Different types of trial design and implications for reporting Alison Wearden

Uncontrolled trials Test feasibility and acceptability of an intervention, and whether there are adverse effects May allow preliminary examination of mechanisms of change N-of-1 designs, randomized schedule, can be used to test theory (cause and effect)

Methods of bias control Selection bias Randomized treatment assignment Types of BIAS Methods of bias control Selection bias Randomized treatment assignment Concealment of assignment Bias in study management Standardized operating procedures Training of research personnel Ascertainment bias Blinding of researchers to treatment assignment Bias introduced after randomization Intention-to-treat analysis Publication bias Prospective registration of trials Publication of negative trials From D. Wang & A Bakhai (Eds) Clinical Trials. A Practical Guide to Design, Analysis and Reporting. 2006, Remedica: London, pp 57

The hegemony of pharmacological trials Blinded randomised controlled trials considered gold standard for testing drugs Heavily influenced guidelines for RCTs Are drug-RCT issues always appropriate for complex behavioural interventions? Can attempts to eliminate bias distort implementation and assessment of intervention?

Types of CONTROLLED TRIALS What factors do you need to take into account choosing a trial type and design? What is your research question? Are there ethical constraints? Where might you want to publish your trial? Efficacy Explanatory Parallel arms/ factorial? Effectiveness Pragmatic Patient preference Equivalence Non-inferiority

Explanatory vs pragmatic Efficacy vs effectiveness Explanatory, efficacy Does this intervention work in people who receive it under carefully controlled conditions? How does it work? Pragmatic, effectiveness What would be the outcome if this intervention were implemented in usual clinical practice? Broad, inclusive sample More flexible implementation Qualitative studies, e.g. to determine why people dropped out Tightly defined sample Fixed protocol Compliance measured Process measures See Zwarenstein et al., 2008, BMJ, 337, 1223-6

Intention to treat analysis Preserves the benefits of randomization, ie minimizes bias. Withdrawals may be more common in one arm of trial Ecological validity (people do drop out) It gives a pragmatic estimate of the effectiveness of a treatment, rather than just a report of the efficacy of the treatment It requires a method for dealing with missing data

Equivalence trials Instead of having a “no treatment” control, equivalence trials test whether a new intervention is as good as (or not worse than) an established treatment with proven efficacy New intervention may have some advantage (convenience, cost) over established, but is it at least as efficacious?

Randomisation issues Unit of randomisation (individual, group) Simple or block design (to even up group numbers) Stratification or minimisation to achieve groups balanced on key baseline characteristics Patient preference designs Zelen Wennberg Rucker Comprehensive cohort

Preference designs Patient preference may be an issue refusal to participate reduced compliance with non-preferred arm Consent may be taken AFTER randomization, e.g. Zelen design: Eligible patients Randomized Novel Treatment B Consents Receives B Declines Receives A Standard Treatment A (May be analysed as randomized)

Some trial designs Simple, parallel design Is A better than control? Factorial design Are effects of A and B additive or interactive? A & B each controlled Cross-over design (Unlikely – useful with small samples) A Eligible Randomised to A or control control Randomized to 1,2,3 or 4 A Control for A B 1 = A+B 2 = B + control for B 3 = A + control 4 = Double control Eligible Randomised to A THEN B or B THEN A A B B A

Analysis issues Pre-specification of primary outcomes if there are multiple measures Analysis plan pre-specified and published Intention to treat vs per protocol analysis How are missing data dealt with?

Where do you want to publish? Which audience do you want to reach? Does impact factor matter? Uniform requirements for “prestigious” medical journals Trial must be registered Protocol must be published CONSORT guidelines must be adhered to Important issue of publication bias

Special considerations for reporting different types of trials Standard CONSORT guidance is at http://www.consort-statement.org/consort-statement/ designated primary outcomes sequence generation allocation concealment …. CONSORT guidelines have been elaborated for trials with complex interventions (cf. drug trials) and non-simple designs

Reporting of psychological trials should include Components of the intervention(s), how they were individualised (if applicable), how they were standardized How therapist fidelity to treatment assessed How experimental intervention and comparator(s) were implemented Description of therapists (specialism, experience etc) and setting Boutron et al., 2008, Ann Int Med, 148, 295-309

Reporting pragmatic trials Explanatory vs pragmatic “attitude” Sample description and eligibility criteria – is sample typical? Is setting typical? CONSORT diagram should explain reasons for non-participation if known Particular attention to clinical interest of findings Description of key aspects of setting which affected findings Zwarenstein et al., 2008, BMJ, 337, 1223-6

Issues with equivalence, or non-inferiority, trials People may be carrying out equivalence trials without realising it. Analysis with respect to a pre-stated margin of non-inferiority (smallest clinically interesting difference) ITT analysis may increase risk of type 1 error Choice of outcomes important People may be carrying out equivalence trials without realising it. Piaggio et al., 2006, JAMA,295,1152-1160

Reporting equivalence trials Need to reference established efficacy of “standard” treatment Hypotheses should be framed in terms of non-inferiority “Margins of equivalence” should be reported