Factorial Clinical Trials: Design and Analytical Issues
Dear Sir,
At the outset I would like to congratulate BMJ & Prof. Christine
Stabell Benn1 and her group for a brilliant research paper on Fast Track
Basis.
The research paper had a prior hypothesis that vitamin A
supplementation would be particularly beneficial when provided with the
BCG vaccine conducted as factorial clinical trial. We must however
remember that even though factorial designs2 are clinical trials to assess
effect of more than one treatment in one clinical trial, their utility is
limited to two circumstances. When two or more treatments do not have
interaction (biologically, statistical or clinically), the effects can be
assessed with a relatively smaller sample size as compared to two or more
separate parallel armed studies. Thus we get two or more trials for price
of one. A factorial trial design is the only trial design to assess
interaction between two or more treatments as groups with all combinations
allow a direct comparison between them with larger sample size than
individual parallel group trials. Interaction is said to be present when
effect of treatment A is affected by presence of treatment B. Interaction
which is based on assumption of biological concepts is termed as
Biological interaction. But, many clinical trials concentrate only on
statistical interaction which is secondary to the biological plausibility
of presence of interaction. Much depends on the scale of measurement. It
can be Qualitative where presence of Treatment B reverses the treatment
effect of A or Quantitative where there is no change of direction.
In the present research paper, the p value for interaction was not
significant, probably due to insufficient sample size as described by
authors. Hence the presence of interaction is difficult to rule it out.
If sample size can not be increased then interaction would be at least
twice as large for main effects to be detected with same power and same
precision, which is very rare3.
Furthermore, in this paper, only vitamin A treatment results are
presented. Analytical issue is also a prime concern in factorial clinical
trials. Primary analysis should address the research question. Along with
reporting the results of each intervention, combination of two treatments
should also be reported which is commonly not done 3,4. This issue has
also been highlighted by many like McAlister 4 and CONSORT 5. As McAlister
describes in case of absence of interaction between two treatments, “at
the margin” analysis is appropriate while in presence of interaction
between two treatments, “inside the table” analysis is appropriate.
Interaction ratio with 95% confidence interval may also be required to
assess significant interaction. Use of multivariate regression models (if
required with interaction terms) may also be required. The results need
to be adjusted to avoid error due to multiple testing5. Thus I would like
to conclude that design and analytical issues are prime concern in a
Factorial Clinical trial.
Dr.Pankaj Shah
Associate Professor,
Dept. of Community Medicine,
Sri Ramachandra Medical College & Research Institute,
Sri Ramachandra University, Porur, Chennai-116,
Tamil Nadu, India.
1.Christine Stabell Benn, Ane Bærent Fisker, Bitiguida Mutna Napirna
et.al. Vitamin A supplementation and BCG vacination at birth in low
birthweight neonates: two by two factorial randomized controlled trial BMJ
2010;340:c1101 (Accessed online http://www.bmj.com/cgi/content/full/340/mar09_1/c1101 on 12.05.2010- as
Full article online Published on 9th March 2010.)
2.Steven Piantadosi, Factorial Designs In: Clinical trials: a methodologic
perspective. A John Wiley & Sons Inc. Publication Second Edition
2005: 501-514
3.Alan A Montgomery, Tim J Peters, Paul Little Design, analysis and
presentation of factorial randomized controlled trials BMC Medical
Research Methodology 2003, 3:26
4.McAlister; Sharon E. Straus; David L. Sackett; et al. Analysis and
Reporting of Factorial Trials: A Systematic Review JAMA. 2003;289(19):2545
-2553
5.David Moher, Sally Hopewell, Kenneth F Schulz CONSORT 2010 Explanation
and Elaboration: updated guidelines for reporting parallel group
randomised trials CONSORT 2010 Explanation and Elaboration: updated
guidelines for reporting parallel group randomised trials BMJ
2010;340:c869
6.Stephanie Green, Ping-Yu Liu, Janet O’Sullivan Factorial Design
Considerations Journal of Clinical Oncology, Vol 20, No 16 (August 15),
2002: pp 3424-3430
Competing interests:
None declared
Competing interests:
No competing interests
22 May 2010
Pankaj B Shah
Associate Professor,
Dept of Community Medicine, SRMC & RI, Porur, Chennai-116, Tamil Nadu, India
Rapid Response:
Factorial Clinical Trials: Design and Analytical Issues
Dear Sir,
At the outset I would like to congratulate BMJ & Prof. Christine
Stabell Benn1 and her group for a brilliant research paper on Fast Track
Basis.
The research paper had a prior hypothesis that vitamin A
supplementation would be particularly beneficial when provided with the
BCG vaccine conducted as factorial clinical trial. We must however
remember that even though factorial designs2 are clinical trials to assess
effect of more than one treatment in one clinical trial, their utility is
limited to two circumstances. When two or more treatments do not have
interaction (biologically, statistical or clinically), the effects can be
assessed with a relatively smaller sample size as compared to two or more
separate parallel armed studies. Thus we get two or more trials for price
of one. A factorial trial design is the only trial design to assess
interaction between two or more treatments as groups with all combinations
allow a direct comparison between them with larger sample size than
individual parallel group trials. Interaction is said to be present when
effect of treatment A is affected by presence of treatment B. Interaction
which is based on assumption of biological concepts is termed as
Biological interaction. But, many clinical trials concentrate only on
statistical interaction which is secondary to the biological plausibility
of presence of interaction. Much depends on the scale of measurement. It
can be Qualitative where presence of Treatment B reverses the treatment
effect of A or Quantitative where there is no change of direction.
In the present research paper, the p value for interaction was not
significant, probably due to insufficient sample size as described by
authors. Hence the presence of interaction is difficult to rule it out.
If sample size can not be increased then interaction would be at least
twice as large for main effects to be detected with same power and same
precision, which is very rare3.
Furthermore, in this paper, only vitamin A treatment results are
presented. Analytical issue is also a prime concern in factorial clinical
trials. Primary analysis should address the research question. Along with
reporting the results of each intervention, combination of two treatments
should also be reported which is commonly not done 3,4. This issue has
also been highlighted by many like McAlister 4 and CONSORT 5. As McAlister
describes in case of absence of interaction between two treatments, “at
the margin” analysis is appropriate while in presence of interaction
between two treatments, “inside the table” analysis is appropriate.
Interaction ratio with 95% confidence interval may also be required to
assess significant interaction. Use of multivariate regression models (if
required with interaction terms) may also be required. The results need
to be adjusted to avoid error due to multiple testing5. Thus I would like
to conclude that design and analytical issues are prime concern in a
Factorial Clinical trial.
Dr.Pankaj Shah
Associate Professor,
Dept. of Community Medicine,
Sri Ramachandra Medical College & Research Institute,
Sri Ramachandra University, Porur, Chennai-116,
Tamil Nadu, India.
Email : drpankajsshah@yahoo.co.in
References:
1.Christine Stabell Benn, Ane Bærent Fisker, Bitiguida Mutna Napirna
et.al. Vitamin A supplementation and BCG vacination at birth in low
birthweight neonates: two by two factorial randomized controlled trial BMJ
2010;340:c1101 (Accessed online
http://www.bmj.com/cgi/content/full/340/mar09_1/c1101 on 12.05.2010- as
Full article online Published on 9th March 2010.)
2.Steven Piantadosi, Factorial Designs In: Clinical trials: a methodologic
perspective. A John Wiley & Sons Inc. Publication Second Edition
2005: 501-514
3.Alan A Montgomery, Tim J Peters, Paul Little Design, analysis and
presentation of factorial randomized controlled trials BMC Medical
Research Methodology 2003, 3:26
4.McAlister; Sharon E. Straus; David L. Sackett; et al. Analysis and
Reporting of Factorial Trials: A Systematic Review JAMA. 2003;289(19):2545
-2553
5.David Moher, Sally Hopewell, Kenneth F Schulz CONSORT 2010 Explanation
and Elaboration: updated guidelines for reporting parallel group
randomised trials CONSORT 2010 Explanation and Elaboration: updated
guidelines for reporting parallel group randomised trials BMJ
2010;340:c869
6.Stephanie Green, Ping-Yu Liu, Janet O’Sullivan Factorial Design
Considerations Journal of Clinical Oncology, Vol 20, No 16 (August 15),
2002: pp 3424-3430
Competing interests:
None declared
Competing interests: No competing interests