Edited by: Elena Succurro, University of Magna Graecia, Italy
Reviewed by: Cristina Bianchi, Azienda Ospedaliero-Universitaria Pisana, Italy; Maria Grazia Dalfra’, University of Padua, Italy
*Correspondence: Camilla Festa,
This article was submitted to Obesity, a section of the journal Frontiers in Endocrinology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
The primary aim of this study was to assess insulin requirements and carbohydrate to insulin ratio (CHO/IR) in normal weight, overweight, and obese pregnant women with type 1 diabetes across early, middle, and late pregnancy.
In this multicenter, retrospective, observational study we evaluated 86 of 101 pregnant Caucasian women with type 1 diabetes under pump treatment. The women were trained to calculate CHO/IR daily by dividing CHO grams of every single meal by insulin units injected. Since the purpose of the study was to identify the CHO/IR able to reach the glycemic target, we only selected the CHO/IR obtained when glycemic values were at target. Statistics: SPSS 20.
We studied 45 normal weight, 31 overweight, and 10 obese women. Insulin requirements increased throughout pregnancy (p < 0.0001 and <0.001 respectively) in the normal and overweight women, while it remained unchanged in the obese women. Insulin requirements were different between groups when expressed as an absolute value, but not when adjusted for body weight. Breakfast CHO/IR decreased progressively throughout pregnancy in the normal weight women, from 13.3 (9.8–6.7) at the first stage of pregnancy to 6.2 (3.8–8.6) (p = 0.01) at the end stage, and in the overweight women from 8.5 (7.1–12.6) to 5.2 (4.0–8.1) (p = 0.001), while in the obese women it remained stable, moving from 6.0 (5.0–7.9) to 5.1 (4.1–7.4) (p = 0.7). Likewise, lunch and dinner CHO/IR decreased in the normal weight and overweight women (p < 0.03) and not in the obese women. The obese women gained less weight than the others, especially in early pregnancy when they even lost a median of 1.25 (−1 −1.1) kg (p = 0.005). In early pregnancy, we found a correlation between pregestational BMI and insulin requirements (IU/day) or CHO/IR at each meal (p < 0.001 and p = 0.001, respectively). In late pregnancy, a relationship between pre-gestational BMI and CHO/IR change was found (P = 0.004), as well as between weight gain and CHO/IR change (p=0.02). The significance was lost when both variables were included in the multiple regression analysis. There was no difference in pregnancy outcomes except for a higher pre-term delivery rate in the obese women.
Pre-gestational BMI and weight gain may play a role in determining CHO/IR during pregnancy in women with type 1 diabetes under pump treatment.
Pregnant type 1 diabetic women have a high rate of obstetric and fetal complications such as preeclampsia, stillbirths, neonatal mortality, congenital malformations, and neonatal morbidity (
A poor glycemic control during pregnancy, along with an inadequate pre-conceptional care, increases maternal–fetal complications (
Carbohydrate counting and carbohydrate-to-insulin ratio (CHO/IR) are a valuable tool in the management of type 1 diabetes in improving glycemic control and flexibility in eating habits (
Our previous observations confirmed a progressive CHO/IR decline over time at each meal in women with type 1 diabetes under continuous subcutaneous insulin infusion (CSII) therapy during pregnancy (
In the last decades, all countries have been experiencing a dramatic rise in obesity due to several environmental factors, also affecting type 1 diabetic patients from childhood (
CSII use in pregnancy is still debated, and results obtained from the largest European RCT (
Our hypothesis is that pre-gestational BMI class and weight gain may influence the CHO/IR trend throughout pregnancy. A deeper understanding of the role of these two variables may contribute to improving the accuracy of algorithms used as prediction models in new technologies.
The aim of this retrospective study was to assess insulin requirements and CHO/IR in well-trained normal weight, overweight and obese pregnant women with type 1 diabetes across early, middle, and late pregnancy.
From 2006 to 2012, 101 pregnant Caucasian women with type 1 diabetes were followed in four Italian centers dedicated to the management of diabetes in pregnancy (
HbA1c was checked at the first visit and monitored monthly, with the purpose of achieving values <6.0% (42 mmol/mol).
At the first visit all patients were trained on carbohydrate counting and on how to achieve and maintain strict metabolic control, with monthly refresher training sessions. Moreover, they received dietary recommendations with special reference to caloric intake and glycemic index, according to Italian Recommendations (
The diet provided 45–50% carbohydrates, 20% proteins, and 30–35% fats, divided between breakfast (10–15%), lunch (20–30%), dinner (30–40%) and three snacks (5–10%; mid-morning, mid-afternoon, and before bedtime). A caloric surplus was assigned in the second (340 kcal) and third (450 kcal) trimester of pregnancy to ensure adequate energy reserves and normal fetal growth (
All women reported GAD65 Ab titer positivity at the diabetes diagnosis.
Twenty-eight patients (32.6%) switched to CSII during pregnancy (at 11 ± 2.8 weeks of gestation), and the remaining patients used CSII therapy before conception.
Paradigm REAL-Time or Paradigm VEO (Medtronic Inc); Animas (West Chester, PA, IR 1200/2020) and ACCU-Chek Spirit (Roche Diagnostics) insulin pumps were used and only short-acting insulin analogs (lispro/aspart) were adopted.
The women were divided into three sub-groups according to their pre-gestational BMI (normal weight, overweight, and obese).
Insulin requirements were expressed both as total I.U. per day and I.U. per body weight kg, as it currently happens in clinical practice/trials, giving information in women with different pregestational BMI and recommended weight gain.
The women measured CHO/IR daily at each meal by dividing the grams of CHO in the meal by insulin units needed. As the purpose of our study was to identify the CHO/IR able to reach the glycemic goals, we considered CHO/IR when the glycemic values were at target only.
In particular, the measured CHO/IRs were included in the analysis when the Fasting Capillary Blood Glucose (FCBG) ranged from 70 to 90 (acceptable up to 100 mg/dl), 1-h post-prandial BG values ranged from 100 to 130 (acceptable up to 140 mg/dl) and when CHO grams and insulin dose were correctly reported and congruent with medical prescription. During follow-up visits, the CHO/IR setting was verified and updated if necessary.
CHO/IR values were collected from patients’ diaries only. We could not use data from management system downloads as at that time only 26.7% of the women used sensors.
We only considered blood glucose levels written in the diaries and verified on a glucose meter.
The difference between CHO/IR at three different stages of pregnancy (Early:13–14th g.w.; Middle: 27–28th g.w.; Late: 33–35th g.w.) was expressed as “Delta CHO/IR.”
The correction factor was calculated on the basis of the current 1,800 rule (
In agreement with our clinical practice, at the time of the first visit a written informed consent related to the use of clinical data (anonymously) was obtained from all women attending our outpatient offices.
All statistical analyses were performed using SPSS statistics version 20.0 (SPSS, Chicago, IL). Data are shown as median (25th and 75th percentiles) or mean (SD) according to distribution and as numbers (percentage). Kruskal–Wallis or ANOVA/Fisher tests were used to compare groups according to data distribution. Wilcoxon signed-rank and Friedman tests were used to analyze data longitudinally. Spearman correlation, univariate and multiple linear regression analysis were used to investigate relationships between variables. A p ≤ 0.05 was considered significant.
We enrolled 86 women with type 1 diabetes, aged 33.2 ± 5.2 yrs, disease duration 14 yrs (9–21) periconceptional HbA1C 52 (47–62) mmol/mol, BMI 24.4 (22.2–27.1) kg/m2, and weight gain at 33–35 g.w. of 14 (9.8–18) kg.
The patients were divided into three groups according to their pre-gestational BMI: 45 normal weight, 31 overweight and 10 obese women whose main clinical characteristics are shown in
Main clinical characteristics.
Normal weight n. 45 | Overweight n. 31 | Obese n. 10 | p-value | |
---|---|---|---|---|
Age ± SD (yr) | 33.2 ± 5.2 | 29.5 ± 5.5 | 31.7 ± 6.3 | ns |
Pre-gestational BMI (kg/m2) | 22 (20.4–23) | 26.7 (25.7–28.3) | 34.4 (31.9–37.2) | ns |
Disease duration (yr) | 13 (6.5–21) | 15.6 (10.2–19.5) | 14 (11–22.7) | ns |
HbA1c (mmol/mol) | 49.7 (43.7-59.9) | 49.2 (45-60) | 49.5 (38.8-58) | ns |
At 33–35 g.w., obese women gained less weight than the normal weight women (see
The main maternal-neonatal outcomes are reported in
Pregnancy Outcomes.
Normal weight% (n) | Over-weight% (n) | Obese% (n) | Tot% (n) | p-value | |
---|---|---|---|---|---|
CS | 90 (36/40) | 92.6 (25/27) | 100% (10/10) | 92.2 (71/77) | ns |
Macrosomia | 26.2 (11/42) | 25 (7/28) | 0 (0/9) | 22.8 (18/79) | ns |
LGA (>90th centile) | 34.1 (14/41) | 33.3 (9/27) | 11.1 (1/9) | 31.1 (24/77) | ns |
Preterm birth (<34 g.w.) | 0 (0/41) | 0 (0/28) | 40 (4/10) | 5.1 (4/79) | 0.001 |
Preterm birth (<37 g.w.) | 24.4 (10/41) | 28.6 (8/28) | 50.0 (5/10) | 29.1 (23/79) | ns |
hypoglycaemia | 15.8 (6/38) | 25.9 (7/27) | 22.2 (2/9) | 20.3 (15/74) | ns |
hypocalcemia | 0 (0/38) | 1 (1/27) | 0 (0/9) | 1.3 (1/74) | ns |
Jaundice | 13.1 (5/38) | 29.6 (8/27) | 11.1 (1/9) | 18.9 (14/74) | ns |
Female | 46.3 (19/41) | 46.4 (13/28) | 20.0 (2/10) | 43.0 (34/79) | ns |
CS, Cesarean Section.
Glycemic goals were generally reached before and after meals (see
In the normal weight women, insulin requirements, however expressed (24 h, basal or bolus total insulin and unit/kg, basal or bolus), significantly increased in late pregnancy (p < 0.0001). A similar trend was observed in the overweight women (<0.001; <0.001; 0.002, and 0.0006, respectively).
In the obese women, the insulin requirements, however expressed, did not change significantly during pregnancy (total insulin p = 0.14 and UI/kg, p = 0.1; total boluses p = 0.2, boluses/kg p = 0.4; total basal p = 0.1; basal/kg p = 0.27). Full data are reported in
Insulin requirement and weight gain: comparison among the three groups.
EARLY 13–14th g.w. | MIDDLE 27–28th g.w. | LATE 33–35th g.w. | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
BMI | Normal(n = 38) | Over(n = 29) | Obese(n = 10) | p-value* | p-value** | Normal(n = 45) | Over(n = 31) | Obese(n = 10) | p–value* | p–value** | Normal(n = 45) | Over(n = 31) | Obese(n = 10) | p–value* | p–value** |
|
0.0048 | N |
0.01 | N vs O.02 | 0.04 | N vs Ob.015 | |||||||||
Median | 3.1 | 2.25 | −1.25 | 11 | 7.3 | 6 | 14 | 13.7 | 8.0 | ||||||
25th–75th Percentile | 2–5 | 1–5.2 | −3.1 −1.1 | 8–13 | 5–12 | 2.1–8.2 | 12–18.1 | 10–18.9 | 2.9–15 | ||||||
|
0.0005 | N vs Ob.001 |
0.004 | N vs Ob.01 | ns | ns | |||||||||
Median | 33.30 | 41.15 | 53.5 | 38.63 | 46.20 | 58.08 | 49.8 | 56.45 | 62.64 | ||||||
25th–75th Percentile | 23.6–51 | 23–67.8 | 46.4–59.2 | 22.5–97.6 | 39.1–75.9 | 38.6–73.7 | 32–101.6 | 26–102.3 | 38.9–83.2 | ||||||
|
ns | ns | ns | ns | ns | ns | |||||||||
Median | 0.56 | 0.55 | 0.55 | 0.55 | 0.57 | 0.52 | 0.68 | 0.7 | 0.61 | ||||||
25th–75th Percentile | 0.4–0.8 | 0.31–0.7 | 0.49–0.8 | 0.4–1.3 | 0.49–1.05 | 0.38–0.9 | 0.4–1.3 | 0.47–1.13 | 0.37–1.0 | ||||||
|
0.008 | N vs Ob.01 |
0.008 | N vs Ob.005 | 0.02 | ns | |||||||||
Median | 17.2 | 24.1 | 30.88 | 21.13 | 25.35 | 33.58 | 24 | 30.45 | 38.92 | ||||||
25th–75thPercentile | 10.2–36.1 | 8.4–36.9 | 21.6–36 | 6.5–45.7 | 19.2–40.2 | 19.6–41.5 | 16–47 | 22–48.3 | 19.9–44.6 | ||||||
|
ns | ns | ns | ns | ns | ns | |||||||||
Median | 0.32 | 0.31 | 0.32 | 0.30 | 0.32 | 0.30 | 0.33 | 0.36 | 0.39 | ||||||
25th–75th Percentile | 0–0.6 | 0.11–0.53 | 0.23–0.47 | 0.1–0.4 | 0.25–0.5 | 0.19–0.5 | 0.3–0.5 | 0.28–0.56 | 0.19–0.49 | ||||||
|
0.0006 | N vs Ob.01 |
0.05 | ns | ns | ns | |||||||||
Median | 16.1 | 17.05 | 22.62 | 17.5 | 20.85 | 24.5 | 25.8 | 26 | 23.97 | ||||||
25th–75th Percentile | 6.5–28.1 | 9–39.0 | 22.0–25 | 9.9–69 | 13.8–38 | 19–32.2 | 8.4–69 | 13.8–54 | 19–43.1 | ||||||
|
ns | ns | ns | ns | ns | ns | |||||||||
Median | 0.23 | 0.23 | 0.26 | 0.26 | 0.26 | 0.22 | 0.35 | 0.31 | 0.22 | ||||||
25th–75th Percentile | 0.1–0.5 | 0.11–0.53 | 0.2–0.33 | 0.1–0.9 | 0.17–0.5 | 0.19–0.4 | 0.1–0.9 | 0.17–0.56 | 0.18–0.5 |
N, Normo–weight; Ov, Over–weight; Ob, Obese; ns, non-significant.
*Kruskal–Wallis, **Post hoc Analysis Bonferroni corrected p–values (S<0.016).
In early pregnancy, insulin requirements increased according with BMI, being lower in the normal weight women and higher in the obese women (normal weight
These differences, although smaller, persisted in middle pregnancy (normal weight
When insulin requirements were adjusted for body weight (IU/kg), no difference was found in any phase of pregnancy (see
In early pregnancy, we found a significant correlation between pregestational BMI and insulin requirement expressed as total IU/day at each meal (p < 0.001), as confirmed by univariate linear regression analysis (
At each meal, CHO/IR declined significantly throughout gestation in the normal (breakfast p = 0.01; lunch p = 0.03; dinner p = 0.0009) and overweight women (breakfast p = 0.001; lunch p = 0.006; dinner p < 0.0001) (see
CHO/I in normal weight, over–weight and obese women with type 1 diabetes across early, middle, and late pregnancy.
Early 13–14 g.w. | Middle 27–28 g.w. | Late 33–35 g.w. | *p–value | **p–value | CHO/IR Delta | |
---|---|---|---|---|---|---|
|
||||||
CHO/I breakfast | 13.3 (9.8–6.7) | 7.5 (5.5–9.5) | 6.2 (3.8–8.6) | 0.01 | EvsM 0.004 |
7.4 (0–11.4) |
CHO/I lunch | 12.4 (9–16) | 10 (9–13.3) | 7.9 (5.6–2.9) | 0.03 | EvsM 0.08 |
3.7 (0–6.4) |
CHO/I dinner | 14 (9.7–16.8) | 10 (7.1–11.7) | 6.9 (5.5–10) | 0.0009 | EvsM 0.1 |
6.1 (3.2–10.3) |
|
||||||
CHO/I breakfast | 8.5 (7.1–12.6) | 7.3 (5–9) | 5.2 (4–8.1) | 0.001 | EvsM 0.08 |
3.4 (1.5–5.4) |
CHO/I lunch | 11.3 (7.4–13.5) | 8.5 (6.2–11) | 7.3 (5.4–9.8) | 0.006 | EvsM 0.01 |
3.1 (1.5–5.7) |
CHO/I dinner | 12.4 (8.9–16) | 9.2 (7.5–12) | 8.1 (6.7–13) | <0.0001 | EvsM 0.02 |
1.7 (1–7.3) |
|
||||||
CHO/I breakfast | 6.0 (5.0–7.9) | 5,9 (4,7–8) | 5.1 (4.1–7.4) | 0.7 | EvsM >0.9 |
1.4 (–3.4–3) |
CHO/I lunch | 7.8 (6.5–9.2) | 7.1(4.8–9.3) | 7.3 (4.1–10.7) | 0.7 | EvsM 0.5 |
0.7 (–3.4–3.5) |
CHO/I dinner | 9.3 (8.2–10.8) | 7.3 (6.8–8.5) | 7.2 (4.8–9.4) | 0.2 | EvsM 0.1 |
3 (–0.5–6.3) |
Median (interquartile range).
Longitudinal analysis for each group: *p–value Friedman, **p–value Wilcoxon signed rank. E, Early, M, Middle, L, Late.
In addition, in order to assess if this overall CHO/IR decrease was a continuum process, we compared the CHO/IR changes happening between early and middle pregnancy or between middle and late pregnancy (see
In early pregnancy, the normal weight women had the highest CHO/IR at breakfast (
In early pregnancy, we found a significant correlation between pregestational BMI and CHO/IR at each meal (p = 0.001), as confirmed by univariate linear regression analysis (beta −0.48; 95%CI (−0.78 to −0.21); P < 0.001; R2 0.32).
In late pregnancy we found a significant correlation between pregestational BMI and CHO/IR change at breakfast and dinner (Spearman test, p = 0.03 and 0.048, respectively) as well as a correlation between CHO/IR change at breakfast and weight gain (p = 0.02). The significant relationship between BMI or weight gain and CHO/IR change at breakfast (univariate simple regression) was lost when both variables were included in a multiple regression analysis (see
BMI, weight gain and CHO/IR decrease at breakfast: univariate and multiple regression analysis.
|
IC 95% | p | R2 | |
---|---|---|---|---|
Model 1. Dependent: Delta CHO/IR | ||||
BMI | c0.341 | −0.65 to -0.028 |
|
0.11 |
Model 2. Dependent: Delta CHO/IR | ||||
Weight gain | 0.38 | 0.05 to 0.55 |
|
0.14 |
Model 3. Multiple regression analysis. Dependent: Delta CHO/IR | ||||
BMI | −0.2 | −0.56 to 0.16 | 0.27 | 0.17 |
Weight gain | 0.22 | −0.07 to 0.5 | 0.14 |
Delta CHO/IR is defined as the difference between CHO/IR in early pregnancy and CHO/IR in the late pregnancy.
The bold values of p provided are ≤0.05.
The calculated correction factor was higher in the normal weight women, lower in the obese women and intermediate in the overweight women in early (p = 0.0005) and middle (p = 0.0008) pregnancy. As pregnancy progressed, the calculated correction factor significantly decreased in the normal and overweight women only, while it remained unchanged in the obese women. As a consequence, in late pregnancy the three groups had a similar insulin sensitivity (see
Correction Factor across early, middle, and late pregnancy, in each group and between groups.
Early 13–14 g.w. | Middle 27–28 g.w. | Late 33–35 g.w. | *p–value | |
---|---|---|---|---|
BMI < 25 kg/m2 | 53.7 (47.6–62.8) | 45.2 (34.9–53.6) | 36.14 (29.0–44.7) | <0.0001 |
BMI 25–29.99 kg/m2 | 45.3 (38.4–51.1) | 39.0 (29.3–47.1) | 31.8 (23.5–40.7) | <0.0001 |
BMI ≥ 30 kg/m2 | 33.8 (30.7–37.7) | 30.6 (24.9–36.5) | 28.7 (23.8–32.4) | 0.4 |
**p-value | 0.0005 | 0.008 | 0.3 |
Median (interquartile range) *p-value Friedman, **p-value Kruskal–Wallis.
In this study we found that insulin requirements, however, expressed (24-h total units or units/kg), increased from early to late pregnancy in the normal and overweight women by about 1.5 and 1.4 times, respectively. Similarly, CHO/IR decreased in both the normal and the overweight women.
In contrast, insulin needs did not change in the obese women, nor did CHO/IR (
However, considering the whole cohort the insulin increase rate was similar to that observed in some studies (
Analogously, the CHO/IR decrease was smaller in our study with respect of others such as that of Zagury et al. (
We measured insulin requirements and CHO/IR from the 13–14th g.w., as the women had reached a stable metabolic control (
During pregnancy, the insulin requirement increases with reference to the physiological increase in insulin resistance (
We think that a relevant role is played by weight gain as we observed significant relationship between weight gain and CHO/IR change (univariate simple regression). Previous study showed that the changes in insulin sensitivity from the time before conception through early pregnancy were inversely correlated with the changes in maternal weight gain in lean women with normal and abnormal glucose tolerance (
Even if we did not report data from the beginning of pregnancy, we think that in obese women, the higher insulin resistance already documented in early pregnancy is mitigated by a lower weight increase—even weight loss in early pregnancy—compared to normal weight and/or overweight women in all pregnancy phases. This evidence may explain the differences in insulin requirements and CHO/IR among groups at baseline in contrast to late gestation, when the obese women achieved a lower weight gain. Moreover, the “quality” of weight gain could also have an impact on the insulin resistance increase rate. In fact, although we do not have data on the body composition and on the physical activity of our population, evidence shows that lean mass increases more in obese women than in normal weight women during pregnancy (
Since significant relationship between pregestational BMI or weight gain and CHO/IR change was lost in multiple regression analysis, we conclude that pregestational BMI and weight gain do not play an independent predictive role, but together contribute to CHO/IR change.
All the women received a dietary prescription as well as general training on the function of the glycemic index and macronutrient distribution in meals. Therefore, we think that a medical nutritional therapy appropriate in quantity and quality (
However, this hypothesis is not fully documented in our work since the patients’ eating habits were not investigated prior to the first visit. Moreover, the results of the study show that the obese pregnant women lost weight in the first trimester, despite the application of the abovementioned diet recommendations. Given that adherence to the diet was verified through the self-monitoring diary where women recorded the meal, the amount of carbohydrates, and the insulin used, we speculated that these findings can be derived from three factors: 1—MNT during pregnancy (
These data obtained with old technologies were not influenced by an automated insulin suspension system and CHO/IR was selected only when pre- and post-prandial glucose levels were at target.
Finally, other variables not investigated in the present study could affect insulin requirements and CHO/IR. The first of these, fetal hyperinsulinemia, by lowering fetal glycemia increases the concentration gradient of glucose across the placenta, influencing the glucose flux from mother to fetus and reducing maternal glucose levels (
The main study limitation is that this is a retrospective analysis sharing the same electronic database used in a previous study (
However, to date, few longitudinal data on CHO/IR in women with type 1 diabetes across early, middle, and late pregnancy are available and, to our knowledge, none with measured CHO/IR. This is the first study performed to assess CHO/IR in normal weight, overweight and obese pregnant women.
Deeper information might help in the management of pregnancy in type 1 diabetic women under insulin pump treatment, by predicting the CHO/IR trends during gestation. Therefore, a better practice in using technology may produce better outcomes.
Pre-gestational BMI and weight gain contribute to determining the CHO/IR trend during pregnancy in pregnant type 1 diabetic Caucasian women under insulin pump treatment.
Although we cannot draw definitive conclusions as these retrospective findings need to be confirmed in a larger population, we think that these phenomena in obese women with type 1 diabetes in pregnancy deserve to be prospectively studied and considered in therapeutic algorithms of automated and manual insulin infusion systems in pregnancy.
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
Ethical review and approval were not required for the study on human participants in accordance with the local legislation and institutional requirements. The patients/participants provided their written informed consent to participate in this study.
CF and AN: conceptualization, software, validation, writing— review and editing, and visualization. AN: methodology, supervision, and project administration. OB and CG: formal analysis. RF, MB, CS, CF, AN, NV, and CG: investigation and data curation. CF: resources and writing—original draft preparation. All authors contributed to the article and approved the submitted version.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The Supplementary Material for this article can be found online at: