The example of confounding described below is of potential interest both from a teaching perspective and in the field of epidemiological research on the risk of head injury in moped riders. We took data from the Spanish Registry of Traffic Crashes with victims to study the strength of association between position of the rider on the moped (the driver or the passenger) and risk of head injury in all 187 353 moped riders involved in a traffic crash with victims between 1990 and 1999 in Spain, and for whom information about helmet use was available. In the crude analysis (Table), the frequency of head injury was similar for drivers and passengers; accordingly, the crude odds ratio (OR) for the association between being the driver and receiving a head injury was only 1.06. But when we stratified this estimate depending on helmet use, the corresponding values were considerably higher in both helmeted (1.40) and non-helmeted riders (1.41). As a result the OR estimate adjusted by helmet use (using unconditional logistic regression) was 1.41. Careful assessment of the role of helmet use showed that this variable satisfied all three classical requirements for a confounder: * It is causally related with the effect: for both drivers and passengers, the risk of head injury was much higher for non-helmeted than for helmeted riders. * It is associated with exposure, in the absence of effect. The OR for the association between helmet use and being the driver of the moped was 2.61 in non-head-injured riders, a value similar to that obtained for head-injured riders (2.59). * It is not an intermediate step in the causal path between exposure and effect, as the strength of the association between being the driver of the moped and the risk of head injury was the same for helmeted as for non-helmeted riders. Put simply, drivers of mopeds are at an intrinsically higher risk of suffering head injury than are passengers, but this increase is masked (confounded) by helmet use, because helmets (a device which strongly reduces the risk of head injury) are more frequently used by drivers than by passengers. We have found no studies of the risk of head injury in moped riders that took this possible source of confusion into account. It is disappointing that we must usually resort to hypothetical data to teach our students how a confounding factor operates. For these reasons we hope that the real data presented in this letter will be useful for both teaching and research purposes. (Author/publisher)
Samenvatting