• Originally from Germany, moved to U.S. for graduate school - and never left 😄.
• PhD in Physics at Georgia Tech, Postdoc in Computational Biology at Emory.
• Since 2009, Assistant/Associate/Full Professor in Epidemiology & Biostatistics, University of Georgia.
• Data analysis and modeling of infectious diseases on the population and individual host levels.
• Mainly influenza and norovirus, some TB and other bugs. Recently also a lot of SARS-CoV-2.
• More information:
  • https://www.andreashandel.com
  • https://handelgroup.uga.edu

• After 15+ years of basic science/research, I want to move towards more applied work.
• To get that process started, I want to do a short term (approx. 8 month) industry “internship”.
• This presentation is part of my looking for something effort.

• Part 1: Research on dosing for vaccines
• Part 2: Various other projects (R packages, online courses)
• Part 3: Q&A

• Inoculum dose is an important determinant of infection and vaccination outcomes.
• For infection, higher dose is often associated with greater risk of infection (ID50) and more severe outcomes (LD50).
• For vaccines, dose is thought to impact both immunogenicity/efficacy and side effects (morbidity).

• Trade-offs between availability/safety and efficacy likely exist.
• Impact of dose on vaccine efficacy might not be monotone.

• Impact of dose on homologous or heterologous protection might not be monotone.

• A few doses are explored in phase 1 and 2 trials.
• Based on those data, one of the doses is chosen for phase 3 trials (and beyond) in a non-rigorous manner.

Example of BioNTech/Pfizer COVID Vaccine

Claim: Knowing in more detail how dose impacts host response following vaccination might help optimize vaccines.

• Open cohort of individuals who were vaccinated (some repeatedly) during the 2014/15 - 2018/19 flu seasons.
• The default vaccine was the trivalent or quadrivalent standard dose (SD, 15µg) Fluzone vaccine.
• Individuals >=65 years were offered the high-dose (HD, 60µg) trivalent vaccine.
• We evaluated immune response following vaccination, specifically antibodies as measured by hemagglutination inhibition assay (HAI).

Question: What is the impact of standard dose (SD) versus high dose (HD) vaccine on HAI antibodies?

We investigate 4 antibody titer outcomes for each strain:

• Titer increase following vaccination
• Post-vaccination titer
• Seroconversion, defined as pre-vaccination HAI titer <1:10 (limit of detection) and post-vaccination titer >= 1:40 OR a >=4-fold increase
• Seroprotection, defined as post-vaccination HAI >= 1:40

All HAI titer dilution values are converted to a scale from 0 - N with limit of detection = 0, lowest dilution (1:10) = 1, etc. up to highest dilution (1:20480) = 12.

Same 4 outcomes as before, but now computed per overall vaccine:

• For Seroprotection/Seroconversion: Sum across all strains, standardized (fraction from 0-1).
• For post-vaccination titer and titer increase: The average across all vaccine strains.

• Linear or logistic multivariable regression
  • Outcomes as just explained
  • Main predictor/exposure is dose
  • Covariates are age, pre-existing HAI titer, sex and race
• Hierarchical Bayesian framework
  • Vaccine level
  • Strain level
• Accounted for repeaters indirectly through pre-vaccination antibody titers
• Looked at difference or odds ratio between doses

• HD seems to induce overall better homologous and heterologous responses.
• Difference between HD and SD is not that large.
• A good bit of variability is noticeable.
• Only 2 dose levels do not allow for deeper analysis.
• This is a secondary analysis of an observational cohort study.

For this project, the goals were to:

• Develop a conceptual framework combining data with mechanistic models to investigate the impact of dose on infection dynamics.
• Link the models to vaccine outcomes (protection and morbidity).
• Illustrate how to use this framework to predict outcomes for a large range of doses.

We focused on viral infections as a stand-in for live attenuated vaccines.

\[ \begin{aligned} \textrm{Uninfected cells} \qquad \dot{U} & = - bUV \\ \textrm{Infected cells} \qquad \dot{I} & = bUV - d_I I \\ \textrm{Dead cells} \qquad \dot{D} & = d_I I \\ \textrm{Virus} \qquad \dot{V} & = \frac{pI}{1+s_F F} - (d_V V + k^{'}_{A}A + b^{'} U)V\\ \textrm{Innate response} \qquad \dot{F} & = p_F - d_F F + \frac{g_F (F_{max} - F)V}{V+h_V} \\ \textrm{B cells} \qquad \dot{B} & = \frac{F V}{FV+h_F} g_B B \\ \textrm{Antibodies} \qquad \dot{A} & = r_A B - d_A A - k_{A}AV \\ \end{aligned} \]

We want to know protection. We can map antibodies to it. We use this relation: \(P=1−1/(1+e^{k_1(log(A)−k_2)})\)

Coudeville et al 2010 BMC Med Res Meth

We want to know morbidity (side effects/symptoms). We can map the innate response to it. \[ M = \int \frac{aF^c}{b+F^c} \]

Hayden et al 1998 JCI

Conceptual model suggests that protection (and morbidity) might be peaked (left flu, right HPIV).

• Well-calibrated mechanistic models might be useful to predict dose impact and optimize vaccines.
• Immune responses vary for different infections/vaccines, thus models will likely need to be tailored.
• Getting the right kind of data to properly build and calibrate models will be tricky.

• Handel et al 2018 PLoS Comp Bio
  
• Li & Handel 2014 JTB
• Moore et al 2020 BMB
• Rhodes et al 2016 Vaccine
• Rhodes et al 2019 JTB

• Dynamical Systems Approach to Immune Response Modeling
  • Website
  • Online
• Dynamical Systems Approach to Infectious Disease Epidemiology (Ecology/Evolution)
  • Website
  • Online
• modelbuilder
  • Website
  • Online

• Simmulation Modeling in Immunology
  • See also Handel et al 2020 Nat Rev Imm
• Infectious Disease Epidemiology (Ecology)(Evolution) - A model-based approach
• Modern Applied Data Analysis

• Collaborators:
  • Prior work: See published papers
  • Ongoing work: Yang Ge, Zane Billings, Ye Shen, Ted Ross, others
• Funding:
  • NIH

https://phdcomics.com/

• Slides: https://www.andreashandel.com/presentations/
• Contact: https://www.andreashandel.com