Sensitivity and specificity were reported in two studies [18,21], and could be calculated from the trees in one study . There was wide variation in sensitivity (8.3–78.6%) and specificity (25.7–96.1%). Positive and negative predictive values (PPV and NPV) were only reported in one study, and ranged from 10.9 to 51.4% for PPV and from 71.3 to 95.7% for NPV . In another study, PPV and NPV could be derived from the classification trees; PPV ranged from 36.7 to 68.0% and NPV from 84.2 to 96.9%) . Calibration and explained variation were reported in only one study, and there was no indication that models were not adequately calibrated (i.e. the Hosmer-Lemeshow P value was >0.05). The explained variation for the most elaborate model was 22% .
In only two studies, models were presented in such a way that it was possible to calculate a predicted risk for a new individual child, either by following the steps in the classification trees , or by using the reported age-specific and sex-specific PPVs of the corresponding childhood risk factor . One study presented a full regression equation , but as the predictor ‘age at first diagnosis of hypertension’ is naturally not available at the time point of prediction, this model cannot be applied in a new individual. Another study presented a regression equation in the supplementary material, but without providing the intercept, hampering the calculation of a predicted risk .
Risk of bias assessment
Table 4 shows the results of the risk assessment for selection bias, information bias, and bias related to the analysis in the included studies. Two studies [17,22] scored ‘high’ for risk of selection bias related to participant selection; for both studies this was because of limited or unclear information on recruitment and selection of the study population, and because of limited information on key characteristics and predictors for the study sample. For selection bias related to sample attrition, all five prospective studies [17–19,21,22] reported the amount of loss to follow-up, and four [17–19,21] of these scored ‘moderate,’ as loss to follow-up was higher than 20%. With regard to risk of information bias for predictors, two studies [20,22] scored ‘high,’ which was because of: unclear definitions/measurements of predictors [20,22], possibly less valid and reproducible measurements , lack of standardized measurements , different timing of predictor assessment within the study sample , and/or unavailability of predictors at the intended time point of use of the model . For the risk of information bias related to the outcome, four studies [17–19,22] scored ‘high,’ because of lack of information on blinding for predictor information [17–19,22], unclear definitions of outcome and measurement of outcome [17,22], possibly less valid and reproducible measurements [17–19,22], and/or different assessment (including timing) within the study sample [18,19,22]. Lastly, for risk of bias related to the analysis, four studies [17–19,22] scored ‘high’ and the other two studies [20,21] ‘moderate.’ Continuous predictors were sometimes categorized without this being part of the research question [17,19], and missing values were often not clearly reported and only complete cases seem to have been included [18–20,22]. Furthermore, the number of events per variable was too low in one study , and unclear in another . Most notably, as mentioned before, none of the studies accounted for overfitting and optimism, that is, no internal validation and/or external validation was performed [17–22].
We identified six studies meeting the inclusion and exclusion criteria, in which a total of 18 eligible models were presented. These models predict, in childhood, the risk of hypertension in later life, mostly in adulthood. However, no studies were identified that aimed to translate the prediction model into practice, which is reflected in the results of our systematic review. First of all, in each study, multiple combinations of predictors were investigated (e.g. age-specific models and models comparing categorized and continuous predictors), without presenting one final prediction model as the one most optimal for use in clinical practice, in a format that could be applied by others. Related to this, in the majority of the included studies, the development of models was based on a few predictors chosen a priori, such as sex, BMI, overweight or earlier (high) blood pressure. In fact, only one study considered a large number of candidate predictors in order to find the best combination for the prediction model . For clinical or public health practice, it will possibly be more useful to have one optimal prediction model incorporating information on all of the most relevant predictors. Secondly, we saw that only two out of six studies presented prediction models in a format that would allow for application in new individuals [17,18], and that in one study, a predictor was part of the models that would not be available at the intended time point of use . Finally, the lack of attention for the application in practice might also explain the limited information on performance for most models, as well as the lack of validation. Our results, therefore, underline the need for further development and validation studies of childhood prediction models for future hypertension, in order to pave the way for early targeted primordial prevention.
The completeness of reporting varied across the included studies. Inadequate reporting makes it more difficult to assess the quality of the prediction models, and to draw conclusions about the reliability, validity and generalizability of the prediction models . We identified the following important aspects related to reporting and to model development that deserve attention. First of all, sample selection procedures were not always described clearly, and most studies dealt with loss to follow-up whereas failing to report on differences in key characteristics between the sample for analysis and the participants that were lost to follow-up. Secondly, all studies had some limitations with regard to the outcome assessment, such as different follow-up durations for participants, and details on blinding the outcome assessment were often lacking. With regard to the quality of model development, in most studies some aspects could be improved, such as the use of continuous variables, the handling of missing values and the prevention of overfitting (e.g. having an adequate number of events per variable, and not performing univariable preselection). Categorization of continuous variables can lead to a loss of predictive information, and should ideally be avoided . Performing a complete case analysis instead of imputing missing values, might lead to a loss of statistical power and to incorrect estimates of the predictive performance of the model and the predictors .
For most models, the information about the performance was very limited, making it difficult to evaluate and compare the capability of the models to predict high blood pressure in later life. A direct comparison was also difficult because the prediction models in the different studies were targeted at different age groups, and the age at outcome assessment also varied widely. Furthermore, as none of the models was internally or externally validated, and it cannot be determined how optimistic the presented performance is, that is, how well these models would perform in slightly different populations. The AUCs reported in two studies (AUC 0.71–0.74) showed that discrimination between children who did and who did not have hypertension in adulthood was reasonable [19,20]. These results support the idea that the development of a reasonably performing prediction model for high blood pressure might be possible.
A strength of this review is that we applied a comprehensive search strategy in both Medline and Embase, two databases that together cover the majority of the medical scientific literature. By hand searching the reference lists of relevant reviews and the included studies, no new studies were identified. We, therefore, consider it unlikely that we have missed a relevant prediction modelling study for this topic. Nevertheless, it should be noted that we restricted our search to English publications, and although we did not identify relevant non-English publications through hand searching, we cannot fully exclude that we might have missed studies on prediction models written in another language. Another strength is that we used the CHARMS checklist to systematically extract the data on key characteristics of the study population, the model development, and the final models. A limitation might be that, because of the large amount of references identified with the search strategy, title and abstract screening was primarily performed by one reviewer (M.H.). Two other reviewers (M.K. and Y.V.) together checked a random 10% sample, and as the agreement was over 99%, and the first reviewer (M.H.) was being more inclusive of titles/abstracts, we consider the selection to be adequate. Another limitation could be that during data extraction, the two reviewers M.H. and M.W. were not blinded for journal and author details. However, both M.H. and M.W. do not have any affiliations with any of the authors or journals and, therefore, we do not believe it to have had any effect on the data extraction and evaluation.
On the basis of the results of this review, we would recommend to perform additional analyses (including validation and/or adaptation) on the existing models in order to move towards implementation in practice, or the development of a new model to identify children at high risk of hypertension. When developing such a new model, it is important to carefully choose the target age (or age range) for application of the prediction model, the age for outcome assessment, and the candidate predictors. For the latter, it can be recommended to consider a broad range of candidate predictors (e.g. based on literature) if the sample size and the number of events allow for it. On the basis of the results of this systematic review, the following predictors can be considered relevant: weight status, blood pressure, parental hypertension, parental occupational status, sex, and age. With regard to weight status, in the studies both continuous measures (BMI) and categorized measures (overweight/obesity) were used [17,19–22]. As dichotomization of continuous predictors can lead to a loss of valuable information [23,26], using BMI instead of overweight/obesity is to be preferred. Moreover, in children it might be better to use standardized measures of BMI, such as z-scores or standard deviation scores relative to age and sex, because of normal changes in BMI that occur as children age . For blood pressure, similar considerations are important. For example, a continuous measure of blood pressure, especially SBP, was shown to be a more important predictor than childhood prehypertension, a dichotomized predictor . SBP might be more predictive of future hypertension than DBP [17,19,20]. The use of blood pressure standard deviation scores or percentiles (based on sex-specific and age-specific reference values) as a continuous variable might also be considered [18,28]. Parental occupational status was identified as a predictor in one of the studies , but other socioeconomic indicators such as parental educational level or income might also be relevant . Next, helpful methodological resources are available to further improve prediction model development [23,25]; these discuss appropriate model selection methods, dealing with missing values, and internal and external validation methods. Also, it can be recommended to report the methods and results with help of the ‘Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis’ (TRIPOD) statement and the TRIPOD explanation and elaboration document, in order to improve the quality and transparency of reporting . Lastly, it is important that prediction models are presented in such a way that they can be applied in new individuals by other researchers or healthcare professionals interested in the model. This will also allow for external validation (by others), which is essential to evaluate the generalizability and applicability of a prediction model in other populations than the study population.
In conclusion, several prediction models were identified that predict, in childhood, the risk of hypertension in later life, mostly in adulthood. Important predictors were weight status (BMI or overweight/obesity), current high blood pressure, SBP, DBP, sex, age, parental hypertension, and socioeconomic status. In general, the quality of reporting and model development was suboptimal, and as none of the identified models were validated, it is not possible to assess their value for practice and to recommend the use of any of these models. Because of the lack of validation, the reported estimates of the performance are likely to be too optimistic. The results of this review indicate that there is some potential for a prediction model for future hypertension based on multiple characteristics in childhood.
We would like to thank Wichor Bramer, biomedical information specialist at Erasmus Medical Center Rotterdam, for assisting with the development of the search strategy and for performing the search.
Statement of financial support: This study is part of larger project aiming to develop prediction and decision tools for childhood overweight and cardiometabolic risk factors, funded by The Netherlands Organization for Health Research and Development (ZonMw grant no. 200500006).
Conflicts of interest
There are no conflicts of interest.
1. Lim SS, Vos T, Flaxman AD, Danaei G, Shibuya K, Adair-Rohani H, et al. A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990–2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet
2. Expert panel on integrated guidelines for cardiovascular health and risk reduction in children and adolescents
. Expert panel on integrated guidelines for cardiovascular health and risk reduction in children and adolescents
: summary report. Pediatrics
2011; 128 (Suppl 5):S213–S256.
3. Berenson GS, Srinivasan SR, Bao W, Newman WP 3rd, Tracy RE, Wattigney WA. Association between multiple cardiovascular risk factors and atherosclerosis in children and young adults. The Bogalusa Heart Study. N Engl J Med
4. Strong JP, Malcom GT, McMahan CA, Tracy RE, Newman WP 3rd, Herderick EE, et al. Prevalence and extent of atherosclerosis in adolescents
and young adults: implications for prevention from the Pathobiological Determinants of Atherosclerosis in Youth Study. JAMA
5. Raitakari OT, Juonala M, Kahonen M, Taittonen L, Laitinen T, Maki-Torkko N, et al. Cardiovascular risk factors in childhood and carotid artery intima-media thickness in adulthood: the Cardiovascular Risk in Young Finns Study. JAMA
6. Hartiala O, Magnussen CG, Kajander S, Knuuti J, Ukkonen H, Saraste A, et al. Adolescence risk factors are predictive of coronary artery calcification at middle age: the cardiovascular risk in young Finns study. J Am Coll Cardiol
7. Din-Dzietham R, Liu Y, Bielo MV, Shamsa F. High blood pressure trends in children and adolescents
in national surveys, 1963 to 2002. Circulation
8. Weintraub WS, Daniels SR, Burke LE, Franklin BA, Goff DC Jr, Hayman LL, et al. American Heart Association Advocacy Coordinating Committee; Council on Cardiovascular Disease in the Young; Council on the Kidney in Cardiovascular Disease; Council on Epidemiology and Prevention; Council on Cardiovascular Nursing; Council on Arteriosclerosis; Thrombosis and Vascular Biology; Council on Clinical Cardiology, and Stroke Council. Value of primordial and primary prevention for cardiovascular disease: a policy statement from the American Heart Association. Circulation
9. Piepoli MF, Hoes AW, Agewall S, Albus C, Brotons C, et al. 2016 European Guidelines on cardiovascular disease prevention in clinical practice: the Sixth Joint Task Force of the European Society of Cardiology and Other Societies on Cardiovascular Disease Prevention in Clinical Practice (constituted by representatives of 10 societies and by invited experts) Developed with the special contribution of the European Association for Cardiovascular Prevention & Rehabilitation (EACPR). Atherosclerosis
10. Kavey RE, Daniels SR, Lauer RM, Atkins DL, Hayman LL, Taubert K, et al. American Heart Association guidelines for primary prevention of atherosclerotic cardiovascular disease beginning in childhood. Circulation
11. Chen X, Wang Y. Tracking of blood pressure from childhood to adulthood: a systematic review and meta-regression analysis. Circulation
12. Moher D, Liberati A, Tetzlaff J, Altman DG, Group P. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ
13. Flynn JT, Kaelber DC, Baker-Smith CM, Blowey D, Carroll AE, Daniels SR, et al. Clinical practice guideline for screening and management of high blood pressure in children and adolescents
14. Whelton PK, Carey RM, Aronow WS, Casey DE Jr, Collins KJ, Dennison Himmelfarb C, et al. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA guideline for the prevention, detection, evaluation, and management of high blood pressure in adults: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension
15. Moons KG, de Groot JA, Bouwmeester W, Vergouwe Y, Mallett S, Altman DG, et al. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist. PLoS Med
16. Smit HA, Pinart M, Anto JM, Keil T, Bousquet J, Carlsen KH, et al. Childhood asthma prediction models: a systematic review. Lancet Respir Med
17. Ferrer M, Fernandez-Britto JE, Bacallao J, Perez H. Development of hypertension in a cohort of Cuban adolescents
. MEDICC Rev
18. Juhola J, Magnussen CG, Viikari JS, Kahonen M, Hutri-Kahonen N, Jula A, et al. Tracking of serum lipid levels, blood pressure, and body mass index from childhood to adulthood: the Cardiovascular Risk in Young Finns Study. J Pediatr
19. Juhola J, Oikonen M, Magnussen CG, Mikkila V, Siitonen N, Jokinen E, et al. Childhood physical, environmental, and genetic predictors of adult hypertension: the cardiovascular risk in young Finns study. Circulation
20. Vos LE, Oren A, Bots ML, Gorissen WH, Grobbee DE, Uiterwaal CS. Does a routinely measured blood pressure in young adolescence accurately predict hypertension and total cardiovascular risk in young adulthood? J Hypertens
21. Li L, Pinot de Moira A, Power C. Predicting cardiovascular disease risk factors in midadulthood from childhood body mass index: utility of different cutoffs for childhood body mass index. Am J Clin Nutr
22. Sun SS, Grave GD, Siervogel RM, Pickoff AA, Arslanian SS, Daniels SR. Systolic blood pressure in childhood predicts hypertension and metabolic syndrome later in life. Pediatrics
23. Steyerberg EW, Vergouwe Y. Towards better clinical prediction models: seven steps for development and an ABCD for validation. Eur Heart J
24. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med
25. Royston P, Moons KG, Altman DG, Vergouwe Y. Prognosis and prognostic research: developing a prognostic model. BMJ
26. Royston P, Altman DG, Sauerbrei W. Dichotomizing continuous predictors in multiple regression: a bad idea. Stat Med
27. Giussani M, Antolini L, Brambilla P, Pagani M, Zuccotti G, Valsecchi MG, et al. Cardiovascular risk assessment in children: role of physical activity, family history and parental smoking on BMI and blood pressure. J Hypertens
28. Miersch A, Vogel M, Gausche R, Siekmeyer W, Pfaffle R, Dittrich K, et al. Blood pressure tracking in children and adolescents
. Pediatr Nephrol
29. Shi Y, de Groh M, Morrison H. Increasing blood pressure and its associated factors in Canadian children and adolescents
from the Canadian Health Measures Survey. BMC Public Health
30. National High Blood Pressure Education Program Working Group on High Blood Pressure in Children and Adolescents
. The fourth report on the diagnosis, evaluation, and treatment of high blood pressure in children and adolescents
2004; 114 (2 Suppl 4th Report):555–576.
adolescents; blood pressure, children; hypertension, prediction; review, risk assessment
Supplemental Digital Content
Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.