Results

Parts of this chapter are reproduced from Frost et al.1 © The Author(s) 2019. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The text below includes minor additions and formatting changes to the original text.

Recruitment

Recruitment of general practices

As outlined in General practice recruitment and selection, the West of England and Wessex CRNs invited general practices to express an interest in taking part in the trial. A total of 83 practices expressed an interest, of which 64 performed an initial search (Figure 1). On performing the search, 18 practices were excluded because they did not have sufficient numbers of patients and 16 other practices did not progress further. Practices were recruited between July 2018 and May 2019 (Figure 2). After having conducted contractual agreements and consent processes with practices, we randomised 30 general practice sites. These comprised 32 general practices, owing to one collaboration of three practices operating a hub-and-spoke model and sharing nurse resource, and therefore randomised as a single site for the trial.

FIGURE 1 The CONSORT flow diagram. (Figure not reproduced here; see the original article.)

FIGURE 2 Cumulative number of practices randomised over time. (Figure not reproduced here; see the original article.)

Characteristics of participating general practices

At the time of recruitment, practices provided estimated list sizes; these ranged from 7600 to 48,623 patients (mean 19,576) (see Appendix 1, Table 39). Although routinely available data on general practices in this region do not allow us to identify and merge smaller practices operating as a hub-and-spoke model as participating practices have, participating practices reflect some of the larger practices in the region [regional median practice size in June 2019: 9440 patients (IQR 6429 to 13,239)]. Intervention practices, on average, recruited 31 patients each and control practices recruited 43 patients each, on average.

Patient-level socioeconomic data were not available for all registered practice patients. General practices were therefore characterised using area-level deprivation, which was estimated for the hypothesised catchment area for the practice (estimated as the MSOA). Deprivation scores ranged from 4.22 to 33.62, and scores were slightly higher in the usual-care arm [mean 16 (SD 8.39)] than in the intervention arm [mean 11 (SD 5.00)], indicating higher levels of socioeconomic deprivation.

Patient screening and randomisation of sites

When participating practices conducted a detailed database search, as outlined in Patient screening and invitation, they identified 7872 potentially eligible patients. A random selection of 160 patients from two practices were not screened as their practices were very large and they had already committed to screening a set number of patients, and 97 patients were not screened owing to practice capacity. Manual screening by GPs of the remaining 7615 patients used the eligibility criteria outlined (see Appendix 1, Table 38). Reasons for ineligibility are outlined in Table 3.

Once general practices completed their manual screening of patient lists, sites were randomised. The randomisation process, which was minimised by centre, practice size and area-level deprivation of the practice, resulted in 17 sites being randomised to deliver the intervention (3171 eligible patients) and 13 sites being randomised to deliver usual care (2397 eligible patients). Practices themselves, however, were not informed of their allocation until patient screening had been completed and the invitation letter mail-out uploaded (see Patient screening and invitation).

Patient recruitment

Letters of invitation were sent to 4808 of the 5568 eligible patients (intervention practices, 2516; usual-care practices, 2292). Randomly selected eligible patients were not included in this mail-out (intervention practices, n = 655; usual-care practices, n = 105), as a maximum number was specified for mail-out from individual sites to avoid disproportionate recruitment at larger practices (see Patient screening and invitation). Expressions of interest were returned by 2300 patients [intervention practices, n = 1234 (49%); usual-care practices, n = 1066 (47%)], of whom 1671 [intervention practices, n = 919 (74%); usual-care practices, n = 752 (71%)] confirmed their interest in participating. Using available age data on those who were sent letters of invitation (Table 4), we observed that those who expressed an interest in participating in the trial were similar in age to those who said that they were not interested (mean age 68.7 and 71.0 years, respectively). Those who did not get in touch to confirm their interest were much younger (mean age 62.7 years).

Those who expressed an interest in participating were approached for screening by the CRN team. A total of 71 patients were not contactable at this stage; of those whom the CRN teams were able to reach, 281 were deemed ineligible, 26 were unwilling to proceed further and 1293 were found to be eligible. Men who were eligible at CRN screening and were willing to participate in the trial tended to be slightly younger than those eligible but unwilling to participate, although comparable to those ineligible at screening (mean age 68.3 years vs. 74.1 years vs. 70.7 years) (Table 5).

Patients who were found to be eligible at the CRN screening stage were invited to participate in the trial and 1077 consented [intervention practices: n = 524 (74%); usual-care practices: n = 553 (94%)]. The men who were eligible and consented to participate tended to be slightly older than those who were eligible and did not consent (68.7 years vs. 66.5 years, respectively) (Table 6).

Patient recruitment over time is illustrated in Figure 3 and recruitment by practice is presented in Appendix 1, Table 40.

FIGURE 3 Cumulative patient recruitment over time. (Figure not reproduced here; see the original article.)

The CONSORT flow diagram (see Figure 1) describes how and when participants were lost to follow-up between consent and 12 months. Ultimately, 154 participants were lost to follow-up and 915 of the recruited 1077 participants (85.0%) provided primary outcome data at 12 months.

Pilot phase

The 4-month internal pilot phase of the trial was successfully completed in November 2018. A total of 142 participants were recruited, exceeding the target of 120 participants. Sixteen practices had formally agreed to take part in the trial during the pilot phase, which, although slightly short of the target of 18, was sufficient for progression to the main phase of the trial. The key changes that were made to the conduct of the trial between the pilot phase and the main phase of the trial were an increase in the maximum number of patients invited by a single site from 150 to 220, and confirmation that 30 sites would be required to maintain the trial power calculation in the light of variation in the number of patients recruited per site. The decision was also taken to conduct only a single invitation mail-out from each site, rather than inviting additional patients in a second mail-out to achieve site targets as necessary, to avoid introducing any bias post randomisation. Although screening of all patients for both mail-outs would have been conducted before randomisation, rescreening would have been required to check patient status before a later, second, mail-out, when sites would have been aware of their allocation.

Baseline data

The baseline characteristics of the participants of the TRIUMPH trial are presented in Table 7. The mean age was 69 years in the intervention arm and 68 years in the usual-care arm, although the trial included some younger men who were in their thirties. In both arms, participants were overwhelmingly white, with only 13 participants overall describing themselves as being from another ethnic group. In both arms, most men were married or in civil partnerships. Socioeconomic status was determined using area-level IMD based on the home postcode; participants were from all five quintiles of deprivation, although the proportion from the least deprived quintile was in both arms higher than the 20% we would expect if participants were equally distributed by quintile (intervention arm, 49.0%; usual-care arm, 42.9%). Participants self-reported height and weight; mean values of both were comparable between treatment arms.

Electronic medical records were used to gather further information on the clinical characteristics at baseline. As not all practices were able to provide such data, more data were missing in the intervention arm. Approximately one-third of participants had no record of comorbidities, one-third had one comorbidity and one-third had more than one comorbidity. The distribution was comparable between treatment arms. The median number of GP consultations in the 12 months before baseline was the same in both arms (4 consultations), but the proportion with a referral to urological services in that period was slightly lower in the intervention arm than in the usual-care arm (intervention arm, 2.93%; usual-care arm, 3.49%).

We also used primary care EMRs to assess recent urine and kidney function tests. Few men in either arm reported having undergone urine analyses in the 6 months before baseline (intervention arm, n = 79; usual-care arm, n = 52), and few of these had abnormal test results (intervention arm, 1.27%; usual-care arm, 3.85%). Kidney function testing was more common (intervention arm, n = 170; usual-care arm, n = 215), and the most recent eGFR measure was similar in both arms. The majority of men in both arms who had such testing had chronic kidney disease stage 1 or 2 (intervention arm, 67.1%; usual-care arm, 71.63%).

International Prostate Symptom Score, the primary outcome measure, was measured at baseline; the total score was slightly lower in the intervention arm (mean total score 13.62 points) than in the usual-care arm (mean total score 14.59 points). When examining individual symptoms, the symptoms with the highest mean scores were frequency, nocturia and urgency. The IPSS-QoL scores were very similar in both arms (intervention: 3.47 points; usual care: 3.55 points).

The ICIQ-UI-SF total symptom scores were similar in both groups (intervention arm, mean of 3.6, usual care arm, mean of 3.9). When asked when men leaked urine, the most common responses were before getting to the toilet (intervention, 39.2%; usual care, 42.86%), and when finished urinating and dressed (intervention arm, 33.46%; usual-care arm, 37.07%). Approximately one-third of men in each arm said they never leaked (intervention arm, 35.37%; usual-care arm, 29.29%). Men in the intervention arm completed a bladder diary at baseline to help inform intervention delivery; they most frequently reported nocturia (85.06%) and urgency (71.79%). The EQ-5D-5L data were comparable in both arms, as were the B-IPQ data.

To explore whether or not the disease-specific IPSS-QoL score yields information different from the EQ-5D-5L, we plotted the baseline values of one versus the other (Figure 4). The correlation coefficient of –0.22 suggests a weak to no linear association between the two measures, that the IPSS-QoL score yields information different from that of the EQ-5D-5L and that the use of both is not redundant.

FIGURE 4 Relationship between the EQ-5D-5L and IPSS-QoL scores at baseline. (Figure not reproduced here; see the original article.)

The IPSS does not capture all urinary symptoms; leakage is captured only in the ICIQ-UI-SF. Therefore, we explored how IPSS at baseline correlated with ICIQ-UI-SF scores at baseline. A scatterplot of the two variables is presented in Figure 5. There was little evidence that the two measures were strongly correlated (p = 0.28), suggesting that those with a higher IPSS do not necessarily have higher ICIQ-UI-SF scores, and vice versa, thus confirming the value of both measures.

FIGURE 5 Baseline IPSS and baseline ICIQ-UI-SF scores. (Figure not reproduced here; see the original article.)

As well as using EMRs to identify GP consultations and clinical test results, we also explored the prescriptions that patients were issued in the 3 months pre baseline with a potential indication for treating LUTS (Table 8). Relevant medications for LUTS are alpha-adrenergic antagonists (alfuzosin, doxazosin, tamsulosin) and phosphodiesterase-5 inhibitors (tadalafil) to treat bladder outlet obstruction, and antimuscarinics (oxybutynin, solifenacin, tolterodine, trospium) and beta-3 agonist (mirabegron) to treat overactive bladder syndrome. Approximately 40% of men had at least one prescription of this type in this period. The most common prescription by far was tamsulosin, prescribed to 30.5% of men in the intervention arm and 34.01% of men in the usual-care arm. Oxybutynin and doxazosin were the next most commonly prescribled treatments; both were more common in the usual-care arm.

Intervention delivery

Seven out of 16 West of England sites and 10 out of 14 Wessex sites were randomised to the intervention arm. The intervention was delivered by the TRIUMPH trial research nurses at nine sites (three West of England, six Wessex), by practice nurses/HCAs at six sites (three West of England, three Wessex) and by both at two sites (one West of England, one Wessex) (Table 9). Overall, compliance with the intervention visit and follow-up contacts was very high (see Table 9), as was adherence to patient preferences in the format of their follow-up visits (Table 10).

Participants were directed to different sections of the intervention booklet based on their baseline data, including the bladder diary. We recorded the sections of the booklet that each participant was referred to in the trial case report forms, and compared the baseline IPSS and ICIQ-UI-SF scores between groups (see Appendix 1, Table 41). The most commonly referred booklet sections related to ‘drinks and water intake’, ‘controlling an urgent need to pass urine’ and ‘reducing sleep disturbance caused by needing to pass urine’.

Protocol deviations

No serious protocol deviations were experienced during the trial. Protocol deviations are summarised in Table 11 and detailed in Appendix 1, Table 42.

Follow-up rates and International Prostate Symptom Score data completeness

Follow-up rates were monitored by the practice over the trial period and are described in Table 12. All had at least 80% follow-up for the 12-month visit. Completeness of the primary outcome (IPSS at 12 months) ranged from 75.8% to a high of 95.5%. Overall completeness of the primary outcome was > 80% in both arms.

Twenty-one participants withdrew from the trial, 17 from the intervention arm (Table 13). The most common reasons for withdrawal were ill health and not having sufficient time/finding the trial too inconvenient/not wanting to continue. The nature and level of individual withdrawals from the trial are outlined in Appendix 1, Table 43.

Data quality

Every effort was made to ensure that outcome data were collected at the appropriate month from baseline. In both arms, 12-month questionnaires were completed near the follow-up time point as intended (mean number of months since baseline: intervention arm, 12.30; usual-care arm, 12.18).

The pattern of missing data was explored by identifying variables recorded at baseline that were associated with ‘missingness’ of IPSS at 6 and 12 months (primary outcome). Appendix 1, Table 44, presents the frequency and proportion of missing IPSS data at 6 months by baseline characteristics. We observed that missingness seems to be associated with marital status, as those with complete data were more likely than those with missing data to be married. There was no evidence of a difference in missingness by baseline IPSS, but those with missing IPSS at 6 months tended to have slightly higher B-IPQ and ICIQ-UI-SF scores. These differences, however, were not observed when looking at missingness of IPSS at 12 months (see Appendix 1, Table 45).

Statistical outcomes and estimation

Primary outcome

The mean IPSS at follow-up was lower than at baseline in both treatment groups. When comparing groups, scores at follow-up were slightly lower in the intervention arm [mean 11.60 (SD 6.21)] than in the usual-care arm [mean 13.88 (SD 6.84)], suggesting that symptoms were worse in the usual-care arm (Table 14). After adjusting for baseline IPSS and minimisation variables, the between-arm difference was –1.81 (95% CI –2.66 to –0.95; p < 0.001), strongly suggesting a treatment effect. The treatment effect was similar when adjusting only for baseline values and not variables used in the randomisation.

A summary of symptom-specific scores for the IPSS and ICIQ-UI-SF at each time point is presented in Appendix 1, Table 46.

Secondary outcomes

Urinary symptoms over time

International Prostate Symptom Scores were studied at 6 months as a secondary outcome and, as with the primary outcome analysis, there was evidence that there was a greater improvement in symptoms in the intervention arm than in the usual-care arm (Table 15). The treatment effect, however, was slightly smaller than for the primary outcome (–1.68, 95% CI –2.34 to –1.02; p < 0.001).

Similar results were observed when a repeated-measures approach was used to study IPSS using both 6- and 12-month follow-up data (Table 16).

Trajectories in IPSSs over time were also described by tabulating the number and percentage of men who move between IPSS categories between baseline and 6 months or between 6 and 12 months (see Appendix 1, Table 47). Nearly half of all men (47.5%) stayed in the same category and 37.1% improved over the 12 months. Improvement was slightly more common in the intervention arm.

Incontinence

Patient-reported incontinence symptoms were assessed using the ICIQ-UI-SF at 6 and 12 months. At both time points, scores were higher, suggestive of worse symptoms, in the usual-care arm (Table 17). After adjusting for baseline ICIQ-UI-SF scores and variables used in the minimisation, scores were slightly lower in the intervention arm than in the usual-care arm at both time points (difference in means at 6 months: –0.53, 95% CI –1.02 to –0.04; p = 0.033; difference in means at 12 months: –0.74, 95% CI –1.15 to –0.33; p < 0.001).

Quality of life

Lower urinary tract symptom-specific quality of life was assessed using the IPSS-QoL measure at 6 and 12 months. Scores range from 0 to 6, with higher scores reflecting worse quality of life. At both time points, scores were slightly higher in the usual-care arm; this was also observed in models adjusting for baseline scores and variables used in the randomisation (Table 18).

Lower urinary tract symptom perception

Patient perception of LUTS was assessed at 6 and 12 months using the B-IPQ, where scores range from 0 to 90, with higher scores reflecting a more threatening view of their symptoms. The mean scores at baseline were 38.7 in the intervention arm and 39.4 in the usual-care arm. A greater decrease was seen in the intervention arm at 6 and 12 months (Table 19). At 6 months, the between-group difference after adjusting for baseline values and minimisation variables was –5.34 (95% CI –6.69 to –3.99); at 12 months the difference was slightly smaller (–4.78, 95% CI –6.31 to 3.25).

Secondary care referrals

Overall, 7.6% of men received a referral to secondary care (urology) during the 12-month follow-up period. There was little difference between treatment arms after adjusting for minimisation variables and whether or not a referral had been made pre baseline (odds ratio 0.91, 95% CI 0.51 to 1.62) (Table 20).

Statistical ancillary analyses

Subgroup analyses

We characterised the nature of LUTS at baseline for participants by taking the ratio of the IPSS voiding to storage scores. Including this as a continuous interaction term in our model of IPSS at 12 months provided no evidence that the treatment effect differed according to this ratio (p = 0.971). Similarly, when we distinguished between those who received the intervention via a TRIUMPH nurse (n = 249) and those who received it via a practice nurse (n = 190), we found no strong evidence that the treatment effect differed (p = 0.387). There was, however, very weak evidence (p = 0.094) that the treatment effect differed by how intervention participants preferred follow-up by the clinical team (telephone, n = 310; text or e-mail, n = 210).

Subgroup analysis by the number of contact visits received was complicated by the fact that there was little variability in the data (Table 21). Only five intervention participants had only one follow-up contact, 25 had two follow-up contacts and the rest had three. Unsurprisingly, the results for those who received three follow-up contacts reflects the primary analysis results.

A post hoc subgroup analysis was conducted to assess whether or not treatment effects for the primary outcome (IPSS at 12 months) and IPSS-QoL (12 months) differed depending on whether the data were collected before or during the COVID-19 pandemic. The results are summarised in Tables 22 and 23. We observed no strong evidence that the treatment effect on IPSS-QoL differed by COVID-19 period (p = 0.383), although there was some difference in the primary outcome (p = 0.010), in that the treatment effect was slightly greater in the pre-COVID period. These results must be interpreted with caution, however, as these are exploratory and the study was not powered to detect such effects.

Sensitivity analyses

Imbalances between treatment groups at baseline

There was no evidence of imbalance between the groups in terms of baseline characteristics. As a result, the planned sensitivity analysis additionally adjusting for imbalance was not conducted.

Clustering by nurse/healthcare assistant

The primary analysis was repeated using a practice + nurse cluster to allow for clustering by nurse/HCA at a given practice. The results for this model are presented alongside the original primary outcome analysis in Table 24. Accounting for this clustering slightly reduces the treatment effect from an adjusted mean difference of –1.81 to –1.79, but there remains strong evidence that this differs from the null.

Per-protocol analyses

Five per-protocol analyses were prespecified based on varying ‘amounts’ of the intervention delivered to the participant. The vast majority of participants in the intervention arm received the intervention as intended (Table 25).

As the vast majority of participants received the full (or near full) intervention, some of the analyses were restricted to a very small number of participants, and thus were extremely underpowered. The fifth per-protocol analysis was restricted to those who received the full intervention as intended and included 344 intervention participants (Table 26). The treatment effect in this analysis was slightly larger than that observed in the ITT analysis.

A CACE analysis was planned, but as only nine participants in the intervention arm were deemed non-adherent according to our prespecified definition, the analysis was not conducted.

Excluding participants who were later found to be ineligible

Three participants were recruited to the trial and began follow-up before it was clear that they were, in fact, ineligible for the trial. We repeated the primary outcome analysis excluding these three participants and found that this made no difference to the primary analysis, as shown in Appendix 1, Table 48.

Impact of COVID-19

Half (50.5%) of primary outcome data for the intervention arm were collected pre COVID, compared with only 36.2% of the data from usual-care practices. When comparing the mean IPSS at 12 months pre COVID with the data collected during the pandemic, we see that scores in the usual-care arm were higher pre COVID (Table 27). There was also some evidence that the rate of missing primary outcome data was slightly higher during the COVID-19 pandemic (17.5%) than before it (9.4%).

A sensitivity analysis adjusting for whether the primary outcome was collected pre COVID or post COVID showed no difference from the primary analysis (Table 28).

Sensitivity analyses to examine the impact of missing data

Table 29 presents the results of the complete-case ITT analysis using 887 participants alongside analyses for which missing data were imputed using three different methods. In the first method it was assumed that those for whom data were missing had the lowest possible symptom scores (best-case scenario); this analysis yielded very similar results to the primary complete-case analysis. In the worst-case scenario, it was assumed that those in whom data were missing had high symptom scores, equal to the mean plus 2 SDs [12.78 + (2 × 6.84)]. The difference in means was smaller than that observed in the ITT and best-case scenario, but remained greater than the null (p = 0.017). In the multiple imputation analysis (MICE), the treatment effect was slightly smaller than in the complete-case analysis, yet still indicated a greater improvement in the intervention arm compared with the usual-care arm.

Post hoc sensitivity analysis allowing for clustering within general practice groups

In the case of the general practices that formed a single collaboration, we performed a sensitivity analysis allowing for clustering at the collaboration level (see Appendix 1, Table 49). The results, however, were very similar to the primary analysis, although the estimated ICC was slightly smaller.

Adverse events

Using EMRs, we sought to identify whether patients had LUTS-related adverse events or other urinary ‘expected’ adverse events. We found that a low number of patients reported such events, and reporting was similar in both arms (Table 30).

When considering all SAEs identified through the trial SAE reporting procedures (regardless of whether they were related to LUTS), we found that there were 94 SAEs in total (47 in each arm) and there were three deaths (two in the intervention arm and one in the usual-care arm). All but two SAEs that did not result in death were resolved (Table 31). Individual SAEs are outlined in Appendix 1, Table 50. All were unrelated to the intervention, with the exception of five that were deemed unlikely to be related to the intervention.

Statistical results summary

The trial was successful in meeting general practice site recruitment targets, and patient recruitment exceeded the updated sample calculations. The treatment groups were broadly balanced on baseline characteristics, but the primary outcome measure (IPSS) was slightly lower in the intervention arm than in the usual-care arm (intervention arm, 13.62; usual-care arm, 14.59).

The intervention was successfully delivered: 98% of men received their intervention booklet and 90% had all three contacts forming the intervention.

The primary outcome analysis found that there was a reduction in IPSS, suggesting improved symptoms, between baseline and 12 months in both arms, but the reduction was greater in the intervention arm (–1.81, 95% CI –2.66 to –0.95). The difference was smaller than the difference of 2 points in overall IPSS that we sought to detect in the sample size calculation, and smaller than the minimal clinically important difference of 3 points. The improvement in IPSS, however, was mirrored by improvements in the ICIQ-UI-SF, the IPSS-QoL measures and the B-IPQ. The primary analysis estimates were robust to ways of accounting for clustering and multiple imputation of missing data. High levels of adherence to intervention delivery meant that per-protocol analyses largely reflected the primary analysis.

Although there was no strong evidence that the treatment effect was modified by the nature of LUTS at baseline, by whom it was delivered or by how contact was made in the intervention, a post hoc analysis suggests that the treatment effect may have been stronger in the pre-COVID period.