Development and validation of a quantitative choline food frequency questionnaire for use with drinking and non-drinking pregnant women in Cape Town, South Africa

Fetal alcohol spectrum disorders (FASD) comprise a continuum of alcohol-related neurodevelopmental disorders ranging from the most severely affected children with fetal alcohol syndrome (FAS) to nonsyndromal children who also exhibit neurocognitive and/or behavioral deficits but may lack the facial features or growth deficits seen with FAS [1, 2]. Worldwide, a significant number of women drink heavily during pregnancy despite public health advisories and psychosocial interventions [3‐5]. In the Western Cape Province of South Africa, where rates of heavy drinking during pregnancy are endemically high among women from the Cape Coloured (mixed ancestry) community [6‐8], the prevalence of FAS is as high as 80 per 1000 [9].

A growing body of studies in FASD animal models has demonstrated that optimal maternal choline status can mitigate some of the teratogenic effects of alcohol [7, 10‐16]. Thomas and colleagues have shown protective effects of pre- and postnatal choline supplementation on hippocampal development and related neurobehavioral outcomes in rats [10, 11], including reversal of alcohol-related deficits in eyeblink conditioning, which we have also shown to be profoundly affected in children prenatally exposed to alcohol [7, 12, 13]. We have recently extended these findings to humans in an exploratory randomized, double-blind, controlled trial, which demonstrated that high-dose choline supplementation initiated early in pregnancy can mitigate adverse effects of heavy drinking on infant eyeblink conditioning, cognition, and post-natal growth [14, 15].

Choline is an essential nutrient that is a constituent of the neurotransmitter acetylcholine and a precursor to phosphatidylcholine and sphingomyelin, which are major components of cell membranes and play an important role in cell membrane integrity, trans-membrane signaling, and triglyceride turnover from the liver and blood [16]. In addition, it serves as a methyl-group donor needed for homocysteine metabolism and DNA methylation, a critical mechanism in epigenetic processes that has been implicated in alcohol teratogenesis. Despite the fact that choline can be produced endogenously, it is classified as an essential nutrient [17, 18], and dietary intake, principally from eggs, liver, wheat germ, and milk, is imperative to meet physiological needs. The demand for choline is especially high during pregnancy, when it is actively transported to the fetus against a concentration gradient [19] and depletes maternal stores [20, 21]. In a recent randomized, double-blind, controlled trial comparing two 3^rd-trimester dietary regimens – 480 mg choline/day (just above the adequate intake level of 450 mg/day [19]) vs. 930 mg/day (just over twice adequate intake level), the higher choline intake arm was associated with faster infant processing speed through age 13 mo [22].

We have recently demonstrated that, although the diet and body composition of heavy drinking pregnant women is similar to that of abstaining/light-drinking pregnant women in socio-economically disadvantaged areas in Cape Town, South Africa, in repeated 24-hr dietary recall interviews, almost 90% of the women reported inadequate intake of choline [23]. High rates of inadequate choline intake during pregnancy have also been reported in the U.S. (90%) [24, 25], Canada (77%) [26], and New Zealand (84%) [27]. These studies emphasize the importance of assessment of dietary choline intake in groups at risk for inadequate intake, such as pregnant women, to inform dietary choline interventions, as well as the use of choline supplements. The use of repeated 24-hour recall interviews to assess usual daily intake of micronutrients has been validated in the U.S. and in resource-poor settings but is both labor and time-intensive [28]. One alternative is to use a dietary intake assessment method that focuses specifically on quantification of choline content using a quantitative nutrient-indicated food frequency questionnaire (QFFQ). This methodology is increasingly applied in assessment of usual intake of single nutrients (e.g., vitamin D [29], iron [30], and vitamin K [31].

We developed a quantitative choline food frequency questionnaire (Choline-QFFQ) to estimate usual dietary choline intake (mg/day) by participants in our Cape Town randomized clinical trial conducted to assess feasibility and efficacy of a maternal choline supplementation intervention conducted with heavy drinking women during pregnancy [14, 15]. The current study combines data from this trial and from our larger prospective longitudinal cohort on the effects of prenatal alcohol exposure on development that included both heavy drinking and abstaining/light-drinking pregnant women from the same community as the women in the trial [8, 32]. Our aims were (1) to develop a Choline-QFFQ that can be used to estimate usual daily intake of dietary choline and (2) to test its reliability and validity in both heavy drinkers and abstainers/light-drinkers.

In this prospective longitudinal cohort of heavy drinking pregnant women and abstainers/light-drinkers, we developed and demonstrated reliability and validity of a QFFQ that assesses average daily choline intake and requires minimal time (~10-15 minutes) and resources (food identification cards and portion estimation props) to administer. To our knowledge, this is the first Choline-QFFQ to be developed and validated for use with pregnant women and the first to be used in the Cape Coloured community in South Africa. While this choline quantitative FFQ presented is specific to the Cape Coloured population, the development and validation steps we employed may be used to develop population-specific choline quantitative screening FFQs in other populations. Furthermore, the Choline-QFFQ performed well in our study population, suggesting that a Choline-QFFQ (and the development and validations steps we employed) can be used even in populations in which education is poor and/or socioeconomic status low. Despite extensive prevention efforts and guidelines (ACOG, 2011) and 4 decades of scientific research demonstrating teratogenic effects of prenatal alcohol exposure, women worldwide continue to drink during pregnancy. FASD comprise the most common preventable cause of neurodevelopmental disabilities, with prevalence estimates of 1.1-5.0% in the US and Western Europe [40, 41] and 13.6-20.9% in South Africa [9]. Evidence from both animal [10, 11, 42, 43] and human [14, 15] studies has demonstrated that high-dose choline supplementation can mitigate many of the teratogenic effects of alcohol. It is thus of concern that over 80% of the both heavy drinkers and abstainers/light-drinkers in this study reported choline intake below the AI in all assessments, as has been reported in other populations such as the U.S. [25], Canada [26], and New Zealand [27]. Given our reliability and validity findings, the QFFQ developed in this study may be used to screen alcohol-using pregnant mothers in Cape Town to identify women who may benefit from choline interventions, whether they be dietary counseling or choline supplementation programs. Given our finding that abstaining/light-drinking women in this cohort had similarly high rates of dietary choline inadequacy, this QFFQ may also be used to identify non-drinking pregnant women at risk for choline deficiency, which poses its own potential harms to the mother and developing fetus, including neural tube and craniofacial defects, as well as possible effects on neurocognitive development [18, 24, 44, 45].

The steps of development and validation we employed can be used to develop population-specific choline quantitative screening FFQs in other populations. Development and validation of a nutrient-specific screening QFFQ involved several steps, including 1) development of the local food list; 2) determining the most appropriate recall period, frequency options and portion size estimation method(s), based on the aim of the dietary intake assessment, the target population, and time available for the administration; 3) identifying an appropriate reference method; and 4) identification of interpretable statistical tests for examining the reliability and validity of the newly developed QFFQ. The food list for our Choline-QFFQ was developed based on the integration of the best known dietary sources of choline and knowledge of commonly consumed foods in the Cape Coloured community in Cape Town by a panel of nutrition researchers with expertise on the dietary patterns of the target population. The recall period was set at the previous week based on recommendations by Willett [28]. Portion size estimation was based on portions used in the FoodFinder program, which was developed for use in this population, and modified based on pilot interviews with women from this community. Administration of the Choline-QFFQ using the picture sort method proved to be simple and quick (~10-15 minutes), which is ideal for implementation as a screening procedure in a community healthcare setting. The dietary reference method used was three 24-hour recall interviews, the method most strongly recommended for validation of food frequency questionnaires [32], especially in communities where literacy and education issues may preclude the use of food diaries, such as this socioeconomically disadvantaged community in Cape Town.

We used the statistical tests and interpretations recommended by Lombard et al. [37] to assess different features of reliability and validity. Lombard recommends three group-level comparison tests (comparison of means using the Wilcoxon signed rank test, percentage difference, and Bland Altman Spearman test for bias) and four individual-level tests (Spearman correlation coefficient, same- and opposite-tertile classification, and the weighted kappa statistic for tertile classification). We added a fourth group-level test, the Bland Altman limit of agreement index (as recommended by Kowalkowska et al. [38]), and a fifth individual-level test, a kappa statistic examining the classification agreement for whether a woman reported intake below the choline AI, to provide a measure of clinical relevance. Group level tests are most relevant for research studies, where errors for a single woman are less important as long as group-level associations are acceptable-to-good, and individual-level tests are most relevant for clinical practice, in which a woman’s reported values will guide targeted recommendations and/or interventions. The majority of the five individual level statistical tests indicated acceptable to good reliability of the QFFQ for all comparisons conducted for both heavy drinkers and abstainers/light-drinkers, except for comparisons of visits 1 and 3 among drinkers, suggesting that reliability may decline over time. At the group level, reliability was good-to-acceptable for 2 of 4 tests, namely percentage difference in reported values between visits (≤ 20% = acceptable) and the Bland Altman limit of agreement index, a measure of how many outliers are present. Reliability was poor for the Wilcoxon signed rank test, a measure of how similarly ranked women are between visits, and the Bland Altman test for bias, which demonstrated a greater difference between reported intake between visits if the reported intake was higher. When examining the validity of the Choline-QFFQ compared with the mean of three 24-hour recalls, all group-level tests were acceptable-to-good, as were all individual level test results, except for classification into tertiles for the alcohol users for all comparisons conducted. Of note, the weighted kappa statistic for tertile classification was acceptable across almost all comparisons. The weighted kappa assesses agreement excluding chance and thus may provide a more accurate measure of agreement than the raw tertile agreement % measures.

Reliability and validity are measured separately, but results of the tests should be interpreted in light of each other. Although the QFFQ demonstrated both reliability and validity, tests of validity were more consistently acceptable-to-good than tests of reliability, particularly at the group level, as reflected in the Wilcoxon and Bland Altman bias tests, and the individual-level, when examining agreement as to whether a woman had inadequate intake. The relatively poorer reliability seen in these tests is likely due to the fact that women reported lower choline intake at each visit. This trend may reflect interview fatigue, as has been reported in several other studies [46]; after familiarization with the procedure, women may have chosen fewer cards from the QFFQ card sorting as a result of fatigue and/or a conscious attempt to shorten the interview. Indeed, reliability was better from visits 1-2 than 2-3, but this difference may also reflect the longer time period between interviews (8 weeks for visits 2-3 vs. 4 weeks for visits 1-2). An actual reduction in food and thus choline intake towards the end of the pregnancy must, however, also be considered. The mean energy intake per day decreased somewhat from each visit to the next (2294 kcal/day at visit 1, 2280 kcal/day at visit 2, 2194 kcal/day visit 3). As repeated administration of the QFFQ did not significantly improve validity the Choline-QFFQ screener could be used as a single-visit screener.

This study had limitations common to other longitudinal studies of nutrition. 24-hr dietary recall interviews can yield inaccurate estimates of usual intake due to recall errors. However, we have previously reported that intraclass correlations and within-subject coefficients of variation for dietary nutrient intakes in this cohort, including choline, were similar to those of NHANES and other peer-reviewed epidemiologic studies in the U.S. [23], indicating that random error in this study did not exceed levels generally accepted in the nutritional epidemiology community. Furthermore, we found that energy intake predicted gestational weight gain, further supporting the validity of the 24-hr recall data. Where possible, the method of triads, in which two dietary assessment methods and a biochemical marker of nutritional status for a given nutrient are compared, is recommended for validation of FFQs [46]. We did not employ this method because free choline and other choline metabolites are tightly regulated and relatively unresponsive to dietary changes, as homeostatic mechanisms, such as estrogen-induced endogenous choline production by the PEMT enzyme during pregnancy act to keep plasma choline values in the normal range; lack of correlation between diet and biochemical values is thus common, particularly in cohorts without high rates of severe choline deficiency [47, 48]. Differences between true and estimated levels of maternal alcohol consumption are likely small, given the validity of the interviewing techniques demonstrated in this community in relation to meconium levels of fatty acid ethyl ester metabolites of alcohol [49], infant and child behavior [7, 35, 50], somatic growth [51], and brain structure [52‐54] and function [55].

To our knowledge, this is the first quantitative choline food frequency screening questionnaire to be developed and validated for use in both heavy and non-drinking pregnant women and the first to be used in the Cape Coloured community in South Africa. Given the high prevalence of inadequate choline intake in this and other communities worldwide and the growing evidence that maternal choline supplementation may benefit infant neurodevelopment and can mitigate adverse effects of prenatal alcohol exposure, this tool may be useful for both research and for future clinical outreach programs aimed at identifying pregnant women at risk for choline deficiency. The Choline-QFFQ development and validation methodology we employed can be used in other communities as well, which may be of significant public health utility given the widespread public health burdens of dietary choline inadequacy and maternal alcohol consumption during pregnancy. Our findings demonstrating validity of the QFFQ for use in this highly disadvantaged, poorly educated population supports its potential utility for use in a broad range of social contexts.