Maternal dietary patterns and preterm delivery: results from large prospective cohort study

Dietary information

Data from the Norwegian Mother and Child Cohort Study food frequency questionnaire (www.fhi.no/dokumenter/011fbd699d.pdf) were collected from February 2002 onwards. The study food frequency questionnaire is a validated semi-quantitative questionnaire designed to compile information on dietary habits during the first four to five months of pregnancy. ^{29 30} The food frequency questionnaire was validated in a subsample of cohort participants, showing that relative to a dietary reference method and several biological markers, it produces a realistic estimate of the habitual diet and is a valid tool for ranking pregnant women according to high and low intakes of energy, nutrients, and foods.^{31 32} The food frequency questionnaire was read optically. We converted food frequencies into daily intakes (g/day) and used FoodCalc to calculate nutrient and energy intakes.³³ The food frequency questionnaire covered 255 food and beverage items. We compiled these items into 58 non-overlapping food groups, based on nutrients, common characteristics, or culinary use, as described in more detail previously.²⁴ Dietary energy density is the ratio of energy (kcal) to weight (g); this ratio remains constant regardless of the amount consumed. We calculated a variable denoting energy density by dividing each woman’s daily energy intake by the weight of all foods consumed, excluding all beverages.³⁴

Extraction of dietary patterns

We used principal component analysis to extract dietary patterns. This is a data driven technique that reduces the dimension of the data and groups correlated variables, in this case the food intake variables, to form new linear factors or components. The components (dietary patterns) derived by principal component analysis reflect the combinations of foods consumed by individual participants. The coefficients defining the components are called factor loadings and describe the correlation between each food variable and the components. (table 1​1)) As described above, we aggregated food intakes from the food frequency questionnaire into 58 non-overlapping food groups and standardised them to reduce the influence of food intakes with large variances at the expense of those with minor variances. The number of components retained was based on interpretation of the factor loading matrix after orthogonal (varimax) rotation and a scree plot (fig 2​2)) that shows the proportion of the variance in total consumption of the food variables.¹⁸ We tested the analysis with the Bartlett test of sphericity and Kaiser-Mayer-Olkin tests. We considered food and drink items with loadings of 0.25 or higher on a factor to be important for the interpretability of each pattern.

Open in a separate window

Fig 2 Scree plot for identification of dietary patterns (components) by principal component analysis. Food intakes (g/day) were aggregated into 58 food groups and used as input variables. Factors considered appropriate for patterns shown in table 1​1 are the three factors with eigenvalues >2

We named the new linear components after the nature of the input variables with the highest factor loadings. We assigned participants factor scores for each of the dietary patterns (dietary patterns scores). We derived these factor scores by multiplying factor loadings by the corresponding standardised intake value of each food item and adding all these items. The mean factor score for each pattern is zero. Positive factor scores indicate higher consumption of foods and drinks in that pattern, and negative factor scores indicate low consumption. We divided each extracted dietary pattern into thirds, with the lowest third as the reference.

Preterm delivery

The primary outcome was preterm delivery, defined as delivery between gestational week 22+0 and 36+6. We calculated gestational age on the basis of ultrasound measurements at 17-18 weeks of gestation in 64 846 cases. In 1154 cases, information about gestational length from ultrasound was missing, so we calculated it from last menstruation period. Information about gestational age at birth came from the Medical Birth Registry of Norway.²⁸ We divided preterm delivery into spontaneous or iatrogenic, defining iatrogenic preterm delivery as induced or caesarean delivery, on maternal or fetal indications. We also divided preterm delivery into late (34+0 to 36+6 weeks), moderately (32+0 to 33+6 weeks), and early preterm (22+0 to 31+6 weeks).

Covariates

We selected 10 covariates for their known association with preterm delivery. Information about maternal age and previous preterm delivery came from the Medical Birth Registry of Norway. We treated maternal age as a continuous variable, except in the case of descriptive statistics, for which we divided it into four groups (<20 years, 20-29 years, 30-39 years, and ≥40 years). We analysed history of previous preterm delivery as a dichotomous variable (yes/no). In questionnaire 1, women reported their pre-pregnancy weight and height, from which we calculated body mass index. We included only women who reported weight in the range 35-180 kg and height above 1.40 m. We used body mass index as a categorical variable, divided into four categories (<18.5, 18.5-24.9, 25-29.9, and ≥30). We analysed height as categorical data and divided into thirds (<1.66 m, 1.66-1.70 m, and ≥1.71 m). Information about marital status came from questionnaire 1; we categorised it as either living alone or cohabiting. Information about parity came from questionnaire 1 and from the Medical Birth Registry of Norway. We divided parity into two categories (nulliparous or parous). We categorised smoking as non-smoker, occasional smoker, or daily smoker on the basis of women’s answers in questionnaire 1. We divided maternal education into three categories (<13 years, 13-16 years, and ≥17 years), regardless of the kind of education. We used total energy intake (kJ) as a continuous variable but present it as quarters in table 2​2.. We divided household income into three categories (both partners <300 000 Norwegian Krone (£29 648; €35 660; $48 406) per year, participant or partner ≥300 000 Norwegian Krone per year, and both partners ≥300 000 Norwegian Krone per year) on the basis of information provided in questionnaire 1 and analysed it as categorical data.

We also ran analyses for additional confounders. We analysed physical activity during pregnancy as categorical data divided into one of four categories (none, less than weekly, 1-2 times weekly, and ≥3 times weekly), as reported at gestational week 17. We analysed information from the food frequency questionnaire about nausea and vomiting during pregnancy as dichotomous data (yes/no) and energy density calculated from the food frequency questionnaire as continuous data. We used alcohol intake as both categorical and continuous data (g/day) and collected information from the food frequency questionnaire. We analysed passive smoking as dichotomous data (yes/no) from questionnaire 1 and used information about planned pregnancy as dichotomous data (yes/no) collected from questionnaire 1.

Dietary patterns in relation to maternal characteristics

The distribution of mean dietary pattern scores differed according to maternal characteristics (table 2​2).). The prudent pattern score increased with increasing maternal age and education and was lower in women with higher body mass index. Furthermore, the prudent pattern score was slightly higher in nulliparous women, in non-smokers, and in women with no previous preterm delivery. The Western pattern score was higher in younger women, in smokers, in women with less education, in parous women, and in those with previous preterm delivery. Both underweight and overweight women had higher scores on the Western pattern than did normal weight women. For the traditional pattern, mean factor scores were higher in the young (<20 years) and the older (≥40 years) age groups than in the intermediate age groups. The traditional pattern scores also increased with decreasing education and were higher in parous women, in smokers, and in women with a history of preterm delivery. The mean factor scores for all three patterns increased with increasing energy intake.

Deliveries

Among the 66 000 pregnant women, preterm delivery occurred in 3505 (5.3%) cases, of which 2003 (3.1%) were spontaneous and 1414 (2.2%) were iatrogenic. In 88 (0.13%) cases, information about start of delivery was missing. Late preterm delivery occurred in 2558 (3.9%) of total deliveries, 478 (0.7%) were moderately preterm, and 469 (0.7%) were early preterm.

Strengths

The major strengths of this study are the large number of participants from all over Norway, the prospective design, and the detailed information on a wide range of potential confounding factors. All age and socioeconomic groups are represented in the study group. Women were also unaware of the outcome of pregnancy when they filled in the food frequency questionnaire, so outcome could not affect the reporting. In the food frequency questionnaire, women were asked to report intake since the beginning of pregnancy. The food frequency questionnaire covers the period from becoming pregnant until gestational week 22, meaning that responses covered embryogenesis, a period that is important when it comes to epigenetic mechanisms and for possible later preterm delivery.³⁶ The food frequency questionnaire used in our study has been extensively validated and shown to be a valuable tool for ranking pregnant women according to high and low intakes of energy, nutrients, and foods.^{29 32 37} One of the most important risk factor for preterm delivery is previous preterm delivery,³⁷ which we adjusted for in this study. Smoking also increases the risk of preterm delivery,³⁸ and the fact that smoking is commonly underreported, especially in pregnant women,³⁹ constitutes a major challenge in all epidemiological studies. In this study, we divided smoking into non-smoking, occasional smoking, and daily smoking. This categorisation of the smoking variable has been validated in a subsample of 2997 women in the Norwegian Mother and Child Cohort Study with plasma cotinine as an objective reference, showing a high degree of agreement between plasma cotinine and self reported smoking during pregnancy.⁴⁰ Our findings are likely to be generalisable outside of Norway and thus contribute to the general body of research on diet and health. Studies do not show that a typical Nordic diet, either in the general population or in pregnant women, contain more “prudent” foods than elsewhere.^{41 42}

Limitations

Our study also has limitations. The results are observational, and no causal inference is possible. Despite careful consideration of known risk factors and potential confounding factors, residual confounding cannot be ruled out. The low participation rate (40%) in the Norwegian Mother and Child Cohort Study is a concern. Women in the study are older, better educated, and less often smokers compared with the general pregnant population, which might have affected the outcome in our study. However, the incidence of preterm delivery in the general population is comparable to that in the study cohort.²⁷ Furthermore, a study focusing on the potential self selection bias in the Norwegian Mother and Child Cohort showed no significant differences between eight evaluated exposure-outcome associations (of which preterm delivery was one) in the cohort and the total pregnant population in Norway during the same period.⁴³ Bias might also have arisen in the exclusion process, but the prevalence of preterm delivery was actually higher in the study population than in the 17 386 excluded pregnancies (5.3% (3505 cases) v 3.8% (660 cases)) owing to exclusion of women participating in the study with more than one pregnancy, resulting in a higher proportion of nulliparous women in the study sample.

Possible explanations for findings

The association between the prudent dietary pattern and a reduced risk of preterm delivery was stronger in nulliparous women than for the whole group. Women who were pregnant with their first child had significantly higher scores on the prudent dietary pattern than did the multiparous women. This difference may be explained by nulliparous women being more health conscientious than parous women or by diet being of less importance in later pregnancies than in the first, as the overall risk of preterm delivery is significantly lower in women who have experienced a previous pregnancy.⁴⁴

Furthermore, after stratification according to time of delivery, we found only a significantly reduced risk of late preterm delivery related to high adherence to the prudent dietary pattern. Although results point in the same direction, no significant associations were found between this pattern and moderately or early preterm delivery. This might be due to the low number of cases in these subgroups, or more serious mechanisms might underlie early and moderately preterm delivery—for example, amniotic infections that a prudent dietary pattern cannot prevent. An alternative explanation is that the prudent dietary pattern might reduce the progression to preterm delivery only marginally, in the order of a few days, and that the effect is therefore most easily detectable in late preterm delivery, but this is only speculation.

We did not find any independent association between high scores on the Western pattern and preterm delivery in this study. This pattern explained only 5% of the variance in total food intakes. A low score on the prudent pattern is interpreted as an unwholesome diet, as the dietary pattern scores are negative in this group (see tables 4​4,, 5​5,, and 6​6).). Foods perceived to be unhealthy tend to be underreported to a larger degree than foods regarded as healthy.⁴⁵ The interplay between dietary behaviour, body mass, and health outcome is complex, and the effects are difficult to disentangle. Other studies have shown the importance of limiting the intake of food items typical of the Western pattern.^{46 47 48} In a previous study in the same study population, we found that high intakes of both sugar sweetened and artificially sweetened drinks were associated with an increased risk of preterm delivery.¹³ Sugar sweetened drinks loaded positively on the Western pattern. Although no independent association with preterm delivery was observed for this pattern, we cannot rule out the possibility that high consumption of single items in a pattern may be associated with the outcome. However, as found in a previous study on healthy food intake and mortality in women,⁴⁹ the overall findings indicate that focusing on regular consumption of healthy food is more important than focusing on reducing consumption of unhealthy food.

Delivery starts as a result of a complex paracrine and autocrine dynamic, biochemical event in involving the hypothalamic-pituitary-adrenal axis, resulting in increased adrenal cortisol production and subsequent inflammation.^{32 50} Altering the hypothalamic-pituitary-adrenal axis is one way that food can influence gestational length. Animal studies have found that a high fat diet can act as a stressor in the hypothalamic-pituitary-adrenal system, increasing production of glucocorticosteroids.⁵¹ Whether high adherence to a prudent pattern can have an anti-stressor effect on the hypothalamic-pituitary-adrenal axis is as yet unknown, but it is an interesting hypothesis. A diet characterised by high intake of vegetables, fruit, and berries and rich in antioxidants and vitamins can reduce both systemic and local inflammation (through gut flora),⁵² which might also be one explanation for the reduced risk of preterm delivery associated with adhering to a prudent diet.^{53 54 55 56 57} This dietary pattern is also low in saturated fat, which is also associated with reduced inflammation.⁵⁸

Comparison with other studies

To the best of our knowledge, no previous study has investigated dietary patterns in relation to preterm delivery by principal component analysis. Most other studies have focused on specific foods or nutrients in relation to preterm delivery, but distinguishing the foods in a mixed diet that are responsible for the effect can be difficult. Therefore, studying overall dietary patterns in relation to outcomes might be an important alternative approach.

Some food items found to be associated with the risk of preterm delivery were part of the food pattern identified by principal component analysis. Milk and yoghurt products enriched with probiotic Lactobacillus (probiotic milk products) have been reported to be associated with a reduced risk of preterm delivery,^{15 59} and probiotic milk products had positive loading on the prudent dietary pattern in our study. Also, two large cohort studies found aspects of Mediterranean-type dietary patterns, characterised by high intake of fruit, vegetables, olive oil, and fish, to be associated with a reduced risk of preterm delivery.^{11 60} We found these food items to be positively loaded on the prudent pattern, associated with reduced risk of preterm delivery. A prudent diet comparable to ours has been found in studies to be associated with other beneficial obstetric outcomes. Brantsaeter and co-workers found that a similar dietary pattern was associated with reduced risk of pre-eclampsia in nulliparous women.²⁴ In addition to preterm delivery and pre-eclampsia, an overall healthy diet is also beneficial for other obstetric outcomes, such as to prevent excessive weight gain and reduce the incidence of gestational diabetes, especially in overweight women.^{61 62}

The reduced risk of preterm delivery related to high scores on the prudent dietary pattern was apparent only in the normal weight group. Different findings in normal weight and overweight women were also observed in a study by Magnus et al, in which physical activity was associated with a reduced risk of pre-eclampsia in normal weight, but not in obese (body mass index ≥30), women.⁶³ The negative effect of overweight may have been much greater than the positive effect of a prudent diet in our study. Overweight is a well known risk factor for preterm delivery.^{64 65} In our study, pre-pregnancy overweight and obese women had significantly lower scores on the prudent dietary pattern than did normal weight women (table 2​2)) and might therefore not have benefited from this pattern. Also, overweight pregnant women tend to underreport dietary intake to a larger degree, which may be another explanation for the results.⁶⁶