The Importance of Office Blood Pressure Measurement Frequency and Methodology in Evaluating the Prevalence of Hypertension in Children and Adolescents With Type 1 Diabetes: The SWEET International Database

The prevalence of hypertension in children without diabetes is reported to be 2–4% and that of elevated blood pressure (BP) 14.8% (1). More recent reports show an increase in the prevalence of hypertension to 6.6–10.6% and of elevated BP to 6.6–16.3%, depending on the criteria used (2004 Fourth Report on the Diagnosis, Evaluation, and Treatment of High Blood Pressure in Children and Adolescents [https://www.nhlbi.nih.gov/files/docs/resources/heart/hbp_ped.pdf], European or American Academy of Pediatrics [AAP] guidelines 2017), the BP thresholds, and the measurement methodology (1,2). Other factors influencing variability in the prevalence of hypertension were sex, age, size of the studied population, and the presence of obesity (2).

The prevalence of office hypertension is higher in adolescents with type 1 diabetes (T1D) than in children without diabetes (3,4). Recent studies using the new AAP 2017 criteria reported an increase in the estimated prevalence of hypertension (5). Screening for hypertension in children and adolescents with diabetes is mandatory for early diagnosis and treatment, aiming at preventing cardiovascular damage, since it has been shown that systolic BP in childhood predicts hypertension later in life (6–8).

The accuracy of BP evaluation is crucial for the reliable diagnosis of hypertension, and recent guidelines emphasize the need to implement the appropriate methodology (6,7). Improper evaluation of BP leads to overdiagnosis or underdiagnosis and may, therefore, result in an imprecise estimation of the prevalence of hypertension (6,7). A recent review identified 27 sources of inaccuracy in BP measurement related to the observer, the device, the procedure, and the patient, which may result in misclassification of hypertension status (9). White coat hypertension is the most common reason for falsely elevated BP and hypertension overdiagnosis (6). A recent survey of 52 diabetes centers of the SWEET (Better control in Pediatric and Adolescent diabeteS: Working to crEate CEnTers of Reference) Consortium showed that 74% of centers were using appropriate methodology for BP assessment (10). However, the BP measurement methodology used was highly inconsistent. To date, adjustment for the BP methodology and technology used in large-scale studies aiming to assess the prevalence of elevated BP and hypertension in children and adolescents with diabetes has not been performed.

This analysis of the SWEET international database aims to 1) estimate the prevalence of elevated BP and hypertension in a large cohort of children and adolescents with T1D, and 2) investigate the impact of the BP measurement methodology on the prevalence of hypertension.

SWEET (www.sweet-project.eu) is a network of mainly large (>150 patients) pediatric diabetes centers from different countries around the world. The project was approved by the ethics committee of the AUF DER BULT Diabetes Center for Children and Adolescents, Hannover, Germany (11). All centers are responsible for their respective ethical approval and informed consent (11). The SWEET platform hosts medical data under one unified anonymized diabetes database, which are sent to Ulm University, Institute of Epidemiology and Medical Biometry, ZIBMT, Ulm, Germany, for central evaluation and statistical analyses.

As of October 2020, the SWEET database comprised 76,987 children, adolescents, and young adults with diabetes. Fig. 1 depicts the flowchart for selection of patients. As stated in the SWEET database under “additional treatment,” patients receiving antihypertensive treatment for hypertension or micro/macroalbuminuria were excluded from the analysis to avoid misclassification, as it was not clear whether patients receiving ACE inhibitors for micro/macroalbuminuria had hypertension or not.

Data were collected from all children and adolescents with T1D aged 5–18 years with BP data in 2019 (the most recent year with complete data; those receiving antihypertensive treatment were excluded), from 112 centers in 52 countries. Data from all visits within 2019 were taken into consideration. The regional distribution of centers was as follows: 56 centers from 25 countries in Europe, 7 centers from Australia/New Zealand (Oceania), 7 centers from 6 countries in South America, 8 centers from North America/Canada, and 34 centers from 17 countries in Asia/Middle East/Africa.

Systolic and diastolic office BP measurements were collected from the electronic database. Only one systolic and diastolic BP value could be recorded in the database per visit; therefore, each BP value represents either one BP measurement or the average of two or three measurements per visit. Centers have reported that in case of an abnormal BP level, more measurements were taken. Age, sex, diabetes duration, height, height-SD scores (SDS), weight, BMI-SDS, daily insulin requirements, insulin regimen (pump treatment or not), and hemoglobin A_1c (HbA_1c) were also recorded.

To evaluate the effect of the number of measurements on elevated BP and hypertension prevalence, the participants were divided into groups with measurements obtained at three or more visits and fewer than three visits, according to the BP data available in 2019. Furthermore, the prevalence of elevated BP/hypertension was investigated in participants ≥13 and <13 years of age, and in boys and girls with BP measurements at three or more visits (6).

To examine the impact of BP measurement methodology and technology on the prevalence of elevated BP/hypertension, a subanalysis was performed on 15,742 patients from 50 centers, which had answered a questionnaire on the BP methodology during the same study period, and with available BP data (9). The questionnaire included detailed information on 1) the BP measurement methodology used at each center regarding age at onset of evaluation, frequency of BP measurements, setting, resting time before measurement, body position (lying, sitting) and posture (arm, back, feet), talking during measurements, number of readings per occasion, and 2) BP measurement technology, including device type and brand, availability of cuffs, and cuff selection (10). An overall score for BP measurement methodology and technology was calculated for each center and was used in the current analysis (0 for worst and 4 for best performance); however, for ease of presentation, the final score was transformed into a scale of 100 (highest score 100%) (10). A score of ≥3 (75%) was considered as “very good performance,” 2–2.99 (50–74%) as “reasonable,” and <2 (50%) as “poor performance.”

Normalcy tables for office BP measurements of 2017 U.S. pediatric hypertension guidelines were used for defining BP categories, with elevated BP and hypertension being classified using the percentiles provided (6). Stage 1 and stage 2 hypertension groups were analyzed as a single group as there were few children with stage 2 hypertension. To classify a patient into a BP category, the median of measurements of all annual visits was calculated for systolic and diastolic BP separately. Then an SAS macro tool was used to calculate the categories (https://sites.google.com/a/channing.harvard.edu/bernardrosner/pediatric-blood-press/childhood-blood-pressure). The diagnosis of elevated BP or hypertension was made if the median BP for either systolic or diastolic BP was ≥90th to 95th or ≥95th percentile respectively, according to the percentile-based thresholds of the 2017 U.S. guidelines.

For the calculation of BMI-SDS and height-SDS, the WHO reference values were used (12).

HbA_1c values were determined in each center and standardized according to the Diabetes Control and Complication Trial reference range of 4.0% to 6.0% (13).

Descriptive data are presented as median (interquartile range [IQR]) or percentages. Wilcoxon rank sum test was used for pairwise comparisons of continuous variables, and the χ² test was used for binary variables. Kruskal-Wallis was used for comparisons among multiple groups for continuous variables. The Bonferroni-Holm method was used to adjust P values for multiple testing. Linear regression models were used to investigate the possible relationship between systolic BP and total center score after adjusting for different covariates. For P value adjustments due to multiple group comparisons, the Tukey-Kramer method was used. Outcomes of regression models are presented as estimated least-squares means with 95% CIs. In the models investigating the relationship of BP as a dependent variable with other independent variables, correlation coefficients and variance inflation/toleration were used to check for possible multicollinearity. All correlation coefficients were <0.8 (not high), and variance inflation/toleration values were also within the reasonable bounds to exclude multicollinearity.

Fit statistics with the Akaike information criterion (AkIC) were used to evaluate how well a model fits the data it was generated from. AkIC was used to compare different possible models and determine which one is the best fit for the data. The lowest AkIC value indicates the best model. Analyses were performed using the SAS 9.4 statistical program (SAS Institute, Cary, NC).

In total, 27,120 patients (cohort A) were included in the analysis (Fig. 1), 10,440 individuals with BP measurements at three or more visits in 2019, and 16,680 with fewer than three visits (Table 1).

For the subgroup analysis of centers with center score available for evaluation of their BP measurement methodology, 15,742 patients (cohort B) were analyzed, 7,072 with BP measurements at three or more visits and 8,670 with fewer than three visits (Table 1).

Demographic characteristics of participants in the abovementioned cohorts and subgroups are shown in Table 1.

Participants without data on BP measurement methodology (n = 11,378) did not differ from those in cohort B regarding age, age at diabetes diagnosis, disease duration, BMI-SDS, and sex distribution. They had, however, higher HbA_1c (P < 0.001), higher daily insulin requirements (P < 0.001), and lower height-SDS (P < 0.001).

Patients without BP data (n = 6,051), who were excluded (Fig. 1), were younger than those with BP data (P < 0.001), they had shorter disease duration (P < 0.001), lower BMI-SDS (P < 0.001) and HbA_1c (P < 0.001), and different regional distribution (P < 0.001), yet with no difference in sex and height-SDS. As the proportion of children without BP data was 18%, this is not expected to influence considerably the study results.

In the group of patients with BP measurements at three or more visits, 7,485 had normal BP, 1,829 had elevated BP, and 1,126 had hypertension, and in those with BP at fewer than three visits, 11,186 had normal BP, 2,551 had elevated BP, and 2,943 had hypertension. The prevalence of hypertension in the group of patients with BP at three or more visits was lower compared with those with fewer than three visits (10.8% vs. 17.5%, P < 0.001), whereas the prevalence of elevated BP and normotension was higher (17.5% vs. 15.3% [P < 0.001] and 71.7% vs. 67.1% [P < 0.001], respectively) (Fig. 2).

Participants with hypertension or elevated BP and BP data at three or more visits were older compared with those with normal BP (median age [IQR]: 14.5 [11.0–16.4] and 14.9 [12.4–16.4] vs. 12.6 [10.2–14.9] years, respectively [hypertensive to normal and elevated BP to normal: P < 0.001, hypertensive to elevated: P = 0.01]) and had significantly longer disease duration (5.0. [2.6–8.3] and 5.4 [2.9–8.8] vs. 4.1 [1.8–7.0] years [hypertensive to normal and elevated BP to normal: P < 0.001]). Patients with hypertension had higher HbA_1c compared with normotensive participants (8.33% [7.2–9.1] vs. 8.14% [7.1–8.7], P < 0.001; or 67.5 [54.9–75.8] vs. 65.5 [53.8–72.0] mmol/mol). Furthermore, there were more male patients in the group with hypertension or elevated BP compared with those with normal BP (55.1% and 55.4% vs. 50.4%, respectively [hypertensive to normal: P = 0.023 and elevated BP to normal: P = 0.002])]and fewer pump users in the group of patients with hypertension compared with the group with normal BP (50.3% vs. 55.2%, P = 0.023).

In the group of centers with a score and BP at three or more visits, 4,903 patients had normal BP, 1,328 had elevated BP, and 841 had hypertension, and in those with BP at fewer than three visits, 5,482 had normal BP, 1,513 had elevated BP, and 1,675 had hypertension. The prevalence of hypertension in the group of centers with a score and BP at three visits or more was lower compared with those with fewer than three visits, at 11.9% vs. 19.3% (P < 0.001), whereas the prevalence of elevated BP and normotension was higher at 18.8% vs. 17.4% (P = 0.03) and 69.3% vs. 63.2% (P < 0.001), respectively, indicating that performing BP measurements at fewer than three visits overestimated the prevalence of hypertension (Fig. 2)

Owing to age and BMI-SDS differences between participants with BP at three or more visits or fewer than three visits in all centers (cohort A) and between the subgroup of centers with an available score (cohort B) (Table 1), systolic and diastolic BP values were adjusted for age, BMI-SDS, and region. The differences between the adjusted and unadjusted systolic and diastolic BP values in the A and B cohorts and subgroups were very small (0.4–2.3 mmHg). Results on adjusted systolic and diastolic BP levels in the various BP categories in all centers and in centers with score are available in Supplementary Fig. 1.

The prevalence of hypertension was investigated in patients ≥13 or <13 years of age in participants from all centers with BP at three or more visits. Demographic characteristics of both groups are available in Supplementary Table 1. There were no significant differences regarding BMI-SDS and the percentage of male participants between the age-groups. However, patients ≥13 years had significantly longer disease duration, higher HbA_1c, and lower height-SDS, and there were fewer pump users among them compared with patients <13 years in addition to the expected difference in insulin requirements (Supplementary Table 1).

Systolic and diastolic BP were adjusted for disease duration, HbA_1c, and region; however, the differences were very small (0.0–0.3 mmHg). Participants ≥13 years of age had significantly higher systolic and diastolic BP compared with younger individuals (P < 0.001 for both). Results regarding systolic and diastolic BP levels in two age-groups are available in Supplementary Fig. 2.

Hypertension and elevated BP were significantly more frequent in patients ≥13 years of age compared with those <13 years (P < 0.001) (Fig. 3).

Characteristics of patients stratified by sex are shown in Supplementary Table 1. There was no difference in age between boys and girls; however, girls had longer disease duration; higher BMI-SDS, HbA_1c, and daily insulin requirements; and lower height SDS, and there were more pump users among them compared with boys.

Systolic and diastolic BP were adjusted for disease duration, BMI-SDS, HbA_1c, and region; however, the differences between the unadjusted and adjusted values were very small, ranging from −0.9 to 0.2 mmHg. Systolic BP was lower and diastolic higher in girls compared with boys (both P < 0.001) (Supplementary Fig. 2).

Hypertension and elevated BP were more frequent in boys compared with girls (P < 0.001) (Fig. 3).

The center score was significantly higher in the group of patients with normal BP compared with the group with elevated BP or hypertension in the cohort of patients from centers with an available score of BP measurement methodology (cohort B), and BP measurements at three or more visits (median center score [IQR] 2.74 [2.67–3.05] vs. 2.73 [2.64–3.05] [normal BP to elevated BP P < 0.001] or vs. 2.71 [2.64–3.05] [normal BP to hypertensive group P < 0.001], respectively), suggesting that better BP measurement methodology results in more accurate diagnosis of hypertension.

In a linear regression model with systolic BP as a dependent variable and center score as an independent variable, center score remained an independent predictor of systolic BP after adjusting for multiple covariates, namely age, sex, diabetes duration, height-SDS, HbA_1c, and region in various models with different combinations. Three of the most representative models are available in Table 2. Performing fit statistics showed that the best of three models with the lowest AkIC was the one that included as covariates age, sex, diabetes duration, BMI-SDS, HBA_1c, region, and total center score. All of the abovementioned variables had a significant impact on systolic or diastolic BP. Removing HbA_1c from the model resulted in higher AkIC. Furthermore, replacing BMI-SDS with height-SDS did not give a lower AkIC; therefore, BMI-SDS was considered more suitable to be included in the rest of the models. Removal of adjustment for region resulted in disappearance of the significance of the center score.

Similar results to systolic BP were found for diastolic BP (Supplementary Table 2).

This analysis of BP measurements from 27,120 children and adolescents with T1D attending pediatric diabetes centers from 52 countries that are members of the SWEET international database showed that 1) the prevalence of hypertension in children and adolescents with T1D was 10.8% and that of elevated BP was 17.5%; therefore, a considerable proportion of young patients (28.3%) may be at risk for future morbidity and mortality due to elevated BP; 2) measuring BP on fewer than three occasions overestimates the prevalence of hypertension; 3) BP measurement methodology is an important determinant of BP levels, which affects the prevalence of hypertension, and this should be taken into consideration in future studies and in clinical practice. To our knowledge, very limited data exist regarding the impact of BP measurement methodology on hypertension prevalence in a large multinational pediatric T1D cohort.

The prevalence of hypertension in our study is much lower than that of the study of Dost et al. (5) (44.1%) and closer to older studies that have reported a prevalence of hypertension ranging from 4 to 16% in children and adolescents with T1D (4,14–18). Compared with the study of Dost et al. (5) children in our cohort were younger, had shorter disease duration, younger age at diabetes onset, and lower percentage of male participants, and data were collected at different time periods. The above differences may partly explain the difference in hypertension prevalence between the two studies using the same AAP 2017 criteria for the definition of hypertension.

The considerable variability in hypertension prevalence among the studies might be attributed to differences in several contributors, such as disease duration, obesity prevalence, period of investigation, population differences in HbA_1c, ethnicity, the normative data, and BP thresholds used and BP measurement methodology (4,5,14–19).

Although recent guidelines indicate that at least three BP measurements at different occasions are necessary to diagnose elevated office BP and hypertension (6,7), most hypertension prevalence studies have used only one visit for the evaluation of BP (4,14,15,17,20), very few two (18) or three (5), and in others, the exact number of visits was not reported (16). Τhe current study has shown that hypertension prevalence based on data from fewer than three visits may be overestimated. The variability of office BP, the white coat reaction, and the regression to the mean all lead to a decline in BP levels, with more measurements taken in repeated office visits, preventing, thereby, the overestimation of office BP level and overdiagnosis of hypertension (6). Nambam et al. (18) reported that the prevalence of hypertension in the Exchange registry was 4% in two visits versus 17% in a single visit. A recent meta-analysis in children and adolescents without diabetes showed that when compared with visit 1, the prevalence of elevated BP decreased substantially by 53.7% during visit 2 and by 77.7% during visit 3 (21).

Boys had higher prevalence of hypertension compared with girls. The difference in systolic BP between boys and girls could not be attributed to age or to BMI-SDS; however, boys had higher height-SDS, which could partly explain the higher prevalence of hypertension in boys. Other reasons could be hormonal differences, possible ethnicity disparities, or factors unrelated to diabetes, such as differences in diet or sodium intake (15,22–24). In children without diabetes, there is consistently greater prevalence of high BP in boys (15–19%) than in girls (7–12%) (6). Findings similar to ours were reported by other researchers (5,14). A recent study showed that the prevalence of hypertension was higher in teenaged boys ≥15 years of age compared with the girls (5); however, this difference was quite small in younger children (5). The authors suggested that the sex difference depends highly on the normative data and criteria used (5).

Diastolic BP was higher in girls in our study compared with boys, which could be attributed, at least in part, to higher BMI-SDS, daily insulin requirements, HbA_1c, and longer disease duration observed in girls. It has been reported that diastolic BP can be less accurate in children because of the potential difficulty in distinguishing fourth and fifth Korotkoff sounds, probably more so in girls with smaller arm size (22). In another study including children and adults, a significant observer effect was present for diastolic BP <90 mmHg (25). However, the reasons for the dichotomy in systolic and diastolic BP prevalence between the sexes are not clear.

Reclassification of children according to their BP status with the application of the new AAP 2017 criteria showed that children reclassified as hypertensive were more likely to be boys and slightly taller than normotensive children (5,19). Other studies have failed to show difference in the prevalence of hypertension between sexes (4).

Children ≥13 years of age had significantly higher prevalence of hypertension compared with those <13 years in our study. Studies have suggested that pubertal growth is associated with profound increases in systolic BP in children without diabetes, with noticeably greater increments in boys than in girls (22). Different studies have shown that the prevalence of hypertension is higher in older age-groups, although the definition of age-groups varies among studies (5,14).

An important finding of this study is that the center score evaluating BP measurement methodology and technology is an independent determinant of systolic and diastolic BP level after adjusting for multiple covariates. A wider use of this center score could possibly help to explain differences in hypertension prevalence among diabetes centers for better interpretation of research results. Another important conclusion of this work was that with more accurate BP evaluation, overestimation of hypertension could be avoided and, thereby, unnecessary long-term treatment. A higher percentage of normotensive patients was detected from centers with a higher center score.

It should be mentioned that the accuracy of BP measurement was raised in the National Health and Nutrition Examination Survey studies, and quality-control measures were suggested to minimize observer error (26). A recent position statement emphasized that to minimize observer’s error in BP measurement, standardization of measurement practices with proper training and regular certification in BP assessment are necessary (26).

In the multiple regression models evaluating the relationship between BP and center score, the covariates age, disease duration, sex, BMI-SDS, and HbA_1c were independently related to systolic and diastolic BP. The impact of BMI and HbA_1c on BP has been shown previously (3,6,15,20,27). Furthermore, the effect of hyperglycemia on the vascular wall is complex, involving a variety of mechanisms (28). Parameters such as ethnicity, dietary factors, smoking, physical inactivity, albuminuria, family history, low birth weight, and others, which may also influence BP, were not analyzed in the current study (6,20,27).

Of 27,335 participants, 215 were excluded from the analysis because they were receiving antihypertensive medications. If these participants are added to the 1,126 with documented office hypertension (elevated BP in three or more visits), the proportion of individuals with hypertension who received treatment is very small (16% [215 of 1,341]), suggesting that most individuals with office hypertension remain untreated. It should be mentioned, however, that the diagnosis of hypertension requires confirmation with ambulatory BP monitoring aiming to identify individuals with white coat hypertension who may not need drug treatment (6). It is possible that underreporting of antihypertensive treatment from some centers could have contributed to the high percentage of untreated patients. Undertreatment of children with hypertension has been previously reported, and the percentage of treated individuals varies among different studies from 0.3 to 52% (4,5,15,17,29–31). The abovementioned low percentages suggest that there is either therapeutic inertia or underdiagnosis of hypertension, or most probably both, which raises concerns for future morbidity and mortality. Moreover, in a recent study, participants’ adherence to treatment with ACE inhibitors declined from 84 to 53% after 48 months (32).

Strengths of this study are that 1) it provides real-world data from a very large number of clinics across multiple regions, 2) it estimates more accurately office hypertension prevalence based on measurements on at least three different occasions, and 3) it is one of the few studies investigating the impact of measurement methodology on hypertension prevalence in a large cohort of children with T1D.

One limitation of our study is that the BP measurement reported in the database was representing either one or the median of two or three measurements per occasion. Therefore, there was no homogeneity in this parameter. However, the number of measurements per occasion was taken into consideration for evaluation of the center score, which was used to estimate the impact of BP methodology on the prevalence of hypertension and elevated BP.

It should be stressed that within our initial intention was also to raise awareness on BP methodology measurement and improve the performance among the SWEET centers. Furthermore, although the findings of the study that more BP measurements reduce the prevalence of hypertension are well known and that current guidelines emphasize the need to use the optimal measurement methodology, this study provides real-world data confirming these findings in a very large cohort of T1D across multiple sites, stressing the urgency of guidelines implementation.

Another potential limitation is that there are no data on terminal digital preference as a confounding factor. However, a previous report from the SWEET centers showed that 71.5% of them use automated devices for BP measurement, which eliminates the terminal digit preference, yet a small observer bias cannot be excluded (10,33).

Another limitation of our study was that ethnicity and smoking and their impact on the prevalence of hypertension could not be evaluated, because they are not formal variables recorded in the SWEET database.

Finally, it might be argued that patients on antihypertensive treatment should have been included. However, additional analysis including patients on antihypertensive treatment, showed results very similar to those after excluding them. By including treated children (three or more visits), the prevalence of hypertension was 10.9%, of elevated BP was 17.5%, and of normotension was 71.6% in cohort A and 12.0%, 18.8%, and 69.2%, respectively, in cohort B. The latter analysis has the limitation that the prevalence of hypertension would be slightly overestimated, as children on ACE inhibitors for micro/macroalbuminuria without hypertension would be classified as hypertensive.

This study has shown that in the SWEET international database, ∼30% of children and adolescents with T1D have hypertension or elevated BP, which is a modifiable cardiovascular risk factor and that its early diagnosis and treatment is of great importance for prevention of future cardiovascular events and improvement of life expectancy. Subjects with T1D and increased office BP must be evaluated with ambulatory BP monitoring, and many of them may have hypertension confirmed, requiring long-term antihypertensive drug treatment that should not be delayed. Another important finding is that the BP measurement methodology is a significant determinant of systolic and diastolic BP levels, affecting the prevalence of hypertension and must be routinely considered in both research studies and in clinical practice.

Acknowledgments. The authors are thankful to the following individuals for their support of this work: Sascha R. Tittel, for the data management, as well as Andreas Hungele and Ramona Ranz, for the SWEET DPV software (all Ulm University, Germany), Michael Witsch (Centre Hospitalier de Luxembourg, Luxembourg) for center integration, Thomas Danne and Olga Kordonouri (Kinder- und Jugendkrankenhaus AUF DER BULT, Hannover, Germany), for initiating the SWEET collaboration, and Tanja Riebe, Katharina Klee (Kinder- und Jugendkrankenhaus AUF DER BULT, Hannover, Germany), and Reinhard Holl (Ulm University, Germany), for their invaluable support. Finally, the authors would like to thank all participating centers of the SWEET network, especially the collaboration centers in this investigation. A detailed list can be found in the Supplementary Data (Appendix).

Duality of Interest. This work was supported by the SWEET corporate members, namely Abbott, Boehringer Ingelheim, Dexcom, Insulet, Lilly, Medtronic, and Sanofi. The content is solely the responsibility of the authors and does not necessarily represent the official views of the corporate members. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.V. wrote the manuscript, designed the analysis, and prepared the final draft after incorporating the comments of the rest of the authors. A.V., N.H.B., O.K., V.I., B.P., B.S., A.P.L., S.S., and D.M.M. contributed with research data. S.R.T. performed the statistical analyses. S.R.T., N.H.B., O.K., V.I., B.P., and D.M.M. critically reviewed the manuscript and participated in the discussion with valuable comments. B.S., A.P.L., and S.S. reviewed the manuscript. G.S. contributed to the design of the analysis and the interpretation of the results and provided guidance for the preparation of the first draft. All coauthors approved the final version of the manuscript. A.V. and S.R.T. 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.