The research team, led by A. Nicole Ferguson of Kennesaw State University together with colleagues at Drexel University, Duke University Medical Center, Clemson University, and Cincinnati Children’s Hospital Medical Center, analyzed records from 25,779 singleton infants born between 24 and 30 weeks of gestation. All infants were admitted to a NICU within two days of birth, were born without major congenital anomalies, and came from the Pediatrix Clinical Data Warehouse covering the years 2013 to 2018. This dataset is one of the largest of its kind, which matters because previous attempts to study mortality within fine birthweight strata have been hampered by small sample sizes, forcing investigators to pool gestational ages and obscure age-specific patterns.

The methodological innovation at the heart of the study is a simple but powerful reclassification. Using the Olsen intrauterine growth curves, which provide sex- and gestational-age-specific percentiles for birthweight, the researchers first assigned each infant to the traditional SGA (below the 10th percentile), AGA (10th to 90th percentile), or LGA (above the 90th percentile) category. They then subdivided these groups: SGA infants were split into extremely small (eSGA, below the 3rd percentile) and moderately small (mSGA, 3rd to below 10th percentile); LGA infants were split into moderately large (mLGA, above the 90th to 97th percentile) and extremely large (eLGA, above the 97th percentile); and the AGA group was divided into eight equally sized subgroups, labeled 3-AGA through 9-AGA, ranked from smallest to largest. The result was a twelve-category system replacing the traditional three.

The headline result is striking. Under the traditional classification, SGA infants had a mortality probability of 17.3 percent, while AGA and LGA infants clustered around 6 to 7 percent and were statistically indistinguishable from each other. But when the SGA group was split apart, the picture changed completely. Infants classified as eSGA had a mortality probability of 28.9 percent, the highest of all twelve groups, whereas mSGA infants fared far better at 9.9 percent. In other words, the elevated risk long attributed to being small for gestational age is driven overwhelmingly by the smallest infants, those below the 3rd percentile. The mSGA group, meanwhile, had mortality rates similar to the smallest AGA subgroups, blurring the boundary that the traditional system draws between small and appropriate size.

Perhaps the most novel contribution of the study is its demonstration that risk varies even within the AGA category, which has historically been treated as the low-risk middle ground. In the pooled sample, infants in the smallest AGA subgroups, 3-AGA through 5-AGA, had mortality probabilities of roughly 7.7 to 8.5 percent, significantly higher than the 5.0 to 5.1 percent observed in the largest AGA subgroups, 9-AGA and 10-AGA. At 25 weeks of gestation, the smallest AGA subgroups showed mortality rates around 20 percent, comparable to moderately small infants. This means that a baby sitting just above the 10th percentile cutoff, technically AGA and therefore often assumed to be at ordinary risk, may in fact carry a mortality burden closer to that of an SGA infant. Finer granularity within AGA, the authors argue, can identify additional at-risk infants who would otherwise slip through the standard screening net.

By contrast, subdividing the LGA category added little predictive value. Extremely large infants showed a slightly but not significantly higher mortality probability than moderately large infants, 7.8 versus 5.2 percent in the pooled sample, and neither LGA subgroup differed consistently from the larger AGA groups. The authors caution that low prevalence and low mortality rates, particularly at more mature gestational ages, make these estimates unstable, but the practical message is that the LGA category does not need to be refined for mortality prediction in this population.

Beyond the categorical analysis, the team employed cubic spline models to treat birthweight as a continuous variable, expressed as a sex-adjusted z-score, and to estimate mortality probability as a smooth function of size for each gestational week. These curves offer clinicians a point-of-care tool: a 25-week infant whose birthweight sits exactly at the median for gestational age, a z-score of zero, has roughly a 17 percent chance of dying during the NICU stay, with a 95 percent confidence interval of 15 to 19 percent. At the extreme, eSGA infants at 24 weeks faced mortality probabilities as high as 62.4 percent, while the same category at 28 to 30 weeks carried a risk of 8.8 percent. The spline approach also yields tighter confidence intervals than the categorical models, because it estimates mortality at a specific size rather than averaging across a range of sizes, narrowing the interval from roughly ten percentage points to about four.

The study also surfaces a subtle and uncomfortable methodological problem in neonatal medicine: survival bias in the growth curves themselves. The Olsen curves were built from infants born between 1998 and 2006, and like all intrauterine growth references they exclude infants who died before hospital discharge, precisely the outcome this study measures. The 2013 to 2018 cohort analyzed here was skewed smaller than the curve-building population, especially at the most immature ages. The authors suggest that because the infants at greatest mortality risk were eliminated from the reference sample, current mortality prediction necessarily concentrates on the very smallest categories included in those curves, below the 3rd percentile. As survival of extremely vulnerable infants improves over time, this bias is likely to grow, meaning growth curves may progressively underrepresent the highest-risk babies.

The cohort itself reflects the diversity and clinical complexity of modern American neonatal care. Most infants, 88.3 percent, were exposed to at least one dose of antenatal steroids, and 68.9 percent were delivered by cesarean section. Just over half were male, and an equal proportion had at least one minor anomaly, most commonly a patent ductus arteriosus or a small septal defect. Pre-eclampsia complicated nearly one in five pregnancies, and 61.2 percent of infants were non-White. Median time to death ranged from four days for 30-week infants to nine days for 25-week infants, underscoring that mortality in this population is concentrated early in the NICU course, when interventions and monitoring decisions matter most.

The authors acknowledge limitations. The database captured only in-NICU mortality, without cause of death or linkage to maternal records that would allow evaluation of intrauterine growth restriction, placental insufficiency, or prenatal care. Gestational age was recorded in completed weeks as the neonatologist’s best estimate, and misestimated dates could partly explain the apparently lower mortality among LGA infants, since larger babies may simply be more mature than their charts suggest. Still, the study’s scale, diversity, and week-by-week stratification give its central conclusion considerable weight: mortality risk in preterm infants is better identified by subdividing SGA and AGA birthweight categories, while LGA subdivision adds nothing. The mortality curves and twelve-category framework could be incorporated into electronic medical records, giving clinicians an infant-specific baseline risk estimate drawn directly from a number, the birthweight z-score, that is already available on every admission. For the smallest and youngest babies, that extra resolution could mean earlier recognition of who truly needs the closest watch.

Subject of Research: Birthweight subclassification and mortality risk prediction in preterm NICU infants

Article Title: Impact of SGA, AGA, and LGA birthweight subclassification on NICU mortality risk identification: a cohort study

Article References: Ferguson, A. N., Stanton, J. R., Jones, B. D., Olsen, I. E., Clark, R. H., Bible, J., Reith, M., & Woo, J. G. (2026). Impact of SGA, AGA, and LGA birthweight subclassification on NICU mortality risk identification: a cohort study. Journal of Perinatology. https://doi.org/10.1038/s41372-026-02903-7

Image Credits: AI Generated

DOI: 10.1038/s41372-026-02903-7

Keywords: neonatology, NICU mortality, birthweight percentile, small for gestational age, appropriate for gestational age, large for gestational age, preterm infants, growth curves, birthweight z-score, logistic regression, Pediatrix Clinical Data Warehouse, risk prediction

The research team, led by Ahlan B. Lima and Humberto M. Carvalho of the Federal University of Santa Catarina and Lund University, systematically searched MEDLINE, Web of Science, and Scopus from database inception to October 2025. Following PRISMA guidelines and a protocol pre-registered on PROSPERO, the team screened thousands of records and ultimately included 28 longitudinal studies containing 41 independent samples spanning nine sports, from football and rugby to tennis, volleyball, and freestyle wrestling. Study durations ranged from 30 weeks to 11 years, with sample sizes between 10 and 849 participants and between three and twelve repeated measurement occasions per athlete. Only male athletes were analyzed, because the number of eligible studies including female athletes was insufficient to support separate modeling.

What sets this synthesis apart is its statistical machinery. Rather than averaging single snapshots of performance, the researchers applied Bayesian multilevel growth curve modeling to the repeated-measures data, fitting polynomial functions up to the cubic term to capture the nonlinear character of adolescent development. The multilevel structure acknowledged that repeated measurements are nested within individuals, individuals within studies, and studies within sports, while weakly regularizing priors kept estimates within biologically plausible ranges. From the fitted curves, the team derived the age at peak velocity—the moment when each fitness component improves fastest—using the first and second derivatives of the growth functions. Sensitivity analyses across alternative model specifications confirmed that the headline results were robust.

The central discovery is a striking split in developmental patterns. Vertical jump performance followed a pronounced curvilinear trajectory, rising from roughly 27 centimeters at age 10 to about 41 centimeters at age 17, with peak velocity occurring at 13.2 years. Endurance performance traced a similarly curved path, with peak velocity at 13.8 years. Both ages overlap the typical window of peak height velocity in adolescent boys, the period of most rapid growth in stature. Sprint performance, by contrast, improved in an almost perfectly linear fashion across ages 12 to 17, with 10-meter sprint times falling steadily from about 1.98 seconds to 1.83 seconds, without any distinct acceleration period corresponding to the pubertal growth spurt.

The physiological logic behind this divergence is compelling. Explosive power and endurance are tightly coupled to the biological upheavals of mid-puberty: surging muscle mass, increasing stature, enhanced cardiac stroke volume, greater capillary density, and improved oxidative enzyme capacity all peak in their developmental rate around the same years. Short-distance acceleration, however, appears to depend more on gradual neuromuscular adaptation and the slow accumulation of technical refinement in running mechanics—qualities that build consistently with training exposure rather than surging with a maturational trigger. The authors are careful to note that the temporal coincidence between fitness peaks and peak height velocity is a descriptive observation, not proof of causation, since biological maturity was not assessed in most included studies and training exposure could not be quantified.

Sport-specific curves added another layer of nuance. Rugby players posted the highest absolute vertical jump values, climbing from roughly 35 centimeters at age 10 to nearly 49 centimeters at 17, a profile consistent with the collision demands and resisted accelerations of their sport. Football and handball occupied intermediate positions, while basketball and volleyball—despite being jump-heavy sports—showed lower absolute values, likely reflecting the technical specificity of sport-specific jumping mechanics rather than inferior power development. Freestyle wrestling began with the lowest values but showed steep progression. Crucially, however, the shape of the developmental curves was broadly similar across sports within each fitness component, and sprint development showed minimal sport differences at all, suggesting that short-acceleration capacity is governed by general neuromuscular maturation shared across disciplines.

The practical payoff comes in the form of reference values that coaches can apply immediately. A countermovement jump without arm swing is expected to increase from approximately 25.5 centimeters at age 10 to 41.6 centimeters at 17; Yo-Yo IR1 distance rises from about 893 meters at age 12 to 1,610 meters at 17. The reported 68 percent credible intervals capture the range within which roughly two-thirds of athletes of the same age and sport are expected to fall, mirroring the convention used in standard pediatric growth charts. Performance persistently below the lower bound may warrant closer monitoring, while values above the upper bound may simply signal early maturation or advanced training exposure rather than exceptional talent.

The study is candid about its limitations. Aligning data by chronological age rather than biological maturity means individual variation in the timing of the growth spurt—roughly plus or minus one year around the population mean—is only partially captured. Publication bias assessments revealed asymmetry for sprint and endurance outcomes, so those pooled trajectories deserve somewhat more caution, whereas the vertical jump estimates showed no such asymmetry. Attrition is another concern: coach-driven deselection of weaker athletes could progressively inflate performance estimates at older ages. And because the analysis is restricted to males, the authors stress that female athletes, who follow distinct maturational timelines, require dedicated longitudinal evidence that does not yet exist in sufficient quantity.

Even with those caveats, the implications for youth sport are hard to overstate. The findings argue against early selection based on current performance levels, since a 12-year-old’s ranking may reflect nothing more than where he sits on his own maturational clock. Instead, they support long-term development models that prioritize progressive, individualized adaptation: expecting the fastest jump and endurance gains between ages 12 and 15, valuing steady year-on-year sprint improvement without demanding pubertal leaps, and interpreting every test result against age-referenced norms rather than against the teammate standing nearby. For a field long dominated by intuition and cross-sectional snapshots, this synthesis of nearly three decades of longitudinal data offers something rare—a rigorous, probabilistic map of what normal athletic development actually looks like.

Subject of Research: Longitudinal development of vertical jump, sprint, and endurance performance in male youth athletes

Article Title: Development of Physical Fitness in Young Athletes: A Longitudinal Meta-analysis

Article References: Lima, A. B., Quinaud, R. T., Reis, G. F., Karasiak, F. C., Ageberg, E., Clemente, F. M., Gonçalves, C. E., & Carvalho, H. M. (2026). Development of Physical Fitness in Young Athletes: A Longitudinal Meta-analysis. Sports Medicine – Open, 12(1), Article 144. https://doi.org/10.1186/s40798-026-01109-7

Image Credits: AI Generated

DOI: 10.1186/s40798-026-01109-7

Keywords: youth athletes, physical fitness, longitudinal meta-analysis, Bayesian modeling, vertical jump, sprint performance, endurance, peak height velocity, adolescent development, growth curves, sports medicine, athlete monitoring