Assessing number-specific error in the recall of onset of last menstrual period

The goal of this investigation was to determine whether women who did not report preferred numbers for their last menstrual period (LMP) may be a group of women who are particularly careful in keeping track of their menstrual cycles and therefore have more accurate LMP dating--based on a comparison with ultrasound examinations. We also sought to estimate the frequency with which preferred numbers are reported in different sources of data and for different subgroups of women. First, we examined the 1987 California birth certificates in which LMP was collected at the time of birth (n = 504853). We also examined the records of 43880 women participating in the California Alpha-fetoprotein (AFP) Screening Program between 1986 and 1987, for whom gestational ages based on both early ultrasound examination and LMP were collected before 20 weeks of gestation. In the 1987 California birth certificates, seven numbers--1, 5, 10, 15, 20, 25 and 28--were recorded more frequently than expected. An estimated 12.9% of these records had preferred numbers. The most frequently recorded number was 15, occurring 2.5 times more often than expected (P < 0.01). In the data of the AFP Screening Program, the same seven numbers were preferred, and approximately 7.9% of records were affected by number preference. Comparisons with measurements of gestational age based on ultrasound demonstrated that LMP-based gestational ages in which non-preferred numbers are reported for the LMP are slightly more accurate than those in which preferred numbers are reported (P < 0.01). In most cases, number preference appears to introduce small errors into measurements of gestational age, probably as a result of rounding. Thus, the effect of number preference may be primarily of interest to research studies in which small errors in the measurement of gestational age will have a significant impact on findings.