The Partial Least Squares Path Modeling (PLS-PM) is a method meant to estimate a network of causal relationships defined according to a theoretical model. The complexity of the theoretical construct is studied by taking into account the relationships among non measurable indicators (latent variables), represented by a set of observed variables (manifest variables). PLS-PM aims to estimate, through a system of interdependent equations based on simple and multiple regressions, the network of relations among the manifest variables and their own latent variable, and among the latent variables inside the model. The causal relationships among variables are represented through a Path Diagram, in which the latent variables are enclosed in circles and the manifest variables are enclosed in boxes. PLS-PM involves three sets of relations: 1) structural or inner model, 2) measurement or outer model, 3) the weight relations upon which latent variable scores can be calculated. The first model takes into account the relations among the latent variables and the second takes into account the relations between manifest variables and the corresponding latent variable. In the structural model each endogenous (dependent) latent variable is linked to the others by a multiple regression model. The structural design only assumes recursive models, i.e. the path diagram takes the form of a causal chain with no loops. Different types of measurement models exists, depending the kind of relationship: 1) reflective model (observed variables are considered being caused by the latent variable (i.e., indicators reflect the construct; the latent variable is considered as the cause of the manifest variables and each manifest variable is an effect of the unique corresponding latent variable); 2) formative model (the latent variables are considered as being caused by its manifest variables); and 3) MIMIC model (multiple effect indicators for multiple causes, it represents a mixture of both the reflective and the formative models within the same block of manifest variables). Independently from the type of measurement model, the standardized latent variable scores are computed as a linear combination of its manifest variables and outer weights (the so-called weight relation). Once the theoretical model is specified, the next phase in PLS-PM is the estimation of the model parameters. The PLS algorithm consists of three stages. The first stage is an iterative procedure of ordinary least squares regressions taking into account the relationships of the structural and measurement model, in order to calculate weights required to give final estimates for each latent variable. This first stage is the “core” stage in the PLS algorithm. Subsequently, the second and third stage involve the non-iterative estimation of the coefficients of the structural and measurement model, respectively. The structural model coefficients (path coefficients) are calculated by ordinary least squares regressions between latent variables. The measurement model coefficients (loading coefficients) are also estimated by regressions but taking into account the kind of mode to be used (reflective or formative). PLS-PM has been widely used in economical (the customer satisfaction is a typical example) and psychological settings. In biomedical context, the published articles are scanty and generally published in open access journals. The aim of this study was to apply the PLS-PM in a different field, since it has been widely used in economical (the customer satisfaction is a typical example) and psychological setting. In biomedical context, the published articles are scanty and generally published in open access journals. I used the PLS-PM method in order to analyze the adherence of the procedures provided for diagnosis, treatment (surgical and medical), and follow-up of breast cancer through a set of indicators. Indeed, the used approaches in this field since oversimplify the complex problem since they do not consider simultaneously multiple aspects of the diagnostic, therapeutic and follow-up pathways. This method has several strengths, as PLS-PM allows the reduction of dimensionality of several health indicators into a smaller number of latent variables (and more interpretable), and then allows to study causal relationships between these latent variables, representing the different aspects of the diagnostic, therapeutic and follow-up pathways. This method also requires no distributional assumptions with respect to the variables included in the model. The limit of this method is the bias deriving from the a priori selection of the relationships among latent variables and of the indicators used to characterize the latent variable. Although the limited sample size makes the analyses explorative-orientated only, the present study represents an unique example of PLS-PM application in the biomedical research, in particular in the evaluation of the adherence of the diagnostic and treatment procedures for breast cancer.

PARTIAL LEAST SQUARE PATH MODELING APPROACH IN BIOMEDICAL RESEARCH / V. Rosato ; advisor: A. Decarli. DIPARTIMENTO DI SCIENZE CLINICHE E DI COMUNITA', 2015 Jan 15. 27. ciclo, Anno Accademico 2014. [10.13130/v-rosato_phd2015-01-15].

PARTIAL LEAST SQUARE PATH MODELING APPROACH IN BIOMEDICAL RESEARCH

V. Rosato
2015

Abstract

The Partial Least Squares Path Modeling (PLS-PM) is a method meant to estimate a network of causal relationships defined according to a theoretical model. The complexity of the theoretical construct is studied by taking into account the relationships among non measurable indicators (latent variables), represented by a set of observed variables (manifest variables). PLS-PM aims to estimate, through a system of interdependent equations based on simple and multiple regressions, the network of relations among the manifest variables and their own latent variable, and among the latent variables inside the model. The causal relationships among variables are represented through a Path Diagram, in which the latent variables are enclosed in circles and the manifest variables are enclosed in boxes. PLS-PM involves three sets of relations: 1) structural or inner model, 2) measurement or outer model, 3) the weight relations upon which latent variable scores can be calculated. The first model takes into account the relations among the latent variables and the second takes into account the relations between manifest variables and the corresponding latent variable. In the structural model each endogenous (dependent) latent variable is linked to the others by a multiple regression model. The structural design only assumes recursive models, i.e. the path diagram takes the form of a causal chain with no loops. Different types of measurement models exists, depending the kind of relationship: 1) reflective model (observed variables are considered being caused by the latent variable (i.e., indicators reflect the construct; the latent variable is considered as the cause of the manifest variables and each manifest variable is an effect of the unique corresponding latent variable); 2) formative model (the latent variables are considered as being caused by its manifest variables); and 3) MIMIC model (multiple effect indicators for multiple causes, it represents a mixture of both the reflective and the formative models within the same block of manifest variables). Independently from the type of measurement model, the standardized latent variable scores are computed as a linear combination of its manifest variables and outer weights (the so-called weight relation). Once the theoretical model is specified, the next phase in PLS-PM is the estimation of the model parameters. The PLS algorithm consists of three stages. The first stage is an iterative procedure of ordinary least squares regressions taking into account the relationships of the structural and measurement model, in order to calculate weights required to give final estimates for each latent variable. This first stage is the “core” stage in the PLS algorithm. Subsequently, the second and third stage involve the non-iterative estimation of the coefficients of the structural and measurement model, respectively. The structural model coefficients (path coefficients) are calculated by ordinary least squares regressions between latent variables. The measurement model coefficients (loading coefficients) are also estimated by regressions but taking into account the kind of mode to be used (reflective or formative). PLS-PM has been widely used in economical (the customer satisfaction is a typical example) and psychological settings. In biomedical context, the published articles are scanty and generally published in open access journals. The aim of this study was to apply the PLS-PM in a different field, since it has been widely used in economical (the customer satisfaction is a typical example) and psychological setting. In biomedical context, the published articles are scanty and generally published in open access journals. I used the PLS-PM method in order to analyze the adherence of the procedures provided for diagnosis, treatment (surgical and medical), and follow-up of breast cancer through a set of indicators. Indeed, the used approaches in this field since oversimplify the complex problem since they do not consider simultaneously multiple aspects of the diagnostic, therapeutic and follow-up pathways. This method has several strengths, as PLS-PM allows the reduction of dimensionality of several health indicators into a smaller number of latent variables (and more interpretable), and then allows to study causal relationships between these latent variables, representing the different aspects of the diagnostic, therapeutic and follow-up pathways. This method also requires no distributional assumptions with respect to the variables included in the model. The limit of this method is the bias deriving from the a priori selection of the relationships among latent variables and of the indicators used to characterize the latent variable. Although the limited sample size makes the analyses explorative-orientated only, the present study represents an unique example of PLS-PM application in the biomedical research, in particular in the evaluation of the adherence of the diagnostic and treatment procedures for breast cancer.
15-gen-2015
Settore MED/01 - Statistica Medica
DECARLI, ADRIANO
Doctoral Thesis
PARTIAL LEAST SQUARE PATH MODELING APPROACH IN BIOMEDICAL RESEARCH / V. Rosato ; advisor: A. Decarli. DIPARTIMENTO DI SCIENZE CLINICHE E DI COMUNITA', 2015 Jan 15. 27. ciclo, Anno Accademico 2014. [10.13130/v-rosato_phd2015-01-15].
File in questo prodotto:
File Dimensione Formato  
phd_unimi_R09565.pdf

Open Access dal 08/07/2016

Tipologia: Tesi di dottorato completa
Dimensione 1.64 MB
Formato Adobe PDF
1.64 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/2434/253801
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact