1 / 31

Discussion of Dynamic Treatment Regimens

Discussion of Dynamic Treatment Regimens. Dean Follmann NIAID. Thoughts. Most clinical trials are 2 arms, maybe 2x2 factorial. Simple, clear, familiar. Dynamic Treatment Regimens Not

Télécharger la présentation

Discussion of Dynamic Treatment Regimens

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. Discussion of Dynamic Treatment Regimens Dean Follmann NIAID

  2. Thoughts • Most clinical trials are 2 arms, maybe 2x2 factorial. • Simple, clear, familiar. • Dynamic Treatment Regimens • Not • DTRs efficiently address treatment strategies for settings where multiple sequential treatment decisions are made • More efficient than multiple separate clinical trials • Randomized design and analysis is best. • Selling DTRs • Randomization quality inference where we efficiently address multiple questions

  3. Kidwell

  4. Collins Continue Med (SG1) Response Medication Response Enhance (SG2) Enhance (SG5) Non-Response R R Non-Response R Augment (SG3) Augment (SG6) Continue BMOD (SG4) BMOD Med=G1+SG2+SG3 versus BMOD=SG4+SG5+SG6 2 arm trial E=SG2+ SG5 versus A=SG3+SG6 Stratified clinical trial Two strata are 1) Medication NR 2)BMOD NR What is the best overall strategy? Med + E for NR, Med + A for NR BMOD + E for NR, BMOD + A for NR. An Uncovered 4 arm trial!

  5. Farm Story • Seed company interested in • Manure versus anhydrous fertilizer • Chemical versus mechanical weeding for weedy fields • In March, randomize fields to 1 of the 4 strategies • March Manure + chemical if weedy in June • March Manure + mechanical if weedy in June • March Anhydrous + chemical if weedy in June • March Anhydrous + mechanical if weedy in June • A standard 2 x 2 factorial clinical trial

  6. DeKalb Seed Trial, sequentially No weeding few weeds Manure few weeds Mechanical weeding Mechanical weeding weedy R R weedy R Herbicide Herbicide No weeding Anhydrous March June October

  7. Moral of the Story • Factorial trials have a rich literature • Parsimony in analysis: Main effects, low order interactions • ``Two studies for the price of one” • Fractional factorials • If randomize completely in March, factorial analysis of the 4 strategies is very clear.

  8. Factorial Analysis Intended Action for Weedy Fields Herbicide Weeding Anhydrous Manure Compare Anhydrous versus Manure using 2 strata: Herbicide & Weeding And vice versa

  9. Two Analyses Herbicide Weeding Weedy Fields Anhydrous Manure Compare Anhydrous versus Manure using 2 strata: Herbicide & Weeding Compare Herbicide to Weeding in the weedy fields using 2 strata: Anhydrous weedy fields & Manure weedy fields (it’s legit!)

  10. Best Strategy? • Four strategies • Med + Enhancement for Nonresponse • Med + Augment for Nonresponse • BMOD + Enhancement for Nonresponse • BMOD + Augment for Nonresponse • Best way to pick the winner? • Pick the best of the 4 means. • Pick Med/BMOD then Pick Enhancement/Augment • Use factorial methods

  11. Best Strategy? • Four strategies • Med + Enhancement for Nonresponse • Med + Augment for Nonresponse • BMOD + Enhancement Method C for Nonresponse • BMOD + Augment Method D for Nonresponse • Best way to pick the winner? • Pick the best of the 4 means. • Pick Med/BMOD then Pick Enhancement/Augment • Use factorial methods

  12. DTRs • Seem an exciting, rigorous, & efficient way to address multiple questions when we need to titrate a treatment strategy. • When equivalent to factorial trial, sister literature may be somewhat helpful. • If on the fence fall to a factorial? • Complication can be daunting to collaborators. • Emphasize efficiency of design • Find simple ways to convey results

  13. Backup

  14. Responders Non-Responders & Augmented Non-Responders & Enhanced ,lwd=3) MED BMOD Post Randomization Subgroups for Responders. Strata for Non-responders

  15. Responders Mean response For BMOD+AUGNR Non Responders & Augmented x X NonResponders & Enhanced ,lwd=3) BMOD MED

  16. Mean response For BMOD+AUGNR x X BMOD 1st line treatment MED 1st line treatment

  17. The uncovered 4 arm trial of strategies 2nd line augmented x X 2nd Line enhanced O O BMOD 1st line treatment MED 1st line treatment

  18. Complexity • DTRs can become complicated with multiple trajectories. • Desirable to model mean structure with parsimony e.g. main effects. • Unclear if conventional wisdom about interactions applies here. • But worth exploring • Must fit separate means to uncovered trial if no common treatments.

  19. Lynch • Appealing design to efficiently ask several questions. Best initial therapy. Best 2nd therapy for Responders & Non-responders. • Initial analyses respect randomization. • Subsequent analyses extracts randomized trials for 8 strategies. • (1) initial D2 if R then R1 if NR then NR1 • … • (8) initial D2 if R then R2 if NR then NR2 • Can compare the 8 strategies

  20. One of 8 strategies

  21. Above has D = +1,-1 5 day, not RespTx = +1, -1 NTX ONLY, not NRespTx = +1, -1 CBI+MM+PLA, not Just as we would for an old-school 2x2x2 factorial trial Does RespTX = -1 identify everyone else? i.e. Responders who got NTX+TDM and all Nonresponders?

  22. Moodie • Detailed development of strategies to optimize warfarin dosing. • Comes up with suggested combinations that can be tested in clinical trials. • Practicality of algorithms important

  23. Kidwell

  24. Dynamic Treatment Regimens • Interested in efficiently answering many sequentially appropriate questions. • Subsequent Randomizations can depend on previous responses. • # possible Trajectories is large 2 x 2 x 2 Randomizations Early Randomizations Responses

  25. Vanilla 2x2 Factorial Trial in would-be responders D2=R D5=? AND D2=? D5=R

  26. 2x2x2 DTR is likeTwo Old School 2x2 Randomized Trials Vanilla 2x2 Factorial in would be D2 or D5 Responders D2 Responders are from Principal Stratum 2 & 4 D5 Responders are from Principal Stratum 3 & 4

  27. 2 by 2 Factorial D2 + P D2 R D2 + T R D5 + P D5 R D5 + T R = Randomization

  28. Vanilla 2 by 2 Factorial in Would be D2 or D5 responders D2 + P D2 R D2 + T R D5 + P D5 R D5 + T R = Randomization

  29. Factorial Studies • How to analyze vanilla 2x2 trials? • Long history. Emphasis on parsimony & main effects • Y = b0 + b1 Z1 + b2 Z2 + b3 Z1 Z2 + e • Z1 = 1 if D2 (0 otherwise) • Z2 = 1 if TDM (0 otherwise) • a

  30. Analysis

More Related