Statistics from Altmetric.com
Editor,—A reading of the recent article by Eggleston et al (Gut1998;42:13–16) left several unanswered questions and concerns. The first issue was the authors’ statement that clinical equivalence between cisapride, ranitidine, and omeprazole had been established for uncomplicated gastro-oesophageal reflux disease (GORD) in standard practice. To my knowledge, there are no published studies which have reported a similar finding of clinical equivalence of omeprazole to either cisapride or ranitidine. A quick review of the literature revealed 11 studies in which omeprazole was statistically significantly better than cisapride or ranitidine in GORD.1-11
In a review of the reference abstract12 in which the case for clinical equivalence is postulated, the conclusion of equivalence was highly dependent upon the authors’ definition of treatment failure. I take exception to the authors’ definition that use of medication for longer than three months was a therapy failure. A retrospective database analysis does not have the ability to distinguish between continuing a medication in order to maintain successfully achieved symptom resolution and continued use of medication due to unsuccessful resolution of symptoms. Initiation of maintenance therapy after successful symptom resolution would not require any change in dosage in the case of omeprazole as 20 mg once daily is the recommended dose for healing and maintenance. Defining success as any use of medication for less than three months that did not involve a change in drug also opens up the possibility of misclassification. Those patients who stop using a drug as a result of treatment failure and who do not return for further care would not only be misclassified in this retrospective analysis as a success but could erroneously contribute to a lower economic burden in the therapy group. In the Methods, the authors state that case records for a total of 257 patients were evaluated. However, there is a discrepancy between the abstract and the manuscript on the number of eligible patients. The abstract talks about 279 patients and the manuscript refers to 257 patients. Additionally, why do the authors, with a database like MediPlus™ UK, which has more than one million patients, end up with a sample size of only 257 patients?
In the Discussion, the authors state that if more severely affected patients were selectively treated with omeprazole, one would expect to see increased non-drug resource use in this group. Since this did not occur, the authors consider this as supportive evidence that they had effectively controlled for this selection bias. An alternative explanation for the finding is patients taking omeprazole were more severely ill due to physician selection bias but omeprazole has superior clinical efficacy in GORD compared with rantidine and cisapride thus the omeprazole patients resource utilisation level is a reflection of a reduction from what would be the case with any other treatment.
The authors do not state whether the general practitioner (GP) consultations attributed to the drug treatment groups were all specifically for GORD. No information was provided as to how the authors have separated out GORD related GP consultations from visits for other illnesses. Such a process should be accomplished by a reviewer blinded to treatment allocation, if undertaken. The absence of information leads to a conclusion that the count of GP consultations in table 1 represents visits for all health reasons. As treatment would have had a positive effect on reducing GORD related visits only, the possibility of observing any difference between treatment groups would be diminished, resulting in the finding displayed in table 1.
The authors have based their analysis on an assumption of clinical equivalence which, in this case, is beyond the power of retrospective analysis to prove. Even if the three medications were clinically equivalent, there are a number of other unanswered questions from the manuscript that make it difficult for an unqualified acceptance of the authors’ conclusions.
Editor,—We welcome the quantity and quality of debate generated by our paper and in this respect the paper has fulfilled its principal objective. Many of the issues raised in the resulting correspondence, however, relate to methodological limitations of database analyses that were well rehearsed in the original paper and therefore will not be addressed in detail. For example, we recognise the potential difficulties arising from the non-randomised nature of the study but feel that restriction is essential in enabling the analysis to reflect reality more accurately. The paper also explicitly acknowledges the potential for selection bias in the choice of drugs and specifically examines its potential implications for the interpretation of the results obtained. The patient cohorts seemed to be closely similar in both demographic characteristics and gastrointestinal and non-gastrointestinal medical history. In addition, a selection bias would be expected to lead to a greater intensity of health care resource consumption in patients receiving omeprazole. This did not occur and therefore no evidence is available to support the charge of selection bias.
The limitations of the study are difficult if not impossible to overcome with currently available data and should not distract us from the main focus of the paper. This was to undertake an initial comparison of the therapeutic strategies available to clinicians in the treatment of mild to moderate GORD. The data contained in the paper is merely used to illustrate the general principle that optimum decision making in this therapeutic area, as in any other, requires an informed and sensitive balance of the costs and benefits arising from each strategy.
The clinical outcomes underlying the cost analysis have been reported in a separate publication and provide the necessary clinical springboard for our analysis. The analysis simply indicates that for this particular group of patients (first episodes of uncomplicated GORD) the hypothesis of short term outcome equivalence in practice was not disproved by the clinical database analysis. Evaluating long term comparative outcome in practice would require a modelling approach that evaluated in detail quality of life and clinical outcomes together with the long term pattern of resource use for patients following different therapeutic strategies. An extensive and growing literature is available to inform such a model and research in this area is currently under development.
The basis for a clinician’s choice between cisapride, ranitidine and omeprazole is in this study, as in all others, unknown given the complex array of factors determining such a choice. It is specifically because of such complex influences, which are not captured in randomised clinical trials, that analyses of prescribing in practice are so important, given that they implicitly incorporate the myriad of factors that influence real life prescribing behaviour. It is important, however, to recognise that evidence generated in clinical trials and observational studies of this nature should be perceived as being mutually supportive in generating evidence to guide clinical decision making. Obviously the same dataset can give rise to many different interpretations and we are not, necessarily, providing the only possible interpretation of the information generated.
The Nijmegen group question the “inappropriate” use of the term “cost effectiveness analysis” which is often used to refer to all forms of economic evaluation. The term, however, more properly refers to a particular type of evaluation in which outcome is defined and measured in natural units and costs are measured in monetary terms. Our analysis is structured in a manner that facilitates a direct comparison of the costs per patient relieved of symptoms. Such a structure enables outcome variations to be directly incorporated into the analysis as and when such evidence is generated. Information about cost is useful only when considered jointly with information about clinical outcomes and further research concerning the long term cost and outcome of treatment is currently in progress.
The Nijmegen group’s reference to “effect modification” and its potential effect of diluting variations between the therapeutic options is irrelevant in the context of our paper. The aim of good prescribing is to segment general patient populations to ensure that the therapeutic requirements of individual patients are optimally met by the prescriber. Obviously, in their day to day workload primary care physicians also frequently encounter patients with recurrent GORD as well as with H2 refractory and complicated GORD. Cost effectiveness calculations on “undefined” GORD would inevitably favour omeprazole given that it causes the greatest suppression of acid secretion in patients with severe disease. We make no claims concerning the applicability of the analysis to undefined GORD but simply assert that the extent of acid suppression generated by omeprazole may not be necessary for first episodes of uncomplicated GORD. If this is the case, then hard pressed clinicians will be able to conserve their scarce health care resources for investment elsewhere.
Dr Crawley questions the numerical basis for the analysis. Our sample was generated as a random selection from the entire eligible MediPlus database for 1995 using the software random number based sample selection function. The sample chosen was selected on the basis of manageability given that an individual review of each patient record was required. The reduction from 279 to 257 patients was simply to increase the level of patient homogeneity as outlined in the original paper. Both Crawley and Bate also refer to a large body of evidence supporting the clinical and economic superiority of omeprazole. Unfortunately, for our purposes the vast majority of such evidence is irrelevant given that it does not specifically analyse prescribing in practice for new patients presenting at the primary care level with mild to moderate GORD. Dr Crawley’s belief in the superiority of omeprazole due to the lack of need for dose titration is puzzling. The British National Formulary’s recommendation for GORD is for 20 mg daily for four weeks, followed by 40 mg for four to eight weeks as required, reduced to 10 mg for long term management, increased to 20 mg if symptoms recur. Dr Crawley’s further belief that patients may respond to treatment failure by simply doing nothing also is difficult to substantiate. It is reasonable to assume that individual patients exhibit a personal threshold of tolerance to the symptoms of GORD above which they seek professional advice. Such a threshold will vary significantly between patients but is likely to be fairly stable within the same patient. In such circumstances it is difficult to see why a patient would not seek further professional advice in cases where the initial prescription did not reduce symptomatology below their personal threshold.
Many of the issues raised reflect the fact that many gastroenterologists only come into contact with the tip of the iceberg with regard to GORD. By supporting 95% of patients without recourse to specialist advice, primary care physicians ensure gastroenterologists encounter a very thin slice of reality with regard to the optimal treatment of GORD. The vast majority of treatments are initiated without prior investigation and where empiricism is successful such patients do not come to the attention of gastroenterologists. Although this may be appreciated intellectually by gastroenterologists, our paper provides powerful re-enforcement of the nature of the division of labour between primary care physicians and gastroenterologists in the treatment of reflux disease.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.