Rethinking Statistical Paradigms, and Applications to Health Disparities

Rachael Aikens Chair
Mathematica
 
Tuesday, Aug 5: 2:00 PM - 3:50 PM
4134 
Contributed Papers 
Music City Center 
Room: CC-210 

Main Sponsor

Health Policy Statistics Section

Presentations

Clinical Research Reporting Paradigms May Incompletely Describe Participant Identities

Reporting of participants' baseline characteristics in clinical research is important for understanding a given study's context and typically occurs in a tabular format. However, this format incompletely and ambiguously describes included participants, as their identities are more fully represented by an intersecting set of sociodemographic characteristics rather than discrete characteristics in a table. Standard tabular reporting practices therefore introduce limitations in assessing a study's representativeness as well as its internal validity and external validity. To address this, we propose the addition of a simple graph that more clearly shows the joint distribution of baseline sociodemographic characteristics in a given study. We also discuss several practical considerations for the implementation of such graphs in the communication of clinical research. 

Keywords

health equity

clinical trial

observational study

internal validity

external validity

data reporting 

Co-Author(s)

Lisa Goldman Rosas, Stanford University
Sherri Rose, Stanford University
Oana Enache, Stanford University

First Author

Oana Enache, Stanford University

Presenting Author

Oana Enache, Stanford University

Evaluating Trends and Disparities: Financial Barriers to Care Before, During, and After COVID-19

Introduction: Recent pre-post studies suggest improved healthcare affordability during the COVID-19 pandemic. However, these studies do not account for temporal trends or racial disparities. In this study, we use interrupted time series (ITS) and difference-ITS (d-ITS) analysis to examine trends in financial hardship.
Methods: We analyzed 2017-2023 NHIS data on U.S. adults, considering financial hardship from and inability to pay medical bills, and delayed and foregone care due to cost. Crude and adjusted ITS and d-ITS models were developed for each outcome, with 2020 as the interruption point.
Results: Black adults have an overall higher risk for all outcomes (p<0.001). The overall odds of foregone care (OR: 1.09) and inability to pay bills (OR: 1.27) were on the rise leading up to COVID-19.  The immediate effect of COVID was a 21% decrease in the odds of delayed care but 86% increased odds in inability to pay. Since 2020, the odds of foregone care and inability to pay have declined by 8% and 20% annually, with no significant differences in trends by race.
Conclusions: Despite overall disparities, barriers declined post-COVID potentially due to pandemic-era safety net policies. 

Keywords

COVID-19

Delayed Care

Disparities

Financial Hardship

Interrupted Time Series

Health Policy 

Co-Author(s)

Caroline Andy, Weill Cornell Medicine
Reed Mszar, Yale School of Public Health

First Author

Fabiola Etienne

Presenting Author

Fabiola Etienne

Geospatial Analysis of Postpartum Visit Adherence for Medicaid-Enrolled Mothers in Arizona

Starting in mid-2022, Arizona Medicaid (AHCCCS) extended postpartum coverage from 60 days to 12 months to address postpartum complications. However, driving time to providers remains an understudied barrier to care. This retrospective study analyzed approximately 30,000 AHCCCS-enrolled mothers to assess the impact of driving time and social vulnerability on postpartum visit adherence defined by the HEDIS Prenatal and Postpartum Care (PPC) measure. We calculated one-way driving time to neonatal abstinence syndrome (NAS) reporting hospitals in 15-minute intervals. Vulnerable mothers were those residing in census tracts with high social vulnerability index (SVI) scores, low food access, no internet, or an opioid use disorder (OUD) diagnosis. Fisher's exact tests compared adherence rates between vulnerable and non-vulnerable mothers within each interval. Vulnerable mothers traveling over 60 minutes were significantly less likely to attend visits, with the largest disparities among those in high SVI or low food access areas. 

Keywords

Maternal Health

Health Services Accessibility

Healthcare Inequities

Social Vulnerability

Rural Health

Medicaid 

Co-Author(s)

Logan Cameron, Arizona State University
Andre Perry II, Arizona State University
Anita Murcko, Arizona State University
George Runger, Arizona State University

First Author

El-Ham Ismail, Arizona State University

Presenting Author

El-Ham Ismail, Arizona State University

Impact of redefining statistical significance on P-hacking and false positive rates.

In recent years, concern has grown about the inappropriate application and interpretation of
P values, especially the use of P<0.05 to denote "statistical significance" and the practice of
P-hacking to produce results below this threshold and selectively reporting these in publications.
Such behavior is said to be a major contributor to the large number of false and nonreproducible
discoveries found in academic journals. In response, it has been proposed that
the threshold for statistical significance be changed from 0.05 to 0.005. The aim of the current
study was to use an evolutionary agent-based model comprised of researchers who
test hypotheses and strive to increase their publication rates in order to explore the impact
of a 0.005 P value threshold on P-hacking and published false positive rates. The results supported the view that a more stringent
P value threshold can serve to reduce the rate of published false positive results.
Researchers still engaged in P-hacking with the new threshold, but the effort they expended
increased substantially and their overall productivity was reduced, resulting in a decline in
the published false positive rate. 

Keywords

significance threshold

p-hacking

agent-based most

simulation

effect size 

Co-Author(s)

Dennis Gorman, Texas A&M University
Caitlin Trombatore, Los Angeles Dodgers

First Author

Ben Fitzpatrick, Loyola Marymount University

Presenting Author

Ben Fitzpatrick, Loyola Marymount University

Incentives, Assessment, and the Reliability of Statistical Significance Assessments of Evidence

This analysis evaluates the implications of researcher hypothesis selection incentives on the inferential value of empirical analyses. It illustrates the strength with which a "statistically significant" outcome objective incentivizes researcher aversion to testing possibly true null hypotheses. Mechanically, such aversion reduces the number of true nulls selected for testing, which in turn reduces type I error rates (i.e., erroneous rejections of true null hypotheses). Left unfettered, it leads to settings wherein researchers almost always opt to test false null hypotheses. That is, studies routinely produce reliable "falsifications" of a priori false hypotheses, a practice that transparently lacks inferential relevance. Collectively, the analysis illustrates the importance of comprehensive understanding of researcher incentives and research assessment practices when evaluating the reliability and relevance of findings obtained from Null Hypothesis Significance Test based examinations of evidence. 

Keywords

Priors, Incentives, Statistical Significance, Error 

First Author

Bill Cready, University of Texas at Dallas

Presenting Author

Bill Cready, University of Texas at Dallas

New Graphical Displays and Related Statistical Measures of Health Disparities Among Groups

Different methods for describing health disparities in distributions of continuous health-related variables among race/ethnic or socioeconomic groups provide more insight into the nature of disparities than comparisons of measures of central tendency. Transformations of the Lorenz curve and analogues of the Gini index used in the analysis of income inequality are adapted to provide graphical and analytical measures of health disparities. Akin to the classical Peters-Belson regression method for partitioning a disparity into a component explained by group differences in a set of covariates and an unexplained component, a new modified Lorenz curve is proposed. We explore statistical properties of these estimators through simulation studies and present three application examples: 1.) disparities in BMI between Non-Hispanic Black and Non-Hispanic White women in the U.S. based on the US National Health and Nutrition Examination Surveys 2013-2018, 2.) blood lead levels of children across race/ethnic groups from NHANES 1988-1994, and 3.) Driving distance to nearest acute care hospital across census block urbanicity levels. 

Keywords

Health Disparity

Lorenz Curve 

Co-Author(s)

Barry Graubard, NCI/ DCEG/ BB
Joseph Gastwirth, George Washington University

First Author

Mark Louie Ramos, The Pennsylvania State University

Presenting Author

Mark Louie Ramos, The Pennsylvania State University

Rethinking Statistical Significance: A Bayesian Alternative to P-Values in Public Opinion Survey Res

Rethinking Statistical Significance: A Bayesian Alternative to P-Values in Public Opinion Survey Results
P-values are widely used in public opinion research to assess statistical significance in binary survey data but face criticism for being overly sensitive to sample size, often misinterpreted, and failing to incorporate prior knowledge. These limitations hinder actionable insights crucial for policy-making. Bayesian methods, particularly Bayesian Logistic Regression (BLR), offer a robust alternative by providing probabilistic interpretations and integrating prior knowledge.
This study applies BLR to binary survey data from a Saudi public opinion survey of 1,300 participants, focusing on financial assistance across income groups. Using the rstanarm package in R, the analysis specifies priors, conducts convergence diagnostics (e.g., Rhat), and performs posterior predictive checks. BLR is expected to reveal significant effects overlooked by p-values, provide probabilistic insights, and improve model fit. This approach advances public opinion research by offering a more nuanced and reliable framework for analyzing binary survey data, supporting informed policy decisions. 

Keywords

Bayesian Logistic Regression (BLR)
Public Opinion Research
Binary Survey Data
Statistical Significance
P-values Criticism
Bayesian Methods 

Co-Author(s)

Norah Alhomiedan, Decision support center
Bayan Alzahrani, Decision support center

First Author

Ghadah Alkhadim, Decision support center

Presenting Author

Ghadah Alkhadim, Decision support center