Back to search

Harokopio University

Διερεύνηση ευαισθησίας και ειδικότητας σύνθετων κλιμάκων αποτελούμενων από διακριτές ή συνεχείς τυχαίες μεταβλητές με εφαρμογές σε μοντέλα προσομοιώσεων

Abstract

dc:description

Background/Introduction: The development of information science with the discovery of techniques and tools at the machine language level has contributed to the further development of many other disciplines such as Medicine, Biology, Biometrics, Statistics etc. (Sim et al., 2012; Wong, 2016). In several cases, the parallel development of two distinct branches has led to the creation of new scientific areas. For example, the application of Informatics and Statistics to Bio-Medical Sciences has led to the distinct but inextricably linked disciplines of Bioinformatics and Biostatistics (Moody, 2004; Indrayan, 2012). Bio-Statistics and Bio-Informatics are relatively recent scientific fields, however, they are rapidly growing in recent years. Biomedical sciences use mathematical scales to measure accurately and reliably a clinical or behavioral characteristic of a person (McDowell, 2006). Composite scales are used in a variety of scientific areas, such as biometrics, psychometry and medicine, and are designed to measure various clinical features, which are difficult to measure quantitatively (Kant, 1996; Streiner & Norman, 2008). Scales are quantitative, continuous or discrete, random variables that follow a known or unknown distribution, intending to evaluate biochemical (eg, Serum Serum Level, Cholesterol Level, etc ), Clinical (eg Sense of Pain, Depression, etc.) or behavioral characteristics (eg General Dietary Standard, Degree of Adoption of Mediterranean Diet, etc.) of individuals of a population (Beck, 1961; Huskisson 2012). Scales are derived from the simple (unweighted) or weighted algebraic sum of a set of m variables. The m variables express distinct components of the feature that the scale aims to estimate. Advanced health indices are also used as healthy/patient segregation tools and are widely used in physic-biological, psychometric and social science research (Bansal & Pepe, 2013). The methodology for the construction of scales is not unambiguously established in the international literature but involves a high degree of subjectivity of the researcher and depends on the type of the characteristic that the scale aims to assess (Panagiotakos, 2009). Aim: The purpose of this dissertation is to quantify the relationship between the diagnostic accuracy of composite scales and their specific structural characteristics, influencing their diagnostic accuracy, using a theoretical and/or analytical proofing process, such as: (A) The k number of partitions/divisions of m discrete component variables, in the case of a discrete composite scale (i.e. an index created by discrete component variables), (Bersimis F. et al, 2013). (B) The l range of the support/field of definition of m continuous component variables, in the case of a continuous composite scale (an index generated by continuous component variables), (Bersimis F. et al., 2016). (C) The performance of weights wi in the component variables of a composite scale. The weighting value expresses the degree of correlation between the component variables and the Y characteristic that the scale aims to evaluate (Bersimis F. et al., 2017). (D) The number m of the composite scale’s variables (Bersimis F. et al, 2017). From the aforementioned individual investigations, the holistic goal of this dissertation is to form a proposed methodological background for the construction of scales with high diagnostic ability and predictive precision. Material-Method: This thesis uses theoretical evidence using algebra and analysis elements, as well as a number of simulated data scenarios. More specifically, in the case of investigating the sensitivity of the number k of divisions/partitions of discrete components for variables that follow an unknown distribution or the uniform, the corresponding propositions using algebra elements (Sequences and Rows) are theoretically proved. Theoretical proof is also used in the case of exploration of the sensitivity of continuous indices for uniformly or normally distributed variables using analysis elements (Differential and Integral Calculus). In the case of investigating the use of weights in composite health indices using simulation (Monte Carlo), a set of 1,000 records were created with a differentiation in the parameters of the distributions used, in terms of the number m of the component variables, in terms of size of the samples (e.g., 100, 1,000, 10,000, etc.). More specifically, uniform and normal distribution were used with a variety of parameterization for each scenario. Also, modified distributions are used to approximate the empirical distributions of various characteristics of the medical-biological space. In the context of this thesis, nine scales Ti, i = 0, 1,2, ..., 8 were constructed with different weighting methodology each. For T0 no weighting was used, so it is derived from the simple algebraic sum of the variables Xj, j = 1,2, .., m. Weights were derived from the application of logistic regression and discriminant analysis with a dependent variable Y expressing the clinical condition of individuals and independent of the explanatory component variables Xj, j = 1, 2, .., m. The diagnostic accuracy of the scales was evaluated by calculating the surface area under the (ROC) receiver operating characteristic curve, Sensitivity (True Positive Ratio), Specificity (True Negative Ratio), Accuracy (True Ratio) Positive and Negative predictive value. Using simulation, their predictive capacity was also evaluated and weighted indices were evaluated in the case of actual dietary data from the ATTICA study and are generated using 5 components, which are randomly selected from the Mediterranean diet. Results: The theoretical evidence, both in the case of discrete indices and in the case of continuous scales, shows that the diagnostic accuracy of a scale measured by the sensitivity function, improves when the number k of the divisions of discrete variables increases or when the width l of the support (field of definition) of continuous variables becomes broader. The degree of correlation between the component variables was found to affect the accuracy of a scale. More specifically, initially a set of simulated data was used from correlated variables with a different correlation coefficient, and then the Principal Component Analysis was used to produce rectangular (uncorrelated) variables. An increase in the capacity of the scales was observed when using the unrelated variables produced. Accuracy of the scale is also improved if the number m of component variables increase too. The accuracy of the weighted scales is higher than the diagnostic accuracy of a non-weighted scale. The above results were applied to a nutritional scale based on MedDieScore component variables in a sample of 981 men and women in terms of cardiovascular risk factors (i.e. obesity, diabetes mellitus, hypertension, hypercholesterolemia) after screening for potential confounders. Conclusions: A composite scale presents high ability when the following conditions are met: A) The number k of the partitions/divisions of m discrete component variables takes the maximum possible value, without the problem of the misclassification of individuals of a population in the case of a discrete composite scale (Bersimis F. et al, 2013). B) The l range of the support/field of definition of the m continuous component variables, in the case of a continuous composite scale, without the problem of the misclassification of the individuals of a population (Bersimis F. et al., 2016). C) The weighting wi in the component variables of a composite scale is derived from a statistical classification method, such as Logistic Regression or Discriminant Analysis (Bersimis F. et al., 2017). D) The number m of composite health indicator variables is the maximum, i.e. all the variables associated with the outcome are used (Bersimis F. et al., 2017). Scales are widely used in various disciplines, particularly in the health sector, to detect features or attitudes that are difficult to be detected (Kant, 1996). Accurate measurement of a characteristic, using a composite scale, offers the possibility of early and accurate diagnosis of a disease, resulting in the most appropriate treatment of the diseased person who suffers and hence the improvement of the quality of his life. Extending past studies, this thesis proposes methods to improve the h scales’ diagnostic capabilities that provide public health benefits, with direct results in the use of less government resources to treat the sick in a population.

Degree

thesis:*
Grantor dc:publisher
Harokopio University
Year dc:date
2019

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Bersimis, Fragkiskos
  • Μπερσίμης, Φραγκίσκος

Subjects

dc:subject × 6

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/45365

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Bersimis, Fragkiskos; Μπερσίμης, Φραγκίσκος. Διερεύνηση ευαισθησίας και ειδικότητας σύνθετων κλιμάκων αποτελούμενων από διακριτές ή συνεχείς τυχαίες μεταβλητές με εφαρμογές σε μοντέλα προσομοιώσεων. Harokopio University, 2019. http://hdl.handle.net/10442/hedi/45365