Perhaps we are using two different meanings of the word “crossover”

In a randomized parallel trial patients are randomly assigned to A vs. B.   Over time patients may crossover from one assignment to the other, or perhaps after the randomized period is done all patients in B will be switched to A if there is a perception that A is superior therapy.   This “crossover” is referring to changes in treatment assignment in ways other than through the randomization process.  Crossover in this context may bias or complicate the analysis.

In a randomized crossover trial the initial design is give all patients a period of A and a period of B.  What is randomized is the order of which period comes first.   The intention is to compare A vs. B within the same patient.  Crossover in this context is a key factor of the primary research design.

We have analyzed many cancer trials with outcomes of progression-free survival but they are randomized parallel trials, often with complicated patterns with multiple stages of care and potentially additional randomizations (one randomization for induction chemotherapy, another randomization for consolidation, etc.).  But this is not the randomized crossover trial in which every patient gets both interventions being compared.

One would generally not compare 2 different chemotherapy regimens by having the same patient experience both regimens.   Whereas one could reasonably expect to compare two different analgesics in this way for a chronic pain.

The goal here is to define critical appraisal criteria for trials designed as randomized crossover trials with an expectation for paired (within-patient) analyses.   This is not about how to manage crossover that occurs in randomized parallel trials.

Unfortunately, these are rare trials…vast majority of (modern) cancer trials are based on clinical outcomes such as progression-free survival (PFS) and employ a cross-over  …standard practice has always relied on intention-to-treat analysis, with which clinicians are never comfortable… my point in raising the issue of cross-over in cancer trials is that it remains a methodologically unresolved problem…most people (as this exchange also demonstrates) dismiss them as not true cross-over trials…but this will not make a) cross-over disappear from cancer trials, b) how do you assess the quality of these trials (indeed, how have you assessed the quality of cancer trials in DynaMed where primary outcome was PFS and the cross-over was employed? hint: if you have appraised them as low quality, in many cases you would not agree with the FDA, which has granted approval for marketing for a number of cancer drugs based on such a design)

ben

Not necessarily.  The quality assessment is specific to the outcome.   So such a trial could provide high-quality data on symptom control (perhaps the reason for the trial) and if differences in survival were discovered between crossover periods one would have to consider if this is potentially biased with period effects (second period patients would be limited to those who survived first period).

Survival is not an outcome for which a crossover trial would be an appropriate design.    There is no way to avoid a carryover effect – washout period won’t reverse death.   A key part of critical appraisal of crossover trials is to determine if the crossover design is appropriate for the research being conducted.

 

 

Brian S. Alper, MD, MSPH, FAAFP

Editor-in-Chief, DynaMed (dynamed.ebscohost.com)

 

From: Djulbegovic, Benjamin [mailto:[log in to unmask]]
Sent: Thursday, July 04, 2013 7:49 AM
To: Brian Alper MD
Cc: [log in to unmask]
Subject: Re: Critical appraisal criteria for randomized crossover trials

 

Brian,

No oncology trial ( with survival endpoint) would meet these criteria.

Best

Ben 

Sent from my iPad

( please excuse typos & brevity)

On Jul 4, 2013, at 5:58 AM, "Brian Alper MD" <[log in to unmask]> wrote:

EBH listserv members

 

Periodically we evaluate the critical appraisal criteria we use for DynaMed.   This listserv has helped us in the past (2008) when we added  critical appraisal criteria for randomized trials in the case of early trial termination.

 

We have found critical appraisal criteria for randomized parallel trials are not optimal for analyzing crossover trials.  The CONSORT statement has numerous extensions but does not have an extension for crossover trials.   The Cochrane Handbook was updated in 2011 (version 5.1.0) to include a risk of bias assessment criteria for crossover trials.

 

In consideration of critical appraisal criteria for randomized trials used today and the risk of bias criteria for crossover trials from the Cochrane Handbook we have developed the criteria listed below as a planned revision to our level of evidence criteria for DynaMed (found at https://dynamed.ebscohost.com/content/LOE).  We welcome your feedback in consideration of refining these criteria before we implement them.

 

DynaMed criteria for level 1 (likely reliable) evidence for interventional conclusions from crossover trials – must meet ALL criteria to be considered level 1 evidence

 

1)      Conducted in patients with condition not expected to change spontaneously during course of trial

2)      Clinical outcome (also called patient-oriented outcomes)

3)      Random allocation method for order of assignment (i.e. not assigned by date of birth, day of presentation, “every other”)

4)      Washout period between interventions long enough to avoid carryover effects between interventions

5)      Adequate duration of intervention and assessment period to represent outcome being measured

6)      Blinding of all persons (patients, treating clinicians, outcome assessors) if possible

7)      Follow-up (endpoint assessment) of at least 80% of trial participants AND adequate such that losses to follow-up could not materially change results

8)      Accounting for dropouts (even if not included in analysis)

9)      Analysis of paired data

10)   Analysis not suggesting period effects (i.e. effect resulting for order of intervention), or period effects if present not materially changing results

11)   Adequate statistical power

12)   No other factors contributing substantial bias, such as subgroup or post hoc analyses

 

Thank you.

 

Brian

Brian S. Alper, MD, MSPH, FAAFP
Editor-in-Chief, DynaMed
Medical Director, EBSCO Information Services
[log in to unmask]
10 Estes Street
Ipswich, MA  01938-0231
Phone 800-356-6500 or 978-356-6500 ext. 2749
Cell 978-804-8719
Fax 978-356-7332
dynamed.ebscohost.com

 "It only takes a pebble to start an avalanche."