A systematic review and meta-analysis of carbohydrate benefits associated with randomized controlled competition-based performance trials

Data of the original search by Colombani et al. [11] were used as starting point. The authors performed a search in “PubMed” up to September 3, 2011 using the following combination of key words: (Exercise OR Sport OR Athlete OR Athletes) AND (Hydration OR Water OR Fluid OR Drink OR Drinks OR Beverage OR Beverages OR Glycogen OR Loading OR Carbo OR Carbohydrate OR Carbohydrates OR Glucose OR Fructose OR Maltodextrin) NOT (Mice OR Mouse OR Pig OR Pigs OR Rat OR Rats OR Horse OR Horses OR Fish OR Dog OR Dogs OR Patient OR Patients OR Disease OR Diseases OR Diabetes OR Obesity OR Obese OR “Cord injury” OR “Wheelchair).

In addition to the systematic search of Colombani and coworkers [11] we searched the electronic databases “Embase” as well as the “Cochrane Central Register Of Controlled Trials” up to February, 2016 and expanded the search in “PubMed” starting September 4^th, 2011 to February, 2016 using the same combinations of search terms with the following exceptions: we used “Human” and “Adult <18 to64 years” as further limitations in the database “Embase”. Hand search was done using the reference lists of two meta-analyses [7, 8], yielding one additional article suitable for this systematic review [13].

In accordance to Colombani et al. [11], the following inclusion criteria were defined:

To yield more homogeneous study designs it was necessary to categorize the studies by defining comparable interventions prior to statistical analysis. Classification of groups was performed according to test mode (cycling, running, soccer), carbohydrate intervention (carboloading vs. no carboloading; carbohydrate mouth rinse vs. placebo mouth rinse; ingestion of carbohydrate containing drinks vs. drinks containing no carbohydrates), type of intervention (TT or S + TT), and outcome [performance as time needed to cover a fixed distance (or a set amount of work); distance covered within a fixed time, or power accomplished within a fixed time (or fixed distance)]. Taken together, this resulted in the following classification of groups:

Furthermore, subgroups were formed in order to address two other research questions:

In the scientific literature, the different mechanisms for ergogenic effects of carbohydrates with respect to short and long lasting physical exercise was explained to be due to different carbohydrate availability. For exercise durations lasting ≤ 90 min there should be sufficient substrate without power loss given the condition of regularly filled glycogen stores [14–16]. Therefore we compared exercise durations ≤ and > than 90 min.

Data were analyzed using the Review Manager 5.3 software provided by the Cochrane Collaboration (http://tech.cochrane.org/revman↗). Differences in means were compared for outlining possible differences between carbohydrates and placebo with a fixed-effect meta-analysis using the inverse-variance method. However, when heterogeneity exceeded the level of 50 %, the random-effects model was used. The Cochrane Collaboration suggests to use meta-analyses in order to synthesize evidence from multiple experiments addressing the same research questions. Checking consistency of the results is of major importance in meta-analyses. Statistical heterogeneity in studies is characterized by 95 % CI that show poor overlap. We used the I² statistic to detect heterogeneity [17]. If considerable heterogeneity is observed (I² ≥ 50 %), fixed-effect models should be avoided, since they underperform in that context. Random effects models provide a more conservative approach yielding better estimates [18].

Descriptive data of included trials are given as mean ± SD. Pooled estimates of the effects size obtained by either comprehensive or subgroup meta-analyses are reported as mean difference together with the 95 % confidence intervals, respectively. P-values < 0.05 were considered to be statistically significant. Moreover, effects sizes are given as standardized mean differences (SMD) for each analysis group as Additional files 1, 2, 3 and 4 (see corresponding Result section).

In the original literature search by Colombani et al. [11] performed in the electronic database Pubmed until September 3^rd, 2011, 16,658 articles were identified. Our own updated search for literature yielded 15,105 articles (4,136 articles from PubMed published between September 4^th, 2011 and February, 2016, 2,916 articles from Cochrane Central Register Of Controlled Trials, and 8,053 from Embase, respectively). Articles which contained sufficient information in the title or abstract to identify them as not eligible were discarded, if this was not the case, the full text was consulted. Furthermore, 12 reviews [3–6, 9, 10, 19–24] concerning this topic were hand-searched for eligible studies, however no additional study fulfilling the search criteria was identified. In total, the full text of 205 articles was examined yielding 24 studies that met the inclusion criteria and are displayed in the systematic review (Tables 1 and 2). 16 of these articles provided enough information to allow for a quantitative evaluation. Steps of article search and selection are summarized as a flow chart in Fig. 1.

Subjects were male with the exception of two studies [29, 34] enrolling both genders. Sample size varied between seven and 16 volunteers, mean age ranged between 19 and 35 years, and mean VO_2max ranged between 50 and 71 ml/kg body weight/min.

For each of the four groups, results of both comprehensive as well as subgroup meta-analyses are given in Figs. 2, 3, 4 and 5, respectively. Please note that classification into subgroups was performed for every group independent of resulting numbers of studies.

Group 1 included six studies [34–39] with 13 interventions in total. Pooled estimates of the effects size for the effects of carbohydrate interventions as compared to placebo on time required to finish a TT are presented in Fig. 2 (forest plot showing pooled SMD is given as Additional file 1). Carbohydrate interventions were associated with a significantly lower amount of time [mean differences −0.9 min (95 % CI −1.7, −0.2), p = 0.02]. Following subgroup analyses, significant performance improvements remained only for those studies administering a concentration of carbohydrates between 6 and 8 % [MD = −1.0 min (95 % CI −1.9, −0.0), p = 0.04].

Group 2 included seven studies [40–46] with seven interventions in total. Figure 3 summarizes the pooled estimates for effect size obtained by a random effects model due to the considerable heterogeneity between studies (I² = 73 %; P = 0.001) (forest plot showing pooled SMD is given as Additional file 2). Average cycling time was faster in subjects ingesting carbohydrates as compared to placebo, however, without being statistically significant [mean difference 2.8 min (95 % CI −7.7, 2.1), p = 0.26]. Subgroup analysis including only studies with an exercise duration shorter than 90 min revealed a marginally higher average cycling time in the carbohydrate groups [mean difference 0.1 min (95 % CI −1.1, 1.2), p = 0.89]. In contrast, subgroup analysis taking into account studies with an exercise duration longer than 90 min resulted in a decreased average cycling time following carbohydrate ingestion when compared to placebo [mean difference −4.6 min (95 % CI −12.4, 3.2), p = 0.25].

Group 3 included five studies [13, 31, 37–39] with nine interventions in total, results of which are summarized in Fig. 4 (forest plot showing pooled SMD is given as Additional file 3). Mean power output was significantly more pronounced in participants subjected to a carbohydrate load as compared to placebo [mean difference 20.2 W (95 % CI 9.0, 31.5, p = 0.0004]. Comparable results could be obtained following subgroup analyses subclassifying carbohydrate interventions into ranges of 6–8 % [mean difference 19.3 W (95 % CI 5.6, 33.0), p = 0.006] and 12–18 % [mean difference 22.1 W (95 % CI 2.5, 41.8), p = 0.03], respectively.

Group 4 included four studies [42, 45–47] with four interventions in total. Meta-analytical data are depicted in Fig. 5 (forest plot showing pooled SMD is given as Additional file 4). Mean power output turned out to be significantly increased in volunteers following a carbohydrate intervention [mean difference 8.1 W (95 % CI 0.5, 15.7), p = 0.04]. Concerning subgroup analysis, performance tended to be higher in both studies with an exercise duration greater than 90 min [mean difference 18.8 W (95 % CI −8.4, 46.0), p = 0.18] and shorter than 90 min [mean difference 7.2 W (95 % CI −0.6, 15.1), p = 0.07] without yielding statistically significant results.

Duration of exercise ≤ 90 min did not result in statistically significant differences between carbohydrate interventions and placebo either in group 2 or in group 4. These findings seem to be in contrast with studies reporting an improved performance via carbohydrate mouth rinsing [10, 48–54]. It has been suggested that oral receptors within the mouth and the digestive tract sense carbohydrates and activate brain regions associated with reward and pleasure which may lead to enhanced performance [5, 10, 48]. However, most mouth rinse studies were conducted in a fasted state [48, 50, 52, 53] or had other limitations such as lack of or improper randomization [51, 54] or uncertain time of last ingested meal [49]. In our systematic review, three studies [27, 33, 38] with an exercise duration less than 90 min could not be included in either groups 2 or group 4. Beelen et al. [27] demonstrated a non-significant performance decline when testing a carbohydrate mouth rinse. Likewise, Acker-Hewitt et al. [38] did not find a significantly better performance subsequent to a carbohydrate solution when compared to placebo, while El-Sayed et al. [33] could detect an increase in performance capacity. Therefore, it seems premature to finally evaluate the potential benefit of ingesting carbohydrates in short-term exercises (less than 90 min), further trials reflecting realistic conditions are necessary.

Subgroup analysis of five trials with a duration time higher than 90 min in group 2 resulted in a trend towards a decreased time needed to cover a fixed distance or a fixed set amount of work. A similar trend could be observed in group 4, albeit with only two trials included in the subgroup. Taking all results under consideration, a performance benefit through carbohydrates might be possible when exercise duration exceeds 90 min. However, similar to subgroups with ≤ 90 min, additional studies are required for evidence-based recommendations.

Irrespective of specific carbohydrate concentrations, meta-analytical results of both groups 1 and 3 yielded statistically significant benefits for carbohydrate supplementation. In general, this might be due to multiple factors including maintenance of blood glucose [55–57] and high levels of carbohydrate oxidation especially towards the end of exercise [58, 59], thus sparing liver glycogen [60–63], as well as a central effect of carbohydrates [48, 52].

Regarding range of carbohydrate concentrations, results were statistically significant in favour of the lower range of 6–8 % of carbohydrate supplementation both in groups 1 and 3, while corresponding results for the higher range was only significant in group 3 (12–18 %), but not in group 1 (10–12 %). However, in both groups, the respective higher concentration range resulted in greater statistical variance when compared to the 6–8 % range. Therefore, one might speculate an impact of the administered carbohydrate type becoming more effective at higher concentrations. A high dose of ingested carbohydrates while exercising may cause gastrointestinal discomfort [64] which subsequently may decrease performance [65]. The maximal rate at which a single type of ingested carbohydrate can be oxidized is 60–70 g/h and ingesting more than this amount will not augment the oxidation rate but rather increases the chance for gastrointestinal discomfort [22]. It has been suggested that the ingestion of carbohydrates that use different, not competing, transporters increases the maximal carbohydrate oxidation rate (up to 105 g/h) [66], which has been verified by numerous studies [24, 39, 67–69]. In the study by Baur and co-workers [39], three different carbohydrate solutions were examined (8 % glucose solution, 12 % glucose solution, 12 % glucose–fructose (2:1) solution). The glucose-fructose solution achieved the greatest performance improvement, while the 12 % glucose solution did not affect performance significantly. Likewise, three different carbohydrate solutions (a 6 % glucose polymer-sucrose-solution (2:1, 37 g/h), a 12 % glucose polymer-fructose solution (2.4:1, 75 g/h), and an 18 % (4.1:1, 111 g/h) glucose polymer-fructose solution) were comparatively investigated in the trial by Mitchell et al. [13]. The best performance outcome was found with the 12 % glucose polymer-fructose solution. Despite no direct measurement of gastrointestinal symptoms, the authors concluded that the 18 % solution caused gastrointestinal distress and therefore the performance enhancement was not as high as with the 12 % solution [13].

Thus, the carbohydrate concentration resulting in optimal performance seems to be dependent on many factors, although our data suggests a more consistent benefit with carbohydrate solutions ranging between 6 and 8 %.

The protocol of the present systematic review was designed to summarize the available evidence on the ergogenic effects of carbohydrate supplementation as an expansion of the results by Colombani et al. [11] focusing on randomized controlled trials investigating the outcomes of their interventions under real-life conditions (no overnight fasting, no time-to-exhaustion tests). Moreover, we decided to categorize trials with respect to types of test and performance measurements. This rigid protocol allows for better comparison between the different trials, it is associated with a number of limitations as well. First of all, the number of studies suitable for meta-analyses turned out to be rather low. All of the 16 trials providing extractable data for meta-analyses used cycling as their exercise mode. Although this might be another aspect increasing the homogeneity of the results, it is not possible to draw any conclusions for other types of exercise such as running. Data on the content of the last meal prior to trials suggest heterogeneous pre-exercise carbohydrate intake between studies. Another common limitation of performance studies is the only low to average power with respect to the number of participants ranging between 16 and 32 volunteers in the present meta-analyses. Since only one trial [39] enrolled subjects with a mean VO_2max that would classify them as elite endurance athletes, the results are most likely not affected by heterogeneity between baseline capacities of study participants. In addition, with the exception of References [29] and [34], all trials were performed with male volunteers hampering transfer of results to female athletes. Following conversion of absolute values into percentage data, results were widely spread yielding improvements in assessed outcomes between 0.2 % [45] –13 % [13] as well as declines ranging between −0.6 % [22] and −7.3 % [41], respectively. This may serve as a potential indicator for the heterogeneous study designs.

This article was supported by the Open Access Publication Fund of the University of Vienna. No other sources of funding to be declared.

The datasets supporting the conclusions of this article are included within the article and its additional files.

PCC acquired part of the data (up to 9-3-2013). GH and LS developed the idea for this systematic review, GH prepared the protocol. Literature search was performed by MP and LS, while data extraction, analyses, and synthesis was done by all authors. GH prepared the first draft of the manuscript. Disagreements were resolved by consensus, all authors read and approved of the final manuscript.

PCC has received honoraria and free products from several companies producing and/or selling carbohydrate-containing sport drinks for diverse purposes (e.g., product development consultations, sponsorships of conferences organized by the authors, talks given to or for the companies). All other authors declare that they have no competing interests.

The datasets supporting the conclusions of this article are included within the article and its additional files.