intro-pnuc.tex

% Introduction to the prevalent new-user cohort design 
% (c) 2021-2024 Malcolm Gillies <malcolm.gillies@unsw.edu.au>
% https://github.com/mbg-unsw/pnuc
%
% This work is licensed under a
% Creative Commons Attribution-NonCommercial-ShareAlike 4.0
% International Licence
\documentclass[aspectratio=169,12pt]{beamer} % XXXX fix AR here
\usepackage[latin1]{inputenc}
\usepackage[T1]{fontenc}
\usepackage{textcomp}
\usefonttheme{serif} % need this with Charter font
\usetheme{Berlin}  % using default now
\usecolortheme{beaver}  % using default now
\usepackage[libertine]{libertine} % not using osf (old-style figures)
\usepackage[scale=0.9]{tgheros} % scale to match libertine
\usepackage[varqu,varl]{inconsolata}
\usepackage[libertine]{newtxmath}
\usepackage{graphicx}
\usepackage{tikz}
\usetikzlibrary{shadows}
\usepackage{tikzpagenodes}
\usepackage[round]{natbib}
\usepackage{gitinfo2}

\renewcommand{\gitMark}{\color{gray}\texttt{\tiny\gitBranch\,@\,\gitAbbrevHash\,\gitAuthorDate}}

\renewcommand{\bibsection}{} % suppress "References" section

\setbeamertemplate{navigation symbols}{} % remove navigation symbols
\setbeamercolor*{item}{fg=darkred}

\title{UNSW CRAIC: Prevalent new-user cohort design}
\author{Malcolm Gillies}
\institute{\url{https://github.com/mbg-unsw/pnuc}}
\date{5 February 2024}
\usebackgroundtemplate{%
\begin{tikzpicture}[remember picture,overlay]
    \node[anchor=south west,scale=1,rotate=90] at ([shift={(0cm,0cm)}]current page marginpar area.south east) {\gitMark};
\end{tikzpicture}%
}

\newif\ifsidebartheme
\sidebarthemefalse

\newdimen\contentheight
\newdimen\contentwidth
\newdimen\contentleft
\newdimen\contentbottom
\makeatletter
\newcommand*{\calculatespace}{%
    \contentheight=\paperheight%
    \ifx\beamer@frametitle\@empty%
        \setbox\@tempboxa=\box\voidb@x%
      \else%
        \setbox\@tempboxa=\vbox{%
          \vbox{}%
          {\parskip0pt\usebeamertemplate***{frametitle}}%
        }%
        \ifsidebartheme%
          \advance\contentheight by-1em%
        \fi%
      \fi%
    \advance\contentheight by-\ht\@tempboxa%
    \advance\contentheight by-\dp\@tempboxa%
    \advance\contentheight by-\beamer@frametopskip%
    \ifbeamer@plainframe%
    \contentbottom=0pt%
    \else%
    \advance\contentheight by-\headheight%
    \advance\contentheight by\headdp%
    \advance\contentheight by-\footheight%
    \advance\contentheight by4pt%
    \contentbottom=\footheight%
    \advance\contentbottom by-4pt%
    \fi%
    \contentwidth=\paperwidth%
    \ifbeamer@plainframe%
    \contentleft=0pt%
    \else%
    \advance\contentwidth by-\beamer@rightsidebar%
    \advance\contentwidth by-\beamer@leftsidebar\relax%
    \contentleft=\beamer@leftsidebar%
    \fi%
}
\makeatother

\begin{document}

{
%\usebackgroundtemplate{}
\begin{frame}
\titlepage
\end{frame}
}

\begin{frame}{Talk outline}
    \begin{itemize}
	\item What \& why
	\item Literature review
	\item In context
	\item Step by step
	\item Example
	\item Developments
    \end{itemize}
\end{frame}

\begin{frame}{Why \emph{prevalent} new-user?}
    \begin{itemize}
	\item \emph{i.e.} Compare those staying on old tx with switchers to new
	\item Increase sample size if few treatment-naive patients
	\item Better external validity if disease progression
	\item Key feature: conditioning on length of exposure
	\item Alternative to Marginal Structure Models (MSM)
    \end{itemize}
\end{frame}

\begin{frame}{Popularity of the PNUC design}
% https://tex.stackexchange.com/questions/581745/beamer-moving-text-to-the-bottom-of-the-page
\vskip0pt plus 1filll
\begin{itemize}
	\item Lit review of PNUC to 9 May 2022
	\item 19 publications found
	\item 80\% were collaborations with McGill
	\item Noone has published code
\end{itemize}
\vskip0pt plus 1filll
\flushright{\small{\citet{tazare_prevalent_2022}}}
\end{frame}

\begin{frame}{Active comparator new-user (ACNU) cohort design}
    \begin{itemize}
	\item Problems avoided by the ACNU design:
	\begin{itemize}
		\item Confounding by indication (inactive comparator)
		\item Healthy adherer bias (prevalent users)
	\end{itemize}
    \end{itemize}
\end{frame}

\begin{frame}{PNUC design in context}
\begin{tikzpicture}[remember picture,overlay]
    \node[anchor=south east,scale=1,rotate=90] at ([shift={(1cm,-2cm)}]current page text area.north west) {Increasingly restrictive};
    \draw[thick,<-] ([shift={(1cm,1cm)}]current page text area.south west) --
    ([shift={(1cm,-2cm)}]current page text area.north west);
\end{tikzpicture}%
\vskip0pt plus 1filll
	\centering
	\begin{tabular}{ll}
	\hline
		\bf{No-use} & \bf{Traditional no use} \\
		& \bf{No use episodes} \\
	\hline
		\bf{Active-comparator}  & \bf{Generalized prevalent new user} \\
		& \alert{\bf{Prevalent new user}} \\
		& \bf{Hierarchical prevalent new user} \\
		& \bf{Active comparator new user} \\
	\hline
	\end{tabular}
\vskip0pt plus 1filll
\flushright{\small{\citet{wintzell_selection_2022}}}
\end{frame}

\begin{frame}{PNUC vs MSM}
\vskip0pt plus 1filll
	\centering
	\begin{tabular}{lll}
	\hline
		& PNUC & MSM \\
	\hline
		Complexity & Higher & Lower \\
		Computation & Higher & Lower \\
		Precision & Higher? & Lower? \\
	\hline
	\end{tabular}
\vskip0pt plus 1filll
\flushright{\small{\citet{webster-clark_alternative_2022}}}
\end{frame}

\begin{frame}{Estimands}
    \calculatespace%
    \begin{center}
	\includegraphics[height=0.95\contentheight]{ref/estimands.pdf}
    \end{center}
\end{frame}

\begin{frame}{Prevalent new-user cohort design step by step}
    \begin{enumerate}
	\item Defining the base cohort
	\item Forming exposure sets
	\item Estimating the propensity score
	\item Matching and forming the analysis cohort
	\item Estimating the causal effect
    \end{enumerate}
\end{frame}

\begin{frame}{Defining exposure sets and estimating the propensity score}
    \calculatespace%
    \begin{center}
	\includegraphics[height=0.85\contentheight]{ref/suimoodell-fig3.pdf}
    \end{center}
\flushright{\small{\citet{suissa_prevalent_2017}}}

\end{frame}

\begin{frame}{Time-conditional propensity score}
	\begin{itemize}
		\item Measured at treatment time
		\item Positivity verified within exposure set
		\item Score from conditional logistic regression or Cox
		\item Prevalent users can fall into multiple exposure sets
	\end{itemize}
\end{frame}

\begin{frame}{Matching}
    \calculatespace%
    \begin{center}
	\includegraphics[height=0.85\contentheight]{ref/suimoodell-fig4.pdf}
    \end{center}
\flushright{\small{\citet{suissa_prevalent_2017}}}
\end{frame}

\begin{frame}{Informative censoring among comparator group}
    \begin{itemize}
	\item Follow-up of comparator group is censored if study treatment starts
	\item Starting treatment = patient is still alive (but sicker?)
	\item Remove bias through inverse probability of censoring weights (IPCW)
    \end{itemize}
\end{frame}

\begin{frame}{Example study}
\calculatespace%
\begin{columns}
\begin{column}{0.20\contentwidth}
\begin{tikzpicture}
  \node[drop shadow={shadow xshift=.8ex,shadow yshift=-.8ex},fill=white,draw] at (0,0) {\includegraphics[width=\textwidth]{ref/suissa-pages-01_sm.jpg}};
\end{tikzpicture}
\end{column}
\begin{column}{0.70\contentwidth}
	\begin{itemize}
		\item T.~Tran and S.~Suissa. \textbf{Comparing {New}-{User} {Cohort} {Designs}: {The} {Example} of {Proton} {Pump} {Inhibitor} {Effectiveness} in {Idiopathic} {Pulmonary} {Fibrosis}.} \emph{American Journal of Epidemiology}, 190\penalty0 (5):\penalty0 928--938, May 2021. doi:10.1093/aje/kwaa242.
\nocite{tran_comparing_2021}
	\end{itemize}
\end{column}
\end{columns}
\end{frame}

\begin{frame}{What is the paper about?}
    \begin{itemize}
	\item Question: Are PPIs efficacious in idiopathic pulmonary fibrosis?
	\item Data: UK linked GP and hospital data (CPRD-GOLD/HES/ONS), $n=2944$
	\item Outcomes: All-cause mortality, respiratory death, hospitalisation
	\item Covariates: Demographics, comorbidity, medicines (all time-varying)
	\item Compare results from three different study designs
    \end{itemize}
\end{frame}

\begin{frame}{Idiopathic Pulmonary Fibrosis (IPF)}
    \begin{itemize}
        \item Progressive scarring of the lungs, cause unknown
	\item Average life expectancy after diagnosis about four years
	\item Pirfenidone (2008) and nintedanib (2014) slow progression
	\item Proton pump inhibitors also guideline recommended (weak evidence)
	\item Contention: PPI effects seen in earlier studies are due to bias
    \end{itemize}
\end{frame}

\begin{frame}{Results}
    \calculatespace%
    \begin{center}
	\includegraphics[height=0.90\contentheight]{ref/suissa-fig3-mscm.pdf}
    \end{center}
\end{frame}

\begin{frame}{PNUC sub-cohorts}
    \calculatespace%
	\includegraphics[width=0.85\contentwidth]{ref/webster-fig1.pdf}
\flushright{\small{\citet{webster-clark_initiator_2020}}}
\end{frame}

\begin{frame}{Developments}
\vskip0pt plus 1filll
    \begin{itemize}
	\item Matching with/without replacement or weighting
	\item How to condition on ``length of exposure''?
	\item Positivity and matching
	\item Estimating the propensity score
	\item Discontinuation only; intensification
    \end{itemize}
\vskip0pt plus 1filll
\nocite{webster-clark_presentation_2020}
\flushright{\small{\citet{webster-clark_alternative_2022}}}
\end{frame}

\begin{frame}{Discussion}
    \begin{itemize}
	\item Does the PNUC design look useful to you?
	\item Would you use it rather than a marginal structural model?
    	\item What are the barriers to using these methods?
    \end{itemize}
\end{frame}

\begin{frame}{References}
        \tiny\bibliography{pnuc.bib}
        \bibliographystyle{abbrvnat}
\end{frame}

\begin{frame}{Acknowledgments}
    \begin{itemize}
	\item An earlier version was presented to the Medicine
	Intelligence CRE methods journal club on 1 July 2021
    \end{itemize}
\end{frame}

\end{document}