Describing and leveraging interaction effects for HIV prevention Open Access

Delaney, Kevin Payton (2015)

Permanent URL: https://etd.library.emory.edu/concern/etds/0c483j63c?locale=en
Published

Abstract

Interaction can occur at many levels. People interact in conversations, sexual relations or virtually through the internet. In individuals, sometimes one factor can combine with another to cause disease that wouldn't have been observed otherwise; epidemiologists refer to this as causal interaction. In infectious disease epidemiology there is another level of interaction, in that the infected and uninfected populations must interact to transmit disease, and, because of this, at the population level public health interventions can also interact. In my first study I collected data from 2,666 user profiles of men who use a social networking application, mostly to meet other men for sex. Overlapping circles defined by the geolocation data I extracted from the app covered the entire 132.4 square miles in the City of Atlanta and were analyzed with spatial statistics to highlight areas with higher densities of minority and young minority users. This simple method can describe the spatial density of users of a sexual networking app for future behavioral surveys and to identify areas of highest need for targeting prevention resources. In the second study I used simulated data to compare 10 tests for statistical interaction and contrast these with 3 tests for causal interaction. I found that, at sample sizes typical for epidemiologic studies, the power to detect interaction is limited unless exposures have both strong individual effects and their combined effects are closer to multiplicative than additive. The power is even lower for tests specifically designed to detect causal interaction. The aim of my third study was to describe population level interactions of interventions associated with HIV testing, in a model of the sexual networks of gay men. I found more frequent HIV testing will not result in reduced HIV incidence unless combined with improvements in effective HIV care. Only once care and viral suppression become the normative outcome of HIV diagnosis does additional focus on increasing HIV testing as the gateway to this outcome become warranted. These three very different studies evaluate interaction on several levels. Together they emphasize the importance of studying and leveraging interaction effects for HIV prevention.

Table of Contents

Chapter 1: Overview, Objective and Specific Aims. 1

Interaction. 1

Novel methods for HIV Prevention. 3

Objective and Specific Aims. 6

References. 8

Chapter 2: The HIV Epidemic. 14

HIV Epidemic Worldwide. 16

HIV Prevention Strategies/Interventions. 17

References. 21

Chapter 3 - Specific Aim 1. 28

Introduction. 30

Methods. 32

Data collection. 35

Sampling Strategy. 36

Analysis. 37

Results. 40

Discussion. 48

References. 57

Chapter 3: Supplemental Figures. 66

Chapter 4 - Specific Aim 2. 72

Section 1: Introduction. 72

Section 2: A review of proposed methods to assess interaction. 74

When two causes act together. 74

Specifying the Pattern of Disease Risk. 75

Measures of departure from additive risk differences. 77

Methods of risk model estimation and tests for interaction using SAS software. 78

Tests for SCC interaction when exposure effects are not monotonic. 80

Section 3. Monte Carlo Simulations. 81

Section 4: Results. 85

Section 5: Discussion of simulation results and practical recommendations for testing for interaction. 87

References. 93

Supplemental materials for Aim 2. 113

Supplemental Appendix 1: Notation, Definitions and Further description of simulation parameters, distinction between sufficient component cause interaction, definitive interdependence and statistical interaction, and additional results. 113

Notes on Notation. 114

Dictionary of terms related to the concept of interaction and how they are used in our manuscript. 116

Description of model parameters used in each output in the main text and this appendix. 122

Description of the range of risks of disease D assigned for those exposed to both X1 and X2 in the simulated data. 130

Power to detect greater than additive risk differences and the need to use study designs where the number of participants exposed to both exposures can be controlled experimentally. 133

Power to detect SCC interaction is limited to both large sample sizes and interaction effects that are closer to multiplicative on the risk ratio scale than additive on the risk difference scale. 137

More on the differences between detecting interaction, SCC interaction and definitive interdependence. 143

References. 151

Supplemental Appendix 2: Example SAS code for all tests of interaction evaluated. 152

Chapter 5 - Specific Aim 3. 165

Additional background on Infectious Disease modeling. 165

Epidemiologic considerations. 165

Compartmental Models. 166

Agent-based models. 173

Exponential Random Graph Models. 173

References. 175

Aim 3 - Manuscript to be submitted to Lancet HIV. 178

Introduction. 178

Methods. 179

Modeling strategy and source of parameter values. 179

Statistical analysis of outcomes from model simulations. 180

Investigation of the effects of changes in testing frequency on transmission dynamics. 181

Investigation of population level interactions of testing intervention components. 181

Results. 182

Baseline model and the effects of increases in testing frequency. 182

Effects of testing interventions other than increases in testing frequency. 182

Effects of increases in testing frequency under a scenario where viral suppression is improved. 183

Discussion. 184

References.190

Supplemental Materials for Aim 3. 207

Parameterization of the separable temporal random graph model for partnership formation and other aspects of model for transmission dynamics. 207

STERGM Model parameters. 207

Baseline prevalence and serostatus awareness. 210

Entry into and exit from the population. 212

Parameters for partnership durations, and one-time partnerships and comparison of our approach to others in the literature. 212

Discussion of the importance of partnership duration as a model assumption. 214

Probability of having a one-time partner. 216

Evaluation of the underlying rate of HIV testing and its impact on our transmission models. 217

Modeling the impact of men who never test on HIV incidence. 219

Calculating the total circulating HIV viral load over time. 223

Stochastic variation in models and implications for our main outcomes. 229

References. 233

Chapter 6: Summary of findings and next steps. 239

Contributions to HIV Prevention Research. 239

Areas for Future Research. 242

References. 249

About this Dissertation

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Degree
Submission
Language
  • English
Research Field
Keyword
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files