Nutritional Lifestyle Intervention in Obese Pregnant Women, Including Lower Carbohydrate Intake, Is Associated With Increased Maternal Free Fatty Acids, 3-β-Hydroxybutyrate, and Fasting Glucose Concentrations: A Secondary Factorial Analysis of the European Multicenter, Randomized Controlled DALI Lifestyle Intervention Trial

Maternal obesity is associated with higher risk of maternal and fetal complications in pregnancy, such as pregnancy-induced hypertension, preeclampsia, large-for-gestational-age babies (LGA), increased inflammatory processes, and metabolic derangements (1–3). Several randomized controlled trials (RCTs) of lifestyle interventions aimed to decrease the prevalence of gestational diabetes mellitus (GDM) or glycemic parameters among overweight and obese women but reported comparable effects on maternal glucometabolic parameters at the end of pregnancy between intervention and control groups, as well as comparable birth outcomes (4–6). Progressively, the awareness is evolving that next to glucose, other maternal metabolites such as lipids may also strongly influence maternal and fetal development and pregnancy outcome (7,8). Impaired lipid metabolism in pregnancy is associated with a higher risk of maternal and fetal complications, such as GDM, preeclampsia, LGA, and preterm delivery, and with higher fetal fat accretion and neonatal adiposity, a risk factor for childhood obesity (7,8).

Due to hormonal changes, maternal metabolism increases fat storage and lipogenesis at the beginning to middle of pregnancy, whereas in the third trimester, higher rates of lipolysis occur, based on increasing insulin resistance (IR) and lipolytic hormone actions (9–11). Lipolysis is increased in obesity due to higher IR, and increased free fatty acids (FFAs) interfere with insulin action and contribute to increasing IR, facilitating higher glucose levels (7,10,11).

Obese pregnant women have atherogenic lipid profiles, including increased triglyceride (TG), VLDL cholesterol, and FFA and decreased HDL cholesterol (HDL-C) levels as well as higher glucose levels and IR and increased inflammatory parameters, which are all factors that are associated with adverse pregnancy outcomes and increased maternal endothelial dysfunction and cardiovascular disease risk (7,8,12). Maternal TG and FFAs are associated with fetal macrosomia or LGA (1,7,8). Excessive fetal lipid exposure is conjectured to be associated with augmented fetal fat accumulation with higher risk of childhood obesity, fetal hyperinsulinemia, and metabolic disease (7). In addition, a U-shaped association between maternal TG levels in early pregnancy and congenital anomalies and an increased risk of susceptibility of the offspring to atherosclerosis based on epigenetic programming of arterial cells were reported, which might be caused by fetal exposure to increased maternal cholesterol and oxidative byproducts (12,13). Thus, an improvement of lipid profiles through early lifestyle intervention in obese pregnant women might have beneficial effects on maternal and neonatal outcomes.

In this secondary, factorial analysis of a pan-European lifestyle RCT to prevent GDM (6), we aimed to investigate the effect of different lifestyle interventions (healthy eating [HE] and physical activity [PA]) on maternal lipid metabolism from baseline at around week 15 to the end of gestation as well as on fetal lipid metabolism, i.e., in cord blood.

DALI (Vitamin D And Lifestyle Intervention for GDM prevention) was a prospective multicenter RCT comparing different lifestyle approaches that may prevent GDM progression in obese women recruited prior to 20 weeks of gestation, as previously reported (6,14). The study included women (n = 436) from nine European countries aged ≥18 years, singleton pregnancy before 20 weeks of gestation, prepregnancy BMI ≥29 kg/m², and ability to give informed consent. Exclusion criteria included preexisting diabetes, need for complex diets, inability to walk ≥100 m safely, or significant chronic medical conditions. Each local research ethics committee approved the study, and written informed consent was obtained from all participating subjects. DALI was registered in the ISRCTN registry (ISRCTN70595832) and funded by the European Union 7th Framework Programme (grant agreement 242187). Recruitment was conducted between January 2012 and February 2014. Randomization procedures, blinding, and sample size calculations were described elsewhere (6,14). Obese women were randomized to four intervention groups, which were HE (n = 113), PA (n = 110), a combination of both (n = 108), and usual care (n = 105). A 2 × 2 factorial design approach was used for factorial analysis, with generation of groups HE (n = 221) versus no HE (n = 215) and PA (n = 218) versus no PA (n = 218) compiled of the original intervention groups. Supplementary Fig. 1 gives detailed information on group compilation. Information regarding demographics, prepregnancy weight, maternal smoking, and past/current medical obstetric and medication history was obtained by questionnaire.

GDM was diagnosed according to the International Association of the Diabetes and Pregnancy Study Groups/World Health Organization 2013 GDM criteria, and those with GDM at baseline were excluded. Assessments, oral glucose tolerance tests, and blood sampling were performed before 20 weeks, between 24 and 28 weeks, and between 35 and 37 weeks. Venous cord blood was drawn immediately after birth.

Randomized obese women were assigned a lifestyle coach, and each participant received individual sessions with discussion of seven HE and/or five PA messages (6,14,15) (Supplementary Table 1). Techniques inspired by motivational interviewing were used to deliver the HE and/or PA messages (16). HE messages aimed to reduce the intake of sugar/simple carbohydrates in meals and beverages, decrease fat intake, increase protein and fiber intake, and regulate daily calorie intake by watching portion size. PA messages intended to increase PA with incorporation into daily routines by increasing daily activity/steps and reducing sedentary time, improve strength (lower/upper limbs), and increase weekend activity (6,14). In total, five face-to-face sessions and up to four telephone calls or e-mail contacts were conducted. To support messages, a “tool kit” including a participant handbook, educational materials, pedometers (Digiwalker SW-200; Yamax, Tokyo, Japan), and a flexible elastic dynaband (Thera-Band, Akron, OH) was provided.

PA and food intake were self-reported using questionnaires as previously reported (14). In short, PA in pregnancy was assessed by the Pregnancy Physical Activity Questionnaire, which allows an estimate of time spent on PA (i.e., total PA and moderate-vigorous PA [MVPA]) or sedentary behavior (17). In the DALI trial, MVPA assessed by the Pregnancy Physical Activity Questionnaire was significantly increased in the PA group compared with usual care, and sedentary behavior was reduced in the combined HE/PA and HE groups (6). To assess dietary behavior, a 12-item questionnaire was used, on which self-reported frequencies (i.e., days per week) and amounts (i.e., portions per day) of consumed food were specified, that was based upon prior work (18). To obtain a proxy for nutritional components, the total number of consumed food per week of these reported items was calculated for fiber (i.e., summation of vegetable, fruit, and whole-grain bread), protein (i.e., summation of meat/eggs and fish), carbohydrate (i.e., summation of cakes/muffins, fruit, whole-grain bread, fruit juice, nondiet soft drinks, and potatoes/pasta), and fat (i.e., summation of high-fat milk products, cakes/muffins, and fast food). Portion size is the summation of all products that are associated with unfavorable metabolic effects in women with GDM (i.e., cakes/muffins, high-fat milk products, nondiet soft drinks, potatoes/pasta, and fruit juice). Sugar drinks is the summation of fruit juice and nondiet soft drinks. All values are given in total number of items consumed per week.

Blood samples were obtained after an at least 10-h overnight fasting period. Further samples were taken 60 and 120 min after 75-g glucose ingestion. Glucose and insulin analytical procedures have been reported previously (3,6). From fasting samples, total cholesterol (TC), TG, and 3-β-hydroxybutyrate (3BHB) were measured using colorimetric enzymatic assays using reagents from DiaSys Diagnostic Systems (Holzheim, Germany) and were calibrated using secondary standards from Roche Diagnostics (Mannheim, Germany) (TC and TG) and DiaSys Diagnostic Systems (3BHB), respectively. HDL-C was measured with a homogenous assay from DiaSys Diagnostic Systems, and LDL cholesterol (LDL-C) was calculated according to the Friedewald formula (LDL-C = TC − HDL-C − TG/5). Nonesterified fatty acids (FFAs) and 3BHB were analyzed using an enzymatic reagent and standards from Wako Chemicals (Neuss, Germany). All lipid analyses were performed on an Olympus AU640 automatic analyzer (Beckman Coulter, Brea, CA). Leptin was analyzed by solid-phase sandwich ELISA (E05-086-96; EIASON, Graz, Austria). Analytical sensitivity was 1.0 ng/mL; intra- and interassay coefficients of variability (low/high concentrations) were 6.0/6.9% and 11.6/8.7%, respectively.

In venous cord blood, C-peptide was quantified by chemoluminometric solid-phase sandwich immune assay (ADVIA Centaur; Siemens Healthcare Diagnostics, Vienna, Austria). Analytical sensitivity was 0.05 ng/mL and intra- and interassay coefficients of variability (low/ high concentrations) were 3.7/4.1% and 6.1/6.2%, respectively. All assays were performed according to the manufacturer’s instructions.

Continuous variables were summarized by means and SDs and categorical variables by counts and percentages. Assumption of Gaussian distribution of parameters was decided by visual assessment of histograms and calculation of skewness and kurtosis. Skewed data were log transformed before further analysis (threshold ±2). Comparison for continuous variables between intervention groups was performed using Student t test and using χ² test for binary data. Randomized women were compared with excluded women and women who dropped out using Student t test. Correlations were tested using Pearson test.

Comparisons of TG, LDL-C, HDL-C, FFA, and leptin from baseline to periods 24–28 and 35–37 weeks of pregnancy were made between groups using general linear regression models. Models were adjusted for baseline value of the outcome variable only or adjusted for baseline value of the outcome variable, maternal age, BMI at time of examination, gestational week, IR at time of examination (HOMA index), self-reported food intake (portion size), self-reported PA, and maternal smoking at time of examination in a fully adjusted model. Models with weight gain variables were adjusted for maternal BMI at baseline. Dietary components and PA were analyzed after adjustment for baseline levels using general linear models.

Cord blood lipids were analyzed using general linear models, adjusting for gestational age at birth and sex in a simple model and adjusting for gestational age, sex, maternal age, maternal pregestational BMI, maternal baseline lipid levels, and IR (maternal HOMA index and fetal glucose/C-peptide ratio) in a fully adjusted model.

Statistical analysis was performed using SPSS 25.0 (SPSS Inc., Chicago, IL) and GraphPad Prism 7 (GraphPad Software, La Jolla, CA). A two-sided P value <0.05 was considered statistically significant, except for lipid parameters and leptin, in which we used an explorative Simes procedure to adjust for multiple testing. The level of discharge was calculated as P > 0.01.

Baseline characteristics of randomized, excluded women and dropout rates are shown in Table 1. Of 639 women assessed for inclusion, 168 (26.3%) were excluded due to GDM diagnosis before 20 weeks of gestation. Women excluded before randomization had significantly higher levels of prepregnancy weight, prepregnancy and baseline BMIs, waist and neck circumferences, diastolic blood pressure, TG, glucose and insulin levels during the oral glucose tolerance test, and IR. They were more likely to have a history of GDM in a previous pregnancy or previous macrosomia. Women who dropped out after randomization had lower levels of education, more often a history of GDM, lower LDL-C and HDL-C levels, and higher waist circumferences at baseline. No significant differences between intervention and nonintervention groups were found except for baseline 3BHB (P < 0.05), which was increased in HE versus no HE. The study population was mainly of European descent (n = 378 of 436, 86.7%) and living with a partner (n = 410 of 436, 94.0%). Diabetes in a first-degree relative was reported by 23.2% (n = 101 of 436).

Descriptive data of metabolic parameters at 24–28 and 35–37 weeks of gestation as well as birth and cord blood outcomes are shown in Table 2, and data on self-reported dietary intake and weight gain are shown in Table 3. At 24–28 weeks of gestation, significantly lower weight gain and significant higher FFA, 3BHB, and fasting glucose were found in HE versus no HE. At 35–37 weeks of gestation, the HE group had significantly lower weight gain, lower neck circumferences, and higher FFA levels. No differences were found in PA versus no PA groups. In HE, significant improvements in fat intake and portion sizes at 24–28 weeks and carbohydrate, sugary drink, and fat intake and portion sizes at 35–37 weeks of gestation compared with no HE were found after adjustment for baseline levels. Higher intake of sugary drinks at 24–28 weeks and lower fiber intake at 35–37 weeks of gestation were reported in PA versus no PA.

In simple regression models, adjusted for baseline levels, as well as in the fully adjusted models, significant increased FFA levels were found at 24–28 (0.086 [95% CI 0.036; 0.137], P = 0.001) and 35–37 (0.087 [0.015; 0.160], P < 0.02) weeks of gestation in HE versus no HE (Table 4). In PA versus no PA, a significant difference was found in the simple model with lower HDL-C (P < 0.02) at 35–37 weeks of gestation in PA. After adjustment for multiple testing, significant differences (P < 0.01) were only found in FFA levels at 24–28 weeks of gestation in HE versus no HE.

A significant positive correlation of FFA and 3BHB (r = 0.46, P < 0.001) at 24–28 weeks of gestation was found and no significant correlation between FFA and fasting glucose (r = −0.05, P = NS). Weak negative correlations were found between total carbohydrate intake at 24–28 weeks of gestation and FFA (r = −0.12, P < 0.03), 3BHB (r = −0.13, P < 0.02), and fasting glucose (r = −0.11, P < 0.03) at that time, as well as weak significant correlations between weight gain from baseline to 24–28 weeks of gestation and FFA (r = −0.14, P < 0.01) and 3BHB (r = −0.14, P < 0.01). No correlations were found for other nutritional parameters or PA parameters and FFA, 3BHB, and fasting glucose.

At birth, no differences were found in birth weight, LGA and small for gestational age, or gestational week of birth in HE and PA groups compared with no HE/no PA (Table 2). In HE versus no HE, a trend for higher mean differences in cord blood FFA was found (P = 0.057) in a simple statistical model adjusting for gestational week and fetal sex, which became statistically significant (P = 0.009) in a fully adjusted model (Table 4) and was still significant after correction for multiple testing (P < 0.01). Differences in neonatal cord blood C-peptide levels (0.69 ± 0.47 vs. 0.66 ± 0.46 ng/mL, P = NS) between HE and no HE were not found. No significant changes in cord blood lipids or leptin were found in PA versus no PA.

At 24–28 weeks of gestation, significant lower weight gain and increases in FFA, 3BHB, and fasting glucose levels in the HE group compared with no HE were found in this secondary factorial analysis. Negative associations between carbohydrate intake and FFA, 3BHB, and fasting glucose were found. In cord blood, significantly higher FFA levels in the HE versus no HE group were detected. No differences in other lipids or in the PA versus no PA groups were identified.

Recently, we reported a significant weight gain reduction (−2 kg) in the combined HE/PA lifestyle group of the DALI trial compared with the usual care group, but we were not able to find improved metabolic control. We identified increased fasting glucose levels in the HE group versus usual care, which we previously interpreted as a random finding (6). However, now we have an alternative interpretation of this finding. The dietary changes that included decreasing portion size and a reduction in carbohydrate and fat intake, followed by reduced weight gain in pregnancy, might have increased FFA and 3BHB levels in the HE group due to induction of lipolysis in adipocytes of white adipose tissue. This could also have affected IR and gluconeogenesis and, subsequently, increased fasting glucose (9,10,19). Nonetheless, for our analysis, the groups were redistributed, which necessarily impacts on the interpretation of increased fasting glucose in the HE group.

We found weak but significant negative associations of total self-reported carbohydrate intake with FFA, 3BHB, and fasting glucose to support this hypothesis. Moreover, the negative correlations of weight gain from baseline to 24–28 gestational weeks with FFA and 3BHB partly corroborate these findings. A study comparing complex-carbohydrate/low-fat diet with low-carbohydrate/high-fat diet (carbohydrate/fat/protein 60/25/15% vs. 40/45/15%) also found increased fasting glucose levels in the latter and decreased FFA, fasting glucose levels, and lipolysis in adipose tissue and IR in the first group (20). This study had a small sample size and reflects only short-term changes in the last trimester of obese women affected by GDM but demonstrates an association of a low-carbohydrate diet with higher fasting glucose and FFA levels based on increased lipolysis. An older pilot study partly corroborates these findings; after 4 days on a low-carbohydrate (carbohydrate/fat/protein 35/45/20% and 70 g fiber vs. 70/10/20% and 31 g fiber) diet, women with GDM had increased fasting FFA and IR but no changes in fasting glucose compared with women on a low-fat/high–unrefined carbohydrate diet (21). Ketone bodies were not reported (20,21).

Therefore, it seems our dietary intervention may have paradoxically worked against the goal of preventing GDM by inducing lipolysis. These findings highlight the need of further research on the possible induction of lipolysis by dietary interventions in pregnancy.

Prior studies have not reported differences in FFA, other lipids, or leptin in overweight or obese pregnant women between lifestyle intervention and control groups (4,22–24). Differences between our results and other trials may be due to the lesser reduction of gestational weight gain in these studies and therefore putatively a lesser degree of lipolysis. The UK Pregnancies Better Eating and Activity Trial (UPBEAT) reported reduced glycemic index in the intervention group as well as decreased total energy, carbohydrate, and saturated and total fat and increased protein and fiber intake (4), and LIMIT (Limiting weight gain in overweight and obese women during pregnancy to improve health outcomes: a randomised trial) found increased daily fruit and vegetable consumption and fiber intake as well as a reduction of saturated fat intake after lifestyle advice independent of BMI (5), whereas the other trials did not focus their analysis on nutrition and fat intake (24) or report on diet at all (23). A study investigating the effect of lower fat intake by consuming less saturated and total fat and cholesterol found decreased TC, HDL-C, and LDL-C levels at 36 weeks of gestation in the intervention group (25). Another study comparing low glycemic index versus low fat intake found a significant lower increase of cholesterol and TG levels between baseline and 36 weeks gestation in the low glycemic index group (26).

Recent reports state that a reduction in carbohydrate intake also creates a risk of an unbalanced increase in fat intake, which could increase FFA levels and consequently induce IR and increase maternal glucose levels (7,19). However, in our study, we did not observe an increased, but rather a significantly decreased, self-reported fat intake in HE versus no HE at 24–28 weeks of gestation, which was not associated with any of the maternal lipid parameters, fasting glucose, or gestational weight gain (data not shown). The fiber intake patterns do not suggest higher or lower diet quality in one of the groups.

Studies performed in healthy nonobese pregnant women (mean BMI <25 kg/m²) show contrasting findings compared with those studying obese women. Significant decreases of TC and LDL-C throughout pregnancy after dietary intervention starting at 18 weeks of gestation compared with usual diet were reported (25). Contrary to our results, women performing recreational PA in early pregnancy had significantly lower plasma TG levels than inactive women (27). A linear correlation was found showing lower concentrations of TG and cholesterol across groups as activity intensity increased. Another trial investigating the effects of aerobic exercise in the second half of pregnancy in healthy women (mean BMI = 25.5 kg/m²) observed a trend toward lower maternal FFA levels and significant increases in leptin levels compared with control subjects (28). We did not observe any differences in lipids in PA, which might be due to low PA intensity in our obese women.

FFAs are important for fetal development, and increased maternal FFAs are associated with adverse pregnancy and fetal outcomes (7,19). Literature reports that FFAs are not transferred across the placenta to a large extent (9). It is speculated that the maternal metabolic constitution directly affects placental physiology and that placental function and dysfunction are set or programmed very early in pregnancy (29). Thus, the initiation of lifestyle interven-tion close to 20 weeks of gestation might be too late to affect maternal and neonatal metabolism significantly. Lifestyle intervention and reduced weight gain might act on placenta function, and recently, potential beneficial changes by intervention in placental lipid droplet composition were reported (30). However, in the control group (without lifestyle intervention), higher dihomo-ɣ-linolenic acid (DGLA) content was detected. Interestingly, DGLA content in placental lipid droplets was associated with gestational weight gain, and prior studies reported not only associations between higher maternal DGLA levels during pregnancy and maternal insulin resistance and obesity but also positive associations with adiposity measures in early childhood (30–32). Thus, polyunsaturated fatty acids and especially DGLA may play a central role in the development of the offspring. Placentas of obese pregnant women store larger amounts of lipid caused by increased esterification, which might limit nutrient transport to the fetus (33) and even causes an upregulation of placental genes involved in lipid storage and transportation (34). A positive association of neonatal birth weight and neonatal adiposity with placental lipoprotein lipase (pLPL) activity, which hydrolyzes TG into FFAs and facilitates transplacental fatty acid transfer, was found recently (35). Differences in pLPL activity influenced by maternal metabolism might act as a relevant placental mechanism involved in fetal fat accumulation, which highlights opportunities for future studies (35).

However, we are not able to explain higher cord blood FFA levels and can only speculate that higher FFA is either a reflection of higher IR in fetal tissue or a response to higher catecholamines due to increased fetal stress during delivery. However, we did not find differences in cord blood C-peptide levels or area under the curve glucose/C-peptide ratio (data not shown) between HE and no HE, which are difficult to interpret anyway in a nonfasted state at time of delivery. Nonetheless, we could not find differences in cord blood FFA levels of PA versus no PA groups either, which also were exposed to the fetal stress at delivery. These interesting but unclear speculations require further studies.

Sample size calculation was based on the primary outcome, and no formal calculation of statistical power for this secondary analysis was performed. The original DALI design was as a 2 × 2 factorial trial, and interaction was shown in the analyses for the primary outcome weight gain. Hence, each individual cell was reported in the primary analysis. For the lipid analyses, we have continued with the factorial analysis. A positive preselection might have occurred. Many pregnant women willing to change their lifestyle wanted to participate. This positive spirit of change might also have affected the usual care group (Hawthorne effect), limiting the effects of the lifestyle intervention groups. Decreased participation of subjects at high risk due to intervention-related response bias needs to be considered. Furthermore, we are not able to report differences in specific FFAs or changes in placenta physiology. Food intake and PA were self-reported, which can result in under- or overreporting due to social desirability and recall bias. Additionally, our food questionnaire does not allow the calculation of detailed information of macronutrient content but only gives information about servings per week. We are not able to quantify the degree of carbohydrate restriction in numbers (grams/day) due to the design of our study. Our study aimed to deliver lifestyle intervention messages (Supplementary Table 1) built upon behavioral theory principles (16) and included patient empowerment and cognitive behavioral techniques inspired by motivational interviewing (14,15) but was not based on a predefined macronutrient distribution or strategy to install a low-carbohydrate diet. Type of fat is important for glucose homeostasis, and in our analysis we have not addressed it. DALI included only obese women; thus, no normal-weight control group exists, and the effects seen in our study do not imply emergence in normal-weight pregnancies, which needs to be tested. The strengths of this study are that women with hyperglycemia in early pregnancy were excluded and we collected data on dietary intake, different than many prior studies performed.

The HE intervention, with significantly lower gestational weight gain, resulted in increased maternal FFA and 3BHB levels, which suggests stimulation of maternal lipolysis by the intervention. This might have induced the increased fasting blood glucose and may have worked against the goal of preventing GDM. Even neonatal FFAs in cord blood were significantly higher after HE intervention. However, the consequences of these changes are unknown. In pregnancy, HE compared with PA intervention is a powerful tool to impact maternal and fetal metabolism. This supports healthy nutrition for all pregnant women, particularly for pregnancies at risk for fetal overgrowth. However, further studies are needed to test our hypothesis and to investigate the long-term consequences in obese women and their offspring.

Acknowledgments. The authors thank the coaches, research midwives/nurses, and health professionals collaborating in the recruitment. The authors thank all women, their babies, and families participating in the DALI trials. Finally, the authors thank Andre von Assche (Development and Regeneration, KU Leuven, University Leuven, Leuven, Belgium) wholeheartedly, who contributed to the conception, design, and execution of DALI but sadly died before submission. Rest in peace.

Funding. The project described has received funding from the European Community’s 7th Framework Programme (FP7/2007-2013) under grant agreement 242187. In the Netherlands, additional funding was provided by the Netherlands Organisation for Health Research and Development (ZonMw) (grant 200310013). In the U.K., the DALI team acknowledges the support received from the National Institute for Health Research Clinical Research Network: Eastern, especially the local diabetes clinical and research teams based in Cambridge. In Spain, additional funding was provided by CAIBER 1527-B-226.

The funders had no role in any aspect of the study beyond funding.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.H. wrote the manuscript, performed the statistical analyses, and edited and finalized the manuscript. J.H., D.S., G.D., R.C., J.M.A., R.D., S.G., P.D., E.R.M., D.M.J., L.L.T.A., F.D., A.L., M.G.D., A.B., E.W.-O., A.Z., U.M., D.H., J.G.M.J., F.J.S., M.L., C.L., C.W., D.B.-T., H.S., M.N.M.v.P., and A.K.-W. read and corrected draft versions of the manuscript and approved the final manuscript. J.H., S.G., D.M.J., L.L.T.A., M.G.D., A.B., A.Z., U.M., J.G.M.J., M.L., C.L., C.W., D.B.-T., and H.S. executed DALI and provided input for the interpretation of the results. D.S. was the study coordinator. D.S., G.D., R.C., J.M.A., R.D., P.D., E.R.M., F.D., A.L., E.W.-O., D.H., F.J.S., M.N.M.v.P., and A.K.-W. designed and executed the DALI study and provided input for the interpretation of the results. G.D. was the principal investigator of the study. M.N.M.v.P. was the sponsor of the study. J.H. and A.K.-W. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.