Sensitivity and Specificity of Recalled Vasomotor Symptoms in a Multiethnic Cohort

Abstract

Many epidemiologic studies include symptom checklists assessing recall of symptoms over a specified time period. Little research exists regarding the congruence of short-term symptom recall with daily self-reporting. The authors assessed the sensitivity and specificity of retrospective reporting of vasomotor symptoms using data from 567 participants in the Study of Women's Health Across the Nation (1997–2002). Daily assessments were considered the “gold standard” for comparison with retrospective vasomotor symptom reporting. Logistic regression was used to identify predictors of sensitivity and specificity for retrospective reporting of any vasomotor symptoms versus none in the past 2 weeks. Sensitivity and specificity were relatively constant over a 3-year period. Sensitivity ranged from 78% to 84% and specificity from 85% to 89%. Sensitivity was lower among women with fewer symptomatic days in the daily assessments and higher among women reporting vasomotor symptoms in the daily assessment on the day of retrospective reporting. Specificity was negatively associated with general symptom awareness and past smoking and was positively associated with routine physical activity and Japanese ethnicity. Because many investigators rely on symptom recall, it is important to evaluate reporting accuracy, which was relatively high for vasomotor symptoms in this study. The approach presented here would be useful for examining other symptoms or behaviors.

Many epidemiologic studies collecting self-reported symptom data rely on retrospective reporting, with typical recall intervals of 2 weeks or 1 month (1–10). Daily reporting is likely to be more accurate because it requires less recall (11–14), but it is more logistically difficult. Little research has focused on the congruence of retrospective and daily self-reporting.

The Study of Women's Health Across the Nation (SWAN) (15) is one of the few studies to employ multiple measurement frequencies simultaneously. Utilizing data from SWAN, we compared retrospective reporting of any (versus no) vasomotor symptoms—hot flashes and night sweats—in the past 2 weeks from end-of-month surveys with daily reporting carried out during the same interval. Using as the standard any (versus no) vasomotor symptoms in the past 2 weeks based on daily reporting, we estimated sensitivity and specificity for retrospective reporting and identified correlates of sensitivity and specificity. We hypothesized that sensitivity would increase with frequency of vasomotor symptoms—and thus would be positively related to correlates of vasomotor symptoms—and that both sensitivity and specificity would be related to characteristics typically associated with accuracy of self-reporting, such as education.

MATERIALS AND METHODS

Study participants

Details of SWAN's study design and recruitment have been reported previously (15). Briefly, SWAN includes 7 US sites: Boston, Massachusetts; Chicago, Illinois; Detroit, Michigan; Los Angeles, California; Newark, New Jersey; Oakland, California; and Pittsburgh, Pennsylvania. At each site, approximately 430 participants, including non-Hispanic Caucasian women and women from 1 other minority group (African-American, Chinese, Hispanic, or Japanese), were recruited. Eligibility criteria included: being aged 42–52 years; having an intact uterus and at least 1 ovary; having had at least 1 menstrual period and no use of sex steroid hormones in the previous 3 months; not being pregnant or breastfeeding; being able to speak English or another designated language (Spanish, Cantonese, or Japanese); and self-identification with one of the site's 2 designated racial/ethnic groups. During eligibility screening in 1995–1997, a total of 6,517 women met these criteria, and 3,302 women (50.7%) enrolled. The study was approved by each participating institution's review board, and all participants gave informed consent.

At the first annual follow-up, a subcohort of 880 women was recruited into the Daily Hormone Study (DHS), with oversampling of non-Caucasians. Enrollment criteria were the same as those for the parent study, with participants being approximately 1 year older than at baseline. Each year, DHS participants collected first-morning voided urine for an entire menstrual cycle (up to 50 days), concurrently with completion of a daily symptom diary. All SWAN participants also recorded symptoms experienced during the past 2 weeks in an end-of-month survey. The results presented here include data from the first 3 annual DHS collections, which occurred in 1997–2002. To be included, a 2-week interval of vasomotor symptom reporting from the end-of-month survey had to overlap completely with 2 weeks of DHS daily diaries, with no symptom data being missing from either instrument.

Measures

In the end-of-month survey, participants recorded whether in the past 2 weeks they had experienced any hot flashes or flushes and any night sweats. Participants also recorded on a daily basis the occurrence of any hot flashes/night sweats (i.e., combined) in the past 24 hours. For comparability, both reports were collapsed to indicate any hot flashes or night sweats in the past 2 weeks.

Candidate predictors of sensitivity and specificity included those hypothesized to be related to vasomotor symptoms or to reporting accuracy (16–21). Unless specified otherwise, time-varying information was from the annual study visit preceding the DHS collection. Age was computed from birth date and the end date of the 2-week reporting interval. Educational level and financial strain were measured at baseline. Baseline acculturation was defined as an acculturation score related to preferred language for reading, speaking, thinking, and radio/television programs (22). Baseline smoking was assessed with questions from the American Thoracic Society (23). Body mass index was computed as measured weight (kg) divided by the square of measured height (m). Physical activity was assessed from questions adapted from the Kaiser Physical Activity Survey, based on the Baecke physical activity questionnaire (24–26). General symptom awareness was measured at the first annual follow-up visit using the Somatasensory Amplification Scale (27)—the summed score of the degree of awareness of loud noise, hot or cold, hunger, pain, and things happening in one's body, with responses for each item ranging from 1 (not at all) to 5 (extremely true). Depressive symptoms were measured using the 20-item Center for Epidemiologic Studies Depression Scale, which assessed the extent to which each symptom had been experienced in the previous week (28). Sleep difficulty was defined as the number of the following symptoms occurring at least 3 times per week in the past 2 weeks: trouble falling asleep, waking up several times per night, and waking up earlier than planned and being unable to fall asleep again (29, 30).

Menopausal status was defined as “late reproductive” if menses had occurred in the past 3 months with no decrease in regularity, “early menopausal transition” if menses had occurred in the past 3 months with decreased regularity, “late menopausal transition” if the last menstrual period had occurred more than 3 months ago but fewer than 12 months ago, and “postmenopausal” if the last menstrual period had occurred at least 12 months ago without another cause (31–36). These definitions are similar to but not identical to the definitions proposed in recent studies such as the Stages of Reproductive Aging Workshop (34) or the studies by Harlow et al. (35, 36), which were published several years after SWAN data collection began; the definitions used here were based on criteria available before the start of SWAN in 1995 (e.g., World Health Organization criteria) (31–33). For measurement of estradiol and follicle-stimulating hormone levels, blood was drawn annually on days 2–5 (days 2–7 from January 1, 1996, through May 30, 1996) of the menstrual cycle for regularly cycling participants and on a random day within 90 days of the annual visit for others.

Statistical analysis

Participant characteristics at the first DHS visit were summarized using frequencies, means, and standard deviations. Sensitivity and specificity for retrospective reporting, treating 2-week occurrence of any vasomotor symptoms from daily reporting as the standard, were computed at each visit; sensitivity also was computed by the number of days with vasomotor symptoms entered in the daily diaries. Combining data across the 3 annual visits, random-effects logistic regression models (37) were fitted for retrospective reporting of any vasomotor symptoms (a positive match, i.e., true positive vs. false negative) in the subset of participants with any vasomotor symptoms in the daily diaries, and separate models were fitted for retrospective reporting of no vasomotor symptoms (a negative match, i.e., true negative vs. false positive) in the subset with no vasomotor symptoms in the daily diaries, to identify factors related to sensitivity and specificity, respectively. All hypothesis tests were 2-sided.

RESULTS

Of the 880 DHS participants, 313 provided no usable observations for these analyses—that is, no collections with complete overlap of the 2-week reporting interval and no missing symptom data. The remaining 567 participants provided 1–3 observations (mean = 1.5) for these analyses. At the first round of DHS data collection, 363 women had incomplete overlap of the 2-week reporting interval and another 194 had missing symptom data, for a total sample size of 323 at the first round. Reflecting the study design, these 323 participants were predominantly premenopausal or early perimenopausal at the first round of data collection (Table 1), and approximately one-third were Caucasian. Most had at least some college and were highly acculturated. Few were current smokers, and more than half were overweight or obese.

Table 1.

Characteristics of the Sample at Baseline (n = 323), Study of Women's Health Across the Nation, 1997–1999

Characteristic	Mean (SD) or %	Range
Mean no. of days with vasomotor symptoms in daily diaries	2.1 (3.9)	0–14
No. of days with vasomotor symptoms in daily diaries, %
0	60.1
1–2	16.3
3–5	9.7
≥6	13.9
Vasomotor symptoms on last day of 14-day interval, %	19.5
Mean age, years	47.4 (2.6)	43.0–53.5
Ethnicity, %
African-American	20.4
Caucasian	33.1
Chinese	18.6
Hispanic	5.9
Japanese	22.0
Education, %
Less than high school completion	7.1
High school diploma or equivalent	18.9
Some college	31.0
College degree	21.4
Postcollege studies	21.7
Difficulty paying for basics, %
Very hard	6.2
Somewhat hard	28.8
Not at all hard	65.0
Acculturation, %
Low	16.1
Medium	10.2
High	73.7
Smoking status, %
Never smoker	68.9
Past smoker	21.4
Current smoker	9.6
Mean body mass indexa	27.0 (6.6)	16.7–52.9
Body mass index, %
Underweight/normal (<25)	47.5
Overweight (25–29.9)	25.5
Obese (≥30)	27.0
Mean physical activity domain scores (possible range, 1–5)
Routine activity	2.4 (0.8)	1–5
Sports activity	2.6 (1.1)	1–4.8
Household/caregiving activity	2.7 (0.8)	1–5
Mean symptom awareness score (possible range, 5–25)	15.1 (3.6)	5–24
Mean CES-D score (possible range, 0–60)	8.5 (7.9)	0–48
Mean no. of sleep difficulties occurring ≥3 times/week in past 2 weeks (possible range, 0–3)	0.5 (0.8)	0–3
Menopausal status, %
Premenopausal	25.4
Early perimenopausal	74.3
Late perimenopausal	0.3
Mean estradiol level (pg/mL)	72.3 (72.3)	9.7–521.2
Mean follicle-stimulating hormone level (mIU/mL)	26.8 (24.2)	1.7–196.7

Abbreviations: CES-D, Center for Epidemiologic Studies Depression [Scale]; SD, standard deviation.

^aWeight (kg)/height (m)².

Observations excluded because of incomplete overlap of the 2-week reporting interval differed from the analytic sample only in terms of shorter urine collections due to a shorter menstrual cycle. Compared with the analytic sample, women excluded at baseline because of missing data were more likely to be African-American or Hispanic and had lower socioeconomic status, lower acculturation, a lower estradiol level, more depressive symptoms, more anxiety, lower sports-related activity, and a higher body mass index. Differences in factors other than race/ethnicity reflected confounding by race/ethnicity, with the exception of estradiol. To assess the impact of missing symptom data, we carried out the analyses again after interpolating missing symptom data based on reports made before and after a gap; resulting estimates of sensitivity and specificity using complete and imputed data combined were similar (results not shown) to those presented here for women with complete data.

Sensitivity and specificity (Figure 1) were fairly constant across the 3 annual visits. Overall sensitivity, ranging from 78% to 84%, was lower than specificity (85%–89%). Sensitivity was lower in women with fewer symptomatic days reported in the daily diaries; that is, greater underreporting occurred in participants with less frequent symptoms.

Figure 1.

Sensitivity and specificity of retrospective reporting of vasomotor symptoms at each annual study visit, by number of symptomatic days from daily symptom reporting, Study of Women's Health Across the Nation, 1997–2002. Overall sensitivity was 77.8% at the first visit, 80.6% at the second visit, and 83.5% at the third visit. Numbers at the top of each bar indicate the number of observations.

In separate univariate logistic regression analyses conducted for each candidate predictor, few predictors of sensitivity or specificity emerged. The strongest predictor of sensitivity (Table 2) was number of symptomatic days in the daily diaries, with an odds ratio for a positive match of 19.87 for 6 or more symptomatic days as compared with 1–2 symptomatic days. Similarly, sensitivity was higher among women reporting vasomotor symptoms in the daily diary on the day of retrospective 2-week reporting in the end-of-month survey. Specificity was negatively associated with general symptom awareness; that is, overreporting was more common among women with higher symptom awareness and was positively associated with routine physical activity but not sports or household physical activity. Past smokers, but not current smokers, had lower specificity—greater overreporting—than never smokers. Japanese women had the highest specificity. Associations were similar in multivariate analyses that included all of the variables in Table 2 (results not shown).

Table 2.

Unadjusted Odds Ratios for Positive and Negative Matches of Retrospective Reporting With Daily Symptom Reporting of Any (Versus No) Vasomotor Symptoms in a 2-Week Interval (Individual Random-Effects Logistic Regression Analyses), Study of Women's Health Across the Nation, 1997–2002

Characteristic	Positive Match of Retrospective Reporting With Daily Reporting (354 Observations From 264 Women)			Negative Match of Retrospective Reporting With Daily Reporting (491 Observations From 353 Women)
	OR	95% CI	P Value	OR	95% CI	P Value
No. of days with vasomotor symptoms in daily diaries			<0.0001
1–2	Reference
3–5	2.61	1.31, 5.22
≥6	19.87	7.31, 54.03
Vasomotor symptoms recorded in daily diary on last day of 2-week interval			<0.0001
No	Reference
Yes	6.51	3.18, 13.32
Age (1-year increase)	1.11	1.00, 1.24	0.06	1.03	0.93, 1.15	0.53
Ethnicity			0.20			0.07
Caucasian	Reference			Reference
African-American	2.15	0.97, 4.77		0.92	0.41, 2.05
Chinese	1.11	0.46, 2.68		1.67	0.75, 3.70
Hispanic	2.74	0.54, 13.76		0.81	0.21, 3.08
Japanese	1.98	0.88, 4.43		3.20	1.30, 7.85
Education			0.45			0.25
Less than high school completion	0.45	0.14, 1.46		0.55	0.15, 2.06
High school diploma or equivalent	Reference			Reference
Some college	1.26	0.55, 2.87		0.67	0.28, 1.57
College degree	1.12	0.43, 2.95		1.59	0.58, 4.33
Postcollege studies	0.86	0.35, 2.13		1.08	0.40, 2.91
Difficulty paying for basics			0.22			0.20
Very hard	0.50	0.18, 1.39		0.70	0.20, 2.44
Somewhat hard	1.31	0.68, 2.50		0.57	0.31, 1.06
Not at all hard	Reference			Reference
Acculturation			0.83			0.93
Low	1.03	0.49, 2.16		1.03	0.46, 2.33
Medium	1.45	0.43, 4.82		1.18	0.49, 2.82
High	Reference			Reference
Smoking status			0.27			0.03
Never smoker	Reference			Reference
Past smoker	0.94	0.50, 1.77		0.51	0.26, 0.99
Current smoker	2.43	0.77, 7.62		3.12	0.69, 14.11
Body mass indexa			0.83			0.13
Underweight/normal (<25)	Reference			Reference
Overweight (25–29.9)	1.22	0.58, 2.58		0.66	0.32, 1.35
Obese (≥30)	0.99	0.51, 1.91		0.51	0.26, 1.00
Physical activity domain scores (1-unit increase)
Routine activity	0.93	0.65, 1.31	0.66	1.69	1.11, 2.57	0.01
Sports activity	0.92	0.70, 1.21	0.53	1.16	0.87, 1.55	0.32
Household/caregiving activity	0.96	0.69, 1.36	0.84	0.85	0.60, 1.21	0.38
Symptom awareness score (1-unit increase)	1.00	0.92, 1.08	0.90	0.88	0.81, 0.96	0.004
CES-D score (1-unit increase)	1.00	0.96, 1.04	0.98	0.99	0.96, 1.02	0.53
No. of sleep difficulties occurring ≥3 times/week in past 2 weeks (1-unit increase)	1.02	0.75, 1.39	0.91	0.89	0.60, 1.32	0.06
Menopausal status			0.25			0.98
Late reproductive	Reference			Reference
Early transition	1.28	0.56, 2.89		1.01	0.50, 2.01
Late transition/postmenopausal	2.60	0.81, 8.38		0.89	0.25, 3.2
Log estradiol levelb	0.73	0.51, 1.02	0.07	1.06	0.72, 1.55	0.78
Log follicle-stimulating hormone levelb	1.34	0.94, 1.91	0.11	1.29	0.87, 1.91	0.21

Abbreviations: CES-D, Center for Epidemiologic Studies Depression [Scale]; CI, confidence interval; OR, odds ratio.

^aWeight (kg)/height (m)².

^bOdds ratios were computed for the difference between the 75th percentile and the 25th percentile and were adjusted for day of the menstrual cycle.

DISCUSSION

Both sensitivity and specificity for retrospectively reported vasomotor symptoms in the past 2 weeks were high (78%–89%), similar to findings regarding recall of hormone therapy (38). Accuracy of retrospective reporting was strongly associated with the number of symptomatic days in daily diaries. Women at either extreme—0 or ≥6 symptomatic days in the past 2 weeks—had the highest probability of accurate retrospective reporting, perhaps because these extremes are easier to recall (39). Extreme occurrences are more accurately recalled in other contexts as well, including heavier menstrual bleeding (40) and skipped menstrual periods (41) in studies of menstrual cycling, injurious falls versus noninjurious falls among elders (42), and vigorous rather than moderate physical activity (43–45). Because frequency of vasomotor symptoms has been positively related to severity (46), these results also suggest that women with milder vasomotor symptoms are more likely to underreport retrospectively recalled symptoms.

The higher sensitivity among women with vasomotor symptoms on the day of retrospective reporting is consistent with prior findings that recall of past symptoms is affected by current symptoms (47–52). Higher symptom awareness was associated with overreporting, reflecting the impact of symptom salience on memory (39, 53).

Japanese participants had the highest specificity, as well as lower rates of missing symptom data, suggesting greater general compliance with the study protocol. The lower specificity in past smokers but not current smokers as compared with never smokers may be a spurious finding due to the relatively small number of current smokers. The association of specificity with routine physical activity but not other domains may be due to chance, given the number of statistical comparisons.

One limitation of the study was that by necessity, comparisons of retrospective and daily reporting were restricted to women recording symptoms on a daily basis for at least 13 days prior to the end-of-month report, who thus may have been sensitized to symptom occurrence (or nonoccurrence) (38, 54). The relatively short recall period also may have contributed to increased accuracy (55), as seen for recall of occurrences such as menstrual bleeding (56) and sexual activity (57, 58). A 2-week recall period is frequently used, however, and thus these results are likely to be applicable to many studies of vasomotor symptoms.

In addition, women excluded because of missing data were more likely to have characteristics associated with lower accuracy of reporting. Consequently, estimates of sensitivity and specificity may be inflated in comparison with the general population. However, results including imputations of missing symptom data were similar to those presented here. Moreover, sensitivity and specificity did not differ significantly for women with observed symptom data and women with missing (imputed) symptom data. The odds of a negative match were 1.39 times higher (95% confidence interval: 0.88, 2.20) for women with observed symptom data than for women with imputed symptom data, and the corresponding odds ratio for a positive match was 1.34 (95% confidence interval: 0.82, 2.20).

Other limitations include the lack of measured daily frequency, intensity, or bothersomeness. In addition, because only 23 women had less than a high school education, we had relatively low power to detect a difference between this group and persons with higher levels of educational attainment regarding sensitivity and specificity. No gradient was observed across the higher educational categories, however.

The strengths of this study included the availability of a large and diverse community-based sample with extensive covariate information. The wealth of study data provided a rare opportunity to assess the congruence of retrospective 2-week symptom recall with concurrent daily reporting.

Limiting recruitment to women with frequent or bothersome symptoms, despite the greater accuracy in their symptom measurement, would preclude a description of the full range of women's symptom experiences and decrease the ability to identify correlates or sequelae of symptoms by decreasing sample variability (59). To reduce underreporting without drastically increasing participant burden, more frequent self-reports (e.g., daily reporting over a short interval) could be substituted for retrospective recall, as in the SWAN DHS or the Seattle Midlife Women's Health Study (60). Alternatively, analyses of retrospective reports could collapse “asymptomatic” and “low frequency” subgroups, since the former may include some participants with truly low symptom frequencies. Similarly, given the previously observed correspondence between symptom frequency and severity (46), “asymptomatic” and “mild symptoms” subgroups could be combined in analyses of symptom severity. Accounting in statistical analyses for symptom awareness and occurrence of symptoms on the day of retrospective recall also would be beneficial (49).

Undoubtedly, researchers will continue to rely on retrospective symptom reporting, particularly in large epidemiologic studies. Thus, it is important to improve measurement. Sensitivity and specificity and their correlates may vary across symptoms or behaviors. For example, social desirability may account in part for underreporting of heavy alcohol consumption (61–64) or overestimation of physical activity in participants with higher body mass index (43). The approach illustrated here, however, would be useful for examining accuracy of retrospective reporting for a variety of health-related factors.

Glossary

Abbreviations

DHS

Daily Hormone Study

SWAN

Study of Women's Health Across the Nation