By Mauro Rodríguez-Marín and Luis Gustavo Orozco-Alatorre
When Camila was seven years old, her parents noticed she was still the shortest in her class. “It’s normal, she’ll catch up,” people told them. But something didn’t seem right. After months of waiting to see a specialist, they finally received a diagnosis of growth delay—one that could have been detected much earlier.
This story—common in many households every day—inspired the development of a tool that combines artificial intelligence, open clinical data, and pediatric care.
Through the study Advances in Pediatric Growth Assessment with Machine Learning: Overcoming the Challenges of Early Diagnosis and Monitoring, researchers analyzed a set of biometric and demographic data from public institutions to develop a logistic regression model capable of identifying growth deviations in children with precision, speed, and clarity.
In Camila’s hypothetical case, where a growth disorder was suspected from age seven but took much longer to confirm, the tool described in the study would have enabled a far quicker diagnosis.
Predictive Growth Model
Logistic regression tools are based on machine learning, and their strength lies in delivering interpretable categorical results—such as the presence or absence of a risk. This makes them easier to integrate into clinical and educational settings and supports earlier detection of potential developmental disorders.
Because of its ability to identify risk factors at an early stage, this tool is already widely used in medical research.
In our project, we built on this approach to develop a model that predicts and helps healthcare professionals detect early warning signs of growth-related issues.
The proposed model can identify deviations in children’s height with 94.65% accuracy. Unlike other methods that predict numerical values, this one produces categorical outcomes (for example, “yes/no” or “positive/negative”) using variables such as age, weight, height, and patient history.
It is also explanatory, meaning it can show which of these variables has the strongest influence on the likelihood of a growth deviation.
Importantly, while the model does not diagnose a specific condition, it generates an alert that is both understandable and clinically useful. It flags cases where a child may fall outside the expected growth range, guiding doctors toward a more detailed evaluation.
This explanatory approach marks a key difference from other, more complex algorithms. Here, data does not replace medical judgment—it complements it.
For instance, the tool might detect that a child’s height is below the fifth percentile, considering their age, sex, and population context. That triggers an early alert, which can lead to timely medical consultation.
Clinical Results
Our logistic regression model was trained on a biometric and demographic dataset of over 2,400 children’s records, obtained from the Pediatric Growth Data Set at Stanford Medicine Children’s Health.
The algorithm achieved a sensitivity of 91.03% (its ability to detect positive cases). The dataset is also available on GitHub, making it easier to apply in different hospitals and educational settings, and ensuring the research can be reproduced.
The goal of this research project is to support earlier diagnoses, especially in resource-limited settings or in systems where long waiting lists delay access to specialized pediatric care.
This progress carries both clinical and social implications, since detecting a growth disorder early can mean the difference between effective treatment and a chronic condition.
Early Diagnosis
Child growth follows a predictable sequence, with developmental milestones that generally occur at similar ages across populations. However, many factors can contribute to growth problems in children and adolescents, often stemming from genetic, nutritional, environmental, psychosocial, or hormonal influences.
For example, short stature in families is usually genetic, while delayed growth and puberty often catch up later. Malnutrition and psychosocial stress can also disrupt the process.
Other contributors to growth disorders include endocrine imbalances—such as growth hormone deficiency or hypothyroidism—which can cause growth delays, highlighting the need for a holistic diagnostic approach.
Identifying these causes is essential to implementing effective treatment strategies. Regular follow-ups and adherence to prescribed therapies are crucial for long-term progress (growth phases such as adolescence are critical).
Reference
Rodríguez-Marín, M., y Orozco-Alatorre, LG (2025). Avances en la evaluación del crecimiento pediátrico con aprendizaje automático: Superando los desafíos del diagnóstico y la monitorización temprana. Children, 12 (3), 317
Authors
Mauro Rodríguez Marín. Professor at the Business School of Tecnológico de Monterrey, Guadalajara campus, where he serves as a faculty member, researcher, and consultant in the Department of Marketing and Analytics. He is a Level C member of the National System of Researchers (SNII) of SECIHTI and he holds a postdoctoral degree in Data Analytics from a joint program between Tec and the University of Texas at San Antonio (UTSA). His research areas include demand planning, artificial intelligence applied to business, healthcare, data analytics, and tourism. He has also been a visiting professor at universities such as Yale (USA) and Jean Moulin Lyon III (France).
Luis Gustavo Orozco Alatorre. Professor of Pediatrics at the University Center for Health Sciences of the University of Guadalajara and a Level 1 member of the National System of Researchers (SNI). He served as head of the Pediatric Specialty Program at the New Civil Hospital of Guadalajara, “Dr. Juan I. Menchaca,” accredited by SECIHTI’s PNPC. With over 33 years of private practice, he is certified by the Mexican Board of Pediatrics. He is a former president of the Jalisco Pediatrics Association, A.C., and a member of the Mexican Academy of Pediatrics, A.C. His research areas include child development and growth, pediatric pathologies, infectious diseases, and perinatal medicine
