Figure 9.6. Principles of record linkage in drug safety.



Figure 9.6. Principles of record linkage in drug safety.

association if they are gross or if systematic bias is present. A modest degree of inaccuracy in the data is unlikely to produce a positive finding when no such association exists.

Drug Exposure Data

The record of drug exposure in epidemiological studies is invariably imperfect. In the General Practice Research Database (GPRD) it is a record that a prescription was issued by a doctor. In the MEMO scheme it is a record that the prescription was encashed, which is likely to be substantially preferable. In a study in Tayside, UK, up to 15% of patients did not cash the prescription issued (Beardon et al., 1993). However, in neither case is their any certainty that the patient consumed the prescribed medicine as intended (Sackett et al., 1986)—this can only be assessed by interviewing subjects or by monitoring compliance. The interview approach has some advantages (for example, it can be applied to non-prescription medicines), but because patients are generally not precise observers of their own behaviour, it is not necessarily more accurate than a prescription record. Furthermore, such data are potentially subject to systematic biases, notably recall bias (see above for explanation).

Outcome Data

Similar issues arise regarding the quality and accuracy of the outcome data. It is usually considered necessary to verify outcomes held on a computer record by reviewing the medical records. This will eliminate basic data entry errors and, to an extent depending on the outcome being studied and the diagnostic criteria used, improve the accuracy of the outcome data. It may also allow the time of onset of the disease to be clarified (it is important to know whether or not cases are "incident", i.e. new) and better measurement of potential confounders. However, it should be recognised that there is also a disadvantage inherent in this process, which is that selection of the best-documented cases is being practised. There is therefore potential for selection bias and the incidence of the outcome is likely to be underestimated. It has been argued that these disadvantages may outweigh the benefits of case validation (Evans and MacDonald, 1997) and this is an issue that needs further research.

Selective Use

A particular issue that arises in the design and interpretation of many epidemiological studies of drug safety relates to whether or not the drug of interest is being used selectively in patients who have different characteristics to those using comparators. It is well-recognised that this is a common occurrence and that it may in part relate to marketing strategies. For example, claims that a non-steroidal anti-inflammatory drug has less gastrointestinal toxicity than others may lead to it being selectively used in patients with a history of such problems, a phenomenon known as channelling (Petri and Urquhart, 1991). The result in an observational study is that such a drug will apparently be associated with a higher risk of gastrointestinal toxicity than comparators. This is called ''confounding by indication''. It is analogous to an imbalance at baseline in a randomised trial. The most satisfactory way to deal with this problem is by adjustment for the relevant risk factors, but this is not always feasible, leading to difficulties in data interpretation.

Use of Computerised Databases

For reasons of cost and speed, epidemiological studies of drug safety are increasingly being based on data from computerised databases that have primarily been collected for another purpose (Jick et al., 1991). Use of the databases listed in Table 9.6 has demonstrated the potential of this approach in the investigation of a wide range of drug safety issues. Both major types of epidemio-logical study design have been used, i.e. cohort and case-control designs.

In a cohort design, a defined population is followed forwards in time and the incidence of the outcome in question is measured and compared in individuals exposed or not-exposed to the drug of interest. Both absolute and relative risks can be measured using this approach.

In a case-control design, cases of the outcome of interest are identified along with a group of controls drawn from the same population who do not have the outcome. Prior exposure to the drug(s) of interest is then compared in cases and controls. These studies provide measures of relative risk in the form of odds ratios but they do not directly measure absolute risks. A further approach is for a case-control study to be "nested" within a cohort design. This method has the advantages of both approaches, particularly if it used used to facilitate more detailed control of potential confounding than can be achieved in a complete cohort.

Was this article helpful?

0 0
Drug Free Life

Drug Free Life

How To Beat Drugs And Be On Your Way To Full Recovery. In this book, you will learn all about: Background Info On Drugs, Psychological Treatments Statistics, Rehab, Hypnosis and Much MORE.

Get My Free Ebook

Post a comment