Results

In total, 261 participants, plus 261 study companions, were recruited to the open-label phase of the RADAR trial. Of these, 211 were randomised, with 105 allocated to the intervention group and 106 to the placebo group. Figure 4 presents the CONSORT flow diagram for the trial and summarises participant throughput from eligibility screening, through the open-label phase and randomisation and on to completion of the 3-, 6-, 9- and 12-month follow-ups, as appropriate. The diagram also shows the numbers of participants who did not meet the inclusion criteria.

FIGURE 4 The RADAR CONSORT flow chart. Reproduced with permission from Kehoe et al.79 © 2021 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license. (Figure not reproduced here; see the original article.)

Baseline characteristics and comparisons

The baseline demographic characteristics of the study sample are summarised in Table 5. This table shows that the randomisation process was very effective in achieving balance between the intervention and placebo groups in standard characteristics of age (including different age strata), gender, ethnicity and the number of years of education. Similarly, there was excellent balance between the intervention and placebo arms in the distribution of people who were hypertensive or normotensive. The same could be said for various readings of their sitting and standing systolic and diastolic BP in the case of both participants who were in the mean normotensive ranges of BP specified in the protocol and those who had hypertension and well-controlled BP.

A summary of the baseline characteristics of the study population according to dementia-related outcomes is presented in Table 6. As was evident with the distribution of the general demographic characteristics and general health information, including BP and hypertension status, the randomisation process similarly performed very well in achieving balance between the intervention and placebo arms for the number of years since diagnosis and the status of participants with respect to taking dementia treatments. There was parity between the intervention and placebo arms for various baseline MRI measurements serving as the basis of the primary outcome measure and related to whole-brain structure and volume, as well as measures related to hippocampal volume. Although there was a slight suggestion of imbalance between the trial arms with respect to Scheltens scores of baseline atrophy, our analysis had already included adjustment for this because Scheltens scores were a minimisation variable in our randomisation.

It is also evident from Tables 5 and 6 that there appeared to be excellent balance between trial arms for the various variables needed for the secondary outcomes of the study, including the baseline scores for cognitive assessments, as well as assessments of function and quality of life. These are in addition to the data already mentioned on BP and hypertension status. The imaging data related to subgroups of participants that were used to undertake exploratory analyses of intervention-based change in WMH volume and CBF will be presented and discussed in Secondary outcomes.

Losses to follow-up

We successfully enrolled 211 participants and an equivalent number of study companions (93% of the original recruitment target of 228). Of the 211 subjects recruited, 197 (93% of those enrolled) successfully completed the study, and complete primary outcome data were available for 173 (88% of those randomised). The 7% discontinuation rate was favourable and the total loss of primary outcomes (12%) was just below the 20% we assumed as part of the original power calculations. The impact that recruitment below the original target of 228 participants had on the statistical power of the study was mitigated to some extent by the lower than expected dropout rate and the fact that we secured slightly more primary outcomes than we had assumed. This allowed the study to retain 82% statistical power for our analyses rather than the 84% power that our original assumptions and power calculations for the study were based on.

Adherence to intervention

The study benefited from high levels of adherence to the intervention. This was reflected by a total median adherence for all participants of 96% [interquartile range (IQR) 92–98%], which was almost identical between the two arms of the study [median compliance of 96% (IQR 93–98%) in the control arm, median compliance of 96% (IQR 91–98%) in the intervention arm]. In accordance with our predefined criteria for compliance we determined that 88% of study participants were ‘compliers’. This reflected small differences in the numbers of ‘compliers’ in the control arm (90%) and in the intervention arm (86%).

Primary outcome

The baseline and 12-month vMRI follow-up data for the primary outcome are presented in Table 7. Both the intervention arm and the placebo arm showed some brain atrophy between baseline and the 12-month follow-up. Mean TBV fell from 1022 ml (SD 99 ml) to 1002 ml (SD 98 ml) in the intervention arm and from 1036 ml (SD 111 ml) to 1018 ml (SD 113 ml) in the placebo arm. However, overall the levels of atrophy appeared comparable across both trial arms. When the 12-month cross-sectional brain volumes were analysed, the adjusted mean difference between arms for the total study population (n = 171) with adequate follow-up data amenable to analysis [adjusted for minimisation variables (age and Scheltens score) and baseline brain volume and recruitment centre] was –2.29 ml (95% CI –6.46 ml to 0.89 ml; p = 0.136). This was not significant.

Further analysis of TBV according to boundary shift interval (BSI), which is more sensitive for the measurement of brain atrophy,89 shows that the mean change in TBV-BSI in the placebo arm (19.1 ml, SD 10.3 ml) was almost identical to that in the intervention arm (20.0 ml, SD 10.8 ml). The adjusted mean difference for TBV-BSI between arms of 1.23 ml (95% CI –1.72 to 4.19 ml) was also non-significant (p = 0.41).

Secondary outcomes

A number of secondary outcomes were investigated:

1. We first investigated rates of AD progression as assessed by changes in cognitive assessments [i.e. the 11-item ADAS-Cog,96 the Neuropsychiatric Inventory (NPI)97 and the MMSE98], as well as an examination of change to quality of life (DEMQOL, DEMQOL-proxy)99 and ability to carry out activities of daily living (BADLS100)]. The associated data for baseline and 6- and 12-month follow-up are presented in Table 8.

From Table 8 it appears that, with each successive follow-up visit, all scores worsened in both the intervention and the placebo arms.

It was, therefore, important to examine change over time and so we compared the 12-month assessment scores, conditioned on baseline, for each of the secondary outcome measures. In Table 9 comparisons of the adjusted mean differences between baseline and end of study between the trial arms are summarised. There was little evidence to suggest that losartan afforded any clinical benefit according to the various secondary outcome measures assessed.

1. We also sought to investigate change in the volume of WMH that was measured in our MRI protocol using T2-FLAIR, given that it may predict a 1-year cognitive decline.24 This was possible in only a subgroup of the study sample (15 of the 23 sites), which equated to a total of 105 participants split almost equally between the placebo (n = 51) and intervention (n = 54) arms. The data on WMH volume are presented in Table 10 for baseline and 12-month follow-up, and regression analysis shows that there was no evidence of difference longitudinally in the WMH volumes after 12 months of treatment (p = 0.697).
2. In a small subset of sites (9 of the 24 sites) there was scope to undertake the additional MRI protocol sequences to explore the level of change in CBF using ASL. Given that ASL has not been used widely in dementia trials to date and thus far mainly in single-centre formats, the multicentre format in RADAR was also highly exploratory to gain insights to similar approaches in future studies. A summary of all of the scanner and ASL sequences implemented is given in Table 11.

The retrieved ASL data underwent QA assessment and were processed using the ExploreASL software package104,105 to generate quantitative CBF maps. ASL quality control-checked ASL CBF images were then grouped according to the spatial CoV values. Spatial CoV is a ratio of the SD of CBF to the mean of CBF and has been identified as an additional new approach that allow estimates (proxy) measurements of CBF.106,107

The initial groups were assigned to groups using the following differentiating criteria and stringency towards CBF measurement:

1. predominant CBF contrast: CoV values ≤ 1
2. predominant vascular contrast: 1 < CoV values < 1.5
3. predominantly artefacts CoV ≥ 1.5.

Group 1 data were the most stringent and accurate for the measurement of CBF. Group 2 data, although not as stringent as those in group 1, could serve as a proxy measure for CBF and thus could be used to increase the sample size in addition to group 1 measurements. Group 3 data, because of their largely artefactual nature, were excluded from analyses. All images were then inspected visually by a trained researcher and their categorisation was corrected if needed before further analyses were undertaken.

Owing to the nature of the study design and the intention to measure change over time, we excluded not only data from participants with at least one data set in group 3 (i.e. mainly artefactual), but also data from participants with at least one scan in group 2 (i.e. predominantly vascular contrast) – this participant’s data (even if the other scan was group 1) were assigned to the vascular group.

This approach yielded two data sets:

1. one suitable for CBF perfusion analysis (images with predominant CBF contrast only)
2. one suitable for spatial CoV analysis (images with combinations of predominant CBF contrast and predominant vascular contrast images).

A summary of the distribution of the number of participants per site is presented in Figure 5.

FIGURE 5 Distribution of ASL data according to RADAR site. Site 12 had two scanners in use throughout the study; however, with different specifications and hence was treated separately. (Figure not reproduced here; see the original article.)

An initial look at the data acquired showed large variation in the interpretation of initial ASL protocols; this varied not only between sites with different MRI scanners, but also between sites with comparable scanners. Following the application of our QA approach, we found that this also had a significant effect on the final numbers of available scans that would be amenable to analysis. These involved group 1 scans (which would provide the greatest specificity for CBF) and group 2 scans (which would provide less specificity for CBF but would allow proxy estimations of arterial transit time).

Given the variability of data, we opted to explore the feasibility of a larger combined analysis for the study by first piloting analysis on data from the subset of sites with the largest number of collected viable ASL scans: these sites were 19 and 10.

Scanning site 19 yielded data from 31 participants (as it served more than one recruitment site). The distribution of scans across treatment arms was relatively equal. Based on the QA, data from only 10 of these 31 participants were amenable to CBF analysis (data set 1), whereas data from 23 individuals fulfilled data set 2 criteria and could be analysed as a proxy estimate for arterial transit time. From this site, data from eight participants were excluded because they were deemed to be predominantly artefactual and thus not amenable to analysis. Similarly, at site 10, which had the second largest number of scans available (n = 17 participants), data from only seven participants were amenable to CBF analysis (data set 1), but data from 15 participants were amenable to the arterial transit time analysis by proxy (data set 2) using the CoV approach.

We were then in a position to explore the differences between treatment arms on an individual site basis both at baseline and at 12-month follow-up. Figure 6 presents the data for site 19 and site 10 (in descending order of the number of scans suitable for CoV analysis). We opted to use the validated106,107 CoV analysis approach on data set 2 scans in each case for these exploratory analyses because it provided more data points and thus greater power than the use of data set 1 scans (Table 12).

FIGURE 6 Exploratory comparisons of CBF using spatial CoV analysis. (a) Site 19: grey matter spatial CoV distribution from the baseline and 12-month follow-up ASL scans acquired using a 3-T Siemens Prisma with 2D EPI PASL PICORE Q2T; and (b) site 10: grey matter spatial CoV distribution in the baseline and 12-month follow-up ASL scans acquired using a 3-T Siemens Skyra with 3D GRASE pulsed ASL FAIR QII. 2D, two-dimensional; 3D, three-dimensional; EPI, echo-planar imaging; GRASE, gradient and spin echo; FAIR, flow-sensitive alternating inversion recovery; PASL, pulsed arterial spin labelling; PICORE, proximal inversion with control of off-resonance effects. (Figure not reproduced here; see the original article.)

The information in Figure 6 is presented according to treatment arm and based on data collected during each of the baseline and follow-up visits. Using spatial CoV analyses it would appear that there are no obvious patterns of change between baseline and follow-up in either treatment arm or in either site.

A multiple regression was carried out to test if spatial CoV baseline values and treatment (placebo or intervention) predicted spatial CoV follow-up values for site 19, where the largest number of data survived QA. The results of the regression indicated that the model explained 63.0% of the variance and that the model was significant [F(2,19) = 16.159; p < 0.0005]. It was found that spatial CoV baseline values significantly predicted spatial CoV follow-up values (β₁ = 1.085; p < 0.0005), whereas the assignment of treatment did not add statistically significantly to the prediction (β₂ = 0.163; p = 0.363).

Similar exploratory comparisons were examined (data not shown) for both sites using the more underpowered data set 1. This showed similar patterns but similarly there were no evident changes between baseline and the 12-month follow-up for either treatment arm but there were fewer participants in each (n = 10 for site 19, five in each arm, and n = 7 for site 10, three in the intervention arm and four in the placebo arm.

The change in BP over time was deemed an important consideration in our study given the normal prescribed purpose of the intervention (i.e. to lower BP in the treatment of hypertension). It was also deemed an important positive control for the intervention to demonstrate (1) that the intervention was biologically active (i.e. observable reductions in BP in the intervention arm would be proof the medication was active) and (2) that variations in BP might be associated with the volume of WMH and cerebrovascular health in general and, thus, CBF. Thus, we were interested in the difference in change in BP over 12 months between trial arms and whether or not this had any influence on the primary and secondary outcomes.

Mean systolic and diastolic BP at baseline and 12-month follow-up are presented in Table 13. As described, the mean BP appeared comparable at baseline between the intervention and placebo arms. A slight reduction in mean systolic BP, but not in diastolic BP, was evident in the intervention arm.

Yet on comparison of change in BP over the course of the follow-up period, it is evident that there are statistically significant differences in both systolic and diastolic BP between the intervention and placebo arms (p < 0.001) such that participants in the intervention arm had lower BP over the follow-up period (Table 14).

Given that there were no apparent differences between the intervention and placebo arms of the study at baseline and 12-month follow-up, it was also important to investigate, in line with previous studies,49–51,74 whether or not differences in BP may account for any variation in the primary and secondary outcomes. Correlation analyses showed that there was no strong correlation between systolic or diastolic BP and the primary outcomes, nor was there any evidence of association for any of the assessment-based cognitive, functional or behavioural outcome measures. Although data on the volume of WMH intensities were available for only a subgroup of RADAR participants (45% of randomised participants), we still explored the association with systolic and diastolic BP. Similarly, there was no evidence of any association between volume of WMH and change in systolic and diastolic CBF.

We had also intended to measure the levels of association between the primary outcome of brain atrophy and cognitive decline. This was because brain atrophy has been reported to serve as an objective proxy for cognitive decline in dementia, as previously described. Similarly, it was important to examine the possible association between the rate of brain atrophy and rate of change in our other secondary functional and behavioural outcome measures, predominantly to help inform any future studies where the use of cognitive, functional and behavioural measures would be preferable to using MRI-based outcomes, which, as has been described, can introduce additional complexity and may be more limiting to the identification of recruitment centres.

Table 15 presents the correlations between the primary outcome and the number of secondary cognitive, behavioural and functional outcomes that were assessed. There were variable levels of correlation between the variables presented. The strongest (negative) correlation (–0.56) was between ADAS-Cog scores and MMSE scores. There was also evidence of weak negative correlations between brain volume and ADAS-Cog score (–0.39) and between the functional outcomes DEMQOL-proxy and BADLS (–0.34). By contrast, NPI and BADLS were weakly positively correlated (0.41), as were ADAS-Cog and BADLS (0.36).

One of our secondary outcome measures was an examination of levels of drug compliance and tolerability across the main randomised phase. As already reported, there were high levels of compliance (88%) across all participants, with compliance only slightly lower in the intervention arm (86%) than in the placebo arm (90%) – a difference that was consistent with chance variability.

We also sought to examine the effect of the intervention with respect to SAEs and AEs reported during the study. There was no significant difference between the number of SAEs reported in the intervention arm (n = 22) and the placebo arm (n = 20). A summary of the recorded SAEs is provided in Appendix 2, Table 24. One SAE was reported in the open-label phase and is also summarised in Appendix 2, Table 24.

A summary of the reported AEs over several follow-up visits from the baseline visit is provided in Table 16. As is apparent from Table 16, there were no obvious or statistically significant differences between the intervention and placebo arms in the numbers of AEs. Indeed, the proportion of participants reporting any AE at 12 months was 6% lower in the intervention arm than in the control arm, although this difference was not significant.

Finally, we also explored the results of the primary analysis according to baseline brain volume and according to whether or not participants were taking antidementia drugs. Regarding the analysis to explore the possible interaction between treatment and baseline brain volume, there was no evidence of an interaction (interaction coefficient –0.002, 95% CI 0.039 to 0.036; p = 0.921). Similarly, exploratory analysis of a possible interaction between the primary analysis and taking antidementia drugs revealed no evidence of an interaction (interaction coefficient 8.21, 95% CI –10.18 to 26.59; p = 0.379).

As with all the secondary analyses, the study was powered principally for the primary analyses and so all of these observations should be viewed with appropriate caution.