The Statistical Society of Canada
Research Committee
Position Paper
on
The Canadian Institutes of Health Research
January 8, 2000
1. Introduction
The formation of the Canadian Institutes for Health Research (CIHR) is a dramatic development for health-related research in Canada. The Statistical Society of Canada (SSC) has a strong interest in the success of this initiative. Many members of our large and very active Biostatistics Section develop new statistical methodology specifically intended for health-related contexts in their programs of basic and applied research and apply this methodology in their collaborations with research teams in health-related disciplines. The SSC Research Committee feels the CIHR initiative is of great importance. Statistical scientists should be heavily involved in this initiative because the use of suitable and up-to-date statistical methodology, and hence strong methodological research, is an important component of world class research in many areas of the health sciences.
We understand that investigator-initiated research will be the cornerstone of CIHR-funded research and we fully support this principle. We also fully support the CIHR goal of fostering new synergies among researchers to solve health challenges based on integrated, collaborative and multi-disciplinary approaches. In our view, the primary goal of publicly-funded programs of research support should be to promote excellence. Such a program will encourage outstanding researchers already in Canada to continue to work here, and will encourage others to move to Canada thus enhancing the country's capacity for research.
The Committee has identified two main issues relating to the statistical sciences which are relevant to the CIHR. The first concerns the CIHR application and review process and the second concerns the funding of research in statistical methodology intended specifically for health-related contexts. These are discussed separately in the following sections.
2. The CIHR Application and Review Process
Much of the research carried out in the health sciences involves substantial data collection. This occurs in controlled experiments with animals (toxicology) or human subjects (Phase I, II and III clinical trials), in retrospective and prospective epidemiological studies (case-control, case-cohort, and behavioural/intervention studies), and in population studies. It is in these areas where statistical scientists can make important contributions. Such data-oriented research relies critically on efficient study design and effective statistical analysis to extract the maximal amount of information from the collected data. The inclusion of strong statistical expertise on the research teams for such studies will greatly benefit the quality of health research, as has been long recognised in the NIH programs.
To ensure that research teams integrate statistical expertise in suitable ways, we recommend that:
- Research proposals incorporate a thorough and careful treatment of statistical components relating to study design and proposed analysis. Proposals should identify the member of the research team responsible for providing the statistical guidance. Ideally this individual should be a co-investigator so their expertise is clearly identified. A detailed description of the planned statistical analyses should be included as part of any data-oriented proposal. Explicit guidelines concerning what is expected of research proposals involving the collection, analysis and interpretation of statistical data might usefully be developed.
To ensure that science of the highest quality is carried out, the review process must consider not only the importance of the questions posed for investigation, but also the efficiency with which those questions will be addressed. To promote explicit and thorough evaluation of the quality of the statistical components of proposals, we recommend that:
- Peer review committees seek input from statistical scientists in their evaluation of research proposals. All proposals involving the collection, analysis and interpretation of statistical data should be sent to statistical scientists for review of the study design and analysis aspects. Peer review committees in disciplines with a preponderance of data-oriented research should have a statistical scientist as a member. Other peer review committees might utilize a statistical scientist as an "advisor" during the evaluation process.
We recognize that this second recommendation is already being followed in some degree. For example, the 1999/2000 MRC peer review committees for Clinical Trials and Population Health both have statistical scientists as members, and we are aware of SSC members who serve as MRC reviewers. Our concern is that this may not be happening to the degree it should. For example, peer review committees without statistical scientists as members may have difficulty identifying those proposals which should be subject to a review of their statistical components or the most appropriate statistical reviewers for proposals. Such lost opportunities for the best possible constructive feedback to applicants on aspects of study design and statistical analysis will result in suboptimal utilization of scarce research resources and the quality of the health-related research will suffer as a consequence.
Implementation of these recommendations would ensure rigorous evaluation of the statistical components of research proposals submitted to the CIHR and would lead to greater levels of collaboration by statistical scientists in laboratory, clinical, and epidemiological research. Such involvement brings statistical scientists into contact with important questions arising in health-related research where new statistical methodology is needed. This leads to a "virtuous circle" of enhanced research activity in both disciplines, with more efficient studies and more defensible conclusions in health-related research.
3. Funding of Research in Statistical Methodology
Statistical scientists have played an important role in health-related research since at least the time of Florence Nightingale, but many of the contributions of greatest impact have been made in recent decades. Statistical research on designs for clinical trials brought scientific credibility and greater efficiency to the critical investigation of new therapies. Recent advances in this area include group sequential and other designs to allow planned interim analyses for the possible early termination of a trial so that patients not in the trial may receive the benefit of an effective new therapy sooner than would otherwise be possible. More recently, statisticians have devised novel methods to analyse data collected by epidemiologists in case-control and cohort studies. This has allowed the use of statistically-based techniques for the planned oversampling of cases to achieve great savings, while maintaining efficiency and accuracy. Research in this area is still actively underway.
A breakthrough in health-related research was the development in 1972 by the British statistician, Sir David Cox, of the method of "partial likelihood" to allow the statistical assessment of time-to-event data in the presence of censoring using the proportional hazards model (or "Cox model") to incorporate the influence of explanatory variables. This discovery began an explosion of research activity in ``survival analysis" which continues to the present day and Canadian statisticians have played a prominent role in these developments. This discovery greatly enhanced the amount of information that could be extracted from data collected in many clinical trials. Analysis of such experiments became much more informative, providing more precise evaluations of therapies and greater understandings of disease development. Extensions of the Cox model to analyse experiments involving multiple ``lifetime" events are currently under active development.
The field of survival analysis is only one example of research in statistical methodology specifically intended for health-related contexts. Other areas of recent and current activity include stochastic models to aid the understanding of the future development of the AIDS epidemic, methods for the validation and analysis of biological markers and surrogate endpoints used to increase the efficiency of studies, methods for more comprehensive analysis of quality of life data, methods for the analysis of medical imaging data, and methods for dealing with large measurement errors which can arise, for example, in studies on diet and disease.
Both MRC and NHRDP have occasionally, but rarely, funded statistical projects and researchers. NIH has a long-standing program to fund research in statistical methodology intended for health-related contexts and this explains, at least in part, why the field of biostatistics has flourished to a much greater extent in the US. That several of the very prominent biostatistical researchers working in the US began their professional careers at Canadian universities but subsequently moved to the US may be explained by the better levels of funding available to strong researchers in biostatistics in the US. Similarly, due largely to NIH support of methodological research, promising young biostatistical researchers find a more hospitable research environment south of the border where many statistical research groups affiliated with medical centres receive a large portion of their funding from the NIH. There are no comparable groups in Canada. The NIH has, for many years, provided strong support to research in statistical methodology and clearly sees benefits of this support in the quality of health-related research generally.
To encourage the development of new statistical methodology for health-related contexts, we recommend that:
- The CIHR establish a mechanism to fund research on statistical methodology intended for health-related contexts. As new statistical methodology developed for one health-related context inevitably finds application in other contexts as well, this mechanism should cut across the different institutes that are to be established.
4. Conclusion
The SSC Research Committee strongly supports the CIHR initiative. The above recommendations are intended to enhance the success of the CIHR. Statistical methodology plays a prominent role in research in the health sciences and statistical scientists in Canada have much to contribute to the success of this initiative.
Research Committee, Statistical Society of Canada:
John Petkau, University of British Columbia, Chair
Charmaine Dean, Simon Fraser University
Agnes Herzberg, Queen's University
Louis-Paul Rivest, Université Laval
James Tomkins, University of Regina
Addendum:
On January 13, 2000, President Kalbfleisch forwarded this brief to Dr. Henry Friesen, President of the Interim Governing Council of CIHR, with copies to others at NSERC, MRC and CIHR. He received a response, dated January 25, from Dr. Mark A. Bisby, Director, Programs Branch of the MRC. Dr. Bisby remarks that "the newly constituted health information" peer review committee includes "health statistics and biostatistics" as a major part of its mandate. He asks that it be made known among the members of the SSC that "MRC and its successor CIHR already welcome research proposals in this area."