New Mathematical Model Predicts Which Children With Short Stature Will Respond to Growth Hormone Therapy

For families of children diagnosed with idiopathic short stature, one of the most frustrating questions is also the most fundamental: will growth hormone therapy actually work? The condition, known clinically as ISS, describes children whose height falls substantially below the expected range for their age and sex without any identifiable medical cause such as growth hormone deficiency, chronic disease, or genetic syndrome. Recombinant growth hormone is frequently prescribed to support growth in these children, yet responses vary enormously from patient to patient. Some children experience meaningful catch-up growth and climb steadily up the growth charts, while others show only limited improvement despite months or years of injections. This unpredictability has long made it difficult for clinicians and families to anticipate treatment benefit, set realistic expectations, and decide whether the burden of daily therapy is worthwhile. A new study from South Korea now offers a quantitative framework for answering that question before treatment even begins.

Researchers led by Professor Jung-woo Chae of Chungnam National University developed a mathematical model designed to capture individual differences in growth response over time among children with ISS receiving growth hormone treatment. The work, made available online in the journal Value in Health on July 21, 2026, takes a deliberately different approach from many earlier attempts to predict treatment outcomes. Rather than measuring success simply as height gain in centimeters, the team analyzed changes in age- and sex-adjusted height percentiles. This distinction matters clinically. A child who grows six centimeters in a year may still fall further behind peers if those peers grew eight centimeters, whereas a child who climbs from the first percentile toward the tenth is genuinely catching up relative to children of the same age and sex. Percentile-based measures therefore provide a more intuitive and clinically meaningful picture of catch-up growth than absolute height gains alone.

The study itself was retrospective and single-center, drawing on medical records from 91 prepubertal Korean children with idiopathic short stature, 41 boys and 50 girls, who received recombinant human growth hormone, the formulation known as somatropin, between July 2020 and December 2023. The hormone was administered subcutaneously six to seven times per week, with doses adjusted during routine clinical care rather than under experimental conditions. This real-world treatment context is important, because it means the model was built on the kind of heterogeneous, individually tailored dosing patterns that pediatricians actually encounter in practice, rather than the rigid protocols of a controlled trial. The researchers collected demographic data, growth measurements, parental heights, laboratory values, and treatment information from the records, assembling a longitudinal dataset that could support sophisticated modeling of growth trajectories.

At the heart of the analysis was a Gompertz nonlinear mixed-effects model, a mathematical framework well suited to biological growth processes. The Gompertz function describes growth that is initially rapid and then progressively slows, a pattern that mirrors how children respond to growth hormone therapy, where early catch-up is often followed by diminishing gains. The mixed-effects structure allows the model to separate population-level trends from individual deviations, meaning it can characterize both the typical response across the study group and the substantial variation between individual children. Critically, the researchers incorporated cumulative growth hormone exposure into the model, so that the total dose a child received over time, not just the prescribed daily amount, could be linked to the observed trajectory of height percentile. Over a mean treatment period of 619 days with a standard deviation of 307 days, the mean height percentile in the cohort rose from 1.26 at baseline to 9.16 at follow-up, a striking illustration of what catch-up growth looks like on the percentile scale.

The model estimated a total growth-response parameter of approximately 17.2 percentile points, representing the potential shift in height percentile attributable to treatment under the fitted framework. Yet the more clinically valuable finding was not the average but the spread: responses differed substantially among patients, and the model identified baseline characteristics that predicted where an individual child was likely to fall within that range. According to Professor Chae, three baseline factors were associated with a greater predicted response: higher body mass index, lower levels of insulin-like growth factor-binding protein 3, known as IGFBP-3, and shorter paternal height. Of these, body mass index showed the strongest association with predicted response. Each of these variables is routinely available before therapy starts, which is precisely what makes the finding actionable. A pediatrician weighing whether to initiate growth hormone treatment can already measure a child’s BMI, order an IGFBP-3 assay, and record parental heights without any additional testing burden.

The biological logic behind these predictors is plausible, even though the study establishes association rather than mechanism. Higher BMI has been linked in prior pediatric endocrinology literature to greater growth hormone responsiveness, possibly reflecting differences in metabolism and hormone signaling in children with more adipose tissue. IGFBP-3 is the principal carrier protein for insulin-like growth factor 1, the mediator through which growth hormone exerts much of its effect on the growth plate, and lower circulating levels may indicate a state in which the growth axis has more room to respond to exogenous stimulation. Shorter paternal height, meanwhile, may serve as a proxy for the genetic height potential that treatment is working against, with children whose fathers are shorter having relatively more percentile ground to recover. The model does not prove these causal pathways, but it demonstrates that routinely collected clinical variables carry genuine predictive information about treatment response.

To translate the statistical framework into something usable at the point of care, the team also developed GrowCast, a web-based tool that generates individualized predicted height and percentile trajectories. The tool requires only seven inputs: the child’s age, sex, current height, weight, paternal height, IGFBP-3 level, and the intended growth hormone dose. From these, it produces modeled growth trajectories that clinicians can inspect, compare, and discuss. Professor Chae emphasized that GrowCast enables clinicians to compare modeled treatment scenarios and communicate potential growth trajectories with patients and families. This simulation capability addresses one of the most persistent communication gaps in pediatric endocrinology. Families often struggle to interpret statements about average response rates, and a visualized, individualized trajectory, complete with the uncertainty inherent in prediction, can ground conversations about what therapy realistically offers a particular child.

The two-year simulations built into the study illustrate just how wide the range of predicted outcomes can be. Depending on a child’s body mass index, IGFBP-3 level, paternal height, and cumulative growth hormone exposure, median predicted height percentiles ranged from 4.0 percent in lower-response profiles to 31.0 percent in higher-response profiles. In other words, two children who look superficially similar at the start of treatment could be headed toward very different destinations on the growth chart. This heterogeneity is exactly why a one-size-fits-all description of growth hormone benefit has served families poorly, and why the researchers argue that individual characteristics should be considered when assessing the expected response to therapy. The simulations also allow exploration of different dosage regimens, showing how predicted trajectories shift under various treatment scenarios before any commitment to a particular plan is made.

The researchers are careful about the limits of what they have built. GrowCast is not yet a validated dosing-prescription tool, and the study’s retrospective, single-center design, with its cohort of 91 Korean children, means the model must be tested in larger, prospective, multicenter studies before it can be implemented routinely in clinical practice. Cohort-specific factors, local treatment conventions, and population differences in growth patterns all warrant verification in independent datasets. Nevertheless, the study offers a promising framework for more personalized growth hormone treatment in idiopathic short stature. By combining routinely available baseline characteristics with cumulative growth hormone exposure in a rigorous nonlinear mixed-effects model, the approach may help clinicians estimate individual growth trajectories, support more informed treatment planning, and move pediatric growth hormone therapy away from a trial-and-error enterprise toward genuine personalized medicine.

Subject of Research: Predictive modeling of growth hormone therapy response in children with idiopathic short stature

Article Title: Chungnam National University researchers develop model for predicting growth hormone therapy responses in children with short stature

Article References: Chungnam National University researchers develop model for predicting growth hormone therapy responses in children with short stature. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: idiopathic short stature, growth hormone therapy, Gompertz model, nonlinear mixed-effects modeling, height percentile, pediatric endocrinology, IGFBP-3, body mass index, personalized medicine, pharmacometrics, GrowCast, Value in Health