Real-world clinical databases have become an increasingly important resource for studying disease progression and treatment efficacy in routine clinical practice. Their observational nature, however, introduces significant methodological challenges resulting from hierarchical data structures, repeated measurements, irregular follow-up schedules, and heterogeneous patient trajectories. These issues are particularly relevant in differentiated thyroid cancer, a common endocrine malignancy characterized by excellent long-term survival but requiring prolonged surveillance to early detect disease persistence or recurrence. This thesis develops statistical methods for the analysis of complex longitudinal data collected within the Italian Thyroid Cancer Observatory (ITCO), a nationwide multicenter prospective cohort study established to monitor the clinical management and long-term outcomes of patients undergoing surgery for differentiated thyroid cancer. The overall objective is to improve the characterization of response to treatment by progressively exploiting the increasing amount of longitudinal information available during follow-up. The first contribution addresses the analysis of response to treatment approximately twelve months after initial therapy. A semiparametric multilevel cumulative probit model is proposed to account simultaneously for the ordinal nature of the clinical outcome and the hierarchical structure induced by patients nested within clinical centers. By adopting a finite-mixture representation of the random-effects distribution, the model provides a flexible characterization of unobserved center heterogeneity despite the absence of detailed center-level information. The second contribution extends this framework to repeated response-to-treatment assessments collected during follow-up. Since successive evaluations represent manifestations of the same underlying disease process, the response observed at 12 months cannot be regarded as an exogenous predictor of the response at 36 months. To address this issue, a multilevel endogenous ordinal model is applied, allowing the dependence between repeated clinical evaluations to be modeled explicitly while accounting for center-specific heterogeneity. The final contribution shifts the focus from discrete clinical outcomes to the longitudinal evolution of serum thyroglobulin, a key biomarker in differentiated thyroid cancer follow-up. Functional data analysis combined with fuzzy clustering techniques is employed to identify groups of patients sharing similar biomarker trajectories. The analyses show that clustering based on absolute thyroglobulin concentrations and clustering based on longitudinal variation capture complementary aspects of the disease evolution, highlighting distinct biochemical profiles associated with different response to treatment patterns. Overall, this thesis illustrates how statistical methodology can evolve alongside the increasing availability of longitudinal real-world data. Moving from single ordinal outcomes to repeated clinical assessments and, ultimately, to complete biomarker trajectories, the proposed approaches provide progressively richer representations of disease progression. Beyond their application to differentiated thyroid cancer, the methodologies employed in this work are broadly applicable to observational clinical studies characterized by multilevel structures, ordinal outcomes, and complex longitudinal processes.

Statistical Models for the Dynamics of Response to Treatment in Thyroid Cancer Patients / D'Elia, S.. - (2026 Sep 17).

Statistical Models for the Dynamics of Response to Treatment in Thyroid Cancer Patients

D'ELIA, SILVIA
17/09/2026

Abstract

Real-world clinical databases have become an increasingly important resource for studying disease progression and treatment efficacy in routine clinical practice. Their observational nature, however, introduces significant methodological challenges resulting from hierarchical data structures, repeated measurements, irregular follow-up schedules, and heterogeneous patient trajectories. These issues are particularly relevant in differentiated thyroid cancer, a common endocrine malignancy characterized by excellent long-term survival but requiring prolonged surveillance to early detect disease persistence or recurrence. This thesis develops statistical methods for the analysis of complex longitudinal data collected within the Italian Thyroid Cancer Observatory (ITCO), a nationwide multicenter prospective cohort study established to monitor the clinical management and long-term outcomes of patients undergoing surgery for differentiated thyroid cancer. The overall objective is to improve the characterization of response to treatment by progressively exploiting the increasing amount of longitudinal information available during follow-up. The first contribution addresses the analysis of response to treatment approximately twelve months after initial therapy. A semiparametric multilevel cumulative probit model is proposed to account simultaneously for the ordinal nature of the clinical outcome and the hierarchical structure induced by patients nested within clinical centers. By adopting a finite-mixture representation of the random-effects distribution, the model provides a flexible characterization of unobserved center heterogeneity despite the absence of detailed center-level information. The second contribution extends this framework to repeated response-to-treatment assessments collected during follow-up. Since successive evaluations represent manifestations of the same underlying disease process, the response observed at 12 months cannot be regarded as an exogenous predictor of the response at 36 months. To address this issue, a multilevel endogenous ordinal model is applied, allowing the dependence between repeated clinical evaluations to be modeled explicitly while accounting for center-specific heterogeneity. The final contribution shifts the focus from discrete clinical outcomes to the longitudinal evolution of serum thyroglobulin, a key biomarker in differentiated thyroid cancer follow-up. Functional data analysis combined with fuzzy clustering techniques is employed to identify groups of patients sharing similar biomarker trajectories. The analyses show that clustering based on absolute thyroglobulin concentrations and clustering based on longitudinal variation capture complementary aspects of the disease evolution, highlighting distinct biochemical profiles associated with different response to treatment patterns. Overall, this thesis illustrates how statistical methodology can evolve alongside the increasing availability of longitudinal real-world data. Moving from single ordinal outcomes to repeated clinical assessments and, ultimately, to complete biomarker trajectories, the proposed approaches provide progressively richer representations of disease progression. Beyond their application to differentiated thyroid cancer, the methodologies employed in this work are broadly applicable to observational clinical studies characterized by multilevel structures, ordinal outcomes, and complex longitudinal processes.
17-set-2026
File allegati a questo prodotto
Non ci sono file associati a questo prodotto.

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/11573/1782325
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact