Introduction. The most common approach to trace growth charts is the CG-LMS method. It is based on the estimate of three smoothing splines which model median, coefficient of variation and skewness of the auxological variable distribution as functions of age. In 1999 the EMGF (Extended Mechanistic Growth Function) method was applied, for the first time, to trace Neurofibromatosis Type1 growth charts. This alternative approach is based on a parametric (mechanistic) function, developed for tracing individual profiles (for example double or triple logistic, JPPS or PB1 function), with the inclusion of a few extra-parameters to model dispersion and skewness. The CG-LMS approach is essentially descriptive, whereas the EMGF method is based on an ad-hoc growth function chosen according prior knowledge about the growth process. Aim. The aim of this thesis is to evaluate the differences of performance between the two approaches in tracing height growth charts between 2 and 20 years. Methods. In this simulation study, we defined three different calibrators derived from height data from Central North Italian boys (2-20 years of age). The first calibrator (CG-LMS population) was generated on the basis of the three smoothing splines of CG-LMS model. The second one (E-JPPS population) was generated from the constants of E-JPPS4 function (i.e. an EMGF derived from JPPS function, JPPS plus 4 extra-constants). The third one (POP population) was a mixture of CG-LMS and E-JPPS population in 1:1 proportion. From each calibrator, 500 samples of 5 different sizes (1, 2, 5, 10 and 150 subjects/month) were extracted. The data of each of 3×5×500 samples were fitted with the best CG-LMS model, chosen according to Huiqi Pan criteria, and with E-JPPS4 model. Both models were implemented in SAS software. The average of the 500 values provided by each method for a given centile and a given age was regarded as the expected values of the estimate of that centile. The estimates obtained with the two methods were compared on the basis of their mean squared error or mse [squared of (computed – true population value)], bias (expected – true population value) and variance [squared of (computed – expected value)] separately for each month and averaged across the whole period. Furthermore the estimate of the age at pubertal take-off and pick of height velocity, two milestones of growth process, were computed. Results. The major differences between the two approaches are observed before the age of 4 years, where the E-JPPS4 mse is higher when samples are small (1-5 subjects/month), and during puberty, where the CG-LMS mse is almost always higher. As for CG-LMS, the pattern of the 97th centile bias is the mirror image of the 3rd centile bias, independently of the sample size and calibrator. As a consequence, the interval between the two centiles is underestimated at 14 years and overestimated at 16 years. Surprisingly, the average squared bias is lower for E-JPPS4 than for CG-LMS, even when the samples are extracted from the CG-LMS calibrator. Both methods detect the precession of pubertal inflection points (the pubertal peak of yearly increase): the higher the centile the earlier the inflection point. Discussion and Conclusion. The results of this simulation study indicate that EMGF performs as well as GC-LMS method. The lower squared bias of the estimates obtained with the E-JPPS4 model, independently of the calibrator, suggests the JPPS function properly describes the pattern of growth derived from cross-sectional data. The presence of the precession of the pubertal inflection points, observed also in CDC and UK charts, seems to indicate that this phenomenon is real and not an artifact due to the rather rigid structure of EMGF models.

MODELLI FLESSIBILI O RIGIDI PER TRACCIARE CARTE DI CRESCITA? UNA SIMULAZIONE / E. Spada ; tutor: S. Milani. Universita' degli Studi di Milano, 2012 Jan 23. 24. ciclo, Anno Accademico 2011. [10.13130/spada-elena_phd2012-01-23].

MODELLI FLESSIBILI O RIGIDI PER TRACCIARE CARTE DI CRESCITA? UNA SIMULAZIONE

E. Spada
2012

Abstract

Introduction. The most common approach to trace growth charts is the CG-LMS method. It is based on the estimate of three smoothing splines which model median, coefficient of variation and skewness of the auxological variable distribution as functions of age. In 1999 the EMGF (Extended Mechanistic Growth Function) method was applied, for the first time, to trace Neurofibromatosis Type1 growth charts. This alternative approach is based on a parametric (mechanistic) function, developed for tracing individual profiles (for example double or triple logistic, JPPS or PB1 function), with the inclusion of a few extra-parameters to model dispersion and skewness. The CG-LMS approach is essentially descriptive, whereas the EMGF method is based on an ad-hoc growth function chosen according prior knowledge about the growth process. Aim. The aim of this thesis is to evaluate the differences of performance between the two approaches in tracing height growth charts between 2 and 20 years. Methods. In this simulation study, we defined three different calibrators derived from height data from Central North Italian boys (2-20 years of age). The first calibrator (CG-LMS population) was generated on the basis of the three smoothing splines of CG-LMS model. The second one (E-JPPS population) was generated from the constants of E-JPPS4 function (i.e. an EMGF derived from JPPS function, JPPS plus 4 extra-constants). The third one (POP population) was a mixture of CG-LMS and E-JPPS population in 1:1 proportion. From each calibrator, 500 samples of 5 different sizes (1, 2, 5, 10 and 150 subjects/month) were extracted. The data of each of 3×5×500 samples were fitted with the best CG-LMS model, chosen according to Huiqi Pan criteria, and with E-JPPS4 model. Both models were implemented in SAS software. The average of the 500 values provided by each method for a given centile and a given age was regarded as the expected values of the estimate of that centile. The estimates obtained with the two methods were compared on the basis of their mean squared error or mse [squared of (computed – true population value)], bias (expected – true population value) and variance [squared of (computed – expected value)] separately for each month and averaged across the whole period. Furthermore the estimate of the age at pubertal take-off and pick of height velocity, two milestones of growth process, were computed. Results. The major differences between the two approaches are observed before the age of 4 years, where the E-JPPS4 mse is higher when samples are small (1-5 subjects/month), and during puberty, where the CG-LMS mse is almost always higher. As for CG-LMS, the pattern of the 97th centile bias is the mirror image of the 3rd centile bias, independently of the sample size and calibrator. As a consequence, the interval between the two centiles is underestimated at 14 years and overestimated at 16 years. Surprisingly, the average squared bias is lower for E-JPPS4 than for CG-LMS, even when the samples are extracted from the CG-LMS calibrator. Both methods detect the precession of pubertal inflection points (the pubertal peak of yearly increase): the higher the centile the earlier the inflection point. Discussion and Conclusion. The results of this simulation study indicate that EMGF performs as well as GC-LMS method. The lower squared bias of the estimates obtained with the E-JPPS4 model, independently of the calibrator, suggests the JPPS function properly describes the pattern of growth derived from cross-sectional data. The presence of the precession of the pubertal inflection points, observed also in CDC and UK charts, seems to indicate that this phenomenon is real and not an artifact due to the rather rigid structure of EMGF models.
23-gen-2012
Settore MED/01 - Statistica Medica
growth charts ; LMS method ; EMGF method ; JPPS function
MILANI, SILVANO
Doctoral Thesis
MODELLI FLESSIBILI O RIGIDI PER TRACCIARE CARTE DI CRESCITA? UNA SIMULAZIONE / E. Spada ; tutor: S. Milani. Universita' degli Studi di Milano, 2012 Jan 23. 24. ciclo, Anno Accademico 2011. [10.13130/spada-elena_phd2012-01-23].
File in questo prodotto:
File Dimensione Formato  
PHD_UNIMI_R08164_TESI_1.PDF

Open Access dal 06/01/2013

Tipologia: Tesi di dottorato completa
Dimensione 389.12 kB
Formato Adobe PDF
389.12 kB Adobe PDF Visualizza/Apri
PHD_UNIMI_R08164_TABELLE_2.PDF

Open Access dal 06/01/2013

Tipologia: Tesi di dottorato completa
Dimensione 970.84 kB
Formato Adobe PDF
970.84 kB Adobe PDF Visualizza/Apri
PHD_UNIMI_R08164_FIGURE_3.pdf

Open Access dal 06/01/2013

Tipologia: Tesi di dottorato completa
Dimensione 4.79 MB
Formato Adobe PDF
4.79 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2434/168455
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact