Adjusting for recall bias with the proportional hazards model

In studying falling frequency in the elderly, we observed that having subjects keep a diary led to a larger number of falls reported than had been noted in a previous study in the same population. The previous study asked subjects to report any falls in the previous three months. We considered two related explanations for the observation of lower incidence reports with a 3-month recall survey. First, there may have been under-reporting of falls due to recall bias. Second, the less severe falls (which did not result in injuries) may not be reported. We suggest that the proportional hazards model may be used to adjust studies in which recall is used to determine incidence and time to falls.