BACKGROUND: There is a wealth of epidemiological data that may be used to assess the risk benefit, or the relative safety profile, between medications. Simulation techniques have been applied to clinical trial designs and in other fields such as the motor industry, allowing evaluation of ‘what if’ scenarios for safety outcomes. For drug safety surveillance, simulation techniques may allow us to focus efforts on expected scenarios.
OBJECTIVE: To outline the approach and application of stochastic simulation models in drug safety. An example will illustrate the approach: how heart failure and asthma/COPD exacerbation events can be modeled for glaucoma patients not exposed to treatment, and howevents in a timolol maleate exposed group vary with off-label use.
METHODS: The essence of the approach is to generate random times to events based on the risk of the event occurring in a non-exposed patient group. This first step provides the natural history of the safety events. Drug exposure is incorporated into the models by applying relative risk parameters obtained from epidemiological studies and clinical trials. Thus fictitious cohorts are generated and evaluated. Timolol maleate is contraindicated for glaucoma patients with history of asthma or heart failure due to systemic absorption of this beta-blocker. We modeled the occurrence of these events in a non-exposed cohort and then examined scenarios of differing prevalence of history of asthma/COPD for users of timolol. Prevalence of asthma/COPD in the general population is 12%. Despite the contraindication included in timolol’s label, it is possible that 9% of users have asthma/COPD.
RESULTS: In a non-exposed glaucoma cohort of 10 000 patients, over 5-years around 267 cases of heart failure cases, 94 cases of de novo asthma, 402 asthma exacerbations, and 988 deaths will arise. Exposure to timolol may increase heart failure cases to 353 (32%) cases whilst the number of de novo asthma remains stable. Deaths may increase to 1014 (2.6%). If labeling were adhered to there would be no asthma/COPD exacerbations. However, for the most likely scenario of off label use (prevalence¼9%) there may be 410 asthma exacerbations, whilst if the contraindication had no influence on prescription (asthma/COPD prevalence¼12%) then we may expect 547 cases.
CONCLUSION: This example shows that stochastic simulation modeling may be a good approach to quantify the effect of drug exposure on safety outcomes by incorporating existing epidemiological data in a single framework. However, the complexity of the models makes validation cumbersome.
