Oral health research is often a source of challenging data: statistical errors in the design, analysis and conclusions of dental research are frequently reported; when clinical outcomes are collected, they have an inherently hierarchical structure (e.g. teeth nested within people who are often nested within dental practices) with often non-normal distributions that can be challenging to analyse and present; outcome assessment becomes a challenge due to measurement errors and reliability of the data collected.
IQuaD is a randomised controlled trial (RCT) in dentistry using a complex design and is the motivating example for this PhD. IQuaD uses a split-plot design with cluster-level (68 dental practices) and individual-level (1,877 dental patients) randomisation that presents challenges for sample size calculation, analysis and reporting. IQuaD’s primary outcome is bleeding on probing, a measure of gingivitis collected through clinical assessment.
The PhD will address challenges in split-plot designs and the difficulty and costs of measuring clinical gingival bleeding in a large scale, multicentre pragmatic RCT by investigating the methodological implications for different ways of clinically measuring gingival bleeding and by assessing the diagnostic performance of several new self-reported measures.
The thesis will be focused on the following objectives:
- Conduct a systematic review to identify split-plot designs used in healthcare research and describe their methodology and develop guidance on its report
- Develop simulation methods to calculate a sample size in a split-plot design under a variety of assumptions and provide guidance on how to do it
- Assess the diagnostic performance of a set of new self-reported bleeding questions compared to clinical bleeding on probing and update a systematic review about the diagnostic performance of other self-reported bleeding measures
- Assess the statistical implications of collecting repeated clinical bleeding measures in the IQuaD RCT and inform future trials based on this
Supervision: Professor Craig Ramsay and Professor Graeme MacLennan
Data Collection - Ongoing
Goulão, B., MacLennan, G. & Ramsay, C. 'The split-plot design was useful for evaluating complex, multi-level interventions but there is need for improvement in its design and report'. Journal of Clinical Epidemiology. [ONLINE] DOI: 10.1016/J.JCLINEPI.2017.10.019
Goulão, B. MacLenann, G. Ramsay, C. (2017) Split-plot designs: sample size considerations. Trials Vol 18, Suppl 1.
Goulão, B. MacLennan, G. Ramsay, C. (2016) Split-plot designs in healthcare: an overview. Young researchers using statistics Symposium 2016 Programme.
Implementation research design: an introduction to the split-plot randomised controlled trial. 1st UK Implementation Science Conference. 19th July 2018, London.
Split-plot designs: sample size considerations. International Society of Clinical Biostatistics Conference. Vigo, 9th – 13th July 2017.
Split-plot designs: sample size considerations. Society of Clinical Trials / International Conference of Trial Methodology. Liverpool, 7-10th May 2017.
Split-plot designs in healthcare: an overview, International Conference of Clinical Biostatistics, Birmingham 22nd – 25th Sept 2016
Strengths and limitations of split-plot designs: an application in oral health research, 7th International Conference in Methodological Issues in Oral Health Research, Bergen 11-13 May 2016