A framework for a critical assessment of the quality of epidemiological studies of driver health and accident risk

A large number of studies have tried to assess how various aspects of driver health influence driver involvement in accidents. The objective of this paper is to provide a framework for a critical assessment of the quality these studies from a methodological point of view. Examples are given of how various sources of bias and confounding can produce study findings that are highly misleading. Ten potential sources of error and bias in epidemiological studies of the contribution of driver health impairments to road accidents are discussed:(1)Poor description of the medical conditions whose effects are studied (measurement error).(2)Inadequate control for the effects of exposure on accident rate.(3)Sampling endogeneity with respect to assessment for fitness to drive (outcome-based sampling; self-selection bias).(4)Combined exposure to several risk factors.(5)Poor control for potentially confounding factors.(6)Failure to specify potentially moderating factors (interaction effects).(7)Failure to consider a severity gradient with respect to the effect of health impairments.(8)Failure to specify the compliance of drivers with medical treatments or treatment effectiveness.(9)No data on the population prevalence of various health conditions.(10)The use of multiple study approaches and methods making the comparison and synthesis of findings difficult.Examples are given of how all these items may influence the findings of a single study or make synthesising findings from multiple studies difficult. A checklist for assessing study quality is provided.

► Studies assessing the association between medical conditions and accident involvement tend to vary in quality.
► There is a need for assessing the quality of studies according to a formal checklist.
► A checklist is provided and examples are given of how various sources of error and bias can influence the results of epidemiological studies of driver health and accident risk.