Methods in pharmacoepidemiology : four studies, four settings

Abstract

In the present thesis, four studies using four different settings and designs are presented in which associations between pharmacological treatments and possible adverse drug reactions are investigated; some methodological considerations identified in them are discussed.

In a population-based case-control study, using information from telephone interviews, the association between use of acid-suppressing drugs and the development of acute pancreatitis was investigated. Statistically significant odds ratios were found, associating use with a duration of less than six months of proton-pump inhibitors (OR 3.2; 95% CI 1.4-7.4) or H2-antagonists (OR 4.9; 95% CI 1.6-15) with the outcome. However, confounding by indication was present to some extent, but the association was interpreted as probably causal.

Using the UK General Practice Research Database (GPRD), a case-control study nested in a population of women with a diagnosis of menorrhagia, the relationship between use of tra-nexamic acid, mefenamic acid or norethisterone and the development of venous throm-boembolism (VTE) was evaluated. Use of tranexamic acid was too scarce to allow for certain conclusions; 3 cases and 4 controls exposed to the drug yielded an OR of 3.2 (95% CI 0.6-15.8). Use of mefenamic acid was associated with the outcome with an OR of 5.5 (95% CI 2.1-5.8), as was use of norethisterone, OR 2.4 (95 % CI 1.0-5.8). A diagnosis of anemia, indicating a more severe menorrhagia, was also associated with the outcome, OR 2.2 (95% CI 1.0-4.9); a fact which may indicate that confounding by indication was at hand, and that menorrhagia itself may constitute a risk-factor for VTE.

The safety of antiretroviral treatment (ART) for HIV was analysed in a Swedish cohort-study of 1,072 patients. An association between ART and the development of midline- and inguinal hernias was identified. The hazard ratios (HR) were different depending on duration of ART containing a protease inhibitor (PI). The highest HRs were observed during the second to third year of treatment: HR 10.7 (95%CI 1.3-85.7) for midline hernia, and HR 4.4 (95% CI 1.1-16.6) for inguinal hernia.

By linking a cohort of 5,756 patients treated with isotretinoin for severe acne with the Swe-dish national patient register of hospitalizations, the association between treatment with isotretinoin and hospitalizations for suicide attempt was analyzed. The incidence and Stan-dardized Incidence Ratios (SIR) of suicide attempts before, during and after treatment was estimated by use of a cohort crossover approach. The SIR for suicide attempts started to rise above one already two years before treatment. It was highest within six months after treatment, SIR 1.9 (95% CI 1.1-3.2) for first attempts; within three years after treatment, the observed number of suicide attempts was very close to the expected. It was concluded that a strong association between severe acne and suicide attempt was present, and that a significantly increased risk associated with use of isotretinoin could not be detected on the population level. However, the possibility could not be excluded that certain patients, pre-sumably sensitive to isotretinoin, made suicide attempts as a result of the treatment.

A number of methodological aspects identified in the included studies are discussed in the thesis: confounding by indication was identified, challenged and estimated; the importance of dose-response and duration-response relationships in assessing causality was demonstrat-ed; reversed causality was observed and handled; assessing the presence of recall bias in an interview setting was described; and the effect of a presumed under-estimation of an important confounder was demonstrated using a simulation. Advantages and disadvantages of using the different data-sources, depending on type of treatment and character of adverse events, are discussed.