{"id":{"repo_id":"uiuc","oai_identifier":"oai:www.ideals.illinois.edu:2142/132651"},"canonical_url":"https://search.dev.ndltd.org/etd/uiuc/oai:www.ideals.illinois.edu:2142/132651","repository":{"repo_id":"uiuc","name":"University of Illinois - Urbana-Champaign","base_url":"https://www.ideals.illinois.edu/oai-pmh"},"display":{"title":"Enhancing health equity in antidiabetic medication use and diabetes care outcomes through data science approaches","abstract":"Diabetes remains a major public health concern in the United States, affecting more than 10% of the population, with 90% to 95% of cases attributable to type 2 diabetes (T2D). Antidiabetic medications play a central role in glycemic control and in preventing or delaying diabetes-related complications. Understanding differences in antidiabetic medication use patterns and promoting sustained adherence are essential for reducing disparities and improving diabetes outcomes. The overarching objective of this dissertation is to apply data-oriented analytical and machine learning models to enhance understanding of inequities in antidiabetic medication use and to promote more equitable and effective diabetes care. In the first study, we examined racial and geographic disparities in antidiabetic medication discontinuation among Medicare beneficiaries with T2D residing within or surrounding Diabetes Belt (DB) counties. We defined discontinuation as at least 90 consecutive days without any antidiabetic medication supply, and the overall discontinuation rate in our dataset in 41.2%. The Cox regression model reflected that non-Hispanic (NH) Black individuals had a significantly higher discontinuation risk (HR: 1.22, 95% CI: [1.162, 1.286]) compared to NH White individuals. In contrast, no significant difference was observed between residents of DB and non-DB areas after adjustment for socioeconomic factors (HR: 1.02, 95% CI: [0.973, 1.072]). In the second study, we applied a novel time-series clustering approach to characterize long-term antidiabetic medication use patterns over a three-year period. We defined five different transitions (switch, intensification, de-intensification, discontinuation, and re-initiation) based on changes in the number of concurrently used medications to describe longitudinal medication patterns. The most common patterns included continuous metformin monotherapy (36.3%), recurrent cycles of discontinuation and re-initiation (23.1%), one-time discontinuation (14.2%), switching (3.0%), and intensification (2.8%). These patterns were grouped into five clinically meaningful clusters. Cluster-level comparisons indicated that NH Black patients were more likely to experience discontinuation, whereas female patients were more likely to undergo intensification. Patients who switched or re-initiated therapy tended to be older, had more comorbidities, and higher rates of emergency and inpatient visits. In the third study, we developed deep learning models to predict future medication adherence using patients’ longitudinal medication histories and demographic and clinical characteristics. The baseline models, including long short-term memory (LSTM), LSTM-Attention, and Transformer Encoders, achieved comparable performance, with an accuracy of approximately 82.6%, an F1 score of 0.85, and an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.90. To evaluate how the complexity of medication-use histories affects prediction, patients were divided into low, medium, and high groups using Jenks natural breaks based on their number of medication transitions. Two modeling strategies were then examined to capture the heterogeneous temporal patterns across these groups. The first strategy trained independent sequential models for each group, but this approach did not improve performance and showed reduced accuracy for patients with medium or high number of transitions compared to the baseline models. The second strategy used a shared LSTM backbone with group-specific attention heads, allowing the model to adapt to different levels of transition complexity while maintaining a single architecture. This shared model achieved modest improvements in medium and high group in both accuracy (medium group: 82.9% compared with 82.7%; high group: 78.5% compared with 78.3%) and F1 score (medium group: 0.845 compared with 0.844; high group: 0.821 compared with 0.816) relative to baseline models. These three studies collectively demonstrate that data-driven approaches can connect population-level equity assessment with individualized precision care. By integrating survival analysis, clustering, and machine learning, this work illustrates how data science can reveal structural inequities, identify subgroup-specific challenges, and enable personalized interventions. Advancing health equity in diabetes care requires both population-level policy efforts and patient-centered predictive tools, developed through interdisciplinary collaboration and equitable model design.","abstract_html":"Diabetes remains a major public health concern in the United States, affecting more than 10% of the population, with 90% to 95% of cases attributable to type 2 diabetes (T2D). Antidiabetic medications play a central role in glycemic control and in preventing or delaying diabetes-related complications. Understanding differences in antidiabetic medication use patterns and promoting sustained adherence are essential for reducing disparities and improving diabetes outcomes. The overarching objective of this dissertation is to apply data-oriented analytical and machine learning models to enhance understanding of inequities in antidiabetic medication use and to promote more equitable and effective diabetes care. In the first study, we examined racial and geographic disparities in antidiabetic medication discontinuation among Medicare beneficiaries with T2D residing within or surrounding Diabetes Belt (DB) counties. We defined discontinuation as at least 90 consecutive days without any antidiabetic medication supply, and the overall discontinuation rate in our dataset in 41.2%. The Cox regression model reflected that non-Hispanic (NH) Black individuals had a significantly higher discontinuation risk (HR: 1.22, 95% CI: [1.162, 1.286]) compared to NH White individuals. In contrast, no significant difference was observed between residents of DB and non-DB areas after adjustment for socioeconomic factors (HR: 1.02, 95% CI: [0.973, 1.072]). In the second study, we applied a novel time-series clustering approach to characterize long-term antidiabetic medication use patterns over a three-year period. We defined five different transitions (switch, intensification, de-intensification, discontinuation, and re-initiation) based on changes in the number of concurrently used medications to describe longitudinal medication patterns. The most common patterns included continuous metformin monotherapy (36.3%), recurrent cycles of discontinuation and re-initiation (23.1%), one-time discontinuation (14.2%), switching (3.0%), and intensification (2.8%). These patterns were grouped into five clinically meaningful clusters. Cluster-level comparisons indicated that NH Black patients were more likely to experience discontinuation, whereas female patients were more likely to undergo intensification. Patients who switched or re-initiated therapy tended to be older, had more comorbidities, and higher rates of emergency and inpatient visits. In the third study, we developed deep learning models to predict future medication adherence using patients’ longitudinal medication histories and demographic and clinical characteristics. The baseline models, including long short-term memory (LSTM), LSTM-Attention, and Transformer Encoders, achieved comparable performance, with an accuracy of approximately 82.6%, an F1 score of 0.85, and an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.90. To evaluate how the complexity of medication-use histories affects prediction, patients were divided into low, medium, and high groups using Jenks natural breaks based on their number of medication transitions. Two modeling strategies were then examined to capture the heterogeneous temporal patterns across these groups. The first strategy trained independent sequential models for each group, but this approach did not improve performance and showed reduced accuracy for patients with medium or high number of transitions compared to the baseline models. The second strategy used a shared LSTM backbone with group-specific attention heads, allowing the model to adapt to different levels of transition complexity while maintaining a single architecture. This shared model achieved modest improvements in medium and high group in both accuracy (medium group: 82.9% compared with 82.7%; high group: 78.5% compared with 78.3%) and F1 score (medium group: 0.845 compared with 0.844; high group: 0.821 compared with 0.816) relative to baseline models. These three studies collectively demonstrate that data-driven approaches can connect population-level equity assessment with individualized precision care. By integrating survival analysis, clustering, and machine learning, this work illustrates how data science can reveal structural inequities, identify subgroup-specific challenges, and enable personalized interventions. Advancing health equity in diabetes care requires both population-level policy efforts and patient-centered predictive tools, developed through interdisciplinary collaboration and equitable model design.","abstract_has_math":false,"creators":["Zhang, Peng"],"institution":"University of Illinois Urbana-Champaign","degree_name":"Ph.D.","degree_level":"Dissertation","degree_discipline":"Community Health","degree_department":null,"school":null,"contributors":["Kang, Hyojung","Hallal, Pedro","He, Jing Rui","Quinn, Lauretta"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-12","date_published":"2025-12","updated_at":"2026-07-22T22:25:07Z","subjects":["type 2 diabetes","machine learning","antidiabetic medication adherence","racial and geographic disparities","longitudinal medication patterns"],"languages":["en"],"rights":["Copyright 2025 Peng Zhang"],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/2142/132651","outbound_label":"Handle","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Kang, Hyojung","Hallal, Pedro","He, Jing Rui","Quinn, Lauretta"]},{"key":"dc:creator","label":"Author","values":["Zhang, Peng"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025-12","2025-11-25"]},{"key":"dc:type","label":"Dc Type","values":["text","Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Community Health"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Ph.D."]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Illinois Urbana-Champaign"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["type 2 diabetes","machine learning","antidiabetic medication adherence","racial and geographic disparities","longitudinal medication patterns"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Copyright 2025 Peng Zhang"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://hdl.handle.net/2142/132651"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Diabetes remains a major public health concern in the United States, affecting more than 10% of the population, with 90% to 95% of cases attributable to type 2 diabetes (T2D). Antidiabetic medications play a central role in glycemic control and in preventing or delaying diabetes-related complications. Understanding differences in antidiabetic medication use patterns and promoting sustained adherence are essential for reducing disparities and improving diabetes outcomes. The overarching objective of this dissertation is to apply data-oriented analytical and machine learning models to enhance understanding of inequities in antidiabetic medication use and to promote more equitable and effective diabetes care. In the first study, we examined racial and geographic disparities in antidiabetic medication discontinuation among Medicare beneficiaries with T2D residing within or surrounding Diabetes Belt (DB) counties. We defined discontinuation as at least 90 consecutive days without any antidiabetic medication supply, and the overall discontinuation rate in our dataset in 41.2%. The Cox regression model reflected that non-Hispanic (NH) Black individuals had a significantly higher discontinuation risk (HR: 1.22, 95% CI: [1.162, 1.286]) compared to NH White individuals. In contrast, no significant difference was observed between residents of DB and non-DB areas after adjustment for socioeconomic factors (HR: 1.02, 95% CI: [0.973, 1.072]). In the second study, we applied a novel time-series clustering approach to characterize long-term antidiabetic medication use patterns over a three-year period. We defined five different transitions (switch, intensification, de-intensification, discontinuation, and re-initiation) based on changes in the number of concurrently used medications to describe longitudinal medication patterns. The most common patterns included continuous metformin monotherapy (36.3%), recurrent cycles of discontinuation and re-initiation (23.1%), one-time discontinuation (14.2%), switching (3.0%), and intensification (2.8%). These patterns were grouped into five clinically meaningful clusters. Cluster-level comparisons indicated that NH Black patients were more likely to experience discontinuation, whereas female patients were more likely to undergo intensification. Patients who switched or re-initiated therapy tended to be older, had more comorbidities, and higher rates of emergency and inpatient visits. In the third study, we developed deep learning models to predict future medication adherence using patients’ longitudinal medication histories and demographic and clinical characteristics. The baseline models, including long short-term memory (LSTM), LSTM-Attention, and Transformer Encoders, achieved comparable performance, with an accuracy of approximately 82.6%, an F1 score of 0.85, and an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.90. To evaluate how the complexity of medication-use histories affects prediction, patients were divided into low, medium, and high groups using Jenks natural breaks based on their number of medication transitions. Two modeling strategies were then examined to capture the heterogeneous temporal patterns across these groups. The first strategy trained independent sequential models for each group, but this approach did not improve performance and showed reduced accuracy for patients with medium or high number of transitions compared to the baseline models. The second strategy used a shared LSTM backbone with group-specific attention heads, allowing the model to adapt to different levels of transition complexity while maintaining a single architecture. This shared model achieved modest improvements in medium and high group in both accuracy (medium group: 82.9% compared with 82.7%; high group: 78.5% compared with 78.3%) and F1 score (medium group: 0.845 compared with 0.844; high group: 0.821 compared with 0.816) relative to baseline models. These three studies collectively demonstrate that data-driven approaches can connect population-level equity assessment with individualized precision care. By integrating survival analysis, clustering, and machine learning, this work illustrates how data science can reveal structural inequities, identify subgroup-specific challenges, and enable personalized interventions. Advancing health equity in diabetes care requires both population-level policy efforts and patient-centered predictive tools, developed through interdisciplinary collaboration and equitable model design.","Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2027-12-01","The student, Peng Zhang, accepted the attached license on 2025-11-21 at 14:28.","The student, Peng Zhang, submitted this Dissertation for approval on 2025-11-21 at 15:08.","This Dissertation was approved for publication on 2025-11-25 at 13:24.","DSpace SAF Submission Ingestion Package generated from Vireo submission #22923 on 2026-02-19 at 18:45:55"]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Enhancing health equity in antidiabetic medication use and diabetes care outcomes through data science approaches"]}]}],"canonical_facts":{"dc:contributor":["Kang, Hyojung","Hallal, Pedro","He, Jing Rui","Quinn, Lauretta"],"dc:creator":["Zhang, Peng"],"dc:date":["2025-12","2025-11-25"],"dc:description":["Diabetes remains a major public health concern in the United States, affecting more than 10% of the population, with 90% to 95% of cases attributable to type 2 diabetes (T2D). Antidiabetic medications play a central role in glycemic control and in preventing or delaying diabetes-related complications. Understanding differences in antidiabetic medication use patterns and promoting sustained adherence are essential for reducing disparities and improving diabetes outcomes. The overarching objective of this dissertation is to apply data-oriented analytical and machine learning models to enhance understanding of inequities in antidiabetic medication use and to promote more equitable and effective diabetes care. In the first study, we examined racial and geographic disparities in antidiabetic medication discontinuation among Medicare beneficiaries with T2D residing within or surrounding Diabetes Belt (DB) counties. We defined discontinuation as at least 90 consecutive days without any antidiabetic medication supply, and the overall discontinuation rate in our dataset in 41.2%. The Cox regression model reflected that non-Hispanic (NH) Black individuals had a significantly higher discontinuation risk (HR: 1.22, 95% CI: [1.162, 1.286]) compared to NH White individuals. In contrast, no significant difference was observed between residents of DB and non-DB areas after adjustment for socioeconomic factors (HR: 1.02, 95% CI: [0.973, 1.072]). In the second study, we applied a novel time-series clustering approach to characterize long-term antidiabetic medication use patterns over a three-year period. We defined five different transitions (switch, intensification, de-intensification, discontinuation, and re-initiation) based on changes in the number of concurrently used medications to describe longitudinal medication patterns. The most common patterns included continuous metformin monotherapy (36.3%), recurrent cycles of discontinuation and re-initiation (23.1%), one-time discontinuation (14.2%), switching (3.0%), and intensification (2.8%). These patterns were grouped into five clinically meaningful clusters. Cluster-level comparisons indicated that NH Black patients were more likely to experience discontinuation, whereas female patients were more likely to undergo intensification. Patients who switched or re-initiated therapy tended to be older, had more comorbidities, and higher rates of emergency and inpatient visits. In the third study, we developed deep learning models to predict future medication adherence using patients’ longitudinal medication histories and demographic and clinical characteristics. The baseline models, including long short-term memory (LSTM), LSTM-Attention, and Transformer Encoders, achieved comparable performance, with an accuracy of approximately 82.6%, an F1 score of 0.85, and an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.90. To evaluate how the complexity of medication-use histories affects prediction, patients were divided into low, medium, and high groups using Jenks natural breaks based on their number of medication transitions. Two modeling strategies were then examined to capture the heterogeneous temporal patterns across these groups. The first strategy trained independent sequential models for each group, but this approach did not improve performance and showed reduced accuracy for patients with medium or high number of transitions compared to the baseline models. The second strategy used a shared LSTM backbone with group-specific attention heads, allowing the model to adapt to different levels of transition complexity while maintaining a single architecture. This shared model achieved modest improvements in medium and high group in both accuracy (medium group: 82.9% compared with 82.7%; high group: 78.5% compared with 78.3%) and F1 score (medium group: 0.845 compared with 0.844; high group: 0.821 compared with 0.816) relative to baseline models. These three studies collectively demonstrate that data-driven approaches can connect population-level equity assessment with individualized precision care. By integrating survival analysis, clustering, and machine learning, this work illustrates how data science can reveal structural inequities, identify subgroup-specific challenges, and enable personalized interventions. Advancing health equity in diabetes care requires both population-level policy efforts and patient-centered predictive tools, developed through interdisciplinary collaboration and equitable model design.","Submission published under a 24 month embargo labeled 'U of I Access', the embargo will last until 2027-12-01","The student, Peng Zhang, accepted the attached license on 2025-11-21 at 14:28.","The student, Peng Zhang, submitted this Dissertation for approval on 2025-11-21 at 15:08.","This Dissertation was approved for publication on 2025-11-25 at 13:24.","DSpace SAF Submission Ingestion Package generated from Vireo submission #22923 on 2026-02-19 at 18:45:55"],"dc:format":["application/pdf"],"dc:identifier":["https://hdl.handle.net/2142/132651"],"dc:language":["en"],"dc:rights":["Copyright 2025 Peng Zhang"],"dc:subject":["type 2 diabetes","machine learning","antidiabetic medication adherence","racial and geographic disparities","longitudinal medication patterns"],"dc:title":["Enhancing health equity in antidiabetic medication use and diabetes care outcomes through data science approaches"],"dc:type":["text","Thesis"],"thesis:degree_discipline":["Community Health"],"thesis:degree_level":["Dissertation"],"thesis:degree_name":["Ph.D."],"thesis:institution_name":["University of Illinois Urbana-Champaign"]},"updated_at":"2026-07-22T22:25:07Z"}