Medication prescribing in the elderly and the effect on health related outcomes: an investigation of bias in observational studies using computerised claims databases.

Abstract

Background: This thesis explores the effects of medication prescribing on patient outcomes in an ageing population, specifically, the population of Australian veterans. The primary source of data is the computerised administrative claims database maintained by the Commonwealth Department of Veterans' Affairs. This database is a valuable resource yet knowledge about how these data can be analysed and interpreted to study the effects of medicine use in the Australian setting is limited. An important source of bias in observational studies relating medication prescribing to health outcomes arises from confounding by the reason for prescription, or confounding by indication. The extent to which traditional pharmacoepidemiological studies utilising administrative claims databases can deal with confounding is limited as these data sources often lack information on many potentially important confounders, such as clinical information, life style factors and disease severity. Aim: The aim of this thesis was to investigate the use of two methods, developed to overcome possible bias in observational studies due to unmeasured confounding; instrumental variable analysis and the self-controlled case-series design. To illustrate how these techniques may be used to overcome confounding, I investigate how they apply to the assessment of the adverse effects of antipsychotic prescribing in the elderly. Methods: The instrumental variable analysis was used to compare the risk of death, hip fracture and pneumonia between the antipsychotic classes. The instrumental variable analysis aims to control for unmeasured confounding by attempting to mimic the process of random assignment in a randomised controlled trial. The self-controlled case-series design was used to investigate the risk of hospitalisation for stroke, hip fracture and pneumonia associated with antipsychotic initiation. The self-controlled case-series design uses a patient as their own control, thereby implicitly controlling for constant patient specific confounders, even those that are unmeasured. Results: Using a cohort of 20,205 elderly patients aged over 65 years of age, I have shown that the profiles of patients receiving antipsychotic medicines vary between the class of antipsychotic initiated and those variables that differ are likely to be associated with the reported adverse events of these medicines. This indicates the potential for confounding in observational studies of antipsychotics and suggests that appropriate study designs are required to minimise the effect of confounding in order to get a clear understanding of the potential adverse events of these medicines. The instrumental variable analysis suggested that typical antipsychotics were associated with an extra 24 (95% confidence interval (CI) 18-30) deaths per 100 patients per year compared to atypical antipsychotics, and an extra 10 (95% CI 7-14) deaths per 100 patients per year among nursing home residents. In this analysis I proposed a new instrument, facility prescribing preference, as an alternative to the doctor prescribing preference instrument; the latter which has been used extensively in the pharmacoepidemiological literature. I was able to show that facility preference may be a valid instrument for further work in this area as it was highly correlated with actual treatment (Odds Ratio 19.2; 95% CI 17.1-21.6), provided a good balance of measured patient characteristics and was consistently strong over time. While the instrumental variable analysis can provide information regarding the comparative risk of antipsychotics between the classes it cannot inform about the individual risk of these medicines compared to no treatment. To answer this question I used the self-controlled case-series design to estimate the excess risk of hospitalisation for stroke, hip fracture and pneumonia after initiation of an antipsychotic. Atypical antipsychotics were not associated with an increased risk of stroke, which is consistent with randomised controlled trial evidence. No such evidence is available for typical antipsychotics in the elderly, however, the case-series analysis suggests that there is a small but significantly increased risk of hospitalisation for stroke in the first week after initiation (Incidence Rate Ratio (IRR); 2.1, 95% CI 1.1-4.2). For pneumonia the risk was raised in all periods after antipsychotic initiation. This risk was highest in the first week after initiation and remained significantly raised by 50% with more than 12 weeks of treatment (Typical antipsychotics IRR; 1.5, 95% CI 1.2-1.9, Atypical antipsychotics IRR; 1.5, 95% CI 1.3-1.7). The risk of hip fracture was significantly increased for both classes but this risk was sustained only with long-term typical antispychotic use (IRR; 1.3, 95% CI 1.1-1.6). The self-controlled case-series design has been used extensively in the investigation of vaccine safety. I have found, however, that the application of this design to the study of the effects of medicine prescribing in the elderly may require the addition of an unexposed group to control for the increasing incidence of hospitalisation with age in this population. I also explored the use of risk periods prior to initiating therapy with antipsychotics. Patients were more likely to have had a hospitalisation for stroke in the week prior to initiating typical antipsychotics (IRR; 6.9, 95% CI 4.7-10.0) while atypical antipsychotic initiators had no excess risk in the same period (IRR; 1.2, 95% CI 0.5-2.6). These results suggest that the use of pre-exposure risk periods may be required in medicine outcome studies when the outcome of interest is a hospitalisation event that leads to an increased likelihood of initiating treatment. Conclusion: This thesis has illustrated that identifying and reducing confounding will enhance the validity of observational studies investigating the safety of medicines using computerised claims databases. By employing methods that help to overcome the problem of confounding I was able to demonstrate that antipsychotic use in the elderly is associated with significant harm and the increasing use of these medicines in Australia poses a major public health concern. Randomised controlled trial evidence suggests that for every 100 patients treated with atypical antipsychotics over 12 weeks, only 8 to 33 would show any benefit, however, there would be 1 additional death and 2 additional cerebrovascular events. Using the self-controlled case-series design I estimated that there would be 8 additional pneumonias, and 2.5 additional hip fractures for every 100 patients treated with atypical antipsychotics over 12 weeks. In addition, typical antispychotics were found to be associated with at least equivalent, if not more, harm. The knowledge obtained in this thesis will help to inform how Australian computerised claims databases may be interrogated to examine the safety of medicines that are under investigated in randomised controlled trials. This information will allow prescribers and policy makers to make more informed decisions about the risks of medicines.