The effects of recall on reporting injury and poisoning episodes in the National Health Interview Survey
- 1Office of Analysis and Epidemiology, National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, MD, USA
- 2Office of Research and Methodology, National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, MD, USA
- 3Northern New England Poison Centre, Portland, ME, USA
- Correspondence to: Dr M Warner Office of Analysis and Epidemiology, National Center for Health Statistics, Centers for Disease Control and Prevention, 3311 Toledo Road, Hyattsville, MD 20782 USA; mwarnercdc.gov
- Accepted 21 July 2005
Abstract
Objective: To examine effects of length of time between injury or poisoning and interview on the number of reported injury and poisoning episodes in the National Health Interview Survey (NHIS). (Hereinafter, both injuries and poisonings will be referred to as “injuries”.)
Design: The NHIS collects data continuously on medically attended injuries occurring to family members during the three months before interview. Time between injury and interview was established by subtracting the reported injury date from the interview date. Values were multiply imputed for the 25% of the episodes for which dates were only partially reported.
Main outcome measures: An analysis of mean square error (MSE) was used to quantify the extent of errors in estimated annual numbers of injuries and to compare the contributions of bias and variance to these errors.
Results: The lowest estimated MSEs for annualized estimates for all injuries and for less severe injuries were attained when the annualized estimates were based on 3–6 elapsed cumulative weeks between injury and interview. The average weighted number of injuries reported per week per year was 8% lower in later weeks (weeks 6–13) than in earlier weeks (weeks 1–5) for all episodes, and 24% lower in later weeks than in earlier weeks for contusions/superficial injuries, with both differences being statistically significant. For fractures, however, the averages in the two periods were statistically similar.
Conclusions: The error associated with the estimated annual number of injuries was large with a three month reference period for all injuries and for less severe injuries. Limiting analysis to episodes with up to five weeks between injury and interview has statistical, intuitive, and analytic appeal for all injuries and for less severe injuries.
