Reducing weight and increasing physical activity in people at high risk of cardiovascular disease: a randomised controlled trial comparing the effectiveness of enhanced motivational interviewing intervention with usual care

Objective The epidemic of obesity is contributing to the increasing prevalence of people at high risk of cardiovascular disease (CVD), negating the medical advances in reducing CVD mortality. We compared the clinical and cost-effectiveness of an intensive lifestyle intervention consisting of enhanced motivational interviewing in reducing weight and increasing physical activity for patients at high risk of CVD. Methods A three-arm, single-blind, parallel-group randomised controlled trial was conducted in consenting primary care centres in south London. We recruited patients aged 40–74 years with a QRisk2 score ≥20.0%, which indicates the probability of having a CVD event in the next 10 years. The intervention was enhanced motivational interviewing which included additional behaviour change techniques and was delivered by health trainers in 10 sessions over 1 year, in either group (n=697) or individual (n=523) format. The third arm received usual care (UC; n=522). The primary outcomes were physical activity (mean steps/day) and weight (kg). Secondary outcomes were changes in low-density lipoprotein cholesterol and CVD risk score. We estimated the relative cost-effectiveness of each intervention. Results At 24 months, the group and individual interventions were not more effective than UC in increasing physical activity (mean difference=70.05 steps, 95% CI −288.00 to 147.90 and mean difference=7.24 steps, 95% CI −224.01 to 238.50, respectively), reducing weight (mean difference=−0.03 kg, 95% CI −0.49 to 0.44 and mean difference=−0.42 kg, 95% CI −0.93 to 0.09, respectively) or improving any secondary outcomes. The group and individual interventions were not cost-effective at conventional thresholds. Conclusions Enhancing motivational interviewing with additional behaviour change techniques was not effective in reducing weight or increasing physical activity in those at high CVD risk.


Time to intervention commencement
There were unexpected delays to intervention commencement. Date of intervention commencement was missing for 35 (2.9%) participants. We endeavored to start the intervention within six months of randomisation, and this was possible for 986 (83.2% of 1185) participants: 561 (82.3% of 682) in the group arm and 425 (84.5% of 503) in the individual arm. The time between randomisation and intervention start for participants randomised to each intervention arm is illustrated in Figure S1. Participants randomised to receive the intervention (group or individual arms) waited a mean of 3.12±2.70 months (min=0.16, max=15.21, median=2.07, IQR=0.92-4.99) from randomisation to their scheduled introductory session. The wait time did not differ between the group (3.17±2.78 months, n=682) and individual arms (3.04±2.59 months, n=503; t(1120.5)=0.83, p=0.41).
We investigated the association between delays to intervention commencement and the number of sessions attended. We observed a moderate negative correlation in the group arm (r(680)=-0.20, p<0.001), but not in the individual arm (r(501)=-0.03, p=0.57). The difference in correlations between the arms was significant (z=2.94, p=0.003).

Intervention duration
For those participants attending at least one of Sessions 1-10 (n=782), we calculated the time between Session 0 and the final session attended. Across intervention arms, the mean time between Session 0 and the final session was 330.6±173.8 days (min=6, max=808, median=366, IQR=231-440; dates were unavailable for one patient (0.1%)). Time between Session 0 and final session was significantly longer in the individual arm (355.6±173.1 days, n=373) than in the group arm (307.8±171.4 days, n=408; t(771.4)=3.87, p<0.001).

Participant adherence to intervention
After the completion of each intervention session, the health trainer was responsible for recording: (a) if the patient set a target for that session (yes/no); and (b) if that target was achieved (coded as no/partially/fully). Session 0 was not included as participants were not asked to set goals for that session. Of the 1220 participants randomized to the intervention arms, 774 had adherence data available (63.44%). Of those, 407 (58.39% of those randomized) were in the group arm and 367 (70.17% of those randomized) in the individual arm.
The 774 participants attended a total of 5932 sessions. Table S3 shows the rate of targets set overall and by intervention arm. The overall rate of targets set differed significantly between arms (χ 2 (1)=6.12, p=0.01) but the distribution of targets set at each session did not differ significantly between arms (χ 2 (9)=16.18, p=0.06; data are not shown).

Loss to follow-up
Participants were not seen at follow-up for one of three reasons: (i) they had withdrawn from the study, (ii) they had died, or (iii) they were non-contactable at follow-up due date (attempts were made to contact the participant up to six months after the due date). Participants randomised to either of the intervention arms but who did not take part in the intervention did not necessarily withdraw from the study, and their data are included at follow-up to carry out the ITT analysis. Table S4 shows loss to follow-up at both 12-and 24-month follow-ups and the reasons for withdrawal given. Those who were permanently lost to follow-up at 12 months (withdrawn/died) were not contacted again at 24 months, whereas those who were non-contactable at 12 months were contacted again at 24 months. for the individual arm and 11.9% for the UC arm. Data were collected for 91.6% of all participants for at least one of 12-or 24-month follow-up, and for 79.7% of participants at both 12 and 24 months. The differences in loss to follow-up between the treatment arms were significant at 12 month (χ 2 (2)=17.99, p<0.001) and 24 month follow-up (χ 2 (2)=13.39, p=0.001).

Accelerometer data completeness
The remaining participants did attend follow-up appointments but did not necessarily complete accelerometer wear for collection of PA outcomes. Table S5 provides details on the amount of accelerometer data collected by trial arm at each time point. The required PA data at baseline was at least five days of ≥540 minutes accelerometer wear. A number of participants did not wear the accelerometer for the sufficient amount of time at baseline and, therefore, were randomised in error. In the original analysis plan we stated that we included those with at least four days data in the primary analysis to limit exclusions. However, after discussions with the statistician of the TSC, we included all participants to minimize loss of power and to avoid inducing a potential bias. In agreement with the TSC, we used for all time points the wearable data points if participants wore the accelerometer at least one full day (≥540 minutes).
Otherwise, the data were considered missing.
The total missing activity data across all arms of the study at baseline was 0.5% (n=8) and at 12 showed that there were no differences in estimates of means and standard deviations. Both methods produced unbiased under missing completely at random conditions (with EM imputation sometimes performing slightly better) and hence their use is regarded as standard for intermittent missing days of activity data. 2,3 Day of the week, month, and year were included as indicator variables in the imputation model. Activity data will be imputed within each treatment arm separately. Standard clinical and demographic variables were included as potential predictors.

Predictors of missing outcome data at 12 and 24 months
In addition to missing the follow-up assessment, activity data were also missing because accelerometers were not worn sufficiently long enough. Two patients did not provide weight data at 24 months. An analysis of missing outcome data was therefore done in addition to missing attendance. A stepwise logistic regression with baseline variables, treatment arm, borough at baseline and potential predictors of missing outcome data at 12 and 24 months for weight and PA separately revealed that education, smoking status, PHQ-9 depression score and treatment arm are the most important predictors. However, only little variation was explained by the models (<2.5% pseudo-R 2 ). Education, smoking status, and PHQ-9 depression score were included as predictors of missingness in a sensitivity analyses to assess potential bias due to missing outcome data. Table S6 presents baseline characteristics of participants with missing data at 12-month followup and those who attended 12-month follow-up and participants with missing data at 24-month follow-up and those who attended 24-month follow-up.

Sensitivity Analyses Sensitivity analysis adjusting for imbalances on baseline characteristics
No imbalances were observed on any of the pre-specified baseline characteristics, therefore, a sensitivity analysis adjusting for these variables was not done for either of the primary outcomes.
A total of 14 sensitivity analyses were conducted for each (except for two occasions) of the primary outcomes:  1. Adjusting for partially nested random effect for therapist. 2. Adding therapist and general practice as random factors and exchangeable residual covariance matrix. 3. Removing potential outliers (removed n=12 for weight; n=6 for PA). 4. Only including patients with BMI>25kg/m 2 (analysed n=2298 for weight; n=2159). 5. Adjusting for treatment compliance (including patients in arms A/B that attended at least one intervention session; analysed n=2427 for weight; n=2305 for PA). 6. Adjusting for the delay in intervention start (continuous variable). 7. Adjusting for the unblinding of the research assistant at each follow-up appointment (binary variable). 8. Adjusting for the number of accelerometer wear days at baseline (>3 days, binary variable). 9. Adjusting for the number of accelerometer wear days at baseline (>5 days, binary variable). 10. Adjusting for the number of valid accelerometer wear days at baseline (≥540 minutes, continuous variable). Not completed for weight. 11. Adjusting for the number of accelerometer wear days at each follow-up appointment (continuous variable). Not completed for weight. 12. Adjusting for a BMI score <25 kg/m 2 at baseline (binary variable). 13. Adjusting for a QRisk2 score ≥20.0% at baseline (binary variable). 14. Adjusting for predictors (PHQ-9, smoking status, education) of missing outcome data.
Tables S9-S12 present the output of the primary and sensitivity analyses for each of the primary outcomes, following the numbering above. None of the above sensitivity analyses altered our conclusions for either of the primary outcomes.
The secondary comparison ('group versus individual') was assessed on a 5% significance level.

12-month follow-up
Individual -UC Adjusting for the unblinding of the research assistant at each follow-up appointment (binary variable). 8. Adjusting for the number of accelerometer wear days at baseline (>3 days, binary variable). 9. Adjusting for the number of accelerometer wear days at baseline (>5 days, binary variable). 10. Adjusting for the number of valid accelerometer wear days at baseline (≥540 minutes, continuous variable). Not completed for weight. 11. Adjusting for the number of accelerometer wear days at each follow-up appointment (continuous variable). Not completed for weight. 12. Adjusting for a BMI score <25 kg/m 2 at baseline (binary variable). 13. Adjusting for a QRisk2 score ≥20.0% at baseline (binary variable). 14. Adjusting for predictors (PHQ-9, smoking status, education) of missing outcome data.

Cost-effectiveness
The number and percentage of participants using specific services or groups of services are shown in Table S13. In the 12-month period prior to baseline assessment the use of all services was very similar between the three treatment arms. There were relatively few people attending A&E or day hospitals or being admitted as inpatients. About half of each group had outpatient contacts and nearly everyone had community contacts. This was particularly due to GP visits.
In the 12 months prior to the 12-month follow-up, around two-thirds of the group arm and threequarters of the individual arm received the relevant intervention. Visits to A&E and outpatient attendances had increased slightly from baseline but there were no clear differences between treatment arms. These patterns were again observed in the 12-month period prior to the second follow-up, although there was again a slight increase in the proportions having outpatient attendances.
For those with specific service contacts, the average number of contacts is shown in Table S14.
At baseline, community services were used more intensely than other services. The inpatient contacts refer to number of days in hospital. It can be seen that there are no major differences between arms. The data for the one-year follow-up show that the individual arm had slightly more intervention contacts than the group arm. Those admitted to hospital from the individual arm had more days in hospital than the other two arms. There were no clear differences between arms in the period up to the 24-month follow-up.
Service costs (including zero costs for non-users) were similar for inpatient care, outpatient attendances, and community contacts (Table S15). Costs of services did not differ markedly between arms, although inpatient costs were somewhat higher for the individual arm at the 12month follow-up and lower at the 24-month follow-up. The intervention cost was highest for those in the individual arm.
Compared to UC, total costs at baseline were on average £151 more for the group arm (95% CI=-£27 to £328) and £55 more for the individual arm (95% CI=-£96 to £203). The group arm costs were one average £95 more than those for the individual arm (95% CI=-£93 to £299).
After controlling for baseline in a regression model, the total costs in the 12-month period prior to the first follow-up were on average £89 more for the group arm than the UC arm (95% CI=-£274 to £390), and £409 more for the individual arm than the UC arm (95% CI=-£171 to £1133). The individual arm costs were on average £320 more than those for UC (95% CI=-£170 to £1133). By the second follow-up, and again controlling for baseline, the mean costs for the group arm were £82 more than for UC (95% CI=-£93 to £263) and £39 more for UC compared to the individual arm (95% CI=-£108 to £188). Costs were £121 more for the group arm than for the individual arm (95% CI=-£37 to £309).
Mean (SD) total costs over the whole 24-month follow-up period with year two costs discounted by 3.5% were £2071 (£3363) for the group arm, £2230 (£7645) for the individual arm, and £1852 (£3726) for UC. Compared to UC and controlling for baseline, the group arm had costs that were on average £172 higher (95% CI=-£237 to £599) and the individual arm had costs that were £352 higher (95% CI=-£309 to £1271). For cases where both cost and QALY data were available, the group arm had incremental costs of £173 compared to UC and the individual arm incremental costs of £356. These are the incremental costs subsequently used in the incremental cost-effectiveness ratios (ICERs).
Mean EQ-5D tariff scores were similar for each arm and did not change markedly over time (Table S16). Controlling for baseline utility and compared to UC, the group arm resulted in 0.0150 fewer QALYs (95% CI, -0.0388 to 0.0088) and the individual arm in 0.0039 more QALYs (85% CI, -0.0217 to 0.0294). For cases where both cost and QALYs were available, the group arm produced 0.0149 fewer QALYs than UC; and the individual arm 0.0064 more QALYs than UC. These figures are used as the denominator in the ICERs. Table S13. Number (%) of participants using services in 12 months prior to baseline and each follow-up.  The group arm was less effective than UC and more expensive. As such it was dominated.
Individual was more expensive and more effective. The ICER was £55,625 per QALY (£356 divided by 0.0064 QALYs). Uncertainty around the estimates are shown in the cost-effectiveness planes ( Figures S2 and S3). In Figure Figure S4 shows the cost-effectiveness acceptability curves where all three arms are compared for different values placed on a QALY gain. When a zero value is placed on a QALY, UC has by far the highest probability of being the most cost-effective option (in this situation, only costs are relevant and UC is less expensive). As the value on a QALY is increased, the probability that the individual arm is the most cost-effective option increases steadily while the other two arms see a fall in the probability that they are the most cost-effective. This is to be expected as the  individual arm is more expensive and more effective than the other options and as the effect (i.e. increased QALYs) is valued more it increasingly offsets the cost. However, at a value of £30,000 (above which NICE is likely to decide an intervention is not cost-effective) the individual arm has a 37.4% likelihood of being the most cost-effective option compared to 58.1% for UC. At this value, the group arm has a likelihood of 4.5% of being the most cost-effective option.

Figure S4. Cost-effectiveness acceptability curves.
In sensitivity analyses, the intervention costs were increased and decreased by 25% and 50% respectively to reflect different staff grades delivering the interventions. When intervention costs were reduced by 25%, the group arm had costs that were on average £188 more than for UC while costs for the individual arm were £395 more than for UC. The differences when intervention costs were increased by 50% were £204 and £434 respectively. Not surprisingly  these increased differences mean that the interventions are even less likely to be cost-effective compared to UC.
When intervention costs are reduced by 25%, the group arm has costs that are on average £157 more than for UC. With fewer QALYs, UC is still dominant. Individual is now £317 more than for UC resulting in a cost per QALY of £49,531. With a reduction of 50%, the group arm still has higher costs than UC (by £141) and so continues to be dominated while the individual arm has costs that are £278 higher resulting a cost per QALY of £43,438.