Protein is an important component of infant and child nutrition, as it provides essential amino acids that are required for growth( Reference Heinig, Nommsen and Peerson 1 , Reference Axelsson, Ivarsson and Raiha 2 ). However, a high intake of protein in early childhood has also been associated with the development of obesity( Reference Rolland-Cachera, Deheeger and Akrout 3 – Reference Weber, Grote and Closa-Monasterolo 5 ). Already in childhood, obesity can lead to adverse cardiometabolic health outcomes such as hypertension, high cholesterol levels and insulin resistance( Reference Weiss, Dziura and Burgert 6 , Reference l'Allemand-Jander 7 ). This suggests that high protein intake in children may lead to unfavourable effects on these outcomes. However, studies in adults have reported beneficial effects of protein intake on blood pressure (BP), and insulin and TAG levels( Reference Tielemans, Altorf-van der Kuil and Engberink 8 – Reference Westerterp-Plantenga, Lemmens and Westerterp 11 ).
It is unclear whether dietary protein intake has an effect on BP, insulin sensitivity or blood lipids in children. Since cardiometabolic risk factors in childhood continue into later life and have been shown to predict CVD and type 2 diabetes in adulthood( Reference Morrison, Friedman and Gray-McGuire 12 , Reference Morrison, Friedman and Wang 13 ), it is important to study the determinants of cardiometabolic risk already in childhood. Therefore, we aimed to conduct a systematic review on the associations of protein intake with BP, insulin sensitivity and blood lipids in children. In addition, we aimed to explore whether the reported effects differ between vegetable and animal protein intakes.
Methods
The present systematic review was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement( Reference Liberati, Altman and Tetzlaff 14 ). Ethical approval was not required as this was a secondary data analysis.
Literature search
An extensive literature search was conducted with the help of a medical librarian in the databases Medline (via OvidSP), Embase (via Embase.com) and the Cochrane Library. In addition, we searched PubMed for articles that were not yet available via Medline. All databases were searched from their inception until 31 May 2013. The search strategy consisted of three elements: infants, children or adolescents; protein intake; cardiovascular or metabolic health outcomes. To capture studies that did not explicitly mention protein in the title or abstract, we also included general terms referring to diet and nutrient intake. All elements were searched using both controlled vocabulary terms (Medical Subject Headings and/or Emtree) and free text words in the title or abstract. Limits were applied to include only human studies and to exclude conference papers, editorials and letters. No limits were set on language or year of publication. The full search strategies for all the four databases are provided in online Supplementary Table S1. In addition to the systematic search, we contacted authors and searched reference lists for the most recent 10 % of articles in our review.
Study eligibility criteria
Studies identified from the literature search were selected on the basis of the predefined selection criteria presented below.
Inclusion criteria were as follows:
-
1 Cross-sectional studies, case–control studies, cohort studies and intervention studies
-
2 Studies conducted among children ≤ 18 years old
-
3 Studies reporting total, animal and/or vegetable protein intake, either in absolute amounts (e.g. g/d or kJ/d) and/or relative to total energy intake (e.g. percentage of energy (E%))
-
4 Studies investigating the associations between protein intake and one or more of the following outcomes:
-
a. BP: systolic or diastolic BP; mean arterial pressure; hypertension
-
b. Insulin sensitivity: insulin levels; glucose levels; glucose tolerance; homeostatic model assessment of insulin resistance (HOMA-IR); type 2 diabetes mellitus
-
c. Blood lipids: total cholesterol levels; HDL-cholesterol levels; LDL-cholesterol levels; TAG levels
-
Exclusion criteria were as follows:
-
1 Studies in children with congenital diseases, phenylketonuria, type 1 diabetes or kidney disease
-
2 Studies from which the exclusive effects of protein intake cannot be extracted (e.g. when protein supplements were combined with other nutrients without proper control)
-
3 Letters, conference abstracts, reviews or editorials
-
4 Studies not conducted in human subjects
Study selection
Working in pairs, two authors independently reviewed the titles and abstracts to determine whether the studies satisfied the selection criteria. Any disagreement with article selection was resolved through discussion or with the help of a third reviewer. Full-text articles were retrieved for the selected titles and were assessed again by two independent reviewers. For articles in languages other than English, colleagues fluent in the language assisted with translating.
Data extraction
Data were extracted using a structured data extraction form designed before data collection. Detailed study-level characteristics were collected including study design, study size, study duration, details on exposure and outcome assessment, and characteristics of the study population. We also derived information on the statistical analyses and covariate adjustments. All types of summary statistics were extracted, both for the entire study population and for subgroups; and both for crude models and for adjusted models where applicable. Authors were contacted if insufficient data were published (e.g. if effect estimates were not stated in the paper). A second reviewer checked the data extraction for a random 20 % of the studies.
Quality assessment
Using a predefined scoring system, two reviewers independently evaluated the quality of the included studies. The quality score was developed on the basis of previously used scoring systems and was modified to assess the quality of studies with different study designs( Reference Carter, Gray and Troughton 15 , 16 ). A score of 0, 1 or 2 points was allocated to each of the following five items: study design; study size; exposure assessment; outcome assessment; and adjustment for potential confounders or randomisation. This resulted in a total score ranging from 0 to 10 points, with a score of 10 representing the highest quality. Details on the quality score are presented in online Supplementary Table S2.
Synthesis of evidence
Because of the diversity in study designs, outcome measures and low methodological quality of the studies, a meta-analysis could not be performed. Instead, a qualitative analysis (best-evidence synthesis) was performed to synthesise the results and quality of the included studies( Reference Slavin 17 ). In line with previous systematic reviews, we defined the following four levels of evidence( Reference Singh, Uijtdewilligen and Twisk 18 , Reference Bruijn, Locher and Passchier 19 ): (1) strong evidence is provided if at least two higher-quality studies are available and if these report consistent findings; (2) moderate evidence is provided if one higher-quality study and one or more lower-quality studies are available with generally consistent findings; (3) limited evidence is provided if only one higher-quality study is available or if multiple lower-quality studies report generally consistent findings; (4) insufficient evidence is provided if no higher-quality studies are available or if studies report inconsistent findings. Studies were considered as generally consistent if at least 75 % of the studies showed statistically significant results in the same direction( Reference Slavin 17 ). Studies were considered as higher quality if they had a quality score of 6 or higher. If two or more studies on the same outcome were of higher quality, we disregarded lower-quality studies in the evidence synthesis( Reference Slavin 17 ).
Results
Study selection
In the systematic search, 6636 unique references were identified (Fig. 1). Of these references, 6305 were excluded on the basis of the title and abstract. For the remaining 331 articles, the full text was retrieved and critically reviewed. After the selection process, sixty papers were included, reporting on unique fifty-six studies. Fig. 1 shows the flow chart of the selection process.
Fig. 1 Flow chart of the study selection.
Characteristics of the included studies
Table 1 shows the characteristics of the included studies and study populations. The fifty-six studies included a total number of 22 040 participants (n 19–4508 subjects per study), with a variation in mean age from 0 to 17·5 years. We decided to include three studies (published in four papers) performed in subjects with an age range up to 19 or 20 years as their age ranges were very wide( Reference Glueck, Waldman and McClish 39 , Reference Morrison, Larsen and Glatfelter 55 , Reference Sarría Chueca, Martín Nasarre de Letosa and Lomba García 65 , Reference Sugiyama, Xie and Graham-Maar 73 ). Most studies included both boys and girls, except for one study in boys only( Reference Knuiman, Westenbrink and van der Heyden 48 ) and one study in girls only( Reference Heyman, Berthon and Youssef 42 ). Most studies were performed in Europe( Reference Aeberli, Zimmermann and Molinari 20 – Reference Andersen, Lifschitz and Friis-Hansen 23 , Reference Cowin and Emmett 29 , Reference Garemo, Palsdottir and Strandvik 36 , Reference Gately, King and Greatwood 38 , Reference Gonzalez-Requejo, Sanchez-Bayle and Baeza 40 , Reference Heyman, Berthon and Youssef 42 , Reference Kouvalainen, Uhari and Akerblom 49 , Reference Larsen, Roed and Ibsen 50 , Reference Lucas and Morley 52 , Reference Menghetti, D'Addesa and Censi 53 , Reference Obuchowicz and Obuchowicz 57 , Reference Pistulkova, Pisa and Poledne 59 , Reference Räsänen, Wilska and Kantero 61 , Reference Sanchez-Bayle, Gonzalez-Requejo and Pelaez 64 , Reference Sanchez-Bayle, Gonzalez-Requejo and Pelaez 65 , Reference Ulbak, Lauritzen and Hansen 75 ), the USA or Canada( Reference Berenson, Blonde and Farris 24 Reference Casazza, Dulin-Keita and Gower 26 Reference Casazza, Dulin-Keita and Gower 27 Reference Davis, Ventura and Weigensberg 31 Reference Davis, Alexander and Ventura 32 Reference Frank, Voors and Schilling 34 Reference Frank, Berenson and Webber 35 Reference Glueck, Waldman and McClish 39 Reference Lindquist, Gower and Goran 51 Reference Morrison, Larsen and Glatfelter 55 Reference Nicklas, Webber and Srinivasan 56 Reference Perry, Tremblay and Signorile 58 Reference Potter and Kies 60 Reference Schachter, Kuller and Perkins 66 Reference Sharma, Roberts and Hudes 69 – Reference Suter and Hawes 74 Reference Weidman, Elveback and Nelson 76 ), and Australia or New Zealand( Reference Boulton, Magarey and Cockington 25 , Reference Garnett, Gow and Ho 37 , Reference Hitchcock and Gracey 43 , Reference Jenner, English and Vandongen 46 , Reference Regan, Cutfield and Jefferies 62 ). Other studies were performed in populations in South or Central America( Reference Colin-Ramirez, Castillo-Martinez and Orea-Tejeda 28 , Reference Hermelo, Borroto and Bacallao 41 , Reference Rinaldi, de Oliveira and Moreto 63 ), South Africa( Reference Mia and Vorster 54 , Reference Schutte, Van Rooyen and Huisman 67 , Reference Schutte, Van Rooyen and Huisman 68 ), Korea( Reference Hong, Kim and Kang 44 ), Russia( Reference Il'chenko, Tubol and Dorofeeva 45 ) and Turkey( Reference Keser, Yucecan and Cizmecioglu 47 ), and one study included subjects from the Philippines, Ghana and three European countries( Reference Knuiman, Westenbrink and van der Heyden 48 ). Most studies examined general population-based samples of children. Some studies specifically included high-risk populations, i.e. children with high cholesterol levels( Reference Hitchcock and Gracey 43 , Reference Pistulkova, Pisa and Poledne 59 , Reference Sarría Chueca, Martín Nasarre de Letosa and Lomba García 65 , Reference Starc, Shea and Cohn 72 , Reference Vobecky, Vobecky and Shapcott 77 , Reference Weidman, Elveback and Nelson 79 , Reference Simons-Morton and Obarzanek 80 ) or overweight children( Reference Davis, Ventura and Weigensberg 31 – Reference Duckworth, Gately and Radley 33 , Reference Garnett, Gow and Ho 37 , Reference Gately, King and Greatwood 38 , Reference Hermelo, Borroto and Bacallao 41 , Reference Keser, Yucecan and Cizmecioglu 47 , Reference Obuchowicz and Obuchowicz 57 , Reference Rinaldi, de Oliveira and Moreto 63 , Reference Smith and Rinderknecht 71 , Reference Ventura, Davis and Alexander 76 ).
Table 1 Characteristics of the sixty included papers (fifty-six studies) on protein intake and cardiometabolic health outcomes in children
NR, not reported; SES, socio-economic status; T2DM, type 2 diabetes mellitus; T1DM, type 1 diabetes mellitus; LDL-C, LDL-cholesterol.
* Articles with the same superscript letter used data from the same study population.
† All case–control studies used cross-sectional data.
‡ Denmark, the Netherlands, UK, Greece, Germany, Spain, Bulgaria and the Czech Republic.
Only four of the fifty-six studies were randomised controlled trials comparing high-protein (22·5–30 E%) with low-protein (10–15 E%) diets, for 1 or 6 months( Reference Damsgaard, Papadaki and Jensen 30 , Reference Duckworth, Gately and Radley 33 , Reference Garnett, Gow and Ho 37 , Reference Gately, King and Greatwood 38 ). In three of the four trials, lower- and higher-protein diets were isoenergetic and energy-restricted, with energy from protein being replaced by carbohydrate( Reference Duckworth, Gately and Radley 33 , Reference Garnett, Gow and Ho 37 , Reference Gately, King and Greatwood 38 ). In the fourth trial, protein was also replaced by carbohydrate, but without energy restriction( Reference Damsgaard, Papadaki and Jensen 30 ). The remaining fifty-two studies were observational, of which five had a longitudinal design (follow-up 1·1–7·5 years) and forty-seven were cross-sectional. The mean protein intake in the observational studies ranged from 7·7 to 19·2 E% (see online Supplementary Table S3). Of the included studies, twenty-three investigated the associations of protein intake with BP, fifteen with insulin sensitivity and forty-two with blood lipids. Details on exposure and outcome measurements are presented in online Supplementary Table S3.
The overall quality score of the included studies ranged from 1 to 9 (Table 1), with a mean score of 4·2. Of these studies, fifteen received a quality score of 6 or higher. Because of the large number of cross-sectional studies, most studies scored low on the item study design. Most studies did receive a high score on exposure and outcome assessment methods. The majority of the studies received a score of zero for the item on adjustments, since they did not control for important potential confounders such as age, sex, energy intake and body weight.
Protein intake and blood pressure
Overall, twenty-three studies reported on the associations between protein intake and BP in children, of which ten studies were considered higher quality (Table 2).
Table 2 Reported associations between protein intake and blood pressure (BP) in children
MAP, mean arterial pressure; SBP, systolic blood pressure; NR, not reported; DBP, diastolic blood pressure; SES, socio-economic status; M, male; F, female; E%, percentage of energy.
* The intensive intervention group received most of their foods for free and had a higher adherence to the intervention diets.
† See the column ‘Covariate adjustments’: model 2 additionally adjusted for intake of other nutrients.
‡ Statistical analysis used was not clearly reported.
In four intervention studies, of which three were performed in overweight children, no significant effects of a high-protein diet compared with a low-protein diet on systolic BP, diastolic BP or mean arterial pressure were found( Reference Damsgaard, Papadaki and Jensen 30 , Reference Duckworth, Gately and Radley 33 , Reference Garnett, Gow and Ho 37 , Reference Gately, King and Greatwood 38 ). All the four trials had a quality score of 6 or higher. In the trial by Damsgaard et al. ( Reference Damsgaard, Papadaki and Jensen 30 ), the authors did observe a significant beneficial effect of a high-protein diet in a subgroup of 5- to 18-year-olds who received a more intensive intervention. This subgroup received free foods in addition to dietary instructions and had a higher adherence to the intervention diet. In this subgroup, children who received the high-protein diet had a 1·0 mmHg (95 % CI 0·3, 1·7) lower diastolic BP and a 6·5 mmHg (95 % CI 1·5, 15·0) lower mean arterial pressure than those who received a low-protein diet( Reference Damsgaard, Papadaki and Jensen 30 ).
Of the observational studies, six had a quality score of 6 or higher. One longitudinal and two cross-sectional studies of higher quality reported inverse associations between protein intake and BP in at least one of their subgroups( Reference Jenner, English and Vandongen 46 , Reference Simons-Morton, Hunsberger and Van Horn 70 , Reference Ulbak, Lauritzen and Hansen 75 ). The three remaining higher-quality studies did not find a significant association between protein intake and BP( Reference Casazza, Dulin-Keita and Gower 26 , Reference Sharma, Roberts and Hudes 69 , Reference Sugiyama, Xie and Graham-Maar 73 ), but two did report non-significant inverse associations( Reference Casazza, Dulin-Keita and Gower 26 , Reference Sugiyama, Xie and Graham-Maar 73 ). In contrast, one study with a quality score of 3 observed that children with high BP had a significantly higher intake of protein than those with normal BP( Reference Menghetti, D'Addesa and Censi 53 ). The remaining lower-quality studies showed no significant associations( Reference Aeberli, Spinas and Lehmann 21 , Reference Berenson, Blonde and Farris 24 , Reference Colin-Ramirez, Castillo-Martinez and Orea-Tejeda 28 , Reference Frank, Voors and Schilling 34 , Reference Frank, Berenson and Webber 35 , Reference Hong, Kim and Kang 44 , Reference Il'chenko, Tubol and Dorofeeva 45 , Reference Lucas and Morley 52 , Reference Schachter, Kuller and Perkins 66 , Reference Schutte, Van Rooyen and Huisman 68 , Reference Smith and Rinderknecht 71 , Reference Ventura, Davis and Alexander 76 ).
Protein intake and insulin sensitivity
Of the included studies, fifteen (published in sixteen papers) examined the associations between protein intake and measures of insulin sensitivity in children (Table 3). The studies examined various measures of insulin sensitivity, including fasting insulin or glucose levels, HOMA-IR; or measures of insulin responses following an oral or intravenous glucose tolerance test, such as the insulin sensitivity index and acute insulin response. No studies were identified with type 2 diabetes mellitus in children as outcome.
Table 3 Reported associations between protein intake and measures of insulin sensitivity in children
HOMA-IR, homeostatic model assessment of insulin resistance; NR, not reported; E%, percentage of energy; SDS, standard deviation score; SES, socio-economic status; Si, insulin sensitivity index; AIR, acute insulin response; QUICKI, quantitative insulin sensitivity check index; HOMA-β, homeostatic model assessment of β-cell function.
* The intensive intervention group received most of their foods for free and had a higher adherence to the intervention diets.
Of the fifteen studies, six had a quality score of 6 or higher. In three of these higher-quality studies, a significant association was reported between protein intake and measures of insulin sensitivity. In the previously described trial by Damsgaard et al. ( Reference Damsgaard, Papadaki and Jensen 30 ), again, no effects were observed in the full study population. However, in the subgroup that underwent a more intensive intervention, insulin levels and HOMA-IR were significantly reduced in the high-protein group compared with the low-protein group( Reference Damsgaard, Papadaki and Jensen 30 ). In a cross-sectional study in 7- to 12-year-old children, Casazza et al. ( Reference Casazza, Dulin-Keita and Gower 26 , Reference Casazza, Dulin-Keita and Gower 27 ) observed that, after extensive adjustments, higher protein intake was significantly associated with lower fasting glucose levels, but not with fasting insulin levels or the insulin sensitivity index, and that it was inversely associated with acute insulin response. Data from another cross-sectional study showed that higher protein intake was associated with lower insulin resistance (HOMA-IR) in 9- to 11-year-old children( Reference Sharma, Roberts and Hudes 69 ). In summary, all the three studies show a beneficial effect of protein intake on one or more measures of insulin sensitivity (i.e. lower glucose levels and lower insulin resistance), but one of the studies also reported a harmful effect of protein intake (lower insulin response). The remaining three higher-quality studies did not find significant associations between dietary protein intake and measures of insulin sensitivity( Reference Davis, Ventura and Weigensberg 31 , Reference Davis, Alexander and Ventura 32 , Reference Garnett, Gow and Ho 37 ), neither did any of the lower-quality studies( Reference Aeberli, Spinas and Lehmann 21 , Reference Garemo, Palsdottir and Strandvik 36 , Reference Heyman, Berthon and Youssef 42 , Reference Hong, Kim and Kang 44 , Reference Il'chenko, Tubol and Dorofeeva 45 , Reference Keser, Yucecan and Cizmecioglu 47 , Reference Lindquist, Gower and Goran 51 , Reference Obuchowicz and Obuchowicz 57 , Reference Regan, Cutfield and Jefferies 62 , Reference Ventura, Davis and Alexander 76 ).
Protein intake and blood lipids
Of the fifty-six included studies, forty-two reported on associations between protein intake and levels of one or more blood lipid parameters in children (Table 4). Among these, twenty-two studies (published in twenty-three papers) investigated effects on TAG levels, of which five were of higher quality. None of these higher-quality studies reported a significant effect( Reference Damsgaard, Papadaki and Jensen 30 , Reference Garnett, Gow and Ho 37 , Reference Gately, King and Greatwood 38 , Reference Rinaldi, de Oliveira and Moreto 63 , Reference Sharma, Roberts and Hudes 69 ). One lower-quality study reported that higher protein intake was correlated with higher TAG levels( Reference Keser, Yucecan and Cizmecioglu 47 ). However, this correlation was not adjusted for potential confounders and the study had a quality score of only 2. The remaining sixteen lower-quality studies did not find significant associations( Reference Berenson, Blonde and Farris 24 , Reference Frank, Voors and Schilling 34 – Reference Garemo, Palsdottir and Strandvik 36 , Reference Glueck, Waldman and McClish 39 , Reference Hermelo, Borroto and Bacallao 41 , Reference Hong, Kim and Kang 44 , Reference Il'chenko, Tubol and Dorofeeva 45 , Reference Lindquist, Gower and Goran 51 , Reference Mia and Vorster 54 , Reference Morrison, Larsen and Glatfelter 55 , Reference Perry, Tremblay and Signorile 58 , Reference Potter and Kies 60 , Reference Sanchez-Bayle, Gonzalez-Requejo and Pelaez 64 , Reference Suter and Hawes 74 , Reference Ventura, Davis and Alexander 76 ).
Table 4 Reported associations between protein intake and blood lipid levels in children
NR, not reported; TC, total cholesterol; LDL-C, LDL-cholesterol; HDL-C, HDL-cholesterol; E%, percentage of energy; M, male; F, female; SES, socio-economic status.
* The intensive intervention group received most of their foods for free and had a higher adherence to the intervention diets.
† Partly overlapping populations.
‡ Statistical analysis used was not clearly reported.
§ Matched for sex and age.
The relationship between protein intake and HDL-cholesterol levels was reported in twenty-four studies (published in twenty-five papers; Table 4). Statistically significant associations were observed in three studies: two positive( Reference Knuiman, Westenbrink and van der Heyden 48 , Reference Starc, Shea and Cohn 72 ) and one negative( Reference Keser, Yucecan and Cizmecioglu 47 ). However, all the three studies reported only simple correlations without adjustment for potential confounders and had quality scores ranging from 2 to 4. In the remaining studies, including four higher-quality studies, no significant associations were observed( Reference Akerblom, Viikari and Uhari 22 , Reference Boulton, Magarey and Cockington 25 , Reference Casazza, Dulin-Keita and Gower 27 , Reference Cowin and Emmett 29 , Reference Damsgaard, Papadaki and Jensen 30 , Reference Garnett, Gow and Ho 37 , Reference Glueck, Waldman and McClish 39 – Reference Heyman, Berthon and Youssef 42 , Reference Hong, Kim and Kang 44 , Reference Il'chenko, Tubol and Dorofeeva 45 , Reference Kouvalainen, Uhari and Akerblom 49 , Reference Morrison, Larsen and Glatfelter 55 , Reference Nicklas, Webber and Srinivasan 56 , Reference Potter and Kies 60 , Reference Rinaldi, de Oliveira and Moreto 63 , Reference Sarría Chueca, Martín Nasarre de Letosa and Lomba García 65 , Reference Sharma, Roberts and Hudes 69 , Reference Starc, Shea and Cohn 72 , Reference Suter and Hawes 74 , Reference Ventura, Davis and Alexander 76 ).
Associations between protein intake and total and/or LDL-cholesterol levels were investigated in thirty-eight studies (Table 4). Of these, five studies reported significant associations between protein intake and cholesterol levels; however, they all had a quality score of 5 or lower and there was no consistency in the direction of the effect( Reference Berenson, Blonde and Farris 24 , Reference Boulton, Magarey and Cockington 25 , Reference Mia and Vorster 54 , Reference Pistulkova, Pisa and Poledne 59 , Reference Vobecky, Vobecky and Shapcott 77 ). The remaining studies, including five higher-quality studies, did not find significant associations( Reference Aeberli, Zimmermann and Molinari 20 , Reference Akerblom, Viikari and Uhari 22 , Reference Andersen, Lifschitz and Friis-Hansen 23 , Reference Cowin and Emmett 29 , Reference Frank, Voors and Schilling 34 – Reference Garnett, Gow and Ho 37 , Reference Glueck, Waldman and McClish 39 – Reference Il'chenko, Tubol and Dorofeeva 45 , Reference Keser, Yucecan and Cizmecioglu 47 – Reference Larsen, Roed and Ibsen 50 , Reference Mia and Vorster 54 , Reference Morrison, Larsen and Glatfelter 55 , Reference Potter and Kies 60 , Reference Sanchez-Bayle, Gonzalez-Requejo and Pelaez 64 , Reference Sarría Chueca, Martín Nasarre de Letosa and Lomba García 65 , Reference Sharma, Roberts and Hudes 69 , Reference Suter and Hawes 74 , Reference Ward, Melin and Lloyd 78 , Reference Weidman, Elveback and Nelson 79 ).
Vegetable and animal protein
Only four of all the fifty-six studies included investigated the associations between vegetable or animal protein intake and cardiometabolic factors in children. One higher-quality cross-sectional study reported no associations between animal or vegetable protein intake and BP( Reference Schutte, Van Rooyen and Huisman 68 ). Furthermore, three lower-quality studies reported inconsistent results. The first study found no effect of animal or vegetable protein intake on blood lipid levels in overweight children( Reference Keser, Yucecan and Cizmecioglu 47 ). The second study reported that animal protein intake was positively correlated with serum LDL-cholesterol levels (r 0·20, P= 0·0002), whereas vegetable protein intake was not( Reference Mia and Vorster 54 ). In contrast, the last study reported an inverse correlation between the ratios of animal:vegetable protein intake and levels of LDL-cholesterol (r − 0·28, P= 0·05)( Reference Potter and Kies 60 ).
Discussion
To our knowledge, the present systematic review was the first to summarise the published literature on the effects of protein intake on BP, insulin sensitivity and blood lipids in children. In the present review, fifty-six studies (published in sixty papers) on the association between protein intake and one or more of these cardiometabolic outcomes in children were identified. Overall, the literature shows insufficient evidence for an effect of animal, vegetable or total protein intake on BP, insulin sensitivity or blood lipids, due to a lack of high-quality studies and inconsistency in the results.
Protein and cardiometabolic outcomes
Of the ten higher-quality studies that investigated the relationship between protein intake and BP in children, four reported inverse associations in one or more of their subgroups( Reference Damsgaard, Papadaki and Jensen 30 , Reference Jenner, English and Vandongen 46 , Reference Simons-Morton, Hunsberger and Van Horn 70 , Reference Ulbak, Lauritzen and Hansen 75 ), while the other studies found no significant effects. Although these results suggest a possible inverse association, there is insufficient evidence to draw meaningful conclusions. Also, the observed effect estimates for reductions in BP are small and not clinically relevant at an individual level. Nevertheless, they may be relevant at a population level. More studies are needed to verify whether protein intake is associated with BP already in childhood, and whether this effect tracks into adulthood. A possible inverse association between protein intake and BP is in line with meta-analyses of studies in adults( Reference Tielemans, Altorf-van der Kuil and Engberink 8 – Reference Rebholz, Friedman and Powers 10 ). The mechanisms underlying a beneficial effect of protein intake on BP have not yet been clarified( Reference Teunissen-Beekman, Dopheide and Geleijnse 81 ). Proposed pathways include the increased synthesis of cellular ion channels in response to protein intake; increased renal plasma flow and increased glomerular filtration rate; or the vasodilating effects of certain amino acids( Reference Anderson 82 ).
For the association between protein intake and measures of insulin sensitivity in children, three of the six higher-quality studies reported significant results, but in different directions. Therefore, we conclude that there is insufficient evidence for an association between protein intake and insulin sensitivity in children. Two systematic reviews of intervention studies in adults showed inverse associations between protein intake and fasting insulin levels, but no significant effect on fasting glucose levels( Reference Santesso, Akl and Bianchi 9 , Reference Schwingshackl and Hoffmann 83 ). The effects of protein intake on insulin sensitivity are suggested to act through the stimulation of insulin secretion( Reference Tremblay, Lavigne and Jacques 84 ).
Many studies assessed the associations between protein intake and blood lipids in children, but only five had a quality score of 6 or higher, and none of these higher-quality studies reported a significant effect. Meta-analyses of trials in adults also reported no significant effects of protein intake on LDL- or HDL-cholesterol levels( Reference Santesso, Akl and Bianchi 9 , Reference Schwingshackl and Hoffmann 83 ). For TAG levels, one meta-analysis reported that subjects following a higher-protein diet had lower TAG levels than those consuming a diet lower in protein( Reference Santesso, Akl and Bianchi 9 ), while another meta-analysis, which included only trials with a duration of more than a year, observed no effect( Reference Schwingshackl and Hoffmann 83 ).
In children, endpoint measures of cardiometabolic health can usually not be observed, since they occur later in life. Several studies in adults investigated the relationship of protein intake with CVD and type 2 diabetes. Findings from three cohorts were that total and animal, but not vegetable, protein intakes were associated with an increased risk of type 2 diabetes( Reference Sluijs, Beulens and van der 85 – Reference Song, Manson and Buring 87 ). In three large cohort studies in women, total protein intake was found to be inversely associated with the incidence of CHD( Reference Prentice, Huang and Kuller 88 ), stroke( Reference Larsson, Virtamo and Wolk 89 ) and IHD( Reference Hu, Stampfer and Manson 90 ). In contrast, data from two other large cohorts showed that diets high in protein were associated with an increased risk of CVD( Reference Kelemen, Kushi and Jacobs 91 , Reference Lagiou, Sandin and Lof 92 ). Finally, in a Dutch prospective cohort, a U-shaped association was observed between protein intake and cardiovascular events after 6·4 years of follow-up, with incidence rates being higher in subjects with low or high protein intake than in those with median protein intake levels( Reference Halbesma, Bakker and Jansen 93 ).
Types of protein
Studies on the effects of vegetable or animal protein intake on BP, insulin sensitivity and blood lipids in children are scarce, and the results are inconsistent. Studies in adults also reported inconsistent effects of either animal or vegetable protein intake on cardiometabolic risk factors. Meta-analyses of trials and prospective studies reported no differences between the effects of vegetable and animal protein intake on BP( Reference Tielemans, Altorf-van der Kuil and Engberink 8 , Reference Rebholz, Friedman and Powers 10 ). In contrast, prospective cohort studies reported that animal, but not vegetable, protein intake decreases the risk of stroke( Reference Larsson, Virtamo and Wolk 89 , Reference Iso, Stampfer and Manson 94 ), but increases the risk of diabetes( Reference Sluijs, Beulens and van der 85 , Reference Song, Manson and Buring 87 ). More studies, both in adult and child populations, are needed to elucidate the differential effects of animal and vegetable protein intake on cardiometabolic health.
Some of the inconsistencies in the results of the studies included in the present review may be explained by the type of dietary protein intake. If vegetable and animal protein intakes differently affect cardiometabolic risk factors and the ratio of vegetable:animal protein intake varies between populations, this might explain some of the discrepancies in the results for total protein intake. Furthermore, if vegetable and animal protein intakes affect outcomes in opposite directions, their effects might cancel each other out when studying total protein intake.
Not only the ratio of vegetable:animal protein intake, but also the main food sources and therefore the amino acid composition of protein might vary between populations. For example, a study in children reported a positive association of dairy protein intake with BMI and body fat, but no associations of meat or cereal protein intake with these outcomes( Reference Gunther, Remer and Kroke 95 ).
Besides the source and type of protein, also the total amount of protein consumed in the population might have affected the results. Mean protein intake in the studies included in the present review ranged from 8 to 19 E% (see online Supplementary Table S3). A potential relationship between total protein intake and cardiometabolic risk factors might not be linear( Reference Halbesma, Bakker and Jansen 93 ). The effects could therefore be different for various levels of protein intake and might not be identified when statistical approaches are used that assume a linear relationship.
Quality of the included studies
We applied a scoring system with a theoretical range from 0 to 10 to assess the quality of the included studies. Only fifteen studies were regarded as having a relatively high quality (score ≥ 6). Many studies received a low score for the items study design and adjustment for potential confounders. Most of the studies included in the present review (forty-seven of the fifty-six) were cross-sectional, only five were longitudinal and four were intervention studies. In observational studies, even after adjustment for multiple potential confounders, residual confounding may exist. Intakes of several other nutrients, for example, could be correlated not only with protein intake, but also with several other (unmeasured) determinants of cardiometabolic health such as exercise, BMI and dietary patterns. In many studies included in the present review, the results were not at all or not sufficiently adjusted for important confounding variables, which limits the validity of their results.
Important potential confounders or mediators in the association between protein intake and cardiometabolic health are energy intake and measures of obesity. Since protein is a macronutrient, protein intake is strongly associated with energy intake, but energy intake might also be associated with BP, insulin sensitivity and blood lipids. Therefore, not adjusting for energy intake could confound the results. In only twelve of the fifty-two observational studies included in the present review, energy intake was considered in some way (e.g. by expressing protein intake in E% or by adjusting for energy intake)( Reference Aeberli, Spinas and Lehmann 21 , Reference Casazza, Dulin-Keita and Gower 26 , Reference Davis, Ventura and Weigensberg 31 , Reference Davis, Alexander and Ventura 32 , Reference Jenner, English and Vandongen 46 , Reference Regan, Cutfield and Jefferies 62 , Reference Rinaldi, de Oliveira and Moreto 63 , Reference Schutte, Van Rooyen and Huisman 67 , Reference Sharma, Roberts and Hudes 69 , Reference Simons-Morton, Hunsberger and Van Horn 70 , Reference Sugiyama, Xie and Graham-Maar 73 , Reference Ulbak, Lauritzen and Hansen 75 ). Nevertheless, even for the studies that did control for energy intake, we cannot exclude that the observed associations of increased protein intake are, in fact, caused by a decreased intake in energy from carbohydrate or fat. In the four trials included in the present review, a higher protein level of diets was accomplished by lowering the percentage of energy from carbohydrate. The effects of these diets could thus also be ascribed to a decrease in carbohydrate intake rather than to an increase in protein intake.
An important potential mediator in the relationship between protein intake and cardiometabolic health is body weight or body fat. Protein intake has been positively linked to childhood obesity( Reference Rolland-Cachera, Deheeger and Akrout 3 – Reference Weber, Grote and Closa-Monasterolo 5 ), while in adults, it has been inversely associated with obesity( Reference Santesso, Akl and Bianchi 9 ). Since obesity is strongly associated with cardiometabolic health, it is interesting to investigate the association of protein intake with cardiometabolic outcomes both with and without adjustment for measures of body composition. However, in only nine of the fifty-two observational studies, the results were adjusted for a measure of body weight or composition( Reference Aeberli, Zimmermann and Molinari 20 , Reference Casazza, Dulin-Keita and Gower 27 , Reference Davis, Ventura and Weigensberg 31 , Reference Lindquist, Gower and Goran 51 , Reference Lucas and Morley 52 , Reference Schutte, Van Rooyen and Huisman 68 , Reference Sharma, Roberts and Hudes 69 , Reference Smith and Rinderknecht 71 , Reference Starc, Shea and Cohn 72 , Reference Arnberg, Larnkjaer and Michaelsen 96 ). We did not observe any trends after comparing the results from the studies included in the present review that did with those that did not adjust for measures of obesity. Of the included studies, ten were performed in overweight children only( Reference Davis, Ventura and Weigensberg 31 – Reference Duckworth, Gately and Radley 33 , Reference Garnett, Gow and Ho 37 , Reference Gately, King and Greatwood 38 , Reference Keser, Yucecan and Cizmecioglu 47 , Reference Rinaldi, de Oliveira and Moreto 63 , Reference Smith and Rinderknecht 71 , Reference Ventura, Davis and Alexander 76 ) and three studies included both overweight and normal-weight children( Reference Aeberli, Spinas and Lehmann 21 , Reference Hermelo, Borroto and Bacallao 41 , Reference Obuchowicz and Obuchowicz 57 ). The latter three studies reported no clear differences in the associations between protein intake and insulin sensitivity or blood lipid levels among the overweight v. the normal-weight group.
Only four intervention studies met the selection criteria for the present review, of which two were short term (29 and 31 d)( Reference Duckworth, Gately and Radley 33 , Reference Gately, King and Greatwood 38 ), limiting the ability to observe an effect. The other two trials had a duration of 6 months, but consisted of dietary advice only( Reference Damsgaard, Papadaki and Jensen 30 , Reference Garnett, Gow and Ho 37 ). In one of these trials, the actual protein intake did not even differ significantly between the two groups among participants who received dietary instructions only (18·6 (sem1·3) v. 17·6 (sem1·3) E%, P= 0·31)( Reference Damsgaard, Papadaki and Jensen 30 ). However, the latter trial also included a subgroup that received free foods and had a higher adherence to the intervention diet (23·7 (sem 1·4) v. 16·9 (sem 1·3) E% protein; P= 0·001) . In this more intensive treatment group, beneficial effects of protein intake on BP and insulin sensitivity were observed, whereas in the total group, no differences were observed( Reference Damsgaard, Papadaki and Jensen 30 ).
Strengths and limitations of the present review
The main strength of the present review is that it gives a comprehensive overview of the currently available evidence on the effects of dietary protein on BP, insulin sensitivity and blood lipids in children. A very extensive literature search in multiple databases was used to identify articles. We aimed to reduce the problem of publication bias by also searching for publications that did not explicitly mention protein intake in their title or abstract, and by contacting authors to identify unpublished studies. Studies were independently screened and data extracted by two reviewers using a predefined and meticulous procedure. We assessed the quality of the included studies with a scoring system in order to more objectively distinguish between the higher- and lower-quality studies. Unfortunately, many of the included studies were of relatively low quality. This made it difficult to draw conclusions regarding the absence or presence of the association evaluated. A meta-analysis was not possible due to the large levels of heterogeneity in study design, outcomes and age range of the children. Moreover, we were limited by a lack of reported effect estimates. Therefore, we conducted only a qualitative synthesis of evidence, in which we took into account the quality score of the included studies and the consistency of the reported results.
Conclusions
The fifty-six studies included in the present systematic review provide insufficient evidence for effects of protein intake on BP, insulin sensitivity or blood lipids in children. Although a substantial number of studies addressed these associations, data from high-quality studies investigating the independent effects of protein intake are scarce. Results from the few high-quality studies were not consistent. Further research of high methodological quality is needed to understand the effects of protein intake on cardiometabolic health in children. Specifically, in order to investigate the independent effects of protein intake, future studies should take into account important potential confounding factors such as total energy intake and other dietary factors, and measures of body weight. A better evaluation of the effect of protein intake on cardiometabolic outcomes in children is important since cardiometabolic risk factors in childhood have been shown to predict CVD and type 2 diabetes in adulthood. Therefore, insight into early-life determinants of cardiometabolic risk may contribute to the prevention of CVD and type 2 diabetes in adulthood.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0007114514003699
Acknowledgements
T. V., A. V., C. L. A., P. K. B., A. B.-L., J. F. F., E. T. M. L., A. S., S. S., A. T., O. H. F. and E. H. v. d. H. work in ErasmusAGE, a centre for ageing research across the life course funded by Nestlé Nutrition (Nestec Limited), Metagenics Inc. and AXA. The funders had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; and preparation, review or approval of the manuscript.
The authors' contributions are as follows: T. V., E. H. v. d. H. and O. H. F. designed the research; W. M. B. and T. V. designed and conducted the literature search; T. V., A. V., C. L. A., P. K. B., A. B.-L., J. F. F., E. T. M. L., A. S., S. S., A. T., O. H. F. and E. H. v. d. H. conducted the study selection; T. V. and E. H. v. d. H. analysed the results; T. V., O. H. F. and E. H. v. d. H. wrote the paper; E. H. v. d. H. had primary responsibility for the final content. All authors critically reviewed and approved the final manuscript.
The authors had no conflicts of interest.
