An observational study of the discrediting of COX-2 NSAIDs in Australia: Vioxx or class effect?

NSAIDs (Non-Steroidal Anti-inflammatory Drugs)have been consistently associated with adverse gastrointestinal (GI) and renal effects [1]. Selective COX-2 inhibitors (COX-2s) that promised to minimize adverse GI effects [1] were first marketed in Australia in 1998, and listed on the national medicines subsidy Pharmaceutical Benefits Scheme (PBS) from 2000. While PBS guidelines suggested that COX-2s should be prescribed only to patients with a history of GI disorders, concomitant use of corticosteroids, anti-coagulants and advanced age, rather than as routine therapy [2], prescriptions for COX-2s increased rapidly, peaking at about 250,000 Australian users in 2004 [3, 4], suggesting these guidelines were not being followed.

The expected advantage of fewer GI side effects for COX-2s compared to non-selective NSAIDs (ns-NSAIDs) was supported by longer-term safety studies, but early studies also showed an increase in cardiovascular (CVD) and renal events [5, 6]. Subsequent studies found a fourfold increase in risk of myocardial infarction (MI) for rofecoxib (Vioxx) users compared to naproxen users (ans-NSAID) [7], an excess of CVD events in trials of COX-2 efficacy in preventing recurrent colonic polyps [8, 9], and a greater risk of coronary heart disease for high-dose rofecoxib (> 25 mg/day) users in observational studies [10]. Safety concerns intensified and rofecoxib was withdrawn by the manufacturer world-wide in September 2004 [11, 12]. While similar concerns were expressed in relation to other COX-2s [13], these medicines were not withdrawn. In Australia, the Therapeutic Goods Administration (TGA) required manufacturers to provide explicit product information warnings about CVD risk and advised that all medicines in the COX-2s class should be regarded as having increased CVD risk [14]. Regardless, two COX-2s, celecoxib and meloxicam, were both among the top 25 highest volume medicines dispensed on the PBS in 2006 [15]. In 2007, the TGA cancelled registration of lumiracoxib, a newly available COX-2, due to concerns about serious liver side effects, further discrediting this drug class [16]. After the withdrawal of rofecoxib, paracetamol (acetaminophen) was widely promoted as first line therapy, especially for older people with arthritis [17], given that ns-NSAIDs have also been associated with increased CVD risk [18‐20].

International examination of the context of the discrediting of the COX-2s and subsequent effects on prescribing practices have shown different medicine switching patterns according to country differences in prescriber characteristics, professional recommendations and therapeutic guidelines [5, 21‐30].

Longitudinal evaluation of a US pharmacy claims database found increased prescribing of other NSAIDs, including those with relative COX-2 selectivity, after the withdrawal of rofecoxib and another COX-2, valdecoxib [24]. A large US nationally representative cross-sectional survey of ambulatory care visits showed relatively stable NSAIDs prescribing from 1999 to 2005; initially, COX-2s substituted ns-NSAIDs, but after withdrawal of rofecoxib, prescribing of ns-NSAIDs and non-narcotic analgesics rose sharply [28], suggesting the perception of a class effect.

In the UK, data from the General Practice Research Database showed that approximately 80% of those using COX-2s stopped within 6 months of the major discrediting event (the withdrawal of rofecoxib) [27]. In Scotland, data from a national prescription database showed that the withdrawal of rofecoxib led to a short-lived initial increase in prescription of celecoxib, and a parallel increase in the prescription of other NSAIDs, suggesting these were prescribed as alternatives to rofecoxib [21]. In Ireland, interrogation of a national prescribing database found prescribers were less likely to switch patients previously receiving rofecoxib therapy to another COX-2 irrespective of age or gender, either in the short or longer term [23]. The authors suggest that prescribers viewed rofecoxib related CVD safety concerns as a class effect [30]. However, another Irish study found evidence of chronic use of COX-2s despite discrediting [22], with female patients, those over 65 years and those at CVD risk more likely to start celecoxib rather than a ns-NSAID, suggesting that prescribers may not have perceived the discrediting as a class effect [22, 30]. Other European studies had varying findings. A small Italian study found that rofecoxib withdrawal resulted in a marked decrease of COX-2 use in general practice, indicative of a belief in a class effect [26]. In contrast, a German study, using data for patients within the statutory health insurance system [29], rofecoxib withdrawal led to initial increased prescribing of other COX-2s (celecoxib, valdecoxib), similar to the situation in Scotland [21]. Subsequent safety warnings on COX-2s and the withdrawal of valdecoxib in April 2005 led to pronounced reductions in COX-2 prescribing, and increased ns-NSAID prescribing. While German physicians responded promptly to safety warnings [29], reduced prescribing of COX-2s led to a simultaneous increase in NSAID prescribing, and especially those NSAIDs with a relative preference for COX-2, which may be associated with COX-2 like safety risks [29]. In the Netherlands, a study of individual prescription patterns of COX-2 users showed that while discrediting led to a significant decrease in use of COX-2s, large numbers of patients appeared not to switch to alternative pain treatments [25].

In Australia, aggregated PBS dispensing data have been used to examine the influence of COX-2s on overall NSAIDs and paracetamol prescribing afterCOX-2 discrediting [2]. Overall NSAIDs use declined, with evidence of slight increases in the use of paracetamol in April 2005. Reduced total COX-2 prescribing following rofecoxib withdrawal suggested prescribers saw discrediting as a class effect. While these aggregated data show the rise and fall of COX-2 prescribing in Australia, individual switching behavior can only be assumed, given the lack of examination of individual level longitudinal data. A study of COX-2 use in an Australian veteran population [5], found that despite the increased vulnerability of veterans receiving heart failure or diabetes medicines to adverse NSAIDs effects, uptake rates of COX-2s were similar to the rest of the veteran population, suggesting that while GI safety messages were interpreted broadly by prescribers, other adverse effects were not considered [5]. No Australian studies have evaluated the effects of COX-2 discrediting on individual level patterns of medicine use for a large general population sample [4].

When a medicine is withdrawn, as in the case of rofecoxib, follow-up at the individual consumer level is uncommon, as it can be difficult and costly. The Australian Longitudinal Study on Women's Health (ALSWH) provides a rare opportunity to examine consumer medicine use behaviours following a discrediting event in a large cohort of older women. The overall objective of this paper was to examine the impact of the major COX-2 discrediting event (withdrawal of rofecoxib) on dispensing of the COX-2 class of medicines. The specific aims were to describe:

1. medicine switching after rofecoxib withdrawal by older Australian women using rofecoxib at30 September 2004;

2. characteristics of women who switched to an alternate COX-2 in the first 100 days after 30 September 2004; and

3. patterns of use of alternate COX-2s taken up in the first 100 days after rofecoxib withdrawal, in the next 100 days after this medicine switch.

Data for this project are from Australian Longitudinal Study on Women's Health (ALSWH) Surveys 3 (S3) and 4 (S4) for the 1921-26 birth year cohort(undertaken 2002 and 2005), individually linked to PBS medicines data. The ALSWH is a national study which began in 1996 with a random sample of more than 40,000 women in three age cohorts [31]. Since 1998, follow-up surveys have occurred on a three yearly staggered cycle by age cohort. The Older cohort of women (n = 12,432) were 70-75 years at S1 in 1996 and 79-84 years at S4 (distributed March 2005, after discrediting). Retention rates over the 10 years of the study have been high (over 80% at each survey). Further information about ALSWH can be found at http://​www.​alswh.​org.​au[32]. The University of Newcastle Human Research Ethics Committee approved all aspects of the study (H-076-0795).

This study explored individual patterns of medicine use and characteristics of COX-2 users following the withdrawal of rofecoxib (Vioxx) from the market, for women with a "concessional beneficiary" status in a national general population sample of older Australian women. The type of medicine switched to and the pattern of medicine use differed significantly depending upon whether a woman was a continuous(high use) or non-continuous (lower use) rofecoxib user in the 12 months prior to the discrediting event. Continuous users overwhelmingly switched to another COX-2 and remained a continuing COX-2 user for at least 100 days post-switch. The switching behaviour observed here suggests that the safety issues leading to the discrediting of rofecoxib may not have been seen as a COX-2 class effect by this group of Australian prescribers and consumers.

Although there has already been considerable exploration of this discrediting event [2, 21‐24, 26‐29], the strength and interest of this study is that it examines medicine use at the individual consumer level. The linkage of national subsidised pharmaceutical database medicines data to longitudinal survey data avoids the issues of consumer recall and allows examination of dispensing patterns in relation to the demographic, health and health service use characteristics of consumers. This study is novel in Australia.

There are some limitations to the study. First, the cohort includes only women. However, prevalence of arthritis is significantly higher for women than for men in the USA [45], Canada [46], the UK [47], and Australia [48, 49], so this is an appropriate group in which to study the use of arthritis medicines. Second, many of the alternate medicines cost less than the PBS general copayment for at least some of the study period, meaning that only women with a "concessional beneficiary" PBS status for all years of interest could be included. This represents 85% of the Older ALSWH cohort who agreed to PBS linkage. Previous analyses have found that Older ALSWH concessional women were quite similar to non-concessional Older ALSWH women - although less likely to see a specialist or be overweight and more likely to have hypertension [35]. A third issue is that the PBS does not capture over-the-counter analgesics, so use of these medicines may be underestimated. However, in other research, where PBS data was compared with medicines data collected from a home visit medicine audit for people aged 65 years and over, PBS data was found to be close to complete, so we might expect the same for this group of concessional status older women [50]. A fourth issue is that the patterns of prescriptions for the non-continuous users group, included as a comparator for the continuous users group, are acknowledged as potentially more heterogeneous than for the continuous users group. The non-continuous users group is defined as women dispensed eight or fewer rofecoxib prescriptions in the 12 months before discrediting, on the premise that these women had a lesser expressed need for these medicines. So, this group could potentially include women who received only one prescription early in the 12 months (where the discrediting event was unlikely to be an issue), or one prescription late in the 12 months, or regular prescriptions over the 12 months of interest. The median number of scripts in this group is two prescriptions (IQR: 1-4) in 12 months, so there would be few women who received regular prescriptions, but time of dispensing could be quite varied. This means that interpretation of behaviour for this group must be circumspect, and may not necessarily be closely tied to the discrediting event. This group is useful, however, as a comparator for the switching behaviour of continuous users.

The patterns of medicine switching varied significantly depending upon whether a woman was a continuous or non-continuous rofecoxib user pre-discrediting. Continuous rofecoxib users overwhelmingly switched to another COX-2, mostly within 100 days of rofecoxib withdrawal, and the majority remained continuing COX-2 users for at least another 100 days. When an alternate COX-2 was not taken up in the first 100 days by erstwhile continuous rofecoxib users, they were less likely to be taken up over the 1200 days of the study. Interestingly, a significant proportion (24%) of non-continuous rofecoxib users also became continuing COX-2 users for at least 100 days after first post-discrediting uptake; however a direct switch to another COX-2 was much less common for this group of less frequent rofecoxib users. It appears from these findings that rofecoxib discrediting was not seen as a class effect by many prescribers to these women. This contrasts with the findings of Barozzi & Tett who suggested that reduced COX-2s prescribing following rofecoxib withdrawal confirmed prescribers saw the discrediting as a class effect [2]. While this previous analysis showed that the rise and fall of the COX-2s did markedly influence overall NSAIDs prescribing in Australia, this was only at an aggregate level, and individual switching behaviour could only be assumed. As shown in the current study, at the individual level, it may be that consumer characteristics and preference played a part in the decision on switching. Certainly, those who had previously been dispensed rofecoxib less frequently before discrediting, those who might be assumed to have a lesser need, were less likely, then to take up another COX-2, but this was not the case for those with higher expressed need, those more frequently dispensed rofecoxib. In an aggregate analysis, this detail can be lost, if there are considerably fewer continuous than non-continuous users.

There were few individual characteristics associated with early uptake of an alternate COX-2. COX-2 switchers were more likely to have a diagnosis of arthritis and to have been a continuous rofecoxib user than were COX-2 stoppers, both these circumstances potentially indicating women who experienced more persistent pain. There is some evidence that consumer satisfaction with the COX-2s is high [51], so women may have preferred to stay with this drug class. Non-continuous users may indeed have only used rofecoxib on very few occasions(as evidenced by a median number of only two prescriptions dispensed in 12 months), so rofecoxib withdrawal may have had only minimal impact on their medicine use behaviour.

There is general international clinical agreement that paracetamol should be the pain killer of first choice for arthritis, as a full therapeutic dose provides adequate analgesia for many people, with less risk of side-effects or interactions compared with NSAIDs and other analgesics [2, 3, 52‐54]. New PBS listings for paracetamol containing preparations since April 2005 allow for use of higher doses and quantities of paracetamol specifically for chronic arthropathies and arthritis pain, with the intention of encouraging use of appropriate, regular doses of paracetamol [2]. In this study we found that continuous rofecoxib users were more likely to switch to paracetamol than were non-continuous users. Uptake of paracetamol was slower, and more sustained, and given this slow uptake, the link with the discrediting event may be somewhat tenuous. Again, the findings of Barozzi & Tett [2] that use of paracetamol was steady across 1997-2004 with a slight increase in April 2005, can be contrasted with the current individual level data, which provides the nuances available at this level of data. (See Figure 4).

While the risk profile that lead to the withdrawal of rofecoxib might be indicative of aCOX-2 class effect [11, 19, 55], this may not have been the understanding of many prescribers (or consumers) for this group of community living older Australian women. This interpretation of COX-2 class risk has also varied across international studies of prescribing following the rofecoxib discrediting event, with some studies demonstrating a prescriber response to a class effect [2, 23, 26, 28, 30], and some not [22, 24, 26]. Further research is appropriate to confirm our findings, given this controversy. There is some evidence for continuing chronic use of COX-2s despite discrediting [22], and it would be of interest to explore if this is the case for this group of women, and consider the potential impact on longer term health outcomes.