Rapid Responses to:
|
|
Rapid Responses: Rapid Responses published:
-
![[Read Rapid Response]](/icons/misc/sectionDOWN.gif)
Correction of facts
- John Parkinson (14 February 2009)
-
Re: Correction of facts
- Richard L Tannen (15 February 2009)
-
Limitations of prior event rate ratio adjustment
- Arlene Gallagher, Frank de Vries, Tjeerd-Pieter van Staa (17 February 2009)
-
Re: Re: Correction of facts
- Richard L Tannen, Mark G. Weiner, and Dawei Xie. (26 February 2009)
|
|
|||
|
John Parkinson, Director, GPRD MHRA, Market Towers, 1 Nine Elms Lane, London SW8 5NQ
Send response to journal:
|
I read with interest the paper by Tannen et al that is based on an old, never refreshed, now 7+ years out-of date, version of GPRD and am disturbed to see listed some apparent limitations of "The GPRD database". In the main these limitations refer to this old copy with the consequence that the information, as published, is highly misleading to the many existing users of the fully supported and up-to-date GPRD and to those considering its use. I am surprised to see that ethical approval is stated as, “not required”. This is contrary to the GPRD governance requirements. Indeed the authors did apply and were granted approval (15/02/2002) by the then GPRD Scientific and Ethical Advisory Group (SEAG) for permission to undertake a series of concordance examinations of observational data in relation to specific clinical trials. The authors could argue that as their protocol approval came in 2002 they have been working for 6 years on data supplied at that time. This may or may not be correct, but, it is not an excuse for making the totally out -of-date statements about GPRD limitations. GPRD contains data on 12.5 million patients, has direct links to laboratory data, contains over 117 million BMI, smoking and alcohol records, is record linked to hospital data and for specific studies can have a direct link to central death data. Competing interests: I am the Director of GPRD |
|||
|
|
|||
|
Richard L Tannen, Professor of Medicine University of Pennsylvania School of Medicine, Philadelphia, PA 19104, USA
Send response to journal:
|
We thank Dr. Parkinson for his response, which correctly addresses two issues not appropriately clarified in the manuscript. In regard to the version of the GPRD database used the original statement in the last version of the manuscript submitted read as follows: "The UK GPRD database USED IN THESE STUDIES contains information from the electronic medical records of Primary Care Practices encompassing a representative sample of approximately 5.7% of the UK population over the time frame from 1990 - 2000 and contains records of over 8 million patients (24,25." Unfortunately the modifier noted above in capitals was eliminated during editing and we missed this important change. We apologize for the confusion this may cause. The statement that ethical approval was "not required" should have been addresed by us more carefully. We had stated in communications regarding the manuscript that "Since Ethics Committee Approval was obtained and cited in all the individual publications used as the primary data for this manuscript, we did not think a statement to that effect would be needed for this publication. This ethics review included both the GPRD ethics board and the University of Pennsylania Human Subjects Institutional Review Board. Again we apologize for any confusion this may have caused. Richard L. Tannen Competing interests: None declared |
|||
|
|
|||
|
Arlene Gallagher, pharmacoepidemiologist (GPRD, London UK) and PhD student (Universiteit Utrecht, Netherlands) 1 Nine Elms Lane, London SW8 5NQ, United Kingdom, Frank de Vries, Tjeerd-Pieter van Staa
Send response to journal:
|
Tannen et al suggested that prior event rate ratio (PERR) adjustment can solve issues with confounding in observational studies. PERR adjusted Hazard Ratio (HR) is the HR during follow-up divided by the HR 'prior' to start of follow-up. [1] We wonder whether full consideration has been given to incomplete prior case capture and immortal time bias [2]. If a patient suffers a fatal event prior to exposure, this will not be counted in the prior event rate. Also, PERR adjustment will be biased if prior events modify the risk of subsequent exposure. Tannen et al recommended simulation studies to evaluate the limitations of PERR adjustment [1]. We conducted two simulation studies. The first study evaluated the effects of case fatality (i.e. patients dying before reaching exposure). A hypothetical population of 500,000 patients was divided into low and high risk groups. They had differential case fatality (+10% to +50%) and likelihood of exposure (with no effect of exposure on outcomes). The second study evaluated the effects of 'prior' case occurrence modifying the likelihood of future exposure (-10% to - 50%). PERR adjustment yielded valid results (PERR HR of 1.0) only when case fatality did not vary between high and low risk patients. Erroneous results were obtained when high risk patients were more likely to die (PERR HRs up to 4). If prior case occurrence modified the likelihood of future exposure, PERR adjustment provided erroneous results (PERR HRs up to 10). We feel that the PERR adjustment may be susceptible to common biases. Arlene Gallagher, pharmacoepidemiologist (General Practice Research Database [GPRD], London UK) and PhD student (Utrecht Institute for Pharmaceutical Sciences [UIPS], Utrecht, The Netherlands) Frank de Vries, pharmacoepidemiologist (GPRD) and assistant professor (UIPS) Tjeerd-Pieter van Staa, head of research (GPRD) and assistant professor (UIPS) References [1] Tannen RL, Weiner MG, Xie D. Use of primary care electronic medical record database in drug efficacy research on cardiovascular outcomes: comparison of database and randomised controlled trial findings. BMJ 2009;338:b81, doi: 10.1136/bmj.b81 published online 27 January 2009. [2] Suissa S. Immortal time bias in pharmaco-epidemiology. Am J Epidemiol. 2008;167:492-9. Competing interests: “The division of Pharmacoepidemiology & Pharmacotherapy employing all authors has received unrestricted funding for pharmacoepidemiological research from GlaxoSmithKline, Novo Nordisk, the private-public funded Top Institute Pharma (www.tipharma.nl, includes co-funding from universities, government, and industry), the Dutch Medicines Evaluation Board, and the Dutch Ministry of Health”. All authors conduct work for the General Practice Research Database. |
|||
|
|
|||
|
Richard L Tannen, Professor of Medicine University of Pennsylvania School of Medicine, Philadelphia, PA 19104, USA, Mark G. Weiner, and Dawei Xie.
Send response to journal:
|
We were aware of the issue of “Immortal Time Bias”, but do not believe that our studies are susceptible to this problem for the following reasons. First the study had a predefined time interval during which Exposed and Unexposed subjects that met the entry criteria could be entered into the study. In the PERR analyzed subset neither the Exposed nor Unexposed subjects took the study medication at any time prior to their designated study start time. In the Exposed group study start time was designated as the date of the first prescription of the study medication. The Unexposed subjects were chosen from a large pool of potential Unexposed subjects by random matching to an Exposed subject of similar gender and age, and their start date was considered the same as the Exposed subject. Thus, in effect, the start date of the Unexposed subjects was randomly assigned. This study design does not lend itself to “immortal time bias”. It is unclear exactly how and what the simulation described by Gallagher and co-workers addressed, so it is not possible to comment critically about their findings. Additionally we would note that we have performed preliminary simulation studies of the PERR method. These simulations show that when unmeasured confounding is present, the results of a standard Cox HR and a PERR adjusted Cox HR differ in the same fashion as was found with the empiric results reported in our publication. Competing interests: None declared |
|||