Contributed Poster Presentations: Section on Statistical Consulting

Shirin Golchi Chair
McGill University
 
Tuesday, Aug 5: 2:00 PM - 3:50 PM
4120 
Contributed Posters 
Music City Center 
Room: CC-Hall B 

Main Sponsor

Section on Statistical Consulting

Presentations

01: Association between Human Trafficking Arrests and Online Ad Volume: A Study of Five U.S. Cities

Human trafficking remains a global issue, with traffickers using online platforms to advertise victims disguised as consensual services. Law enforcement struggles to distinguish between sex work and trafficking, especially on Adult Service Websites (ASWs), which traffickers exploit for anonymity and broad reach (Cotterill, 2023). This study examines whether trafficking-related arrests reduce online prostitution ads in Detroit, Seattle, Chicago, Boston, and Houston. Data from all five cities were filtered for relevant offenses and analyzed using linear regression scatterplots to assess relationships between arrests and ad volumes. The scatterplots and cross-correlation function plots revealed no strong correlation between these variables. Further analysis using autocorrelation functions and change point analysis showed significant shifts in data patterns, corresponding to events such as major police operations. This suggests traffickers quickly adapt to law enforcement actions, though certain events may temporarily disrupt their activity. This analysis aids law enforcement in understanding links between arrests and online ads, providing a basis for further research and strategy. 

Keywords

ASWs

cross correlation function

autocorrelation function

change point analysis

human trafficking

time series analysis 

Co-Author(s)

Xinshu Yi, University of Alabama
Subhabrata Chakraborti, The University of Alabama
Nickolas Freeman, The University of Alabama
Jason Parton, University of Alabama

First Author

Isabella Martin, The University of Alabama

Presenting Author

Isabella Martin, The University of Alabama

02: CART Analysis to Predict Serious Complications with Central Line in Pediatric Patients

Introduction. Serious complications like central line-associated bloodstream infection and venous thromboembolism are associated with central venous access devices and can be life-threatening. In this study, we used Classification and Regression Tree (CART) analysis to model the rate of these complications.
Methods. Children from Pediatric Health Information System (PHIS) 2017-2021 database were included in the study. CART analysis was performed to examine important risk factors of serious complications with central line. Data were randomly split into developmental (50%) and validation samples (50%). Gini index was used as splitting criterion. Parent/terminal nodes were set to be 10/5.
Results. Of 67,830 children hospitalized who had at least one central line placed during study period, 4,688 (6.9%) experienced serious complications. The CART model with disease severity, receiving total parenteral nutrition, central line type and complex chronic condition had an area under the receiver operating characteristic curve of 0.77 with sensitivity of 66.8% and specificity of 77.4% for the developmental samples. 

Keywords

Classification and Regression Tree analysis

Predictive modeling

Sensitivity

Specificity

area under the receiver operating characteristic curve 

Co-Author(s)

Melodee Liegl, Medical College of Wisconsin
Alina Burek, Medical College of Wisconsin

First Author

Amy Pan, Medical College of Wisconsin

Presenting Author

Amy Pan, Medical College of Wisconsin

03: Enhancing Causal Inference: The Comparison of Stratification Over Adjustment in IPTW Analyses

Inverse Probability of Treatment Weighting (IPTW) is a key method in causal inference for estimating treatment effects while addressing confounding. While both stratification and adjustment are used to control for confounders, stratification may be superior when a confounder is strongly correlated with treatment. In this study, we emulated a target trial comparing surgical versus endovascular lower extremity revascularization for major adverse limb events. Given that chronic limb-threatening ischemia (CLTI) strongly influences treatment choice, we compared results between including the CLTI in the propensity score model and stratifying the data by CLTI and then running the propensity score model separately within the strata of CLTI. Our findings highlight the need for stratification when a confounder is strongly correlated with treatment. 

Keywords

Inverse Probability of Treatment Weighting (IPTW)

Causal Inference

Stratification

Adjustment 

Co-Author(s)

Maria Montez-Rath, Stanford University
Tara Chang, Stanford Univeristy

First Author

Sai Liu, Stanford University

Presenting Author

Sai Liu, Stanford University

04: Machine Learning Models for Risk Stratification in Infants with Abnormal Development

Introduction: To address the gap in knowledge of developmental outcomes of infants ≥34 weeks gestation requiring NICU care, 1183 infants were evaluated. Variables collected for each infant included demographics, delivery course, NICU diagnoses and management, and developmental screening results. Of interest was which variables predict abnormal developmental screening. Methods: Traditional logistic regression analyses results were unstable because of the multicollinearity problem between the large number of predictor variables. CART analysis was ultimately used. Other machine learning methods, including random forest, TreeNet gradient boosting, MARS and regularized logistic regression methods, were also used for comparison. Results: NICU variables selected as most important in the CART analysis were PO feeding, number of medications prescribed, and number of follow-up appointments at NICU discharge. Other four machine learning models produced similar results to the CART model with slightly higher AUC and specificity, but lower sensitivity. Conclusion: Machine learning algorithms can provide useful tool in health care to assist providers in making complex clinical decisions. 

Keywords

Machine learning

Prediction

Health care 

Co-Author(s)

Katherine Carlton, Medical College of Wisconsin
Jian Zhang
Erwin Cabacungan, Medical College of Wisconsin
Susan Cohen, Medical College of Wisconsin

First Author

Ke Yan, Medical College of Wisconsin

Presenting Author

Ke Yan, Medical College of Wisconsin

05: Metabolic Heat Profiles in Chondrocytes: A Comparison of Functional and Integrated Data Approaches

Osteoarthritis affects the tissues and cells across the whole joint and results in cartilage degradation. Human chondrocytes are the only cell type in articular cartilage and are responsible for cartilage repair and homeostasis through metabolism. An increase in temperature surrounding these cells results in faster metabolic processes which will in turn generate heat as a byproduct. This study analyzes heat measurements of three-dimensionally encapsulated chondrocytes suspended in agarose hydrogels for four carbon sources. The heat is measured every second for 48 hours for each gel. Heat measurement profiles can be analyzed using the total heat over time by integrating instantaneous heat measurements. Or the heat curves themselves can be analyzed using a functional response linear model, which allows more detailed assessments of when the groups might differ and possibly more power to detect differences. We compare the functional data approach to using a conventional linear model for the aggregated heat responses. Pairwise follow-up tests to assess differences between groups and/or in particular regions of time are considered. 

Keywords

biostatistics

functional data

functional ANOVA

follow-up tests 

Co-Author(s)

Mark Greenwood, Montana State University-Bozeman
Ronald June, Montana State University
Erik Myers, Montana State University
Priyanka Brahmachary, Montana State University
Ross Carlson, Montana State University
Campbell Putnam, Montana State University

First Author

Sarah Mensah, Montana State University

Presenting Author

Sarah Mensah, Montana State University

06: ML Identification of Key Features for 90-Day Readmission & LOS in CAR T-Cell Therapy for R/R LBCL

Chimeric antigen receptor (CAR) T-cell therapy efficacy is limited by toxicities and high cost (median $350k) in relapsed/refractory large B-cell lymphoma (R/R LBCL). This study explores patient features predicting 90-day readmission and hospital length of stay (LOS). A retrospective review of 66 patients with R/R LBCL treated at UC San Diego (2016–2022) included 46 clinical variables (e.g., clinical/visits/readmission parameters & adverse events). The targets were 90-day readmission (29%) and related LOS (mean 18.3 days [SD 19.5]). Data were pre-processed for lasso logistic/linear regression (L1-LG/LR), random forest (RF), extreme gradient boosting (XGB), and support vector machine (SVM) with a 75/25 train/test split, 100 randomly created test sets, GridSearchCV tuning, and 5-fold cross-validation. For 90-day readmission, top predictor was Day-60 Readmission LOS (22%-35% from RF to XGB; L1-LG coefficient 2.5), and highly weighted for predicting the associated LOS (35%-70% across models; L1-LR coefficient 2.9). Large variations (Min F1s: 0.33-0.75) and low R2 (about 0.55) were observed. This study highlights key predictors; larger datasets are needed for clinical generalization. 

Keywords

chimeric antigen receptor (CAR) T-cell

90-day readmission

machine learning

important features 

Co-Author(s)

Aaron Trando, University of California San Diego School of Medicine,
Ah-Reum Jeong, Department of Medicine, Division of Blood and Marrow Transplantation
Dimitrios Tzachanis, Department of Medicine, Division of Blood and Marrow Transplantation

First Author

Philip Yeung, Kansas University Medical Center/ MPKey, LLC.

Presenting Author

Philip Yeung, Kansas University Medical Center/ MPKey, LLC.