國立高雄大學圖資館 |

Statistical Inference on Association and Mediation Under Selection Bias and Interference /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Statistical Inference on Association and Mediation Under Selection Bias and Interference /Ritoban Kundu.
作者:	Kundu, Ritoban,
面頁冊數:	1 electronic resource (349 pages)
附註:	Source: Dissertations Abstracts International, Volume: 87-07, Section: B.
附註:	Advisors: Mukherjee, Bhramar; Song, Peter X. -K. Committee members: Kobayashi, Lindsay; Dempsey, Walter.
Contained By:	Dissertations Abstracts International87-07B.
標題:	Statistics.
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=32477030
ISBN:	9798273310469

Statistical Inference on Association and Mediation Under Selection Bias and Interference /
Kundu, Ritoban,

Statistical Inference on Association and Mediation Under Selection Bias and Interference /Ritoban Kundu. - 1 electronic resource (349 pages)

Source: Dissertations Abstracts International, Volume: 87-07, Section: B.

Statistical inference using large observational datasets is challenging due to systematic errors (such as selection bias, information bias, and confounding bias) and complex causal pathways. This dissertation contributes primarily to two major areas of methodological development for observational data: (1) Addressing selection bias in non-probability samples including Electronic Health Records (EHRs) and (2) causal mediation analysis under complex relational structures, either for network/spatial interference or for sequential multiple mediators.The first area consists of three projects. Project 1.1 introduces a graphical framework using selection-directed acyclic graphs to represent and analyze sources of selection bias in association analysis with binary logistic regression outcome models and evaluates four inverse probability weighted estimators (IPW). Project 1.2 extends IPW methods to settings involving multiple cohorts with differing outcome-dependent selection mechanisms and shared individual-level data. I propose a novel Joint Augmented IPW (JAIPW) estimator that offers double robustness against misspecification of any individual selection model. Project 1.3 develops privacy-preserving IPW estimators: Collaborative Pseudo-likelihood (CPL) and Collaborative AIPW, for distributed integration of EHR data across cohorts without sharing individual-level records, while accommodating cohort-specific selection biases. These methods in the first area are illustrated in two US-based biobanks: The Michigan Genomics Initiative and the NIH All of Us Program. The proposed methods to address selection bias in these projects incorporate external information available at the individual or summary level from nationally representative probability samples such as the National Health and Nutrition Examination Survey (NHANES), the National Health Interview Survey (NHIS), the Surveillance, Epidemiology, and End Results (SEER) Program, and the national health data released by the Centers for Disease Control and Prevention (CDC).The second area includes two projects. Project 2.1 develops a Random Effects Network Structural Equation Model (REN-SEM) for causal mediation analysis under network interference. This approach enables investigating a variety of causal pathways arising from direct or indirect influences of exposures and mediators of neighboring units on an individual's outcome. After establishing the identifiability conditions for causal estimands characterizing both mediation and interference effects, I derive key theoretical asymptotic properties of the proposed maximum likelihood estimators under non-i.i.d settings. The proposed methodology is evaluated numerically through simulation experiments and real-world data from the Twitch Gamers social network. In Project 2.2, I extended the proposed REN-SEM paradigm to spatially correlated areal data, which is motivated to study the influence of political ideology on COVID-19 mortality through vaccine hesitancy as a potential mediator in the USA. This extension allows me to assess spatial spillover effects across over 3,000 U.S. counties. Project 2.3 focuses on causal mediation pathways with multiple sequentially ordered mediators-a setting where traditional tests face challenges due to the composite structure of the null hypothesis. The proposed Sequentially Ordered Mediation Effect Test (SOMET) is to test the existence of a mediation pathway of interest. Theoretically and numerically, SOMET is shown to properly control Type I error and achieve desirable power. This project considers two applications: One uses the ELEMENT cohort to evaluate the effect of prenatal lead exposure on adolescent growth mediated by temporally ordered mediators of birth weight and lipid, and the other utilizes the NIH All of Us database to investigate the mediation effect of family history of diabetes on onset of diabetes through sequentially ordered mediators of physical activity and body mass index (BMI).

English

ISBN: 9798273310469Subjects--Topical Terms:

182057
Statistics.
Subjects--Index Terms:

Electronic Health Records

Statistical Inference on Association and Mediation Under Selection Bias and Interference /
LDR:05681nmm a22004813i 4500 001 693819
005 20260511063029.5
006 m o d
007 cr|nu||||||||
008 260819s2025 miu||||||m |||||||eng d
020 $a 9798273310469
035 $a (MiAaPQD)AAI32477030
035 $a (MiAaPQD)umichrackham:006629
035 $a AAI32477030
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Kundu, Ritoban, $e author. $3 1010531
245 1 0 $a Statistical Inference on Association and Mediation Under Selection Bias and Interference / $c Ritoban Kundu.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2025
300 $a 1 electronic resource (349 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 87-07, Section: B.
500 $a Advisors: Mukherjee, Bhramar; Song, Peter X. -K. Committee members: Kobayashi, Lindsay; Dempsey, Walter.
502 $b Ph.D. $c University of Michigan $d 2025.
520 $a Statistical inference using large observational datasets is challenging due to systematic errors (such as selection bias, information bias, and confounding bias) and complex causal pathways. This dissertation contributes primarily to two major areas of methodological development for observational data: (1) Addressing selection bias in non-probability samples including Electronic Health Records (EHRs) and (2) causal mediation analysis under complex relational structures, either for network/spatial interference or for sequential multiple mediators.The first area consists of three projects. Project 1.1 introduces a graphical framework using selection-directed acyclic graphs to represent and analyze sources of selection bias in association analysis with binary logistic regression outcome models and evaluates four inverse probability weighted estimators (IPW). Project 1.2 extends IPW methods to settings involving multiple cohorts with differing outcome-dependent selection mechanisms and shared individual-level data. I propose a novel Joint Augmented IPW (JAIPW) estimator that offers double robustness against misspecification of any individual selection model. Project 1.3 develops privacy-preserving IPW estimators: Collaborative Pseudo-likelihood (CPL) and Collaborative AIPW, for distributed integration of EHR data across cohorts without sharing individual-level records, while accommodating cohort-specific selection biases. These methods in the first area are illustrated in two US-based biobanks: The Michigan Genomics Initiative and the NIH All of Us Program. The proposed methods to address selection bias in these projects incorporate external information available at the individual or summary level from nationally representative probability samples such as the National Health and Nutrition Examination Survey (NHANES), the National Health Interview Survey (NHIS), the Surveillance, Epidemiology, and End Results (SEER) Program, and the national health data released by the Centers for Disease Control and Prevention (CDC).The second area includes two projects. Project 2.1 develops a Random Effects Network Structural Equation Model (REN-SEM) for causal mediation analysis under network interference. This approach enables investigating a variety of causal pathways arising from direct or indirect influences of exposures and mediators of neighboring units on an individual's outcome. After establishing the identifiability conditions for causal estimands characterizing both mediation and interference effects, I derive key theoretical asymptotic properties of the proposed maximum likelihood estimators under non-i.i.d settings. The proposed methodology is evaluated numerically through simulation experiments and real-world data from the Twitch Gamers social network. In Project 2.2, I extended the proposed REN-SEM paradigm to spatially correlated areal data, which is motivated to study the influence of political ideology on COVID-19 mortality through vaccine hesitancy as a potential mediator in the USA. This extension allows me to assess spatial spillover effects across over 3,000 U.S. counties. Project 2.3 focuses on causal mediation pathways with multiple sequentially ordered mediators-a setting where traditional tests face challenges due to the composite structure of the null hypothesis. The proposed Sequentially Ordered Mediation Effect Test (SOMET) is to test the existence of a mediation pathway of interest. Theoretically and numerically, SOMET is shown to properly control Type I error and achieve desirable power. This project considers two applications: One uses the ELEMENT cohort to evaluate the effect of prenatal lead exposure on adolescent growth mediated by temporally ordered mediators of birth weight and lipid, and the other utilizes the NIH All of Us database to investigate the mediation effect of family history of diabetes on onset of diabetes through sequentially ordered mediators of physical activity and body mass index (BMI).
546 $a English
590 $a School code: 0127
650 4 $a Statistics. $3 182057
650 4 $a Medicine. $3 193819
650 4 $a Biostatistics. $3 339693
650 4 $a Bioinformatics. $3 194415
653 $a Electronic Health Records
653 $a Selection bias
653 $a Doubly robust methods
653 $a Privacy preservation
653 $a Causal mediation analysis
653 $a Sequential mediators
690 $a 0564
690 $a 0715
690 $a 0308
690 $a 0463
710 2 $a University of Michigan. $b Biostatistics. $3 967053
720 1 $a Mukherjee, Bhramar $e degree supervisor.
720 1 $a Song, Peter X. -K. $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 87-07B.
790 $a 0127
791 $a Ph.D.
792 $a 2025
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=32477030