Diagnosing growing pains in children by using machine learning: a cross-sectional multicenter study


AKAL F., BATU AKAL E. D., Sonmez H. E., Karadag S. G., Demir F., Ayaz N. A., ...More

MEDICAL & BIOLOGICAL ENGINEERING & COMPUTING, vol.60, no.12, pp.3601-3614, 2022 (SCI-Expanded) identifier identifier identifier

  • Publication Type: Article / Article
  • Volume: 60 Issue: 12
  • Publication Date: 2022
  • Doi Number: 10.1007/s11517-022-02699-6
  • Journal Name: MEDICAL & BIOLOGICAL ENGINEERING & COMPUTING
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, ABI/INFORM, Applied Science & Technology Source, BIOSIS, Biotechnology Research Abstracts, Business Source Elite, Business Source Premier, CINAHL, Compendex, Computer & Applied Sciences, EMBASE, INSPEC, MEDLINE
  • Page Numbers: pp.3601-3614
  • Keywords: Growing pains, Machine learning, Artificial intelligence, Diagnosis, TESTS
  • Hacettepe University Affiliated: Yes

Abstract

Growing pains (GP) are the most common cause of recurrent musculoskeletal pain in children. There are no diagnostic criteria for GP. We aimed at analyzing GP-related characteristics and assisting GP diagnosis by using machine learning (ML). Children with GP and diseased controls were enrolled between February and August 2019. ML models were developed by using tenfold cross-validation to classify GP patients. A total of 398 patients with GP (F/M:1.3; median age 102 months) and 254 patients with other diseases causing limb pain were enrolled. The pain was bilateral (86.2%), localized in the lower extremities (89.7%), nocturnal (74%), and led to awakening at night (60.8%) in most GP patients. History of arthritis, trauma, morning stiffness, limping, limitation of activities, and school abstinence were more prevalent among controls than in GP patients (p = 0.016 for trauma; p < 0.001 for others). The experiments with different ML models revealed that the Random Forest algorithm had the best performance with 0.98 accuracy, 0.99 sensitivity, and 0.97 specificity for GP diagnosis. This is the largest cohort study of children with GP and the first study that attempts to diagnose GP by using ML techniques. Our ML model may be used to facilitate diagnosing GP.