1 / 24

Cluster Randomized Trials and The Stepped Wedge

Cluster Randomized Trials. Randomization at group level; outcome (typically) measured on individuals within the groupClusters may be large (cities, schools)

vinnie
Download Presentation

Cluster Randomized Trials and The Stepped Wedge

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

    1. Cluster Randomized Trials and The Stepped Wedge Jim Hughes UW Biostatistics

    2. Cluster Randomized Trials

    3. A common error: two communities, flip a coin, one gets intervention; other gets control Underlying differences between communities confounded with treatment effect “Change from baseline” doesn’t solve the problem Key: Effective sample size is number of clusters, not number of individuals measured (though both are important) Cluster Randomized Trials Individuals within a group are not independent – individuals within a group vary less then individuals in different groupsIndividuals within a group are not independent – individuals within a group vary less then individuals in different groups

    4. Key Considerations How is the intervention delivered? To the whole group or a subgroup? Individually or en masse (i.e. radio, billboard) How is the outcome measured? Whole group or subgroup? To those that received the intervention or others?How is the intervention delivered? To the whole group or a subgroup? Individually or en masse (i.e. radio, billboard) How is the outcome measured? Whole group or subgroup? To those that received the intervention or others?

    5. Common Trial Designs Rows represent units X = treatment, O = control In crossover trial, order is randomized; washout period may be necessary; cluster serves as their own control Note: In cross-over trial, I am thinking about the situation in which different people are ascertained at time 1 and 2 HPTN054Rows represent units X = treatment, O = control In crossover trial, order is randomized; washout period may be necessary; cluster serves as their own control Note: In cross-over trial, I am thinking about the situation in which different people are ascertained at time 1 and 2 HPTN054

    6. The stepped wedge design Cook and Campbell – “Experimentally Staged Introduction” Note issue of healthy survivor bias if cohorts are followed Outcome usually measured at time of stepping but maybe not in IRTCook and Campbell – “Experimentally Staged Introduction” Note issue of healthy survivor bias if cohorts are followed Outcome usually measured at time of stepping but maybe not in IRT

    7. Reasons for choosing the Stepped Wedge Design Unethical to withhold intervention?Unethical to withhold intervention?

    8. Some Examples Isoniazid therapy previous shown to be effective in preventing TBIsoniazid therapy previous shown to be effective in preventing TB

    9. Some Examples

    10. Some Examples Outcome measured in repeated cross-sectional samplesOutcome measured in repeated cross-sectional samples

    11. Some Examples

    12. WA State EPT

    13. Statistical Issues - Model Model is for a continuous outcome; ideas extend to binary measures This is simplest model – no time by treatment or time by cluster interactions; more complex models possible. Both ?? and ? could be subscripted by jModel is for a continuous outcome; ideas extend to binary measures This is simplest model – no time by treatment or time by cluster interactions; more complex models possible. Both ?? and ? could be subscripted by j

    14. Statistical Issues - Power

    15. Power – SW vs parallel N = 38 HIV+ per time period SW – 8 communities observed twice; parallel – 16 communities observed onceN = 38 HIV+ per time period SW – 8 communities observed twice; parallel – 16 communities observed once

    16. Power vs RR CV=.3, baseline prev = .05CV=.3, baseline prev = .05

    17. Power vs N per cluster CV=.3, baseline prev = .05 CV=.3, baseline prev = .05

    18. Power vs # of randomization steps 24 counties, 3, 6, 8, 12 at a time, respectively N = 100 Ct tests per county per time interval (baseline prev 5%)24 counties, 3, 6, 8, 12 at a time, respectively N = 100 Ct tests per county per time interval (baseline prev 5%)

    19. Power – Delayed treatment effect Treatment effect of .8, .9, 1 at lag 0, 1, 2 Assumes delay known; if unknown, it is estimable although this would likely reduce power furtherTreatment effect of .8, .9, 1 at lag 0, 1, 2 Assumes delay known; if unknown, it is estimable although this would likely reduce power further

    20. Statistical Issues - Analysis LMM, GEE, GLMM – model based Equal cluster sizes means equal number of outcome observations per cluster per time period Some power lost if there are no time trends and you adjust for them, but better safe than sorryLMM, GEE, GLMM – model based Equal cluster sizes means equal number of outcome observations per cluster per time period Some power lost if there are no time trends and you adjust for them, but better safe than sorry

    21. Summary

    22. Thanks

    23. Alternative models Also possible to write models for … Cluster by Time interaction Cluster by Treatment interaction (treatment effect varies by cluster) Treatment by Time interaction (treatment effect varies with time) Treatment effect varies with time since introduction of intervention

    24. Power – Delayed treatment effect Treatment effect of .5, .8, 1 at lag 0, 1, 2Treatment effect of .5, .8, 1 at lag 0, 1, 2

More Related