The Endurance Performance Nerd Alert
Learn to train smart, run fast, and be strong with Thomas Solomon, PhD 
October 2025
Use this Nerd Alert of the latest exercise science and sports nutrition research to improve your running performance or coaching practice.
The research studies are divided into subtopics —
training methods,
sports nutrition,
supplements,
recovery,
athlete health,
injuries and rehab,
and
female athlete physiology
— but I’ve also provided a deeper dive into 4 studies:
Do regular long runs boost durability?
Do fewer run days boost marathon performance?
Can caffeine offset the training impairments caused by a keto diet?
Does CGM-guided fueling cause steadier blood glucose during endurance exercise?
And, there’s my beer of the month to wash it all down.
All the interesting papers I found this month are immediately below.
Dig in and evaluate the authors’ findings by clicking on the titles to access the full papers.
Evaluate each paper thoughtfully—be sceptical, not cynical. To guide you, consider using the framework I applied when doing my deep dives. This approach will help you assess the quality of a study while also appreciating the complexity and nuance of scientific research.
Share this nerd alert with your people:
General training methods and performance
Regular Long Runs and Higher Training Volumes are Associated with Better Running Economy Durability in Performance Matched Well-Trained Male Runners. Zanini et al. (2025) Med Sci Sports Exerc.
Substrate Oxidation Does Not Influence Middle Distance Running Performance: A Randomized Controlled Crossover Trial. Buga et al. (2025) Nutrients.
Training Volume and Training Frequency Changes Associated with Boston Marathon Race Performance. DeJong Lempke et al. (2025) Sports Med.
A variable sampling interval run sum chart for monitoring multivariate coefficient of variation. Cui et al. (2025) PLoS One.
Tailoring exercise intensity: Acute and chronic effects of constant-speed and heart rate-clamped exercise in healthy, inactive adults. Mazzolari et al. (2025) J Sci Med Sport.
Impact of weekly frequency of high-intensity interval training on cardiorespiratory, metabolic, and performance measures in recreational runners - An exploratory study. Lenk et al. (2025) Physiol Rep.
Influence of Trail Running Footwear Foam on Running Economy and Perceptual Metrics. Muzeau et al. (2025) Eur J Sport Sci.
Persistent Improvements in Running Economy With Advanced Footwear Technology During Prolonged Running in Trained Male Runners. Madsen et al. (2025) Scand J Med Sci Sports.
Can Level Ground Biomechanics Predict Uphill and Downhill Running Economy? Steele et al. (2025) J Appl Biomech.
Prolonged running reduces speed at the moderate-to-heavy intensity transition without additional reductions due to increased eccentric load. Barrett et al. (2025) Eur J Appl Physiol.
Determining physiologic variables for changes in 800-m running and 800-m ski ergometer performance. Gjerløw et al. (2025) Eur J Appl Physiol.
Quantifying Running Economy in Amateur Runners: Evaluating VO(2) and Energy Cost with Model-based Normalization. Lee et al. (2025) J Sports Sci Med.
Comparison of modeled lactate threshold 2 with maximal lactate steady state in running and cycling. Keller et al. (2025) Int J Sports Med.
Persisting elevation of total hemoglobin mass after altitude training in elite swimmers: a potential role of prolonged erythrocyte survival. Carin et al. (2025) Am J Physiol Heart Circ Physiol.
Repeated-sprint training in hypoxia: A review with 10 years of perspective. Faiss et al. (2025) J Sports Sci.
Hypoxic ventilatory decline in young healthy adults persists during moderate-intensity exercise in isocapnic hypoxia. Berdeklis et al. (2025) J Appl Physiol (1985).
Impact of intravenous iron or exogenous erythropoietin on hemoglobin mass, exercise performance, and acute mountain sickness during altitude acclimatization. Bradbury et al. (2025) J Appl Physiol (1985).
Influence of "live high-train low" on hemoglobin mass and post-exercise hepcidin response in female endurance athletes. Kuorelahti et al. (2025) Eur J Appl Physiol.
Exercise with overdressing for heat acclimation: a multilayered approach using biophysical modeling and two randomized crossover trials. Greenfield et al. (2025) J Appl Physiol (1985).
Sports nutrition and hydration
Caffeine enhances performance regardless of fueling strategy, however high CHO availability is associated with improved training speeds compared with ketogenic diet. Burke et al. (2025) Br J Nutr.
Comparative effects of continuous glucose monitoring-informed and traditional interval-based carbohydrate refueling protocols on endurance exercise responses: an exploratory study. Poon et al. (2025) J Int Soc Sports Nutr.
Exploring the non-targeted metabolomic landscape in endurance-trained runners following 10 weeks of different dietary patterns and concomitant training. Kripp et al. (2025) J Int Soc Sports Nutr.
Sports supplements
Physiological effects of spirulina supplementation during lactate threshold exercise at simulated altitude (2,500 m): a randomized controlled trial. Gurney et al. (2025) J Int Soc Sports Nutr.
Dosing strategies for β-alanine supplementation in strength and power performance: a systematic review. Ong et al. (2025) J Int Soc Sports Nutr.
Ergogenic effects of supplement combinations on endurance performance: a systematic review and meta-analysis of randomized controlled trials. Zart et al. (2025) J Int Soc Sports Nutr.
Does Nitrate Supplementation Improve Muscle Strength, Power, and Sprint Performance in Females? A Systematic Review and Meta-Analysis. Meng et al. (2025) Life (Basel).
The Effects of Ketogenic Diets and Ketone Supplements on the Aerobic Performance of Endurance Runners: A Systematic Review. Sun et al. (2025) Sports Health.
Athlete health
Bone mineral density varies throughout the skeleton of athletes dependent on their sport: Allometric modelling identifies the "effective" forces associated with body mass. Nevill et al. (2025) J Sci Med Sport.
Athlete's Mental Health and Quality of Life After Sports Injuries. Albishi et al. (2025) JBJS Rev.
Injury and rehab
Enhanced ground reaction force analyses reveal injury-related Biomechanical differences in runners. Nixon et al. (2025) Sci Rep.
Effects of sport specific unplanned movements on ankle kinetics and kinematics in healthy athletes from systematic review with meta-analysis. Giesche et al. (2025) Sci Rep.
Female athletes and sex differences
Determinants of maximal oxygen uptake in highly trained females and males: a mechanistic study of sex differences using advanced invasive methods. Skattebo et al. (2025) J Physiol.
Is pelvic floor loading in female runners associated with post-run changes in pelvic floor morphometry or function? Berube et al. (2025) BJU Int.
Sex Differences in Foot and Ankle Sports Injury Rates in Elite Athletes: A Systematic Review and Meta-analysis of 25,687,866 Athlete Exposures. Talia et al. (2025) Orthop J Sports Med.
My deep dives
Do regular long runs boost durability?
Zanini et al. (2025) Med Sci Sports Exerc: Regular Long Runs and Higher Training Volumes are Associated with Better Running Economy Durability in Performance Matched Well-Trained Male Runners.
What type of study is this?
This study is an observationalAn observational study is where researchers observe what naturally occurs without intervening — no treatment is assigned. I.e., the researchers watch and learn, but don’t interfere. Observational studies are used in epidemiology and can have different study designs, including cross-sectional, case-control, and cohort study designs. and cross-sectionalA cross-sectional study is a type of observational study where the exposure and outcome are measured at a single point in time, giving a snapshot of a population—what’s happening right now. Cross-sectional studies are used in health surveys, prevalence studies, or for hypothesis generation, and can show prevalence (how common something is) and associations (but not cause and effect). E.g., What percentage of runners currently report using recovery supplements, and is use linked to age or training volume? comparison of two performance-matched runner groups differing in training characteristics.
.
What was the authors’ hypothesis or research question?
The authors aimed to test whether runners who regularly perform long runs and accrue higher weekly mileage show better running economyThe rate of energy expenditure (measured in kiloJoules [KJ], kilocalories [kcal] or oxygen consumption [V̇O2]) per kilogram body mass (kg) per unit of distance i.e. per 1 kilometer traveled. A runner with a lower energy cost per kilometer has a higher economy than a runner with a higher energy cost. durability and smaller neuromuscular decrements during a 90-minute run than performance-matched runners who do not.
What did the authors do to test the hypothesis or answer the research question?
Twenty-six well-trained male runners (age 18–40 years) were pair-matched on 10 km performance and assigned to either a long-distance-training group doing regular long runs ≥90 min (n=13) or a short-distance-training group with all runs <70 min (n=13). after visit-1 testing for lactate thresholdLactate threshold is the exercise intensity at which lactate starts to accumulate rapidly in the blood, signaling a shift toward more glycolytic (glucose using) and anaerobic energy use. It is typically the intensity at which effort begins to feel hard and fatigue builds faster. and V̇O2maxV̇O2max is the maximal rate of oxygen consumption your body can achieve during exercise. It is a measure of cardiorespiratory fitness and indicates the size of your engine, i.e., your maximal aerobic power, which contributes to endurance performance., participants completed, on a separate day, a 90-minute treadmill run at individually determined lactate-threshold speed with metabolic sampling every 15 minutes. Outcomes included running economy as oxygen cost and energy cost, ventilatory and lactate responses, rating of perceived exertion, body mass change, and pre-to-post neuromuscular function via isometric quarter-squat peak force and countermovement jump.
What did the authors find?
Both groups showed a drift (worsening) in running economy (RE) over time, but the short-distance-training group deteriorated earlier and more. By 90 minutes, oxygen-cost increased about +6.0% in the short-distance-training group versus +3.1% in the long-distance-training group (group×time p<0.001; small effect size: partial eta-squaredA measure of effect size used in ANOVA that represents the proportion of variance in a dependent variable explained by an independent variable, while controlling for (excluding) the effects of other variables. It is commonly used in factorial designs and typically yields larger values than regular eta-squared. Values range from 0 to 1, with higher values indicating a greater proportion of explained variance.≈0.14–0.16), with similar patterns for energy-cost change (+5.8% vs +2.9%). Pre-to-post neuromuscular function declined more in the short-distance-training group: isometric squat peak force −19.4% in the short-distance-training group vs −12.2% in the long-distance-training group (group×time p=0.002), and countermovement-jump mean power and height fell in the short-distance-training group (−6.6% and ~−0.03 m; p≈0.005–0.012) but were unchanged in the long-distance-training group. Correlations across all 26 runners linked better durability (smaller RE drift at 90 min) with longer weekly long run distance (r-valuePearson’s r-value represents the correlation coefficient, which is a statistic that measures the strength and direction of a linear relationship between two variables, ranging from -1 to +1. An r value close to +1 indicates a strong positive correlation, close to -1 a strong negative correlation, and around 0 no linear relationship.=−0.67, p<0.001) and with higher weekly training distance (r=−0.48, p=0.0038). Changes in strength or jump performance did not correlate with RE durability.
The authors concluded that regularly performing long runs and maintaining higher weekly mileage are associated with better running-economy durability and smaller neuromuscular decrements during prolonged running, even when runners are matched for performance level.
What were the strengths?
The design cleverly pair-matched runners on recent 10 km performance and standardized the prolonged run at each runner’s lactate-threshold speed, which reduces noise from mixed intensity domains. The physiological outcomes were repeatedly sampled with appropriate gas-exchange methods, body-mass adjustment for energy cost calculations was considered, and statistics included group×time interactions with effect sizesAn effect size is a standardized measure of the magnitude of an effect of an intervention. Unlike p-values, effect sizes show how large the effect is and indicate how meaningful it might be. Common effect size measures include standardised mean difference (SMD), Cohen’s d, Hedges’ g, eta-squared, and correlation coefficients.. The paper clearly described inclusion criteria, informed consent, equipment, calibration, and timing of measurements, and provided transparent conflict-of-interest statements. Basically, the lab work was tidy.
What were the limitations?
It’s observational, not randomized, so training exposure may reflect unmeasured traits. The sample is small (n=26; 13 per group), which can limit power and raise false negative (type II error)Failing to detect an effect or difference when there actually is one. I.e, “a missed detection”. Studies with a small sample size (N, number of participants) are more likely to make false negative results. risk for subtler effects. Only men were studied, so there is no sex comparison and the external validity is limited to well-trained, performance-matched male runners, and cannot be generalised to female athletes. Furthermore, the manuscript doesn’t report pre-registration or blindingBlinding is when people in a study don’t know which treatment they’re getting. It stops expectations or beliefs (from patients or researchers) from skewing the results. “Single-blind” means participants don’t know; “double-blind” means participants and researchers don’t know; “triple-blind” means that the participants, researchers, and data analysts are kept in the dark. The goal is simple: fair tests and trustworthy findings..
How was the study funded, and are there any conflicts of interest that may influence the findings?
The authors state that no funding was received and that they have no conflicts of interest.
How can you apply these findings to your training or coaching practice?
For coaches and self-coached runners, this is useful. If you already run decent weekly volume and add regular ≥90-minute long runs, your running-economy may drift less during long efforts, and your legs may keep more of their pop by the end: bonjour, steadier less-detonation prone marathons. The force is quite strong with this one, but it’s still cross-sectional, so don’t treat it like gospel.
What is my Rating of Perceived scientific Enjoyment?
RP(s)E = 7 out of 10. I experienced moderate scientific enjoyment because while the study is well designed and the outcome is very practical and usable, the very small number of participants (13 per group) and the lack of inclusion of female athletes hold the study back. Furthermore, does the durability come mainly from the long run itself, or just from more miles overall? The authors partly answer both, but the knot isn’t fully untangled. I’m left quietly optimistic—and wondering if one extra long run each month is the small hinge that swings a big door.
Important: Don’t make any major changes to your daily habits based on the findings of one study, especially if the study is small (e.g., less than 30 participants in a randomised controlled trial or less than 5 studies in a meta-analysis) or poor quality (e.g., high risk of biasRisk of bias in meta-analysis refers to the potential for systematic errors in the studies included in the analysis, which can lead to misleading or invalid results. Assessing this risk is crucial to ensure the conclusions drawn from the combined data are reliable. or low quality of evidenceA low quality of evidence means that, in general, studies in this field have several limitations. This could be due to inconsistency in effects between studies, a large range of effect sizes between studies, and/or a high risk of bias (caused by inappropriate controls, a small number of studies, small numbers of participants, poor/absent randomization processes, missing data, inappropriate methods/statistics). When the quality of evidence is low, there is more doubt and less confidence in the overall effect of an intervention, and future studies could easily change overall conclusions. The best way to improve the quality of evidence is for scientists to conduct large, well-controlled, high-quality randomized controlled trials.). What do other trials in this field show? (Follow the link to explore those trials.) Do they confirm the findings of this study or have mixed outcomes? Is there a high-quality systematic review and meta-analysis evaluating the entirety of the evidence in this field? If so, what does the analysis show? What is the risk of biasRisk of bias in meta-analysis refers to the potential for systematic errors in the studies included in the analysis, which can lead to misleading or invalid results. Assessing this risk is crucial to ensure the conclusions drawn from the combined data are reliable. or certainty of evidenceCertainty of evidence tells us how confident we are that the results reflect the true effect. It’s based on factors like study design, risk of bias, consistency, directness, and precision. Low certainty means more doubt and less confidence, and that future studies could easily change the conclusions. High certainty means that the current evidence is so strong and consistent that future studies are unlikely to change conclusions. of the included studies? 
Do fewer run days boost marathon performance?
DeJong Lempke et al. (2025) Sports Med: Training Volume and Training Frequency Changes Associated with Boston Marathon Race Performance.
What type of study is this?
This study is
a cohort studyA cohort study is a type of observational study that follows a group (a cohort) over time to see how exposures affect outcomes. I.e., a cohort study tracks people forward (prospective) to see what happens. Cohort studies are used to study causes and risk factors over time, and can show the association between exposure and outcome or new cases over time (incidence). E.g., Does regular running reduce the risk of heart disease over 10 years?.
What was the authors’ hypothesis or research question?
The authors aimed to test whether training volume, quality (“hard”) sessions, cross-training, and changes in training frequency before the 2022 Boston Marathon were associated with race performance.
What did the authors do to test the hypothesis or answer the research question?
917 runners (495 women, 422 men; 47 ± 14 years old) registered for the 2022 Boston Marathon were emailed one month pre-race and completed a survey on demographics, running history, and training during two windows: 12–4 months and 4–0 months before race day. Multivariate linear regressionMultivariate linear regression builds a model to predict several continuous outcomes at once using one or more predictors: One model, many dependent variables. Why use it? When outcomes are related — say running speed and finish time — it can borrow strength across them and give tighter insight. The output gives you a set of coefficients for each outcome, which tell you how much each predictor contributes to each outcome. was used to adjust for age, sex, years of marathon training, and number of prior marathons, and help predict the outcomes (official race time and WA points) from the predictors: weekly running hours, distance, number of running sessions, number of quality sessions, weekly cross-training duration and sessions, and the change in these between the two windows.
What did the authors find?
In the 12-4-month window, greater running hours (>10 h/week vs lower categories), higher weekly running distance, more weekly runs, and more weekly quality sessions each predicted faster times. E.g., every extra running session/week was linked to ~3.6 minutes faster and each additional quality session/week to ~16.2 minutes faster. In the 4–0-month window, higher running hours, distance, weekly runs, and quality sessions again predicted faster times, and now cross-training sessions also mattered. E.g., each extra cross-training day/week was associated with ~6 minutes faster. Critically, when looking at changes between windows, runners who decreased their weekly running sessions in the final 4 months ran about 3 minutes faster than those who maintained or increased. The effects of sex (males were faster than females) and age (older folks were slower) were consistent across the two time windows. Interestingly, experience variables (having trained for more years or having finished more marathons) did not play a role in performance once you accounted for what the runner actually did in training during those months. I.e., current training mattered more than training history.
The authors concluded that habitually higher training exposure, especially more quality work, plus a relative reduction in running frequency in the final four months, was associated with better Boston Marathon performance.
What were the strengths?
The cohort was large for a single race, and the authors adjusted for key confounders (age, sex, years training, previous marathons) and even tested an interaction between running and cross-training. Statistical reporting was clear and transparent, and the outcomes were clearly defined and measured precisely in all participants.
What were the limitations?
Firstly, causality can’t be inferred from this type of observational study design — it’s associations only — and the results may not generalize beyond mostly white, trained but non-elite runners. Furthermore, a survey-based data collection can suffer from recall biasPeople don’t remember past events perfectly. When a study asks them to report what they did, ate, or felt weeks or months ago, their answers can be off. Those memory errors can skew the results—especially if one group has more reason to remember a detail than another. For example: Runners who had a great race might “remember” training more than they did. Runners who struggled might underreport their training. The mismatch between memory and reality is recall bias.. Data collection also lacked granular measures of training intensity distribution, and important factors like nutrition, injuries, and recovery weren’t measured. For example, did the faster runners simpler nail their race day nutrition? Lastly, while the models were decent, they only explained about half the variance, which is good but not the whole story.
How was the study funded, and are there any conflicts of interest that may influence the findings?
Funding came from the Joe and Clara Tsai Foundation via the Wu Tsai Human Performance Alliance and the Eleanor and Miles Shore Faculty Development Awards Program (Harvard Medical School: Boston Children’s Hospital Department of Orthopaedic Surgery and Sports Medicine Award). None of the authors has any major conflicts of interest.
How can you apply these findings to your training or coaching practice?
For coaches and athletes, the message is practical: (i) build a big aerobic volume while keeping some quality in both the general and specific phases. And, (ii) consider trimming the number of weekly runs in the final months while maintaining overall training quality—essentially, a frequency-focused taper. Does that mean everyone should cut run days? Probably not, but the pattern is hard to ignore. The force is strong with this one, but it’s an observational study with several limitations, so don’t overfit your life to it.
What is my Rating of Perceived scientific Enjoyment?
RP(s)E = 6 out of 10. I experienced moderate scientific enjoyment because despite having a clear and practical takeaway — train a lot; reduce the frequency towards race day — the self-reported training measurements and the lack of data on training intensity distribution, nutrition, injuries, etc creates a lot of unknowns.
Important: Don’t make any major changes to your daily habits based on the findings of one study, especially if the study is small (e.g., less than 30 participants in a randomised controlled trial or less than 5 studies in a meta-analysis) or poor quality (e.g., high risk of biasRisk of bias in meta-analysis refers to the potential for systematic errors in the studies included in the analysis, which can lead to misleading or invalid results. Assessing this risk is crucial to ensure the conclusions drawn from the combined data are reliable. or low quality of evidenceA low quality of evidence means that, in general, studies in this field have several limitations. This could be due to inconsistency in effects between studies, a large range of effect sizes between studies, and/or a high risk of bias (caused by inappropriate controls, a small number of studies, small numbers of participants, poor/absent randomization processes, missing data, inappropriate methods/statistics). When the quality of evidence is low, there is more doubt and less confidence in the overall effect of an intervention, and future studies could easily change overall conclusions. The best way to improve the quality of evidence is for scientists to conduct large, well-controlled, high-quality randomized controlled trials.). What do other trials in this field show? (Follow the link to explore those trials.) Do they confirm the findings of this study or have mixed outcomes? Is there a high-quality systematic review and meta-analysis evaluating the entirety of the evidence in this field? (Follow the link to explore those reviews.) If so, what does the analysis show? What is the risk of biasRisk of bias in meta-analysis refers to the potential for systematic errors in the studies included in the analysis, which can lead to misleading or invalid results. Assessing this risk is crucial to ensure the conclusions drawn from the combined data are reliable. or certainty of evidenceCertainty of evidence tells us how confident we are that the results reflect the true effect. It’s based on factors like study design, risk of bias, consistency, directness, and precision. Low certainty means more doubt and less confidence, and that future studies could easily change the conclusions. High certainty means that the current evidence is so strong and consistent that future studies are unlikely to change conclusions. of the included studies?
Can caffeine offset the training impairments caused by a keto diet?
Burke et al. (2025) Br J Nutr: Caffeine enhances performance regardless of fueling strategy, however high CHO availability is associated with improved training speeds compared with ketogenic diet.
What type of study is this?
This study is a quasi-experimentalA quasi-experimental study tests a cause-and-effect idea without randomly assigning people to groups. Researchers use a comparison group that’s as similar as possible, then apply a change or program to one group and not the other. They often use tools like matching or before-and-after trends to balance out differences. It’s stronger than a simple observational study, but not as airtight as a randomized trial. Think “best feasible test” when random assignment isn’t possible. training-camp study with a randomised controlled trialThe “gold standard” approach for determining whether a treatment has a causal effect on an outcome of interest. In such a study, a sample of people representing the population of interest is randomised to receive the treatment or a no-treatment placebo (control), and the outcome of interest is measured before and after the exposure to treatment/control. (with crossoverCrossover means that all subjects completed all interventions (control and treatment) usually with a wash-out period in between.) for caffeine, and a non-randomised trial (a pre-post study)A trial where the outcome of interest is measured before and after exposure to a treatment, but there is no control group or control intervention. Instead, the participants' baseline measurement is used as the control. This type of study design has a high risk of bias and is prone to producing unreliable findings. parallel-group comparison for diet.
.
What was the authors’ hypothesis or research question?
The authors aimed to confirm whether a ketogenic low-carbohydrate, high-fat diet reduces training quality in elite race walkers and to test if low-dose caffeine can “rescue” performance during a key high-intensity session.
What did the authors do to test the hypothesis or answer the research question?
Twenty-one elite race walkers (15 male, 6 female) completed four 14 km “tempo hill” sessions across three weeks inside a research-embedded training camp. A baseline session was done with high carbohydrate intake, then athletes followed either sustained high carbohydrate availability (n=13 including a merged periodised-CHO subgroup) or a ketogenic low-carbohydrate, high-fat diet (n=8) for three weeks. In weeks 2–3, participants were randomized to a crossover of pre-session caffeine gum (~3 mg/kg) versus placebo. Outcomes were session time, speed normalized to percentage velocity at V̇O2maxV̇O2max is the maximal rate of oxygen consumption your body can achieve during exercise. It is a measure of cardiorespiratory fitness and indicates the size of your engine, i.e., your maximal aerobic power, which contributes to endurance performance. (%vV̇O2max), heart rate, perceived exertion, a cognitive function test, and fingertip blood metabolites.
What did the authors find?
Race-walking performance was worse on the keto diet over time. Within groups, the high-carb group improved from baseline (large effect sizeAn effect size is a standardized measure of the magnitude of an effect of an intervention. Unlike p-values, effect sizes show how large the effect is and indicate how meaningful it might be. Common effect size measures include standardised mean difference (SMD), Cohen’s d, Hedges’ g, eta-squared, and correlation coefficients.: Cohen’s dA measure of effect size that quantifies the difference between two group means relative to their pooled standard deviation. It indicates how substantial the difference is, independent of sample size. Common benchmarks for interpretation are 0.2 (small effect), 0.5 (medium effect), and 0.8 (large effect). =0.83), whereas the keto group was slower than baseline (large effect size; d=1.19). Normalized speed (%vV̇O2max) showed that the the high-carb group were faster than the keto group by +5.2% in week 2 (P=0.01) and +4.4% in week 3 (P=0.03). Caffeine, averaged across both diets (n=21), shaved ~2.4% off hill time (small effect size; d=0.37) and increased %vV̇O2max; and these benefits were similar between the high-carb and keto groups. Caffeine also improved cognition function (more questions correctly answered, post-exercise versus pre-exercise, with faster time per correct response), regardless of diet.
The authors concluded that caffeine modestly boosts performance and cognition regardless of diet, but high carbohydrate availability delivered better training speeds than a ketogenic diet, which diminished the quality of key high-intensity sessions.
What were the strengths?
I like the real-world setting with elite athletes working together in a training camp, with a repeated 14 km hill session as the performance marker. The caffeine arm used a randomized, placebo-controlled crossover, which is solid, and the hill course, timing, heart rate, and cognitive function testing (Stroop test) were all well standardized. The analysis used linear mixed modellingLinear mixed modelling is a way to analyze data with both overall trends and person- or group-specific differences. It handles repeated measures and clustered data without pretending everyone is identical. with V̇O2max as a covariate to account for baseline aerobic fitness differences.
What were the limitations?
The paper doesn’t report a power calculation, protocol pre-registrationPreregistration is when a detailed description of a study plan is deposited in an open-access repository before collecting the study data. It promotes transparency and accountability, and boosts research integrity., allocation concealmentAllocation concealment is the step that hides the next treatment assignment before a patient enters a trial. It prevents staff from guessing or peeking, so they can’t steer patients to one group or another. It happens at enrollment, before blinding, and guards against selection bias. (for caffeine/placebo), or an intention to treat analysisAnalyzes people in the groups they were originally assigned to, no matter what they actually did. This preserves randomization and mirrors real-world use. It guards against many biases and usually gives a conservative estimate of benefit. approach. Diet groups weren’t randomized; athletes largely chose their diet, which invites selection biasWho gets into the study differs from who doesn’t, skewing results. The sample isn’t truly representative. and makes causal claims shaky. The total sample was also small (N=21), which reduces statistical powerStatistical power is the probability that a statistical test will correctly reject a false null hypothesis (i.e., detect an effect if there is one; a true positive). Higher power reduces the risk of a false negative (failing to detect a true effect; or a Type II error). Power is typically influenced by sample size, effect size, significance level, and variability in the data, with a common target being 80% (or 0.8). and bumps up the risk of false negativeFailing to detect an effect or difference when there actually is one. I.e, “a missed detection”. Studies with a small sample size (N, number of participants) are more likely to make false negative results. findings for some outcomes. Generalizability is also somewhat narrow: limited to world-class race walkers on a specific hill loop, following a short three-week window.
How was the study funded, and are there any conflicts of interest that may influence the findings?
The study was funded by the Australian Catholic University Research Funds (ACURF, 2017000034). The authors state that they have no competing interests or industry funding.
How can you apply these findings to your training or coaching practice?
For coaches and endurance folks, the message is simple and kinda practical: if your training relies on repeated high-intensity efforts, a high-carb strategy yields better speeds than going keto, and a small caffeine dose (~3 mg/kg) will likely give you a little extra zip and cleaner decision-making late in the session. The force is not strong with keto here — “These are not the droids you’re looking for” — at least for key quality sessions in elites. On a side note, this study is part of the “Supernova” series of studies from Louise Burke’s lab, which has intensively examined the role of high-carb diets vs. ketogenic diets on performance in elite athletes. I went deep on the studies a few years ago; check it out at veohtu.com/supernova.
What is my Rating of Perceived scientific Enjoyment?
RP(s)E = 8 out of 10. I experienced high scientific enjoyment because despite there being no randomization for diet (yes, i know the caffeine crossover was randomized), no power calculation, no protocol pre-registration, and no explicit intention to treat analysis, there are detailed methods, appropriate stats, clear timing of follow-ups, adjustment for baseline aerobic capacity, and near-complete reporting of outcomes and statistics, as well as transparent data availability. And, the major plus: the Supernova studies include real elite athletes training together in a real training camp doing real things: this is as real-world evidenceFindings drawn from routine care, not tightly controlled trials. It uses sources like electronic health records, insurance claims, registries, training camps, or wearables (Garmin/Strava) to show how treatments work in everyday settings and diverse patients. as you get in sports science, and it is super hard to pull off!
Important: Don’t make any major changes to your daily habits based on the findings of one study, especially if the study is small (e.g., less than 30 participants in a randomised controlled trial or less than 5 studies in a meta-analysis) or poor quality (e.g., high risk of biasRisk of bias in meta-analysis refers to the potential for systematic errors in the studies included in the analysis, which can lead to misleading or invalid results. Assessing this risk is crucial to ensure the conclusions drawn from the combined data are reliable. or low quality of evidenceA low quality of evidence means that, in general, studies in this field have several limitations. This could be due to inconsistency in effects between studies, a large range of effect sizes between studies, and/or a high risk of bias (caused by inappropriate controls, a small number of studies, small numbers of participants, poor/absent randomization processes, missing data, inappropriate methods/statistics). When the quality of evidence is low, there is more doubt and less confidence in the overall effect of an intervention, and future studies could easily change overall conclusions. The best way to improve the quality of evidence is for scientists to conduct large, well-controlled, high-quality randomized controlled trials.). What do other trials in this field show? (Follow the link to explore those trials.) Do they confirm the findings of this study or have mixed outcomes? Is there a high-quality systematic review and meta-analysis evaluating the entirety of the evidence in this field? (Follow the link to explore those reviews.) If so, what does the analysis show? What is the risk of biasRisk of bias in meta-analysis refers to the potential for systematic errors in the studies included in the analysis, which can lead to misleading or invalid results. Assessing this risk is crucial to ensure the conclusions drawn from the combined data are reliable. or certainty of evidenceCertainty of evidence tells us how confident we are that the results reflect the true effect. It’s based on factors like study design, risk of bias, consistency, directness, and precision. Low certainty means more doubt and less confidence, and that future studies could easily change the conclusions. High certainty means that the current evidence is so strong and consistent that future studies are unlikely to change conclusions. of the included studies? I’ve written a deep-dive article on this topic; check it out at veohtu.com/supernova.
Does CGM-guided fueling cause steadier blood glucose during endurance exercise?
Poon et al. (2025) J Int Soc Sports Nutr: Comparative effects of continuous glucose monitoring-informed and traditional interval-based carbohydrate refueling protocols on endurance exercise responses: an exploratory study.
What type of study is this?
This study is a randomised controlled trialThe “gold standard” approach for determining whether a treatment has a causal effect on an outcome of interest. In such a study, a sample of people representing the population of interest is randomised to receive the treatment or a no-treatment placebo (control), and the outcome of interest is measured before and after the exposure to treatment/control. with crossoverCrossover means that all subjects completed all interventions (control and treatment) usually with a wash-out period in between..
What was the authors’ hypothesis or research question?
The authors hypothesised that a continuous glucose monitoring (CGM)-informed carbohydrate refueling protocol would produce a steadier blood glucose profile and improve physiological responses versus a traditional feeding protocol during steady cycling.
What did the authors do to test the hypothesis or answer the research question?
After an overnight fast, 12 healthy, recreationally active men (mean age 32.3 ± 6.3 years) completed two 75-minute cycling trials at 50% of their peak power output. Participants completed the trials in a random order: Trial A, a traditional protocol with fixed carbohydrate intakes every 15 minutes; and, Trial B, a CGM-informed protocol where carbohydrate servings were triggered by predefined falling glucose trends on the FreeStyle Libre. Placebo drinks were given when no fall was detected to match fluid intake. Interstitial glucose was measured via CGM every 5 minutes), finger-prick glucose and lactate were measured pre-, mid-, and end-of-ride, and heart rate and RPE were recorded every 15 minutes.
What did the authors find?
The CGM-informed protocol produced lower glucose variability (moderate effect sizeAn effect size is a standardized measure of the magnitude of an effect of an intervention. Unlike p-values, effect sizes show how large the effect is and indicate how meaningful it might be. Common effect size measures include standardised mean difference (SMD), Cohen’s d, Hedges’ g, eta-squared, and correlation coefficients., Cohen’s dA measure of effect size that quantifies the difference between two group means relative to their pooled standard deviation. It indicates how substantial the difference is, independent of sample size. Common benchmarks for interpretation are 0.2 (small effect), 0.5 (medium effect), and 0.8 (large effect). = 0.76) and glucose exposure (moderate effect size, d = 0.65) during exercise, with higher glucose in the traditional feeding group at 60–75 minutes. Despite the difference in glucose, physiological responses — mean heart rate, RPE, and lactate — didn’t differ between groups. On the other hand, carbohydrate intake was markedly different between groups: 18.1 ± 10.1 grams per hour with CGM versus 51.2 grams per hour with traditional feeding.
The authors concluded that while CGM-informed fueling did not change heart rate, lactate, or perceived exertion across a 75-minute steady ride, it did stabilize glucose more than traditional fixed-interval fueling, hinting at a potential role for personalized strategies.
What were the strengths?
The crossover design with within-person comparison is a big plus here. The fueling cues for the CGM protocol were pre-specified from device trend rules, measurement timing was clear, and statistical reporting was very clear with p-values, effect sizes, and interaction tests. Drinks were blinded by lab staff, which likely reduced expectancy effects when drinks differed between carbohydrate and placebo in the CGM condition.
What were the limitations?
It’s a very small study with only 12 participants, so statistical powerStatistical power is the probability that a statistical test will correctly reject a false null hypothesis (i.e., detect an effect if there is one; a true positive). Higher power reduces the risk of a false negative (failing to detect a true effect; or a Type II error). Power is typically influenced by sample size, effect size, significance level, and variability in the data, with a common target being 80% (or 0.8). is limited and the risk of missing a real effect (a false negative (type II error)Failing to detect an effect or difference when there actually is one. I.e, “a missed detection”. Studies with a small sample size (N, number of participants) are more likely to make false negative results.) is high. There was also no sample size calculation, protocol pre-registrationPreregistration is when a detailed description of a study plan is deposited in an open-access repository before collecting the study data. It promotes transparency and accountability, and boosts research integrity., or intention to treatAnalyzes people in the groups they were originally assigned to, no matter what they actually did. This preserves randomization and mirrors real-world use. It guards against many biases and usually gives a conservative estimate of benefit. plan reported. Furthermore, the clinical/practical significanceReflects how meaningful a change is to a person’s health or performance. A small change can be statistically significant but not clinical significant; but, if a change is big enough to matter to people in real life, then it is clinically significant (and doesn’t just tick a statistical significance box). of the glucose differences is unclear because performance wasn’t assessed. Lastly, the CGM device’s inherent lag and occasional sensor detachment were acknowledged, which could muddle real-time decision timing in the field.
How was the study funded, and are there any conflicts of interest that may influence the findings?
The work was supported by the Chinese University of Hong Kong via United College Endowment Fund Research Grant and Lee Hysan Foundation Research Grant Schemes. The authors reported no potential conflicts of interest.
How can you apply these findings to your training or coaching practice?
For everyday endurance athletes and coaches, the takeaway is simple: CGM-guided fueling can cut glucose swings during a steady 75-minute ride without changing how hard it feels or how your heart rate responds. If you’re prone to GI issues when pushing 50 to 60 grams per hour of carbs down your pie hole, the findings suggest you might get by on less when intensity is modest — at least in the short run. But does a steadier glucose trace actually translate to faster times or stronger finishes when the ride stretches to two, three, four hours? The study doesn’t test performance, so, nothing to see here, move along, move along — yet. A lot more high-quality CGM and sports performance research is needed!
What is my Rating of Perceived scientific Enjoyment?
RP(s)E = 6 out of 10. I experienced moderate scientific enjoyment because the design is neat (a randomized crossover) and the stats are tidy, but the small number of participants, a lack of power calculations, and no performance endpoints leave a high risk of false positive findings and the practical impact hanging.
Important: Don’t make any major changes to your daily habits based on the findings of one study, especially if the study is small (e.g., less than 30 participants in a randomised controlled trial or less than 5 studies in a meta-analysis) or poor quality (e.g., high risk of biasRisk of bias in meta-analysis refers to the potential for systematic errors in the studies included in the analysis, which can lead to misleading or invalid results. Assessing this risk is crucial to ensure the conclusions drawn from the combined data are reliable. or low quality of evidenceA low quality of evidence means that, in general, studies in this field have several limitations. This could be due to inconsistency in effects between studies, a large range of effect sizes between studies, and/or a high risk of bias (caused by inappropriate controls, a small number of studies, small numbers of participants, poor/absent randomization processes, missing data, inappropriate methods/statistics). When the quality of evidence is low, there is more doubt and less confidence in the overall effect of an intervention, and future studies could easily change overall conclusions. The best way to improve the quality of evidence is for scientists to conduct large, well-controlled, high-quality randomized controlled trials.). What do other trials in this field show? (Follow the link to explore those trials.) Do they confirm the findings of this study or have mixed outcomes? I’ve written a deep-dive article on this topic; check it out at veohtu.com/cgmforathletes.
My beer of the month
Boyicao.
Brewed by Basqueland (Spain).
Imperial stout.
12.5% ABV.
Wow. Just wow. Nutellery to the nose, nutellery to the tongue, nutellery down the hatch, and a nutellery aftertaste. "Boyicao" must mean "turning Nutella up to eleven" in Spanish.
RP(be)E(r)(Rating of Perceived beer Enjoyment)
9 out of 10
Access to education is a right, not a privilege
Equality in education, health, and sustainability matters deeply to me. I was fortunate to be born into a social welfare system where higher education was free. Sadly, that's no longer true. That's why I created Veohtu: to make high-quality exercise science and sports nutrition education freely available to folks from all walks of life. All the content is free, and always will be.
Every day is a school day.
Empower yourself to train smart.
Be informed. Stay educated. Think critically.
