American Journal of Epidemiology Copyright © 1997 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved

Vol. 146, No. 9 Printed in U.S.A

Reliability of Reported Age at Menopause

Robert A. Hahn,1 Elaine Eaker,2 and Henry Rolka1 Age at menopause is an important epidemiologic characteristic whose reliability of reporting in the US population is not known. The authors examined four hypotheses about the reliability of reported age at menopause in the United States: 1) women with hysterectomy-induced menopause more reliably report their age at menopause than women who have undergone natural menopause; 2) reliability declines with time since menopause; 3) reliability declines with age; and 4) women with higher educational levels report their age at menopause more reliably than women with less education. The authors used linear regression models among 2,545 women in the First National Health and Nutrition Examination Survey and Followup Study (1971-1984) and compared responses at first and follow-up interviews. Among women who had undergone a natural menopause, 44% reported their age at menopause within one year from the first to second interviews; among women who had undergone a hysterectomy-induced menopause, 59% reported their age at menopause within one year from first to follow-up interviews. Only hysterectomy status and years from menopause to follow-up interview were significantly associated with the absolute difference between age at menopause reported at first and follow-up interviews. The authors conclude that caution in studies involving age at menopause may enhance our understanding of this critical event in the lives of women. Am J Epidemiol 1997;146:771-5. age of onset; data collection; hysterectomy; menopause; reliability and validity

Menopause is an event that directly affects the reproductive capacities and personal lives of women. In addition, the hormonal changes associated with menopause may be related to various chronic conditions, e.g., osteoporosis and cardiovascular diseases (1). Menopause is also the outcome of hysterectomy—the most common nonobstetric surgery in the United States. In recent years, 580,000 women have undergone this operation annually (2). Because of the large number of women who have had hysterectomies and because women who have undergone this operation are not at risk for certain other conditions or procedures, e.g., endometrial cancer or surgical sterilization, information on hysterectomy is basic in estimating populations at risk for such conditions (3, 4). Because age at menopause is an important demographic and epidemiologic determinant, we examined the reliability of age at menopause reported at two

different times using a sample from the United States noninstutionalized population of women aged between 25 and 74 years. We examined four hypotheses:

Received for publication July 13, 1995, and accepted for publication June 26, 1997. Abbreviations: NHANES I, First National Health and Nutrition Examination Survey; NHEFS, National Health Examination Followup Study. 1 Epidemiology Program Office, Centers for Disease Control and Prevention, Atlanta, GA. 2 Marshfield Medical Research Foundation, Marshfield, Wl. Reprint requests to Dr. Robert A. Hahn, Epidemiology Program Office, Mailstop D-01, Centers for Disease Control and Prevention, Atlanta, GA 30333.

MATERIALS AND METHODS

1. Women with hysterectomy-induced menopause will more reliably report their age at menopause than women who have undergone natural menopause. 2. Reliability of recall declines with time elapsed since menopause. 3. Reliability declines with age, independent of time since menopausal event. 4. Women with higher educational levels report their age at menopause more reliably than women with lower levels of education. We also examined digit preference in reported age at menopause, i.e., a preference for reporting ages with years ending with "0," "5," or "2."

From 1971 through 1975, as a part of the First National Health and Nutrition Examination Survey (NHANES I), a probability sample of 14,407 adults aged 25-74 years from the noninstitutionalized civilian population of the United States was interviewed and examined; the sample included 8,597 women (5, 6). During the initial survey period (1971-1974), per771

772

Hahn et al.

sons from regions affected by poverty, older persons, and women of childbearing ages were oversampled. During 1974-1975, an additional sample was drawn, without oversampling, to increase the total sample size. From 1981 through 1984, extensive efforts were made in the National Health Examination Follow-up Study (NHEFS) to recontact, reinterview, and briefly reexamine the NHANES I population (6). Follow-up information was available for 13,318 (92.4 percent) of persons in the original sample; of these initial respondents, 2,022 (15.2 percent) had died. Information on subjects who had died or were disabled was provided by proxies. At initial interview, women were asked, "How old were you when your periods or menstrual cycle started?" They were then asked, "Have they stopped entirely?" and "At what age?" At follow-up, women were asked, "At what age did you have your last period?" and "Did your periods stop naturally, because you had surgery or for some other reason?" In this study, we refer to women who reported that their menstrual cycles had stopped naturally as menopausal and to women who reported that their menopause resulted from surgery as having undergone a hysterectomy. The study sample included women who reported at both intake and follow-up interviews that their menses had stopped. We excluded subjects whose menopausal history was reported by proxies as well as women who were pregnant at initial interview. We also excluded two women in the sample whose reported chronological age at first interview was less than their age at menopause reported at the same interview. We measured the reliability of reported age at menopause by the difference between the age reported by menopausal women during their follow-up interview and the age reported in the NHANES I interview (i.e., age reported at follow-up — age reported at first interview). We focused on the absolute difference as a specific outcome measure. To test the four study hypotheses, we used multiple linear regression models (7) in which the absolute difference between age at menopause reported at first and follow-up interviews was the dependent variable; age, hysterectomy status, years from menopause to final interview, and education were independent variables. Because the hypothesized decline in reliability over time might be different for women with and without hysterectomies, we included an interaction term for hysterectomy and time-since-menopause. Nonsignificant terms (p > 0.05) were removed from the model one at a time, and the model was retested. We also used the Pearson product moment correlation

(8) to compare age at menopause reported at first and follow-up interviews; we compared correlations between women with natural menopause and those with hysterectomy-induced menopause. Because data were collected by using a complex sample survey methodology, sampling weights were taken into account in the analysis by using SUDAAN software (9). The design effect observed in this analysis ranged from 1.92 to 2.29. Thus, had the complex survey design not been taken into account, variances would have been underestimated by approximately one half. Finally, we examined digit preference in reported ages, e.g., a preference for ages ending in "0," "5," or "2." Under the assumption of unbiased reporting, it is reasonable to expect close to 10 percent of the age reports to end in each of the digits "0" through "9." To test this assumption, we used a chi-square goodness of fitness test (10). RESULTS

In the NHANES I cohort of 7,011 nonpregnant women aged 25-74 years who responded for themselves in first and follow-up interviews, 2,545 (36.3 percent) reported being menopausal at both interviews. Among menopausal women, 1,473 (57.9 percent) had undergone natural menopause and 1,072 (42.1 percent) had experienced hysterectomies (table 1). Among women who reported a natural menopause, the median age at menopause was reported at first (and follow-up) interviews to be 50 years. Among women who reported menopause by hysterectomy, the median age at menopause was reported at first interview to be 40 years. Women who had undergone a hysterectomyinduced menopause were younger and had less time

TABLE 1. Characteristics of study population by natural or hysterectomy-induced menopause: 2,545 women in the First National Health Examination Survey and Followup Study, 1971-1984 Type of menopause Characteristic

Natural (/)= 1,473)

Hysterectomyinduced (n= 1,072)

1.5 50.0 48.8

30.3 47.0 22.7

83.9 16.1

81.3 18.7

50 (26-67) 24 (7-59)

40(19-62) 21 (7-56)

Age (years) at first interview 25-^4 45-64 65-75 Education (%) High school Age (years) at menopause reported at first interview, median (range) Follow-up, median (range)

Am J Epidemiol

Vol. 146, No. 9, 1997

Reliability of Reported Age at Menopause

from reported age at menopause to follow-up interview than women who had undergone natural menopause; comparison groups did not differ substantially in educational level (table 1). Median follow-up time both for women with natural and with hysterectomy-induced menopause was between 9 and 10 years. Among women who had undergone a natural menopause, median time from menopause to follow-up was 24 years. Among women who had undergone a hysterectomy, median time from menopause to follow-up was 21 years. Among women who had undergone natural menopause, 26.5 percent reported the same age at first and follow-up interviews, and 8.8 percent reported ages that differed by >10 years (table 2, figure 1). Among women who had experienced hysterectomy-induced menopause, 32.6 percent reported the same age at first and follow-up interviews, and 5.6 percent reported ages that differed by >10 years (table 2, figure 1). The median difference in reported age was approximately zero both for women with natural and hysterectomy-induced menopause. Both for women with natural and hysterectomyinduced menopause, there was no systematic tendency for reported ages at menopause to increase or decrease from first to follow-up interviews. In a regression model including terms for age, hysterectomy status, years from menopause to follow-up interview, the interaction of hysterectomy and years since menopause, and education, only hysterectomy status (/3 = 2.29, p < 0.01) and years from menopause to follow-up interview (0 = 1.92, p = 0.02) were significantly associated with the absolute difference between first and follow-up reported ages at menopause (figure 1). There was no association between either a woman's age or her education and the absolute difference in reported age at menopause. TABLE 2. Distribution of absolute differences in reported age at menopause from first to follow-up interviews: 2,545 women in the First National Health Examination Survey and Followup Study, 1971-1984 Absolute

Natural

Hysterectomy

0 1

26.5 17.5 14.0*

32.6 26.3* 12.4

9.4 5.9

7.3 4.6 4.1 1.8 1.6 1.9 1.8 5.6

2

3

:

Type of menopause (%)

difference (in years)

4 5 6

7.4 3.5

7 8

2.5 2.8

9 £10

8.8

1.7

Median value.

Am J Epidemiol

Vol. 146, No. 9, 1997

773

On average, women with hysterectomy had a 10month smaller absolute difference between ages reported at first and follow-up interviews than women with natural menopause. The correlation of ages given at first and follow-up interviews was 0.57 {p < 0.0001) among women with natural menopause and 0.82 {p < 0.0001) among women with hysterectomyinduced menopause (figure 1). Chi-square goodness of fit test indicated digit preference for reported age at menopause for both the initial survey and for the follow-up {p < 0.0001). The pattern of overall preference was greatest for those numerals ending in "0," followed by "5," "2," and "8." At first interview, women with hysterectomy-induced menopause were 1.2-1.3 times more likely than chance to report an age ending in "0," "5," or "8"; women with natural menopause were 2.1 times more likely than chance to report an age ending in "0," and 1.4-1.5 times more likely to report an age ending in "5" or "2." Only among women with hysterectomyinduced menopause, digit preference increased slightly from initial to follow-up interviews. Comparison of responses to questions regarding reproductive characteristics in the follow-up cohort indicates that the questions asked may have been confusing to respondents. Among 1,749 women who reported at follow-up having had a hysterectomy (and who also answered questions about their womb and ovaries), 221 (12.6 percent) reported that they still had their wombs; of these women, 126 also reported that they still had both ovaries. In addition, of 2,425 women who reported not having had a hysterectomy, 206 (8.5 percent) reported not having their womb. DISCUSSION

Among women who had experienced natural menopause and who were interviewed on two occasions, 44 percent reported their age at menopause within one year from the first to second interviews, while among women who had undergone hysterectomy-induced menopause, 59 percent reported their age at menopause within one year from first to follow-up interviews. Among women who had undergone a natural menopause, 9 percent differed in age at menopause reported in two interviews by ^ 1 0 years, and among women who had undergone a hysterectomy-induced menopause, 6 percent differed in age at menopause reported in two interviews by ^ 1 0 years. The finding that the median difference in reported age was approximately zero suggests that there is no systematic age misclassification, i.e., women do not tend to recall their menopause either earlier or later from first to follow-up interviews. Difference in reported age is not significantly affected by chronological age or level of

774

Hahn et al.

Natural Menopause Group

Hysterectomy Group 8

f

8 is a

8

o

a.

8 -

a.

ai

9

o

20

30

40

50

60

70

Initially Reported Age

20

30

40

50

60

70

Initially Reported Age

FIGURE 1. Association between initially reported age of menopause and age (in years) reported at follow-up in the hysterectomy-induced menopause group (left graph) and the natural menopause group (right graph): 2,545 women in the First National Health Examination Survey and Followup Study, 1971-1984.

education, but increases with time since menopause and is greater among women who had undergone natural menopause than among those who had experienced hysterectomy-induced menopause. There was a marked preference in recalled ages ending with digits "0," "5," and, to a lesser extent, "2." In the absence of medical or physiologic documentation, this study could not address validity—that is, reported versus true age at menopause. With discrepant reported ages, although both may be erroneous, both cannot be true. Because reporting may be more accurate when it occurs soon after an event (rather than later) and because recall generally declines with increasing age, ages reported more proximal to menopausal events may be more valid than ages reported later. The questions asked about menopausal status in first and follow-up interviews were not identical and might have been interpreted differently. The question at initial interview, "Have they [i.e., your periods or menstrual cycle] stopped entirely?" (followed by a question of the age at this event) more clearly indicates

menopause than does the question asked at follow-up, "At what age did you have your last period?" However, the next question in the follow-up interview, i.e., "Did your periods stop naturally, because you had surgery, or for some other reason?", clarifies the reference of the preceding question to menopause. Nevertheless, there is evidence of misunderstanding in responses to some of the questions asked. At followup, 12.6 percent of women who claimed to have had a hysterectomy also reported still having their "womb or uterus," and 8.5 percent of women who claimed not to have had a hysterectomy also reported not having their "womb or uterus." The likelihood that a woman reported having had a hysterectomy and still having a womb did not differ substantially among those with and without a high school education. Based on small, regional samples, two previous studies have assessed the reliability of reported age at menopause. Bean et al. (11) recorded menstrual history events reported near the time of their occurrence and found that 83.4 percent of college-educated women who had undergone a hysterectomy reported Am J Epidemiol

Vol. 146, No. 9, 1997

Reliability of Reported Age at Menopause

their age at hysterectomy within one year from first to follow-up interviews a mean of 10.6 years later; 75.5 percent of women with a natural menopause reported their age at menopause within one year from first to follow-up interviews a mean of 7.6 years later. Paganini-Hill and Ross (12) compared the age at last menstrual period reported by older, affluent women with the medical records of the same women. These authors found that 48.4 percent of women who had undergone a natural menopause agreed on their age at menopause within one year, and 67.7 percent agreed within 3 years. Among women with a hysterectomy, 67.8 percent were within one year of reported age at hysterectomy, and 81.4 percent were within 3 years. The reliability reported here is substantially less than that reported by Bean et al. (11) and slightly less than that reported by Paganini-Hill and Ross (12). The greater reliability reported by Bean et al. may be a result of the college education and/or the frequent interviewing of the cohort whose menstrual histories were updated in ongoing surveys, resulting in repeated prompting to recall menstrual events. The lower reliability may also result from different wordings and different understandings (or misunderstandings) by NHANES and NHEFS subjects. Our results suggest the need for caution in research involving reported age at menopause, and particularly greater caution for reported age among women who have undergone a natural menopause and among women who report their menopause many years since the event. Our results also provide a means for conducting sensitivity analyses in studies using reported age at menopause (11); they indicate the range of variability in reporting that may be expected. Such

Am J Epidemiol

Vol. 146, No. 9, 1997

775

caution may enhance our understanding of this critical event in the lives of women.

REFERENCES 1. Belchetz P. Hormonal treatment of postmenopausal women. N Engl J Med 1994;330:1062-71. 2. Graves EJ. 1992 Summary: National Hospital Discharge Survey. Advance data from Vital and Health Statistics, no. 249. Hyattsville, MD: National Center for Health Statistics, 1994: 1-12. 3. Nolan TF, Ory HW, Layde PM, et al. Cumulative prevalence rates and corrected incidence rates of surgical sterilization among women in the United States, 1971-1978. Am J Epidemiol 1982;116:776-81. 4. Lyon JL, Gardner JW. The rising frequency of hysterectomy: Its effect on uterine cancer rates. Am J Epidemiol 1977;105: 439-43. 5. National Center for Health Statistics. Plan and operation of the Health and Nutrition Examination Survey, United. States, 1971-73. Rockville, MD: US Department of Health, Education, and Welfare, 1976. Vital and Health Statistics, series 1, no. 10a. DHEW publication no. (HRA) 76-1310. 6. Cohen BB, Barbano HE, Cox CS, et al. Plan and operation NHANES I Epidemiologic Followup Study 1982-84. Vital Health Statistics, series 1, no. 22. Washington, DC: Public Health Service, 1987. DHHS publication (PHS) 87-1342. 7. Cox DR. Regression models and life tables. J R Stat Soc [B] 1972;34:187-202. 8. Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic research: principles and quantitative methods. Belmont, CA: Lifetime Learning Publications, 1982. 9. SUDAAN, version 6.34. Research Triangle Park, NC: Research Triangle Institute, 1994. 10. Armitage P, Berry G. Statistical methods in medical research. 3rd ed. Boston, MA: Blackwell Medical Publications, 1994. 11. Bean JA, Leeper JD, Wallace RB, et al. Variations in the reporting of menstrual histories. Am J Epidemiol 1979; 109: 181-5. 12. Paganini-Hill A, Ross RK. Reliability of recall of drug usage and other health-related information. Am J Epidemiol 1982; 116:114-22.