Summary of the Biostatistics Core Meeting, July, 2004,
LSHTM
Participants
Neal Alexander,
LSHTM, neal.alexander@lshtm.ac.uk
Ashraf Chaudhary,
JHBSPH, mchaudha@jhsph.edu
Simon Cousens (second day), LSHTM, simon.cousens@lshtm.ac.uk
Richard Hayes, LSHTM,
richard.hayes@lshtm.ac.uk
Katherine Fielding,
LSHTM, katherine.fielding@lshtm.ac.uk
Larry Moulton, JHBSPH , lmoulton@jhsph.edu
Antonio Pacheco, FIOCRUZ, apacheco@fiocruz.br
Bill Pan, JHBSPH , wpan@jhsph.edu
Babis Sismanidis,
LSHTM, charalambos.sismanidis@lshtm.ac.uk
Andrew Thomson,
LSHTM, andrew.thomson@lshtm.ac.uk
Wednesday July 7
Introductions
10-11
- Updates on activities of past year
The primary activities, of course, focused on the development of study protocols, and integration of them into the main proposal to the B&M Gates Foundation. Recruitment activities resulted in bringing on, as part of the Biostatistics Core, Ashraf Chaudhary, Bill Pan, and Andrew Thomson.
- Updates on CREATE project timelines
Goldmines/Aurum (Thibelo): Will advertise in SA papers in a couple weeks for six key posts, including data base manager, and statistician. Will try for a South African statistician, but may go internationally, due to competition from the private sector for qualified personnel. There will be a 6-month preparation phase before individuals are enrolled into the study, toward the end of which randomization will take place.
ZAMSTAR: Hiring is commencing for the local
statistician and data base developer/administrator. Ready to
start baseline prevalence study in both
- Plans for Design, Analysis
This was a very important part of our meeting. The goal, largely met, was to get suggestions, insights, etc. from everyone to sharpen/improve the design and/or current analysis plans for each of these studies.
11-1
There were several good ideas, including a recalculation of the power analysis, the possibility of a paired randomization strategy, and the handling of actively-found TB cases.
1-2 Lunch: LSHTM Refectory
2-4
There was substantial discussion of the merits of paired vs. stratified designs, and criteria for stratification, for the 14 clusters in this study covered by 3-4 mining companies. Opinion was weighted toward stratification on geography and/or company.
4-6
This study, with 16 clusters in
Thursday July 8
- Allocation of interventions
This presentation covered the practical implementation of a couple schemes, coupled with semi-public randomization, and raised further questions regarding “validity”, and permutation-based analyses.
This program will soon be posted on the website; it allows for constraints both at the stratum and overall levels; random sampling of allowable allocations permits medium-to-large designs to be handled.
- Presentation by Andrew Thomson (Richard’s PhD student) on his research plans on cluster randomized controlled trials—Methods for trials with small numbers of clusters. This included work on implications of data generation assumptions on CV and ICC and their relationships across units, and on simulations for comparison of various analytic techniques.
- Research needs and collaboration (some of which may have been emphasized during project-specific discussions), and how to address them
Many topics were covered. Those that both are not yet fully covered in the literature, and that are likely to be needed for CREATE, include:
Effects of imbalance of covariates on 2-stage analyses
Analysis of stepped-wedge trials (combined cross-sectional Poisson vs.
Cox-type vs. crossover analyses)
Optimal choice of samples from clusters—flat weights may be best
Others avenues of investigation included multiple levels of correlation, time-trend analysis, cost-effectiveness, interim analyses and checking of correlation assumptions.
1-2 Lunch: LSHTM Refectory
2-3 Presentation by Fielding on analysis issues in a phased-enrollment study
3-5 Planning future activities
- Writing of design paper(s)
Two design articles are indicated:
one focused on the stepped-wedge aspects of the
In addition, it was decided that there is a need for a comprehensive text on cluster randomized trials, with more emphasis on infectious disease, and on design. Hayes has an upcoming sabbatical…
- Discussion of available expertise, resource-sharing, communication—with our relatively small group, email-based cc-ing to all involved seems the most efficient.
- Data base management issues
Will be abstracting data from medical records, to place into new data base.
Time constraint—need to train, in the near future, the 29 clinics’ personnel on data base items, forms.
Thibelo TB/Aurum—Fielding has been tracking the various requisite data elements/sources. Will have a student who will go to help out, get experience after graduation.
MS Access has been used; health economists are around who have experience with large data bases. Chris Seebregt is there, is advisor to the study (StudyBuilder experience).
ZAMSTAR—need to organize for baseline mapping, prevalence survey, then routine data for TB/HIV data, data on household intervention (process data); need to see how to capture data from TB registers (treatment results).
August Cape Town meeting will look at data management issues, with PIs. Will discuss standardization of data bases across the two countries.
A general discussion of data base practices led to the conclusion that while all aspects of GCP would not be rigidly enforced, sites might take advantage of the opportunity to implement as many of these as possible, with a concentration on documentation and maintenance of audit trails. Best to have quality control built-in, and have someone designated as being responsible for it, overlooking all processes.
Also, it was decided that it would be good to get as many data base personnel from the various sites together for discussion of data base issues.
- Organization for DSMB reporting
Likely DSMB concerns: timeline/execution; data quality; sufficient event rates; safety—INH resistance (on culture positive cases); toxicity; adherence; TB, mortality, HIV endpoints.
Steering Committees may be formed for one or more of the trials; these would see much of the same data, although more frequently than the DSMB; question as to whether would see data by intervention arm.
As far as possible, want to standardize the form of these reports, so that for similar tables, the DSMB does not have to get used to three different formats.
Not yet decided who will do final collation/sending out; rationale for JHU is that it has the greatest enforcement capability; for LSHTM because most work going on in their (Eastern) hemisphere.
- Subcontract reporting
Quarterly financial; annual effort/results
- September Baltimore meeting: Hayes and/or Corbett will put together some modeling information.
One goal is to try and get the sites to standardize definitions and outcomes as much as possible. Another is to try and itemize the possible ancillary/spinoff studies.
- Next Biostatistics Core meeting
Two possibilities:
a) plan the meeting in conjunction with another stats meeting
b) plan the meeting in one of the three sites
At this juncture, it may be best to do the latter, so that data issues and local teams can be at the forefront of discussions. In the future, it might be nice to organize a session at a stats meeting dealing with our trials and deliberations.
Thus: next summer (Jun/Jul) in