducted to estimate possible variations in treatment effect on SF-36v2 PCS at different points in time, namely at 6 weeks and 6 months. The analysis showed no evidence of such variations, with an arm–time interaction for SF-36v2 PCS adjusted for outcome baseline, sex, age, referral problem and PCT of 0.001 (95% CI −0.0041 to 0.006; p = 0.76).
Secondary analyses
Sensitivity analyses
Adjustment for baseline imbalances
Some imbalance at baseline was observed for the employment status variable, but including it as a covariate in the ANCOVA did not change the conclusion of the original primary analysis, namely no evidence of any difference between PhysioDirect and usual care on SF-36v2 PCS at 6 months ().
Comparison of the estimates of trial arms difference in SF-36v2 PCS at 6 months with and without adjustment for employment status at baseline
Effect of missing primary outcome data
Although only about 15% of the primary outcome data were missing in each trial arm, multiple imputations were conducted to determine what effect those data might have had on the primary analysis. The conclusion that there was no evidence of difference between the trial arms was the same with or without multiple imputations of missing data ().
Comparison of estimates of the difference in primary outcome between trial arms with and without multiple imputation of the missing values
Clustering effects
The possibility of clustering effects due to patients belonging to the same PCT or GP practice was explored through multilevel models with random effects attributed to those clusters. Each one of the models was compared with the original one (fixed effects) for its goodness of fit relative to the data by means of a log–likelihood ratio test or through its Akaike information criterion (AIC, lower value corresponding to a better fit), which accounts both for likelihood and parsimony of the model.
The only model for which there was some evidence of a better fit to the data than the one adopted for the primary analysis was a model accounting for a PCT clustering effect, but it maintained the conclusions of the original model (i.e. equivalence between PhysioDirect and usual care) and the clustering effect appeared to be very modest as indicated by an intracluster correlation (ICC) of 0.005 ().
Estimates of arm difference in primary outcome for models which do or do not account for possible clustering effects
Secondary outcomes
Clinical outcomes
A similar regression analysis as the primary one (same covariate adjustments) was conducted using a set of secondary outcomes, representing different ways of assessing health status. These were:
a patient-generated measure, focusing on specific symptoms (MYMOP)
a broad measure of health-related quality of life (EQ-5D) to be used in the economic analysis
a single question about the extent to which the main problem for which the person had been referred to physiotherapy had improved (global improvement score)
a measure of whether or not the individual had ‘responded to treatment’ using the OMERACT-OARSI criteria, based on a composite measure including physical function, pain and improvement in the main problem.
The results of the analysis of MYMOP showed a statistically significant difference between the trial arms at 6 weeks but this difference was very small and the CIs fell within the limits of equivalence that we specified for the MYMOP measure a priori (0.5 points). Both the global improvement score and the analysis according to the OMERACT-OARSI criteria also suggested that patients in the PhysioDirect arm were more likely to have improved to a greater extent by 6 weeks than those in the usual-care arm. The EQ-5D showed a slight trend towards benefit from PhysioDirect at 6 weeks but the difference was small and may well have been due to chance.
It is notable that all five measures of clinical outcome (the four secondary outcomes in and the primary outcome SF-36v2 PCS) showed a similar pattern of slightly better scores from PhysioDirect at 6 weeks, with minimal difference after 6 months' follow-up.
Secondary health-related outcomes at baseline, 6 weeks or 6 months
It should be pointed out that there is some dependence between measures in that two of the elements in the OMERACT-OARSI ‘response to treatment’ measure were based on improvements in scores on the ‘physical function’ and ‘pain’ scales from the SF-36 questionnaire, and the questions that make up these scales also contribute to the SF-36 PCS score, which is the primary outcome.
The following table () shows the scores for each of the individual scales from the SF-36v2 measure, at both 6 weeks and 6 months. There is no evidence of difference between the trial arms on any measure, at either time point. This table also demonstrates that all of the measures improved over time in both arms of the trial.
SF-36v2 scale scores at baseline, 6 weeks or 6 months
Time lost from work
No evidence was found of a difference between the two trial arms in time lost from work due either to the medical condition or to physiotherapy-related treatment provided through the trial (). This analysis included only patients who were in work.
Time lost from work due to either medical condition or trial-related treatment
Preference for future care
Patients' preference for the type of care they would prefer (usual care, PhysioDirect or having no preference) was recorded both at baseline and at 6 months from randomisation. Although none of the categories was particularly favoured at baseline in either arm, after 6 months usual care was the preferred choice for those patients allocated to the usual-care arm, whereas patients allocated to the PhysioDirect arm appeared almost split in their preference between usual care and PhysioDirect. An analysis of the possible effect on preference by the PhysioDirect intervention by means of multinomial logistic regression provided evidence that it substantially increases the odds of preferring PhysioDirect to indifference (). So, although, normally, patients at 6 months would prefer usual care, PhysioDirect can reduce that tendency in its own favour. This is illustrated by .
Preference for usual care, PhysioDirect or neither
Preference for PhysioDirect or usual care. (a) Baseline. (b) At 6 months.
Patient satisfaction
Some evidence was found of slightly higher satisfaction for patients in usual care than for those in PhysioDirect with regards to both satisfaction with the quality of the consultation and overall satisfaction. No difference is evident with regards to satisfaction with access to the service ().
Consultation, access and overall satisfaction levels
It is important to note that data presented in are based on a comparison only of patients who had had any advice from a physiotherapist, either face to face or on the telephone. People who were still on the waiting list at 6 weeks following randomisation (predominantly in the usual-care group) did not complete the questionnaire, and satisfaction with access to care may be different among this group.
We explored whether or not there was an interaction between baseline patient preference and randomisation group in terms of satisfaction with the service, in other words, whether or not people's satisfaction with the service they received was related to their initial preference. We found no such interaction for any of the three components of satisfaction. We also explored whether or not there was an interaction between patient preference and randomisation group with respect to effectiveness (based on the primary outcome of SF-36v2 PCS). Again we found no such interaction.
Analysis of free-text comments from questionnaires to obtain the views of patients regarding waiting time from referral
We noted above that the people who may have been least satisfied with access were those who had not received any physiotherapy but they would not have completed the satisfaction questions. We observed that many people added free-text comments to the questionnaire. Therefore, in a post hoc analysis we undertook a quantitative analysis of these free-text comments.
In the final section of the 6-week and 6-month follow-up questionnaires, participants were invited to comment on ‘Things that were good about the service’ and ‘Things that could be improved’. These comments were entered in full as part of routine data entry. Content analysis was performed to assess:
Whether or not (from the comment) the patient had received physiotherapy (by telephone or face to face) at that point.
The patient's view of waiting time from referral, coded as positive, negative, both, or not stated. An example of when ‘both’ applied would be when the patient was appreciative of the prompt telephone contact but unhappy with the wait to be seen face to face.
The comments from all 6-week and 6-month questionnaires were coded by one researcher. Validation of the coding was carried out by a second researcher, independently coding a 1 in 10 sample. Where there were discrepancies, the coding was discussed and a final coding agreed. The resulting agreement for whether or not patients had had physiotherapy was 98% at 6 weeks and 97% at 6 months; the agreement for view of waiting time was 98% at 6 weeks and 95% at 6 months.
Results are shown comparing patients in the usual care and PhysioDirect arms of the trial ( and ).
Opinions of waiting time for physiotherapy from free-text comments on 6-week questionnaire
Opinions of waiting time for physiotherapy from free-text comments on 6-month questionnaire
It is clear from the results presented above that the free-text comments provide a different perspective, compared with the ‘access’ questions in the questionnaire, on respondents' views on waiting time for advice and treatment for physiotherapy. More positive comments were made in the PhysioDirect arm and more negative comments were made in the usual-care arm.
Adverse events
Despite requesting reports of AEs from physiotherapists and from GPs, and after reviewing the GP notes of all patients in the trial, no AEs were detected in either arm of the trial. In particular, no AEs were detected from 4323 consultations (1445 by telephone and 2878 face to face) over 6 months in 2249 patients allocated to PhysioDirect.
Subgroup analysis
Possible differences in the effect of PhysioDirect compared with usual care on the SF-36v2 PCS at 6 months between subgroups based on age, main referral problem, deprivation or PCT, were analysed by estimating the interaction term of those factors with assigned treatment. No evidence was found of subgroup differences for the primary outcome ( and ). However, potentially important differences cannot be excluded as the CIs are very wide, reflecting the limited power available for subgroup analyses.
Subgroup analyses of SF-36 PCS score at 6 months according to baseline age and presenting complaint
Subgroup analyses of SF-36 PCS score at 6 months according to baseline socioeconomic status and PCT
Finally, we report descriptive data about the pattern of responses on the primary outcome for patients with different sites of MSK problem (). This demonstrates that different types of problem have different impacts on health status; most improve substantially over time (by up to 8 points, when the SD of the PCS in this population is 8.8 points at baseline); and that most of this improvement had occurred by 6 weeks following randomisation. Patients with thoracic or ‘other MSK’ problems improved the least over the 6 months' follow-up, whereas those with lumbar, limb or widespread pain problems improved the most. We have already demonstrated (see ) that there is no evidence of difference overall between the trial arms in terms of improvement, or any evidence that some conditions benefited to a greater extent than others through a PhysioDirect service (see ).
Descriptive statistics for SF-36v2 PCS at baseline, 6 weeks or 6 months by patient's main referral problem
