Reliability of Maternal Recall of Feeding at the Breast and Breast Milk Expression 6 Years After Delivery

Introduction

Breastfeeding statistics commonly serve as a population-level indicator of maternal and child health, and accurate measurement of breastfeeding is important in research on the etiology of outcomes such as childhood asthma and allergies,^1,2 overweight and obesity,^2,3 cognitive development,^2,4 as well as maternal cancers and type 2 diabetes. ⁵ Most epidemiologic studies use retrospective data about breastfeeding that rely on maternal recall. However, if recall is poor, breastfeeding exposure may be misclassified and effect estimates may be biased as a result.

Several previous studies that examined maternal recall of breastfeeding showed that even though mothers tended to slightly overestimate their breastfeeding duration when queried after several years,^6–13 maternal recall related to any breastfeeding (both for ever/never binary initiation and duration) remains strongly reliable, as evidenced by high correlation statistics.^6,7,14–18 In samples with high rates of breastfeeding, agreement for breastfeeding initiation is almost perfect.^7,8,14,19 Sociodemographic factors may affect recall accuracy, but the literature is mixed in terms of which factors are important. For instance, studies have reported that mothers who are multiparous,^6,15 smokers, ⁶ older, ¹⁰ and have low educational attainment^6,12,17 tend to have larger median recall bias toward overestimation and lower reliability (correlation statistics).

However, infant feeding practices have recently changed from predominantly feeding at the breast to including increasing amounts of maternal milk expression (pumping) and feeding children expressed milk, in addition or instead. ²⁰ To date, no studies evaluating recall have included assessment of maternal milk expression and feeding expressed milk.

Differentiating maternal milk expression and child consumption of expressed milk from feeding at the breast is increasingly important as research emerges to suggest they may have differing effects on maternal and child health.^21,22 The Brief Breastfeeding and Milk Expression Recall Survey (BaByMERS) captures a recalled history of each of these feeding practices, but how well women recall these practices after several years using the BaByMERS has not been examined. ²³ Evaluating recall reliability is an important part of validating such an interview tool for use in future research. Our objective in the present study was to compare maternal responses using the BaByMERS at 12 months and 6 years postpartum to evaluate the reliability of maternal recall using this tool.

Materials and Methods

Results

Of the 497 participants who answered the recorded BaByMERS at 12 months after delivery, 299 completed the recalled questionnaire 6 years after delivery (Appendix Fig. A1), the “follow-up complete” group (Table 1). Participants who were nonresponsive, could not be located, or refused the invitation to the second phase were defined as a “follow-up incomplete” group (n = 198). The 299 participants who completed the follow-up study comprised a diverse sample: 35% mothers had a postgraduate education, half were primiparous at the time of the first phase, 6% smoked during pregnancy, a quarter enrolled in WIC during pregnancy or the first year postpartum, 27% had income <$35,000, and 15% had no health insurance or public insurance; more than one-third of infants were delivered by cesarean section. Mothers who delivered at term, had higher educational attainment, higher annual household income, private insurance, and did not enroll in WIC were more likely to complete the follow-up study 6 years postpartum.

The Kappa statistics showed substantial agreement between the two time points for initiations of feeding at the breast (κ = 0.76; 95% confidence interval (CI): 0.62–0.89), maternal milk expression (κ = 0.67, 95% CI: 0.51–0.83), and breastfeeding or feeding breast milk (κ = 0.71, 95% CI: 0.44–0.98) (Table 2). Only moderate agreement was observed for initiations of child consumption of expressed milk (κ = 0.48, 95% CI: 0.33–0.63).

The reliability of maternal recall for durations of feeding at the breast, and breastfeeding or feeding breast milk, was excellent (ICC = 0.96, 95% CI = 0.95–0.97 and ICC = 0.94, 95% CI = 0.92–0.95, respectively), and maternal recall for durations of maternal milk expression and child consumption of expressed milk was slightly less reliable but still strong (ICC = 0.83, 95% CI = 0.80–0.87 and ICC = 0.77, 95% CI = 0.72–0.81, respectively) (Table 3). More than half of the mothers could recall the durations of all feeding or lactation practices to within 1 month.

The Bland/Altman plots showed the bias between each paired recorded and recalled duration value (Fig. 1). Mothers tended to slightly overestimate the durations of all the feeding or lactation practices when they recalled them (difference = 6.4 days, SD = 41.3 for feeding at breast; difference = 9.6 days, SD = 70.4 for maternal milk expression; difference = 8.7 days, SD = 85.5 for child consumption of expressed milk; difference = 14.8 days, SD = 45.4 for breastfeeding or feeding breast milk).

OPEN IN VIEWER

FIG. 1.

Bland/Altman plots for bias between recorded and recalled durations of each feeding practice, Moms2Moms follow-up (n = 299).

When stratified by participant characteristics, the reliability of maternal recall for the feeding practice duration variables was fairly consistent and high across subgroups with a few exceptions. Mothers who identified as a race other than white/Caucasian or black/African American, Hispanic women, and women with a high school/GED or less education tended to have a poorer recall reliability for the duration of child consumption of expressed milk (ICC = 0.46, 95% CI = 0.19–0.77; ICC = 0.45, 95% CI = 0.13–0.83; ICC = 0.58, 95% CI = 0.35–0.79; respectively), although the CIs overlapped with that of all mothers (ICC = 0.77, 95% CI = 0.72–0.81) (Table 4).

Compared with the full sample (ICC = 0.96, 95% CI = 0.95–0.97, Appendix Table A2), mothers who were black/African American, unmarried/not living with a partner, or with public/no health insurance had slightly lower recall reliability on the duration of feeding at the breast (ICC = 0.85, 95% CI = 0.72–0.92; ICC = 0.88, 95% CI = 0.77–0.94; ICC = 0.91, 95% CI = 0.85–0.95; respectively), but these estimates were close to those for the full sample. On the duration of breastfeeding or feeding breast milk (median of differences for the full sample = 0 days, interquartile range [IQR] = 30 days, ICC = 0.94, 95% CI = 0.92–0.95) (Appendix Table A3), mothers who smoked during pregnancy tended to have a larger median recall bias toward overestimation (median = 30 days, IQR = 31 days), and Hispanic mothers had a lower recall reliability (ICC = 0.75, 95% CI = 0.43–0.91). In contrast with the full sample on the duration of maternal milk expression (ICC = 0.83, 95% CI = 0.80–0.87) (Appendix Table A4), women who identified as a race other than white/Caucasian or black/African American and who had a high school/GED or less education reported less reliably (ICC = 0.59, 95% CI = 0.32–0.80; ICC = 0.60, 95% CI = 0.36–0.79; respectively).

Discussion

In this prospective cohort study, maternal recall of infant feeding and maternal lactation practices was strongly reliable 6 years after delivery, although the recall of maternal milk expression and child consumption of expressed milk was somewhat less than feeding at the breast. More than three-fourths of the mothers were able to accurately recall their recorded durations of feeding at the breast and breastfeeding or feeding breast milk to within 1 month. The reliability of maternal recall was somewhat reduced for some subgroups distinguished by minority race or ethnicity, lower economic status, or smoking. This was not consistent across feeding practices and was generally of small to moderate magnitude.

Amissah et al. found that maternal recall of breastfeeding duration was valid 6 years after childbirth with a 1-week bias toward overestimation (ICC = 0.84); and smoking and low educational attainment stood out as the main factors associated with lower recall validity (ICC = 0.61; ICC = 0.63, separately). ⁶ Natland et al. reported the accuracy of 20-year maternal recall for breastfeeding initiation and duration was high (κ = 0.85, 95% CI = 0.82–0.88; ICC = 0.82), with a modest median overestimation of 2 weeks. ⁷ The review article written by Li et al. also examined the reliability of maternal recall for both breastfeeding initiation and duration after years, based on 11 studies published between 1966 and 2003. ¹⁷ Beyond their studies, we assessed the reliability of maternal recall for each separate feeding and lactation practice using our unique BaByMERS: feeding at the breast, maternal milk expression, child consumption of expressed milk, and breastfeeding or feeding breast milk (combined), on both initiation and duration variables. Demographic factors that may affect maternal recall were also tested in our study.

Around a quarter of mothers recalled the durations of maternal milk expression and child consumption of expressed milk to be more than a month longer than recorded, although the recall reliability remained high. As shown in previous studies, the higher the rates of breastfeeding, the stronger the reliability of maternal recall.^7,14 The prevalence of feeding at the breast in our sample was 98%, whereas the prevalence of maternal milk expression and child consumption of expressed milk was 91%. Furthermore, the median recorded duration of feeding at the breast was 1 month longer than that of maternal milk expression and child consumption of expressed milk. Therefore, it was reasonable that the reliability of maternal milk expression and child consumption of expressed milk was lower than that of feeding at the breast.

Some subgroups defined by maternal race/ethnicity, education, marital status, insurance type, and smoking had somewhat poorer maternal recall reliability, but not consistently across all feeding or lactation practices. Only a few studies have examined how sociodemographic factors could potentially affect the reliability of maternal recall of breastfeeding behaviors.^6,10,15,17 Findings from those studies have been inconclusive. Li et al. reported that differences in recall by maternal race/ethnicity, education, and socioeconomic status may have stemmed from differences in social norms or beliefs or communication quality with the interviewers. ¹⁷ They explained that the reason of overestimation of breastfeeding among wealthier and more educated mothers may be according to their beliefs or those of the interviewers. ¹⁷ It may also be that lower educational attainment was related to poorer understanding of the questions, especially for new concepts such as milk expression. Prior studies have reported that mothers with lower socioeconomic status recalled more poorly on traditional breastfeeding questions.^15,17,28 Alternatively, these sociodemographic factors may simply reflect the fact that women in these subgroups had shorter durations for each feeding practice, on average, and prior research has shown that shorter duration is associated with poorer recall, as mentioned previously.^7,9,14

The present study is unique because it was the first study to examine maternal recall of maternal milk expression and child consumption of expressed milk. It validated maternal recall reliability based on both feeding initiation (binary ever/never) and duration (continuous) variables. It was a longitudinal, prospective cohort study. Also, this was one of the very few studies to assess how sociodemographic factors may affect breastfeeding recall.

A limitation of the study was that the study sample came from a single clinical site, so the results may not be generalizable to all U.S. women. Second, as with any longitudinal study, loss to follow-up affected a portion of the sample, and some background characteristics differed between the original sample and the follow-up sample. Our estimates may have been affected by differential loss to follow-up. Third, we were able to compare self-reported feeding practices at two time points, rather than to compare self-report with an ideal gold standard that would be unaffected by self-report. Thus, inaccurate reporting was a possibility at both time points. However, self-report is the typical method to measure breastfeeding in epidemiologic research, so our data reflect typical practice. Also, it is possible that if the BaByMERS had been administered at the first phase as an interview rather than self-administered questionnaire, reliability may have been even higher. We have previously reported that the interview version of the BaByMERS has high concordance with responses to commonly used breastfeeding questions in federal child health surveys and may perform better for capturing expressed milk feeding. ²⁹ Finally, around a quarter of the mothers continued feeding at the breast, and two mothers continued feeding their child expressed breast milk beyond 12 months postpartum. The durations of these practices at both time points were truncated to 365 days. This could lead to a small overestimation of the accuracy of maternal recall on all feeding durations.

Conclusion

Maternal recall of child consumption of expressed milk, maternal milk expression, and feeding at the breast was strong to excellent after 6 years after delivery using BaByMERS. Overall, BaByMERS exhibits acceptable accuracy in capturing the recall history of these contemporary feeding and lactation practices after several years, and can be used as an interview tool for recording infant feeding and maternal lactation in future research. Future researchers should consider that the reliability was lower for some practices among women from racial and ethnic minority groups, women with less educational attainment, unmarried women, women with public or no health insurance, and those who smoked, although this was not consistent across feeding practices and was generally of small to moderate magnitude.