• Open Access

A retrospective analysis of VIOXX in Australia: using clinical trial data and linked administrative health data to predict patient groups at risk of an adverse drug event

Authors


Correspondence to:
Dr Margaret Whitstock, PO Box 4264, Melbourne University, Victoria 3052. E-mail: whitstock@iinet.net.au.

We now have the judgment from the Australian VIOXX® class action, which declares that the drug approximately doubled the risk of heart attack.1

The potential for adverse drug events (ADEs) is greater when a new drug is prescribed soon after its release onto the market, and rapid uptake can place patients at risk of adverse drug events. This was the case with the early large-scale adoption of Celebrex® and VIOXX by Australian GPs.2 The potential for unexpected ADEs is due in part to the highly selected populations used in clinical trials, and the consequent limited information about the action of the new drug in the presence of comorbidities and co-medications, and in part to the limited time of observation of the new drug and the likelihood that its full range of action is not yet known.

As a retrospective exercise, we applied a newly developed risk identification model to predict the likelihood of ADEs in patients prescribed VIOXX. This model combines publicly available clinical trial data and linked de-identified hospital morbidity and Pharmaceutical Benefits Scheme (PBS) medications data to develop an age- and gender-sorted risk profile for current Australian patient groups who have the morbidity targeted by a new drug, but who are likely to be at risk of an ADE if they are prescribed that new drug in the context of their age, gender, comorbidities and co-medications.

We used data from 14 published VIOXX clinical trials known to be available before VIOXX was included in the PBS in February 2001, plus de-identified linked morbidity and medications data on an Australian subset of the arthritis population that would have been the prospective target population for VIOXX.

Morbidity data were drawn from the W.A. Department of Health Hospital Morbidity Data Collection, and the PBS medications data were supplied by the Department of Health and Ageing, under the Cross-Jurisdictional Linked Administrative Health Data Project managed by the WA Department of Health Data Linkage Unit.

We analysed the 14 VIOXX RCTs across a number of factors, including the comorbidities and co-medications exclusion protocols. Though part of the RCT study design, these are often applied quite strategically. A summary of the exclusion protocols for the 14 trials showed limited participation of patients with osteoarthritis (OA) or rheumatoid arthritis (RA) with one or more of cardiovascular morbidities (Class III/IV angina or CHF, hypertension, requirement for anticoagulant therapy, MI within 1 year prior), cerebrovascular morbidities (TIA within 2 years prior, CVA within 2 years prior), or advanced renal disease (creatinine clearance ≤ 30 mL/min). These exclusions meant that when VIOXX was made available through the PBS, there were effectively no data on its action in patient groups who were being treated for these morbidities.

We then analysed the comorbidities and co-medications of W.A. OA/RA patients aged ≥65 years in 1999 and 2000. Only 2000 cohort data will be reported on here.

There were 58,968 patients aged ≥65 years in the 2000 cohort who were prescribed a high-use medication specifically targeting OA/RA (usage based on dispensing data in Australian Statistics on Medicines). Of these 58,968 patients, 40,495 were also prescribed a medication for treating cardiovascular disease. 68.8% of patients in the 2000 cohort were being prescribed medications for co-existing OA/RA and CVD.

There were 34,269 hospital morbidity records for patients aged ≥65 years in the 2000 cohort. These included records for 2,171 patients with an OA/RA diagnosis only, records for 3,150 patients with an OA/RA diagnosis who were also prescribed an OA/RA medication, and records for patients with non-OA/RA morbidities who were prescribed an OA/RA medication.

The PBS and hospital morbidity records were then merged. Of the 58,968 patients taking an OA/RA medication, there were 3,522 patients with a hospital CVD diagnosis, 666 patients with a hospital CVA diagnosis, and 551 patients with a hospital diagnosis of renal disease. The presence of CVD, particularly in males aged 65–79 years, was clearly evident in the 2000 hospital morbidity data.

Had the clinical trials analysis and the 2000 cohort analysis been available, they would together have sounded a warning to prescribers.

We finished the VIOXX exercise by examining the co-prescribing for all patients prescribed VIOXX between January 2003 and September 2004. Of the 38,010 persons prescribed VIOXX in that period, 25,466 were also prescribed CVD medications. That 67% overlap indicated a lack of awareness of the link between CVD and VIOXX right up to the date of its safety withdrawal.

The increasing burden of ADEs on the Australian health system has been clearly documented, and it is argued that we need new systems to deal with the reality that only limited numbers of highly selected patients are studied before a drug is marketed, and such systems must have the capacity to detect early signals of potential ADEs in large populations.3

The risk identification model using clinical trial data and linked administrative health data, as described above, can provide that level of prediction.

Ancillary