Over February and March 2021, Dr Tianjing Li of the University of Colorado hosted a three day virtual methods symposium looking at advanced methods and innovative technologies for evidence synthesis on behalf of the Cochrane US Network funded by the US Agency for Healthcare Research and Quality. The symposium aimed to disseminate research into systematic reviews as a means to improve the quality, effectiveness, and appropriateness of healthcare by synthesizing existing evidence and facilitating the translation of research findings. Speakers included Rachel Phillips who presented work on the advantages of visualisations to evaluate and communicate adverse event information in randomised controlled trials in the session on synthesising harms of interventions. Other presenters included Dr Evan Mayo Wilson (Indian University Bloomington), Riaz Qureshi (Johns Hopkins Bloomberg School of Health), Dr Roger Chou (Oregon Health and Science University). You can watch all of the talks here:
We are delighted to welcome two new members to ICTU at the start of 2021.
In February we welcomed Dr Leila Janani as a Research Fellow in Medical Statistics who will be working the TREAT trial for treating severe paediatric asthma and working with Professor Tony Gordon on Critical Care research studies.
In March we welcomed Dr Graham Wheeler as Senior Lecturer in Medical Statistics, and Deputy Head of Statistics for ICTU. He joins us from Cancer Research UK & UCL Cancer Trials Centre where he was leading on several early phase oncology studies. His research is focused on novel adaptive designs for clinical trials and improvements for their use in practice.
On 20th – 23rd January, eCCR 2021 held its annual three day Critical Care Conference. Victoria Cornelius joined Fernando Zampieri, Jon Silversides, Andrew Althouse and Marion Campbell for a panel discussion on the often misunderstood area of Pilot Studies in Critical Care trials at the Critical Care Reviews 2020 Conference. The discussion covered what pilot trials are and what they are not, how to determine an appropriate sample size, the differences between internal and external pilots, and how to assess if the trial has been a success. The discussion is covered on day 2 at 7 hours 50mins.
This month we have motivated the department to help us run the length of the UK. We have asked everyone to record how much exercise they do each day, then see how far we travel collectively between us. Our team member Jack Elkes also built an R Shiny app to help us track our progress, follow the link below to check this out!
Wanted: members of the public to advise our clinical trials research
We are setting up a group of public members to discuss how we can improve the information reported from clinical trials, so that it is more meaningful and usable for patients and the public. Often the numbers and statistics reported from trials are not well understood and we would like to understand how we can improve this. For more information on this opportunity and how to get involved please see the poster linked below.
Presentation at UKCRC CTU Network Statistics group meeting
The UKCRC Registered CTU Network held its bi-annual statisticians meeting on 8th October. At this meeting Suzie Cro gave a talk on handling unplanned disruptions and missing data in randomised trials ongoing during Covid-19. Richard Emsley (KCL) also gave a talk on departures from random allocation in trials ongoing during Covid-19. For members of the UKCRC CTU network a recording of the talks can accessed by logging into the below link.
On September 1st, the National Institute of Statistical Science (NISS) held an online meeting to investigate topics related to methods or strategies for mitigating disruptions in trials during covid-19, as part of their Ingram Olkin Forum Series. Presenters included Suzie Cro who explored a strategy for handling unplanned disruptions in the analysis of trials using missing data methods. Other presenters included Mouna Akacha (Novartis), who discussed the value of the estimand framework (ICH 2019) in being able to support the proper analysis and interpretation of trials disrupted by covid-19, and David Murray, (Office of Disease Prevention) who provided examples from the Health Care Systems Collaboratory Project and reviewed how these trials adapted in response to various disruptions.
We are pleased to announce the start of 2 new NIHR fellowships. Ms Eleanor Van Vogt starts a 2-year pre-doctoral fellowship supervised by Dr Suzie Cro, Dr Victoria Cornelius and Dr Karla Diaz-Ordaz. She begins year 1 by undertaking the MSc in health data science and machine learning at Imperial College, school of public health and will undertake a research project with ICTU in year 2.
Dr Suzie Cro starts a 3-year advanced fellowship to develop accessible statistical methods to determine treatments effects that matter to patients and prescribing physicians in randomised controlled trials. This includes statistical development of controlled Multiple Imputation methods for estimating hypothetical estimands in clinical trials. Details on the fellowship schemes are available below.
Recommendations to visualise treatment harm profiles in Randomised Controlled Trials – a national consensus
In collaboration with the UKCRC CTU Statistics Operation group, Rachel Phillips, Victoria Cornelius and Suzie Cro hosted three half day meetings to develop a national consensus of recommendations to display harm events in Randomised Controlled Trials. The meeting was attended by national senior statisticians from academia and industry, and delegates with info-graphic expertise. The groups’ aim is to develop practical guidelines to support clinical trial researchers in their choice of visualisations for communicating AE data in publications. The next stage will seek input from clinicians and patients and the group will be publishing their work for trialists later this year.
Update coming soon
A dose-finding design for early phase trials with continuous re-assessment
As part of the NIHR Statistics – Early Phase Trials Group webinar series on “Practical applications of advanced designs in Phase I Trials” Suzie Cro shared her recent experience of designing a dose finding study. The trial requires two agents to be combined where one has discrete dose levels and the other having doses that are patient specific. In the presentation Suzie Cro and Pavel Mozgunov from Lancaster University explained how they will evaluate the different discrete doses of the primary agent and allow for different ranges of the patient specific doses of the second agent.
As part of the 2020 NIHR Statistics Group virtual conference Rachel Phillips and Victoria Cornelius hosted a workshop on the use of visualisations to communicate information on harms in RCTs. In the workshop they demonstrated the use of four graphical displays and contrasted the visual summaries with data presented in the original published journal articles demonstrating impact for interpretation. Delegates provided feedback on the applications across settings. The responses in this session will contribute to the development of practical guidance for communicating harm data.
Update coming soon
Evidence of unexplained discrepancies between planned and conducted statistical analyses
Different methods of analysis applied to the same clinical trial data set can lead to different conclusions. It is therefore important that the statistical analysis approach for a randomised controlled trial is pre-specified, and any changes or additions to the planned analysis should be flagged alongside trial results. This provides full transparency and allows for a thorough evaluation of the results. In a new review of trials published in high impact journals, we identified a large proportion of trials (61%) with one or more undisclosed change to the planned primary analysis approach. To see more results click the below link for the full open access publication of the review.
New Tutorial: Controlled Multiple Imputation for Missing Data Sensitivity Analysis in Clinical Trials
In clinical trials it is most typical that some participant outcomes will not be available. This may be due to missed participant visits, trial withdrawal or other unplanned and uncontrollable events. Therefore often some required data will be missing from the analysis. When there are missing data, it is important that the primary analysis of the trial is conducted under the most plausible assumption for the missing data. Sensitivity analysis under a range of different credible assumptions should then be undertaken to assess how robust the trial results are. One method which readily enables contextually relevant sensitivity analysis, and has recently seen increased discussion and developments in the statistical literature, is Controlled Multiple Imputation. In this new tutorial article, an overview of Controlled Multiple Imputation procedures, and a practical guide to their use for sensitivity analysis of a continuous outcome is provided. Worked examples and Stata code are included to facilitate adoption of such methods, to enable robust evaluation of clinical trial results.
A caution on the use of IPTW in small randomised trials
Recently Inverse Probability of Treatment Weighting (IPTW) using the propensity score has been proposed as an alternative to standard regression methods for baseline covariate adjustment in the analysis of randomized trials (Williamson et al. 2013). Motivated by some recent small trial analyses we explored the properties of IPTW for covariate adjustment within small trial settings. A simulation study and a re-analysis of a pediatric eczema trial involving 60 children revealed the performance of the IPTW variance estimator proposed by Williamson et al. was sub-optimal with smaller sample sizes. We therefore caution against the use of IPTW in small sample settings without small-sample modifications when the sample size is less than 150 and particularly when the sample size <100. In larger samples IPTW using the propensity score method can however be a useful tool for adjustment. When IPTW is used with large samples we demonstrate how the bootstrap variance may be a simpler route to variance estimation, given this incorporates the estimation of the propensity score.
Anca went to visit the London PHASTAR team in Chiswick for their monthly company meeting to present some of the research that’s been going on at ICTU, as well as her systematic review that investigates how rescue medication is defined, reported, and adjusted for in randomised controlled trials. It was also amazing to hear Jennifer Roger’s presentation describing her experience with her BBC Watchdog appearance, where companies within a different industry were ranked to establish who was the best and who was the worst when considering factors such as complaints, customer satisfaction, and cost.
Do early skincare interventions prevent eczema and food allergy?
The protocol for an individual patient data meta-analysis to establish whether early skin care interventions, such as moisturisers, can prevent eczema or food allergy is now available. Published in the Cochrane database, this protocol provides details of the planned analyses that will be conducted by Suzie Cro and Victoria Cornelius. The primary analysis will estimate the pooled treatment policy effect of early skincare interventions on eczema and food allergy. Sensitivity analysis will also explore the pooled effect of complying with early skin care intervention. Subgroup analyses will explore whether risk for atopy, based on genotype and family history are associated with the pooled treatment effect. Results coming later this year!
PSI have recently launched a new visualisation special interest group (SIG). The group aims are to train statisticians and quantitative scientists about effective visualisations to look beyond tables alone. This group is led by Alexander Schacht of UCB and includes our own Rachel Phillips as a core member. The SIG’s first initiative, “Wonderful Wednesdays” will teach participants about visualisation principles and allow them to apply the techniques they learn to examples from the field of healthcare. To learn more about the initiative and how to get involved follow the link below:
MRC-NIHR Trials Methodology Research Partnership Statistical Analysis Working Group
The new MRC-NIHR Trials Methodology Research Partnership Statistical Analysis Working group (SAWG) is now open to expressions of interest to join. If you are researching/interested in developing and applying methods for the statistical analysis of clinical trials we encourage you to join. This group is being co-led by Richard Emsley (KCL) and Tim Morris (UCL) and includes our own Suzie Cro as one of the core members. Additional details on the groups remit can be found at the below link.
Reference‐based multiple imputation for missing data sensitivity analyses in trial‐based cost‐effectiveness analysis
In randomised trials, sensitivity analyses are required to assess the implications of different missing data assumptions on the trial results, including in cost‐effectiveness analysis. In their new paper, Leurent, Gomes, Cro et al. extend and illustrate the reference‐based multiple imputation approach for cost‐effectiveness sensitivity analysis. Reference‐based multiple imputation provides an attractive approach for conducting such sensitivity analyses, because missing data assumptions are framed in an intuitive way by making reference to other trial arms. The method is illustrated using the CoBalT trial which evaluated cognitive behavioural therapy for treatment‐resistant depression and Stata code is provided.
Results of collaborative trial research: Omalizumab reduces severe atopic dermatitis in children
The Atopic Dermatitis Anti-IgE Pediatric trial (ADAPT) trial has been published in JAMA pediatrics. This randomized controlled trial targeted a pediatric population with severe eczema. After a 24 week course of omalizumab or placebo it was found that the children given omalizumab had reduced eczema and improved quality of life in comparison to those given placebo. This was despite lower potent topical corticosteroid use in the omalizumab group, suggesting omalizumab is a treatment option for difficult-to-manage severe eczema in children.
UKCRC Biannual Statisticians Operations Group Meeting
In November, Imperial Clinical Trials Unit hosted the UKCRC’s Biannual Statisticians' Operations group meeting. The day focused on work undertaken by Victoria Cornelius and Rachel Phillips on adverse event analysis in clinical trials. It was a great opportunity for the team to share the findings from some of their current research projects and discuss priorities for next steps in the field of adverse event analysis with the network.
Presentations from the day are available from the UKCRC's website.
The start of October brought along the 2019 ICTMC conference, and it was quite a busy one for ICTU with several members of the team presenting posters during the conference:
Anca Chis Ster presented her work on a systematic review of the reporting and adjustment of Rescue Medication in RCT publications. A further paper on the topic will follow soon.
Suzie Cro brought to light the importance of estimating hypothetical estimands through her poster discussing Controlled Multiple Imputation.
There was also a poster from Nicholas Johnson looking at the importance of Data Integrity through clear rules of good data management with a feline motif.
Rachel Phillips’ poster presented her experiences in having to Access Individual Patient Data and what opportunities her learnings and this sort of data could offer others.
There was also 2 full presentations each from our own Victoria Cornelius and Rachel Phillips; summaries of these presentations alongside copies of their presentations will follow in more news articles!
Pre-SPEC: A framework for the pre-specification of statistical analysis strategies in clinical trials
We have identified that many trials fail to adequately pre-specify their statistical analysis approach. This is problematic since it allows analysts to choose the method which gives the most favourable result. To ensure methods cannot be chosen based on results we have developed a five point framework for the pre-specification of the statistical analysis strategy in clinical trials (Pre-SPEC). The five points are: (1) Pre-specify before recruitment to the trial begins; (2) Specify a single primary analysis strategy; (3) Plan all aspects of the analysis; (4) Enough detail should be provided so that a third party could independently perform the analysis; and (5) Choices in adaptive analysis strategies should use deterministic decision rules.
A pre-print of the framework, written by Suzie Cro, Brennan Kahan (QMUL) and Gordon Forbes (KCL), is available at the below link and we welcome any comments on this work in progress.
Attendance at 2019 Promoting Statistical Insight (PSI) conference in London
In June 2019, PSI held their annual conference in London. This provided an excellent opportunity for Victoria Cornelius and Rachel Phillips to present their methodological work on adverse event analysis, as well as allowing them to learn about recent methodological developments in the pharmaceutical industry.
Work shared included Victoria Cornelius’ research with Odile Sauzet on novel time-to-event methods to detect signals for adverse drug reactions in both the observational and RCT setting; and Rachel Phillips’ review of statistical methods for adverse event analysis in the RCT setting.
An international individual patient data meta-analysis has commenced to establish whether early skin care interventions, such as moisturisers, can prevent eczema or food allergy. Victoria Cornelius and Suzie Cro will be providing statistical support and conducting the analyses for the project (SCiPAD), led by Robert Boyle and Maeve Kelleher (Imperial).
On June 3rd Suzie attended the SCiPAD investigators meeting at EAACI congress in Lisbon and presented the statistical analysis plans to the international trial collaborators. By combining the data from multiple randomised trials which have assessed the impact of skin care interventions we will be able to obtain a more precise estimate of the intervention effect and perform more powerful subgroup analyses to identify patient factors associated with the intervention effect. We also propose statistical methods to explore the impact of compliance on the intervention effect across trials. Data collection is now in progress and result are anticipated summer 2020.
Results coming in summer 2020
May Measurement Month
Today is the start of May Measurement Month – a global campaign to increase awareness of raised blood pressure. During May volunteers will be running screening sites around the world and measuring individual’s blood pressure. Today also sees the publication of the results of the 2018 campaign in the European Heart Journal, which screened 1.5 million individuals worldwide and identified over 1/3 measured had untreated or inadequately treated hypertension.
To obtain a reliable reading of blood pressure individuals in the study had three consecutive blood pressure measurements taken, and the average of the second and third was used within the analysis. However not all individuals had a second or third reading, therefore we designed and implemented a multiple imputation model to handle the missing data in this large data set. Full details of the study, analysis methods and results are available at the below link.
Results of Collaborative Trial Research identify best dual treatment for Hypertension in Black Africans
Lowering blood pressure reduces cardiovascular morbidity and mortality. It is uncertain which dual therapy is most effective in reducing in Black Africans. In this trial published in NEJM the optimum dual therapies identified were a calcium channel blocker in combination with ACE inhibitor or diuretic, this was more effective than an ACE inhibitor and diuretic in contrast to current guidelines.
Review highlights inappropriate analysis of AE data
Clinical trials provide a valuable source of high-quality adverse event (AE) data. However, this review of high impact medical journals conducted by Rachel Phillips, Victoria Cornelius, Odile Sauzet and Lorna Hazell of the Drug Safety Research Unit found that the analysis of AE data in clinical trials was often inappropriate. Suboptimal practice included ignoring valuable information on repeated events and the inappropriate practice of underpowered multiple hypothesis testing. The open access article detailing the complete findings and recommendations is available at the below link.
Successful application to access individual patient data from the pharmaceutical industry
We are extremely grateful to GSK and the CSDR for approving our application for access to individual patient data. The ClinicalStudyDataRequest.com resource enables access to individual patient data from completed clinical trials from a variety of sponsors and funders. In late 2018, a team led by Rachel Phillips made an application to access IPD from a selection of GSK clinical trials to undertake methodological research. The application was approved in early 2019. To read more on what this data will be used for please follow the link below:
On February 12th at 1-4pm, members of the team will be chatting about their research in clinical trials at the People’s Research Café in the Kathy Dolan (Woodlane) Community Centre, White City. Come join us for a free tea or coffee and to have a chat about the information that would be most meaningful and useful for you to see reported from clinical trials!
What is the feasibility of undertaking a stepped-wedge trial in The Gambia to evaluate a community singing-based intervention for perinatal mental health?
Professor Lauren Stewart from Goldsmiths University assembled a truly multi-discipline team with music, psychology, arts, and science components for this partnership building grant. The study is jointly funded by the Arts and Humanities Research Council and Medical Research Council. As a team we visited The Gambia in January 2019 to meet researchers, understand practical issues and local context. An incredibly informative trip that has enabled us to position ourselves for the next stage.
Update coming soon
No more excuses for using a 3+3 design in a dose-finding trial!
This publication led by Dr. Graham Wheeler (UCL) describes how to design a dose-finding study using the continual reassessment method. It provides a shortcut through the overwhelming and often conflicting literature to make ‘recommendations for key design parameters and advise on conducting pre-trial simulation work to tailor the design to a specific trial. It provides practical tools to support clinicians and statisticians, including software recommendations, and template text and tables that can be edited and inserted into a trial protocol. They also give guidance on how to conduct and report dose-finding studies using the CRM.'
Multiple imputation 40 years on, where are we now?
On December 4th, the RSS Medical Section held a meeting on the use of multiple imputation, 40 years since the seminar paper introducing multiple imputation to handle missing data was published. Talks were given by Ian White (UCL) on “Multiple imputation the universal panacea and its limitations” and Tra Pham (UCL) on “Population-calibrated multiple imputation for a binary/categorical covariate in categorical regression models.” James Carpenter (LSHTM, UCL) and Suzie Cro also gave a talk on “Sensitivity analysis for missing trial outcomes” using multiple imputation, slides for their talk are available here:
In clinical trials, little attention has been given to the appropriate loss of information due to missing data in sensitivity analyses. In a new publication by Suzie Cro, James Carpenter and Michael Kenward it is argued that more attention needs to be given to this issue. It is shown that it is quite possible for sensitivity analysis to decrease or increase the information about the treatment effect with missing data. To address this critical issue, the concept of information‐anchored sensitivity analysis is introduced. This is sensitivity analysis in which the proportion of information about the treatment estimate lost because of missing data is the same as the proportion of information about the treatment estimate lost because of missing data in the primary analysis. The full open access article is available at the below link.
RSS 2018 provided a great opportunity for the team to present some of their methodological work. Including a quick fire presentation from Victoria Cornelius on "Detecting signals for adverse drug reactions (ADRs) using non-constant hazard in longitudinal data".
This meeting will share the findings of the PROMiSe study (Patient Reported Outcome MeasureS using Electronic informed consent and data capture), funded by NIHR’s CTU Support Funding programme to develop a practical framework for collecting patient-reported outcomes as individual-level time series, which can subsequently be linked to clinical data on treatments received. The workshop aims to present some novel research methods, and to stimulate more general discussion of ways to bridge the gap between observational studies and clinical trials in areas such as surgery, where randomised trials can face challenges, for example around equipoise and recruitment.
Attendance at 2018 Promoting Statistical Insight conference in Amsterdam
We attended our first PSI conference 'Breaking boundaries in Drug Development' held in Amsterdam this year. It was a great conference to present our review work on analysis methods used in randomised controlled trials. Next year is London!