Model assumptions and solutions to violations of those assumptions
| Assumption | How the assumption applies to this study | Solution | Example of use of the solution in the literature |
| Subsequent exposures should not be affected by previous events. | We might see a temporary increase in UTIs subsequent to an MI or stroke event, which would bias estimates towards the null. | Apply a prerisk period. | Gibson et al studied the association between prescription drugs and road traffic accidents. As some drugs may be used to treat anxiety or pain caused by the crash, a 4-week pre-exposure period was included.37 |
| As both MI and stroke have relatively high death rates, the length of the observation period is dependent on events, and no further exposures are possible after death. | Use an event-dependent observation period model extension38 and conduct a sensitivity analysis that repeats the analysis, excluding individuals who died within 30 days of the event. | Bruer et al used the event-dependent observation period model extension in their study on the association between antipsychotic drugs and myocardial infarction.39 Langan et al studied the risk of stroke following herpes zoster. They conducted a sensitivity analysis excluding individuals who died within 90 days of stroke.40 | |
| Event rates are constant within defined periods | MI and stroke are more common in older individuals and may be affected by seasonal changes. | Control for age and season effects. | Grave et al studied the association between seasonal influenza vaccination and Guillain-Barré syndrome. They adjusted for calendar month, as the vaccinations are seasonal by design.41 In a study of the association between chickenpox and stroke, Thomas et al adjusted for age in 5-year age bands.42 |
| Events are independently recurrent or rare. | MI and stroke are not independent: once an individual has a first event, they are more likely to have a second. | Study first events only. | Langan et al began the observation period 12 months into follow-up time to ensure first stroke events had been correctly identified.40 |
MI, myocardial infarction; UTIs, urinary tract infections.