Methods

The MXU targets high-risk populations in homeless hostels, day centres, drug and alcohol recovery projects, and street populations accessing soup kitchens across London. The study was restricted to hostel populations as these facilities provided a denominator population (number of residents on the day of screening) on which to reliably assess screening uptake. All homeless hostels in London taking part in MXU screening for active pulmonary TB run by F&T (a National Health Service (NHS) led service) were eligible for inclusion in the study if they had taken part in two previous screening sessions. Historical screening uptake data for the previous two rounds were reviewed and those with uptake levels over 80% were excluded from the study. These sites were excluded as it was felt the peer intervention would be unlikely to further improve screening uptake at these locations. Eligible hostels had not been screened in the 6 months prior to the scheduled MXU screening session. There were no additional exclusion criteria. No data about individuals were collected as the intervention was applied at hostel and not at the individual level.

Hostel managers were invited to a meeting hosted by F&T to explain the purpose of the study, and obtain their agreement and consent for participation in the study. Hostel managers unable to attend the meeting were contacted by email and telephoned, where necessary. Hostel managers were reassured that the study was evaluating a peer intervention and not individual hostel performance.

For intervention and control arms, F&T staff were present on the MXU to encourage uptake and manage onward referrals for suspected cases of active TB (usual practice for F&T). At intervention sites, volunteer peer educators arrived at the start of a screening session. Peer educators introduced themselves to the hostel staff and agreed on a work plan. They then moved around the hostel according to the agreed plan of work, knocking on residents’ doors (in conjunction with hostel staff), speaking to residents in all communal areas and those available close to the hostel location (next to the MXU) in order to encourage them to take up the offer of screening.

Peer educator volunteers were recruited via TB clinics in London (and therefore, had experience of TB) or from F&T, where staff identified or were approached directly, by interested service users. Volunteer peer educators, therefore, had experience of TB, homelessness or both. Training for peer educators was provided by attending a 3-day training session run by Groundswell (a registered charity that exists to enable homeless and vulnerable people to take more control of their lives) in conjunction with the research team and F&T, and provided input around TB and what was expected of the peer educators. Training covered issues such as TB transmission, TB risk groups, how treatment is conducted, the importance of screening for active pulmonary disease, how to maximise screening uptake (based on past experiences of F&T staff) and the additional support available for those undergoing screening. Peer educators also underwent a period of shadowing an existing peer educator to learn how to increase screening uptake. Ongoing support for peer educators was provided by Groundswell, based on their experience of working with homeless and vulnerable people to take more control of their lives and to have a greater influence on services they engage with so as to play a larger role in the community.15 Additionally, there was quality assurance of peer activity during screening sessions, which covered issues such as time keeping, communication and key peer activities, including how actively they encouraged residents to be screened.

The primary outcome for the study was the number of hostel residents screened for active TB on the MXU at each screening session. The number of residents eligible for screening at each hostel was determined from bed lists at each site, which took account of the number of residents actually staying at a venue overnight and excluded persons who were absent on the day of screening due to hospital admission, arrest or whereabouts unknown.

The study was a cluster randomised trial with a 1:1 allocation ratio between intervention and treatment arms. The unit of randomisation was hostel screening venue. A cluster randomised design was chosen as the intervention was aimed at the hostel sites rather than individual clients. Randomisation was carried out by the study research team using a master list of hostels at the beginning of the study. Sites were randomised to the intervention or control group using the internet-based service by SealedEnvelopeTM (http://www.sealedenvelope.com/), which ensured allocation concealment until interventions were assigned. To ensure comparability between intervention and control arms, hostels were stratified on the basis of their size (binary variable indicating whether hostels had more than 43 beds) and previous screening uptake level (binary variable indicating whether hostels had greater than 50% historical uptake).

On the basis of analysis of historical screening data from F&T, it was estimated that the intracluster correlation coefficient for this study was 0.08. This produced a maximum inflation factor of 6.52, assuming an average hostel size of 70 residents. Many changes to the provision of homeless hostel services took place from the time when the study was conceived to when the recruitment started, including a decrease in the average size of hostels from 70 to 50 residents. As a result, the inflation factor was recalculated as 4.84. To detect a 15% difference in screening uptake in the two groups (60% vs 75%) in an individually randomised controlled trial—with 90% power at the 5% significance level—would require 216 residents in each arm. Applying the 4.84-fold inflation factor for the clustered randomised design required at least 1045 individuals or approximately 21 homeless hostels in each arm.

Blinding of participants and observers was not possible due to the nature of the intervention. The study statistician (RWA) conducted analysis blinded to the allocation of intervention or control arms.

Baseline variable analysis included examining geographical location of the two study sites, hostel characteristics, including bed numbers and occupancy, rating of the hostel performance by research staff using a tool designed for this purpose (see online supplementary appendix), whether incentives were given and the historical levels of screening uptake. The rating of hostel performance included scoring of how involved the manager of the hostel was during the screening session, how much encouragement for screening there was from staff, whether posters advertising the screening were on display, how well-organised the event was and whether incentives were provided on screening day. No individual level data were collected as part of the study; therefore, it was not possible to examine baseline differences between the intervention or control groups, nor to identify any individual who may have crossed over between groups. Missing data were treated as an additional category in all analyses.

Exposure and outcome data at each venue were collected by study nurses present at the time of screening by the MXU on paper data collection forms. These forms were then entered into a Microsoft Excel 2010 spreadsheet by the study database manager. Data were then cleaned by the study statistician and analysed in Stata V.12.

Poisson regression analysis was used to analyse outcome events at screening hostels. Bed occupancy level was included as the exposure variable, screening uptake as the outcome (or indicator) variable, and hostel venue as a random effect to account for clustering at each site. The analysis was adjusted by inclusion of the randomisation stratification factors of historical uptake rates and bed size.

We conducted a secondary per-protocol analysis that only included sites if peer educators turned up for screening sessions and included a binary variable describing whether or not the research staff felt screening had gone as planned. Secondary analysis was also conducted by subgroups, for large and small hostels separately, and for low and high previous screening uptake level using the binary categorical variables described previously for these analyses.

The study trial registration number is ISRCTN17 270 334. The study was registered retrospectively as a result of a series of events. The study team initially believed it would be registered with an International Standard Randomised Controlled Trial Number (ISRCTN) by the National Institute for Health Research as a result of it being entered onto their Clinical Research Network (CLNR) Portfolio Database. The CLNR team informed the study team that the trial did not qualify to be on the portfolio (and as result, would not automatically receive free ISRCTN registration) because it did not involve individual informed consent and therefore, we would not be able to report individual patient recruitment to the CLNR portfolio. The time delay between July 2011 and the start of study recruitment in February 2012 unfortunately contributed to the fact that the lack of trial registration was overlooked by the study team. When RWA began to work on the study and pointed out the trial had not been registered, the CLNR portfolio team were approached for a second time in order to consider the trial for adoption on their portfolio. This request was denied on 14 November 2013 for a second time. The study team, therefore, decided to register the trial independently on ISRCTN and pay for it out of other research funding, but by the time this was arranged recruitment for the study had been completed.