ESPE Abstracts (2022) 95 P1-316

ESPE2022 Poster Category 1 Growth and Syndromes (85 abstracts)

Curve matching to predict future growth in patients receiving recombinant human growth hormone: an interpretable and explainable method using big data

Paula van Dommelen 1 , Lilian Arnaud 2 & Ekaterina Koledova 3

1The Netherlands Organization for Applied Scientific Research TNO, Leiden, Netherlands; 2Connected Health & Devices, Global Healthcare Operations, Ares Trading S.A (an affiliate of Merck KGaA, Darmstadt, Germany), Eysins, Switzerland; 3Global Medical Affairs Cardiometabolic & Endocrinology, Merck Healthcare KGaA, Darmstadt, Germany

Background: Prediction models demonstrate potential in predicting growth in patients receiving recombinant human growth hormone (r-hGH) for growth disorders. However, considerable information from patients is needed to calculate a predicted growth curve. The curve matching technique only requires height data. This technique identifies growth curves that are similar (matched) among real-world patients within a database. The growth curves of these ‘matched’ patients can then be plotted on the growth chart of the new patient to show and predict how this patient will grow.

Aim: To investigate the validity of curve matching to predict growth in patients with growth hormone deficiency (GHD) and those born small for gestational age (SGA).

Patients and Methods: Height data from the Easypod Connect Observational Study and data extracted from the easypod™ connect ecosystem were analysed. Patients’ monthly height standard deviation scores (HSDS) by treatment duration (0–48 months) were calculated using the broken stick method, and missing data imputed (matching database). Patients with GHD and those born SGA, HSDS at start (-4, <-1 standard deviation [SD]), age at start (3–<16 years), a yearly increase in HSDS (0–24 months), and no yearly decrease in HSDS (24–48 months) were selected. Validity was determined using the SD from the observed HSDS minus the weighted (Proportion of good database matches) mean predicted HSDS curve of 25 other patients with best matching scores (residual analysis).

Results: Within the matching database, data for 3,213 patients (2,487 GHD; 726 SGA) with monthly HSDS were available. The observed database comprised 2,472 patients (1,897 GHD; 575 SGA) with 16,624 HSDS measurements. The error SD at 12 months (range 10–14) of treatment decreased from 0.3 when only HSDS at start is known to 0.2 when an additional measurement in the first three months of treatment was available. The error SD at 24 months (range 21–27) decreased from 0.3 when a measurement at start and between 4–6 months was available to 0.2 when a measurement at start and between 9–12 months was available. When measurements up until 13–24 months were available, the error SD was 0.3 at 36 months (range 33–39) and 48 months (range 40–48).

Conclusion: Curve matching is a valid technique that provides an interpretable and explainable visualisation and prediction of growth in patients with GHD and SGA. It may contribute to the early detection of unique or deviating growth curves, supporting clinical decision-making.

Volume 95

60th Annual ESPE (ESPE 2022)

Rome, Italy
15 Sep 2022 - 17 Sep 2022

European Society for Paediatric Endocrinology 

Browse other volumes

Article tools

My recent searches

No recent searches.