class: title-slide background-image: url(assets/logo.jpg) <div class="title-whole"><div class="title-top"></div> <div class="title-bottom"> <table class="title-table"> <tr> <td class="col1" style="color:#fee600"><img src="assets/unsw.jpg"></td> <td class="col2"> <h1>A Whirlwind Tour of Alternative Study Designs</h1> <h3>Philip J Clare</h3> </td> </tr> </table> </div> </div> --- # Overview 1. Introduction 1. ‘Standard’ designs 1. Experimental Study Designs 1. Descriptive Study Designs 1. Other designs 1. Conclusions --- # 1. Introduction - Descriptive vs experimental - Most important thing is to use the best design for the specific question being asked --- # 1. Introduction - Descriptive studies: - Involve observing and assessing - Can be used for things where it isn’t ethical to randomise - Experimental studies: - Involve the researchers changing something - Tend to be more difficult and expensive --- class: red-section # 2. 'Standard' Designs --- # 2. 'Standard' Designs - Three standard designs we should all know: - Randomised controlled trials - Cross-sectional surveys - Longitudinal surveys - These tend to form the standard by which other study designs are considered - eg how is a particular design better or worse than an RCT --- # 2. 'Standard' Designs Randomised Controlled Trials - Randomise to group; give each group a different intervention; follow-up and compare the groups - Gold standard for causation - Ignorability - Efficacy vs Effectiveness - Not always appropriate --- # 2. 'Standard' Designs Randomised Controlled Trials .center[  ] --- # 2. 'Standard' Designs Cross-sectional survey - Tends to be exploratory - Hard to design an RCT when you don’t know what the intervention should be - Can be cheap and easy to design, run, analyse and report - Difficult to recruit representative samples --- # 2. 'Standard' Designs Prospective Cohort - In some ways, just an extension of cross-sectional - But there are a range of specific issues that arise in longitudinal research that do not apply in cross-sectional research - There are also specific benefits - Can be either descriptive or analytic - With limitations, can be used for causal inference --- # 2. 'Standard' Designs Descriptive Prospective Cohort .center[  ] --- # 2. 'Standard' Designs Analytic Prospective Cohort .center[  ] --- class: red-section # 3. Experimental Designs --- # 3. Experimental Designs - There are lots of other experimental designs: - Pre/post experimental trials - Cluster RCTs - Stepped-wedge - Cross-over Trials --- # 3. Experimental Designs Pre/post Designs - Measure participants, then administer intervention and measure again - Participants form their own control - Within subjects designs reduces variance, so requires smaller sample sizes - Don’t know what would have happened without the experiment --- # 3. Experimental Designs Pre/post Designs .center[  ] --- # 3. Experimental Designs Cluster RCTs (1) - Instead of randomisation at the individual level, randomises at the cluster eg hospital, school - Good for systemic interventions eg new hospital process - Also good for interventions where there is risk of cross-contamination eg participants talking to each other while in hospital --- # 3. Experimental Designs Cluster RCTs .center[  ] --- # 3. Experimental Designs Stepped Wedge (2) - Instead of randomising clusters to intervention or control, randomises the order that intervention begins - Eventually rolls out to all clusters - Pragmatic design – not great for efficacy trials --- # 3. Experimental Designs Stepped Wedge .center[  ] --- # 3. Experimental Designs Cross-over Trials (3) - Randomise to order; give participants intervention; wait for effect to fade; give participants intervention they didn’t receive in the first round - Very good for experiments with multiple, short-acting interventions - Similar benefit to pre/post – ppnts form their own control - Allows for testing of multiple interventions --- # 3. Experimental Designs Cross-over Trials .center[  ] --- class: red-section # 4. Non-experimental Designs --- # 4. Non-experimental Designs - These include: - Qualitative studies - Case studies - Case-control studies - Ecological studies --- # 4. Non-experimental Designs Qualitative Studies - Very in-depth - Great for explaining WHY something happens - Don't generalise (and aren't intended to) --- # 4. Non-experimental Designs Case Studies (4) - In-depth - Illustrative, rather than representative --- # 4. Non-experimental Designs Case-control Studies (5) - Find ‘cases’ (ie people where the outcome has occurred), then see how they differ from others in whom the outcome did not occur - Good for rare outcomes that might not be expected to occur in a prospective study - Can be unreliable – relies on data that has already been collected --- # 4. Non-experimental Designs Case-control Studies .center[  ] --- # 4. Non-experimental Designs Ecological Studies (6) - Defined by the 'level' of measurement - Namely, measurement is at a macro rather than individual level - Can be used to measure very rare diseases etc - Can lead to the 'ecological fallacy' - Where the nature of the group is falsely used to describe the nature of individuals in the group --- class: red-section # 5. Other Designs --- # 5. Other Designs Interrupted Time Series (7) - Measures the change in a trend that is ‘interrupted’ - Technically involves an intervention – but often out of the researchers hands - Sometimes the only way to assess whether high-level interventions have any effect. - Can be problematic because effects often aren't instant --- # 5. Other Designs Interrupted Time Series .center[  ] --- # 5. Other Designs Linkage Studies (8) - Can be either descriptive or experimental - Can involve linking administrative data to an existing study such as a cohort study, or linking different administrative datasets to each other - Can be very complex, and involve extremely large datasets --- # 6. Conclusions - There are lots of study designs that can be used - Most important thing is to use the best design for the specific question being asked --- # 7. References .small[ <ol start=1> <li>Murray DM, Varnell SP, Blitstein JL. Design and Analysis of Group-Randomized Trials: A Review of Recent Methodological Developments. Am J Public Health. 2004;94(3):423-432. <li>Hemming K, Haines TP, Chilton PJ, Girling AJ, Lilford RJ. The stepped wedge cluster randomised trial: rationale, design, analysis, and reporting. BMJ. 2015;350:h391. <li>Johnson DE. Crossover experiments. WIREs Comp Stat. 2010;2:620-625. <li>Yin RK. Case study research: design and methods. SAGE Publications. 2013. <li>Mann CJ. Observational research methods. Research design II:cohort, cross sectional, and case-control studies. Emerg Med J. 2003;20:54-60. </ol> ] --- # 7. References .small[ <ol start=6> <li>Morgenstern H. Ecologic studies in epidemiology: concepts, principles, and methods. Annu Rev Public Health. 1995;16:61-81. <li>Kontopantelis E. Regression based quasi-experimental approach when randomisation is not an option: interrupted time series analysis. BMJ. 2015;350:h2750. <li>Bohensky MA, Jolley D, Sundararajan V, Evans S, Pilcher DV, Scott I, Brand CA. Data Linkage: A powerful research tool with potential problems. BMC Health Services Research. 2010;10:346. ]