Assessing the effectiveness of a ‘site visit’ on recruitment rates in a multicentre randomised trial: SWAT-1

ion from the HSCIC and Department for Education, National Pupil Database (NPD). Whilst this model of linkage offers the possibility for long term evaluation and at a lower cost, what are the associated problems in measuring the proposed study outcomes? This presentation will examine the data sources utilised in the study, and how linkage is carried out by each organisation. An initial assessment of linkage rates and its quality will be reported. As child maltreatment data may also be sourced from LA departments of social services, we will discuss what might be gained and lost for the study in moving from utilising data from this source (i.e. collecting direct from LAs) to the NPD. We will also discuss the consequences for certain analyses such as the Markov Chain Modelling which requires detail about stages of progression through a process and how the quality of the NPD could be validated by sampling local authorities. O70 Managing the morass. Lessons learned from establishing a data linkage model for long-term follow up of a trial cohort using routine health and education data Fiona Lugg, Rebecca Cannings-John, Gwenllian Moody, Mike Robling Cardiff University, Cardiff, UK Trials 2015, 16(Suppl 2):O70 The Building Blocks trial [ISRCTN 23019866] assessed the short-term impact of an intensive programme of antenatal and postnatal visiting by specially trained nurses to support young pregnant women. This followon study, BB: 2-6, will assess the programmes’ impact on longer-term benefits for mothers and children, through the linkage of routinely collected data, with a particular emphasis on the programmes’ impact upon child maltreatment. Follow up will be by linked anonymous data abstraction from the Health and Social Care Information Centre (HSCIC), Office for National Statistics (ONS) and Department for Education, National Pupil Database (NPD). These information centres (ICs) will match participants to the information held in their databases, and send to a third party safe haven. Re-consenting of all mothers would have resulted in a drastic loss of participants therefore section 251 support was sought to allow the transfer of identifiers to ICs for matching. This support required participants to optout rather than consent in. This project will complete in 2018. What has been achieved thus far is the establishment of a regulatory compliant model of linking health, social care and education data to clinical data. This model of linkage offers the possibility for long term evaluation of trials, at lower cost, and with the potential to extend to other sources of routine data. This presentation will review the approach established in this study, reflect on the dis/advantages of the approach, summarise the conditions of s251 support, information governance compliance, IC requirements and finally the challenges (and solutions) faced thus far. O71 Can we reduce bias in open-label trials when blinded outcome assessment is not possible? An example from the trigger trial Brennan Kahan, Vipul Jairath Pragmatic Clinical Trials Unit, Queen Mary University of London, London, UK; Translational Gastroenterology Unit, Nuffield Department of Medicine, University of Oxford, Oxford, UK Trials 2015, 16(Suppl 2):O71 Blinded outcome assessment is a key component of randomised trials, as unblinded assessment can lead to bias. However, in some circumstances blinded assessment may be difficult to achieve. In these situations, it may be useful to modify the outcome definition to remove the most subjective elements, thereby reducing the risk of bias. This is the approach used in TRIGGER, an open-label cluster-randomised trial in patients with acute upper gastrointestinal bleeding. The primary clinical outcome was further bleeding. Blinded outcome assessment was impossible, as all clinicians throughout a hospital were aware of the treatment allocation due to the use of cluster-randomisation, and given the emergency nature of the condition, it was not possible to compile relevant information to send to an adjudication committee in a blinded matter. We therefore modified the outcome definition to remove subjective events (e.g. if a patient vomited blood, whether it was ‘fresh’ enough to indicate a new bleed), leaving only relatively objective events (the presence vs. absence of blood in the patient’s upper gastrointestinal tract, based on a visual inspection by endoscopy). We collected both outcomes (including vs. removing subjective events) during the trial, and compared the estimated treatment effects from both. Including subjective events led to an odds ratio (OR) of 0.83 (95% CI 0.50 to 1.37), compared to an OR of 0.50 (95% CI 0.32 to 0.78) after removing subjective events. The ratio of odds ratios was 1.66, indicating that including subjective events may biased the treatment effect upwards by 66%. O72 Selective reporting in clinical trials an examination of discrepancy rates in pre-specified and reported outcomes in articles submitted to the BMJ Jennifer Weston, Kerry Dwan, Douglas Altman, Mike Clarke, Carrol Gamble, Trish Groves, Sara Schroter, Paula Williamson, Jamie Kirkham University of Liverpool, Liverpool, UK; University of Oxford, Oxford, UK; Queen’s University Belfast, Belfast, UK; The BMJ, London, UK Trials 2015, 16(Suppl 2):O72 Background: Adding, omitting or changing pre-specified outcomes can result in bias because it increases the potential for unacknowledged or post-hoc revisions of the planned analyses. Journals have adopted initiatives such as requiring the prospective registration of trials to promote the transparency of reporting in clinical trials. Methods: A review of all 3156 articles submitted to The BMJ between September 2013 and July 2014. Trial registry entries, protocols and trial reports of randomised controlled trials published by The BMJ and a random sample of those rejected were reviewed to determine the frequency and type of outcome discrepancies between pre-specified and reported outcomes. Editorial, peer reviewer comments and author responses were also examined to ascertain any reasons for discrepancies. Results: In the study period, The BMJ received 311 trial manuscripts, 21 of which were published by the journal. In trials published by The BMJ, 22% (75/339) of the pre-specified outcomes were not reported and 8% (25/297) of reported outcomes were not pre-specified. Discrepancy rates for rejected articles were similar. The majority of reasons provided by authors for outcome discrepancies were not bias related. Conclusions: Mandating the prospective registration of a trial and requesting that a protocol be uploaded when submitting a trial article to a journal has the potential to promote transparency and safeguard the evidence base against outcome reporting biases as a result of outcome discrepancies. Further guidance is needed with regards to documenting reasons for outcome discrepancies and making these reasons available. O73 What is the effect of patient-reported outcome (PRO) item order on prioritisation of PROs in the development of a core outcome set? Katy Chalmers, Kerry Avery, Karen Coulman, Natalie Blencowe, Rhiannon Macefield, Chris Metcalfe, Jane Blazeby, Sara Brookes University of Bristol, Bristol, UK Trials 2015, 16(Suppl 2):O73 Background: Core outcome sets (COS) are a minimum set of outcomes to be measured and reported in all trials of a specific condition. Questionnaires are often used in their development to enable stakeholders to rate potential outcomes in terms of importance. This study examined the impact of the ordering of patient-reported outcomes (PRO) within the questionnaire on their prioritisation. Methods: The order of questionnaire items in the development of a COS for oesophageal cancer surgery was randomised between PROs appearing first (V1) and last (V2). Comparisons were made between the percentages of items rated as essential (scored 7-9 by ≥70%) in the two versions. Results: 115 patients and 68 healthcare professionals completed questionnaires (98 and 85 randomised to V1 and V2 respectively). The percentage of PRO items rated as essential in V1 was 31.6% and in V2 63.2% Trials 2015, Volume 16 Suppl 2 http://www.trialsjournal.com/supplements/16/S2 Page 19 of 84