Overview of Trial Design Options: Adults(Challenges in Clinical Trials in People for Whom Available Therapies have Failed ) : Overview of Trial Design Options: Adults (Challenges in Clinical Trials in People for Whom Available Therapies have Failed ) Martin Schechter Canadian HIV Trials Network University of British Columbia


WARNING: WARNING more questions than answers ahead


Randomized Controls or Not?: Randomized Controls or Not?


Tuberculous Meningitis: Tuberculous Meningitis universally fatal prior to 1945 1946 - streptomycin - new drug in short supply some treated patients survived randomized controls unnecessary


Tuberculous Meningitis: Tuberculous Meningitis extremely homogeneous patient group mortality outcomes prior outcome pattern fully characterized short term study (avoid secular trends) adherence not an issue


Slide7: Patient Heterogeneity


Increasing Heterogeneity: Increasing Heterogeneity drug history drug exposure intensity genotype phenotype virological status immunological status clinical status


Increasing Heterogeneity 2: Increasing Heterogeneity 2 toxicities malabsorption previous treatment interruptions adherence attitude about treatment unknown confounders


Increasing Heterogeneity 3: Increasing Heterogeneity 3 heterogeneity per se does not matter heterogeneity matters if the variables are strongly prognostic


Selected variables predicting decline of pVL < 400 (Multi-drug Rescue Therapy - Montaner et al): Selected variables predicting decline of pVL < 400 (Multi-drug Rescue Therapy - Montaner et al) * baseline CD4 and resistance to DDI, DLV, IND, NLV, NVP, RIT, SAQ were not significant


Slide12: Patient Heterogeneity


Slide13: Patient Heterogeneity Ability to Control Confounders


Slide14: Salvage Studies Live Here


Can we avoid randomized trials?: Can we avoid randomized trials? historical controls? a very attractive approach lessons from the history of medicine


Gastric Freezing for DU: Gastric Freezing for DU President of ACS - cooled gastric balloons case series very impressive “Since 1961, no patients with duodenal ulcer referred for elective operation have been operated on in the senior author’s service. This circumstance itself bespeaks the confidence in the method by patients as well as surgeons.”


Gastric Freezing for DU: Gastric Freezing for DU led to sale of 2500 gastric freeze machines an estimated 15,000 patients chilled double blind RCT in late 1960’s Outcome = surgery, bleed or intractable pain Sham group 44% Freeze group 51%


VA Study - Estrogen & Prostate Ca: VA Study - Estrogen & Prostate Ca RCT of 2313 patients recruited over 7 years no change in eligibility criteria throughout placebo patients in first 2.5 years had significantly worse survival than estrogen patients in the last 2.5 years


Uncontrolled Phase II Cancer studies: Uncontrolled Phase II Cancer studies uncontrolled series in advanced bowel cancer 20 different cases series of rapid injection 5-FU among 6 largest (n= 40 to 150) response rates ranged from 11% to 55%


Traditional Orthodoxy: Traditional Orthodoxy well known that historical control studies are far more likely to yield positive results Sacks HS, Chalmers TC, Smith H Jr. Sensitivity and specificity of clinical trials. Randomized v historical controls. Arch Int Med 1983; 143(4): 753-5. dominance of the RCT ongoing debate


Role of Observational Data: Role of Observational Data Concato J, Shah N, Horwitz RI, Randomized, controlled trials, observational studies, and the hierarchy of research designs. N Engl J Med 2000; 342:1887-92. RCT case-control cohort


Role of Observational Data: Role of Observational Data well designed case-control and cohort studies are not the same as historical control studies former involve careful selection of controls concurrent could play a role if prognostic variables completely characterized


Non-randomized concurrent comparisons by post-randomization variables: Non-randomized concurrent comparisons by post-randomization variables within-study comparisons based on adherence to a regimen if adherers fare better, is this not proof of efficacy?


Analysis by post-randomization adherence in a lipid lowering trial: Analysis by post-randomization adherence in a lipid lowering trial Medication A * adjusted for 40 baseline predictors of CV mortality


Analysis by post-randomization adherence in a lipid lowering trial: Analysis by post-randomization adherence in a lipid lowering trial Placebo * adjusted for 40 baseline predictors of CV mortality


Salvage Therapy and Non-randomized Controls: Salvage Therapy and Non-randomized Controls very heterogeneous populations some variables measurable, some not strong prognostic factors many factors as yet unidentified variable surrogate marker outcome CAUTION!


Salvage Therapy and Non-randomized Controls: Salvage Therapy and Non-randomized Controls homogeneous definition “worst” category across the board closer to true “salvage” definition well characterized outcomes


Control of Confounders: Control of Confounders randomization large sample size combination therapy of randomization and large sample size likely to distribute known and unknown confounders equally


Remedies for Confounding in Smaller Sample Sizes: Remedies for Confounding in Smaller Sample Sizes stratified randomization 2n strata unwieldy risk index minimization adaptive allocation made to minimize imbalances post-hoc adjustment effect on unknown confounders crude vs. multivariate


There is no within-study remedy for lack of power in smaller sample size studies!: There is no within-study remedy for lack of power in smaller sample size studies!


The Need for Blinding?: The Need for Blinding? it is orthodoxy that blinding is required numerous studies have shown less bias in fully blinded studies lower likelihood of positive results


The Need for Blinding?: The Need for Blinding? Schulz KF, Chalmers I, Hayes R, Altman DG. Empirical Evidence of Bias: Dimensions of Methodological Quality Associated With Estimates of Treatment Effects in Controlled Trials. JAMA 1995;273:408-412. based on 33 meta-analyses of 250 primary trials involving a total of 62,091 participants and 12,030 outcome events.


The Need for Blinding?: The Need for Blinding? Is blinding feasible? Can the artifact of blinding introduce more bias than it prevents?


The Need for Blinding - An example: The Need for Blinding - An example Standard (≤ 4 drugs) vs. Mega-HAART (≥ 5) Standard may benefit through adherence Blinding would mean from 9 to 17 different types of pills Could wipe out the adherence advantage of Standard-HAART Bias from a clinical trial artifact


Slide36: Treatment Received Intent to Treat


Particular Clinical Trial Challenges: Particular Clinical Trial Challenges rapid cross-over/drop-out eg. cessation (multi-drug to interruption) intent-to-treat becomes meaningless treatment-received becomes biased availability of new treatments or strategies e.g. genotypic testing, compassionate access when not built into current protocol


Clinical Trial Challenges: Clinical Trial Challenges rapid cross-over/drop-out offer of early vs. late Rx may induce better protocol adherence e.g. interruption now vs. interruption in X mos participant education - possibility of switch after poor response trigger availability of new treatments or strategies rolling protocols pre-planned randomization of future options


Factorial Designs - Illustration: Factorial Designs - Illustration What is the role of ADV vs. DLV and NLF vs. RTV in NNRTI-naïve, IND treated subjects? SQV sgc + RTV + DLV SQV sgc + NLF + ADV


ACTG 359 (2 x 3 Factorial): ACTG 359 (2 x 3 Factorial) SQV sgc +


OPTIMA (2 x 2 Factorial): OPTIMA (2 x 2 Factorial)


Randomised trial of effects of vitamin supplements on pregnancy outcomes and T cell counts in HIV-1-infected women in TanzaniaFawzi WW et al. Lancet 1998; 351:1477-82.: Randomised trial of effects of vitamin supplements on pregnancy outcomes and T cell counts in HIV-1-infected women in Tanzania Fawzi WW et al. Lancet 1998; 351:1477-82.


Stratified Factorial: Stratified Factorial Combo 1 Combo 2 Combo 4 Combo 3 Combo 1 Combo 2 Combo 4 Combo 3 Factor present Factor absent


Factorial Designs: Factorial Designs Suppose Factor I, II, III (e.g. mutation) Suppose Factor A, B, C Three combinations 1, 2, 3 Full factorial = 3 x 3 x 3 = 27 different cells Latin Square design (basic science, vet science)


Latin Squares: Latin Squares


Factorial Designs: Factorial Designs ideally suited when multiple therapies exist that can be given in different combinations ideally suited when different strategies can be combined Rx combinations, interruptions, genotyping, adjunctive therapy, immunomodulators, complementary therapy


Factorial Designs: Factorial Designs independent treatment effects and interactions symmetrical and efficient more bang for the buck generally underutilized in clinical trial literature


Drug-wise vs. Strategy-wise Evaluation: Drug-wise vs. Strategy-wise Evaluation Suppose new drugs A and B in 2 classes new resistance patterns within class A available for trials now B available in 6 months


Drug-wise Evaluation: Drug-wise Evaluation Trial 1 of switch A vs. non-switch Later Trial 2 of switch B vs non-switch Each could possibly contaminate the other End result: 2 monotherapy switch trials with co-intervention


Strategy-wise Evaluation 1: Strategy-wise Evaluation 1 Start Trial 1 of switch A vs non-switch Pre-schedule second randomization of switch B vs. non-switch when B is available


Strategy-wise Evaluation 2: Strategy-wise Evaluation 2 Wait 6 months Perform 2x2 factorial double switch trial with double sample size test interactive effects of double switch


Miscellaneous Considerations: Miscellaneous Considerations Randomized consent designs Adaptive assignment (e.g. play the winner) rapid endpoints required may be biased violates “simple, rapid, objective, foolproof”


Cost-Effectiveness Considerations: Cost-Effectiveness Considerations Quality of life Economic costs and savings Survival Quality Adjusted Life Years (QALYs) Cost per QALY


Some Comparators (US$ per QALY): Some Comparators (US$ per QALY)