PREPARACIÓN DE LAS ESENCIAS

A key strength of our report was its

comprehensiveness – this report has identified more trials than many previous systematic reviews – and direct integration of the results of the systematic review into the assessment group economic model. In addition, we were able to include a number of direct comparisons between COX-2 selective NSAIDs published recently. The latter, particularly for rofecoxib, celecoxib and lumiracoxib, show similar efficacy between agents but direct comparisons with adequate power, using comparable doses, and of sufficient duration are needed to clearly understand safety issues. Some other limitations in the evidence were identified:

1. Outcomes examined by trials are relatively broad and sometimes poorly defined, particularly for older studies, increasing the potential for bias in the reporting and analysis of data. For example, in most trials, the PUB category did not provide specific data about the frequency of perforations, gastric outlet

obstructions or GI bleeds associated with haemodynamic instability or hospitalisation because of these adverse events.

2. Many studies did not report adverse events adequately or, perhaps worse, mentioned several events in an ad hoc manner, so that, when collated, events may not have reflected their actual occurrence or allowed meaningful comparisons between drugs used.

3. There was a lack of consistency in the reporting of the principle GI safety events of POBs and PUBs. This was due to both different

operational definitions of these events across different studies and also the differential access to outcomes details, for example whether a full trial report was available. This reporting bias appeared to vary across COX-2 NSAIDs. Therefore, considerable caution needs to be applied in directly comparing the pooled POBs and PUBs between drugs.

4. The non-selective NSAID preferred in many studies, naproxen, reflects preferences in the USA, where naproxen is used widely. In England and Wales, diclofenac and ibuprofen predominate. In some studies, the choice and dose of non-selective NSAID comparator, and limited details of the population studied (for example, aspirin use and prior GI history), make it difficult to generalise this evidence base to routine clinical practice.

5. Age restrictions and other exclusion criteria also limit generalisability, supporting the case for more pragmatic studies. A variety of observational data clearly show the limitations of NSAIDs in clinical practice. Trials reported here invariably included individuals who were established and accepting of NSAIDs and indeed required a flare of symptoms on NSAID withdrawal before inclusion. This biases

towards not only inflated figures on drug retention with chronic therapy, but also a greater likelihood of response to any therapy on the basis of spontaneous improvement of symptoms after a flare (regression to the mean). Potential limitations of our review:

1. According to the assessment criteria used, the majority of included trials were judged to be of ‘good’ to ‘excellent’ quality, that is, with

appropriate randomisation and concealment, double blinding and low loss to follow-up. However, despite selective inclusion criteria, there was often considerable attrition in many trials because of adverse events and lack of efficacy. This attrition varied for different drugs, so, for example, in the CLASS study 47

and 41% of patients completed the trial at 52 weeks in the celecoxib and non-selective NSAID (diclofenac and ibuprofen) arms, respectively. As a result, there is less patient ‘exposure’ to non-selective NSAID than celecoxib in the initially randomised groups. By implication, this would favour NSAID patients for GI safety outcomes. This is overcome, however, by presenting data that allow for differing durations of drug exposure. 2. The quality and amount of evidence for newer

COX-2 selective drugs were generally far greater than for older drugs, particularly in terms of long-term GI and CV safety data. This, and the heterogeneity of outcome data for selective and non-selective NSAIDs (indicated by observational studies), raise a question about, conceptually, considering NSAIDs simply as two separate classes of agents. 3. For accuracy, we relied on full study reports for

data. However, trial reports from drug sponsors were not available universally. For example, most celecoxib trials study reports were available, but in contrast no industry study reports were available for etodolac and meloxicam. This may have led to unforeseen biases.

There are a number of potential limitations of the cost-effectiveness analysis undertaken in this report, including issues of model structure and model parameters:

1. The majority of models developed for arthritis specifically exclude consideration of adverse events other than GI events and MI risk and therefore do not take into account differences in GI tolerance or efficacy between drugs. Nor do published models allow for differences between agents in other adverse events such as skin rashes or hepatitis. As an adaptation of the Maetzel model, the AGM is similar in this respect, but the initial (‘switching’) cycle added to our model allows drug switching and therefore does take into account, to some extent, drug changes, including withdrawal for lack of efficacy or adverse events.

2. The model only allows one clinical event possible in each cycle (i.e. an arthritis patient cannot undergo MI and a serious GI event within same Markov cycle).

3. Our model, in common with other published models, does not consider drug compliance and the tendency for many patients to use NSAIDs intermittently rather than continuously. 4. Relatively limited observational data were

available to populate the initial (switching)

5. Clinical GI events and MI risk for comparator NSAIDs used in the model were based on data from patients in CLASS not taking aspirin. In contrast, the model used RRs of clinical GI events and MI for the COX-2 selective agents were based on meta-analysis that includes all trial patients (i.e. both aspirin users and non- users). Nevertheless, evidence from our clinical review indicates that the effect of COX-2 on GI events and MI risk is maintained, regardless of aspirin status.

6. There is uncertainty around the GI protective RR associated with PPI plus non-selective NSAID compared with non-selective NSAID alone. 7. The utility values used are based on those

reported by Maetzel and colleagues305using a

sample of the general public and the standard gamble method. Although this is a recognised approach to the derivation of utility values, it has been pointed out that the method may underestimate the severity of short-term

effects.305