You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
JobMarketPaper/Paper/sections/10_CausalStory.tex

373 lines
19 KiB
TeX

\documentclass[../Main.tex]{subfiles}
\graphicspath{{\subfix{Assets/img/}}}
\begin{document}
%I need to describe separating concerns, e.g.
% Begin by talking about goal, what does it mean? This might need some work prior to give more background.
As I am trying to separate strategic concerns
(the effect of a marginal treatment methodology)
and an operational concern
(the effect of a delay in closing enrollment),
we need to look at what confounds these effects and how we might measure them.
To start, we'll look at the data generating model, the values of interest,
and both the observed and unobserved confounders.
We'll also discuss how the data collected fits the data generating process.
The primary effects one might expect to see are that
\begin{enumerate}
\item Adding more drugs to the market will make it harder to
finish a trial as it is
more likely to be terminated due to concerns about profitabilty.
\item Adding more drugs to the market
will make it harder to recruit, slowing enrollment.
\item Enrollment challenges (i.e. delays) increase the likelihood that
a trial will terminate.
\end{enumerate}
Unfortunately, these causal effects are confounded in many different ways.
Figure \ref{FIG:CausalModel} contains a description of the causal model.
% The first issue is that the severity of the disease and the size of the population
% who has that disease affects the ease of enrolling participants.
% For example, a large population may make it easier to find enough participants
% to achieve the required statistical discrimination between
% control and treatment.
% Second, for some diseases there exists an endogenous dynamic
% between the treatments available for a disease and the
% market size/population with that disease.
% \authorcite{cerda_endogenousinnovationspharmaceutical_2007}
% proposes two mechanisms
% that link the drugs on the market and market size.
% The inverse is that for many chronic diseases with high mortality rates,
% more drugs cause better survivability, increasing the size of those markets.
% The third major confound is that the drugs on the market affect enrollment.
% If there is a treatment already on the market, patients or their doctors
% may be less inclined to participate in the trial, even if the current treatment
% has severe downsides.
%
% There are additional problems.
% One is in that the disease being treated affects the
% safety and efficacy standards that the drug will be held too.
% For example, if a particular cancer is very deadly and does not respond well
% to current treatments, Phase I trials will enroll patients with that cancer,
% as opposed to the standard of enrolling healthy volunteers
% \cite{commissioner_drugdevelopmentprocess_2020}
% to establish safe dosages and (hopefully) obtain some effectiveness data.
% % The trial is more likely to be terminated early if the drug is unsafe or has no
% % discerenable effect, therefore termination depends in part on a compound-disease
% % interaction.
% Another challenge comes from the interaction between duration and termination;
% in that if a trial terminates before closing enrollment for issues other
% than enrollment, then the enrollment will still be low.
% On the other hand, if enrollment is low, the trial might terminate.
% Thus it is impossible to tell if the low enrollment caused the termination
% or if the termination caused the low enrollment.
% Finally, while conducting a trial, the safety and efficacy of a drug are driven by
% fundamental pharmacokinetic properties of the compounds.
% These are only imperfectly measured both prior to and during any given trial.
% Previously measured safety and efficacy inform the decision to start the trial
% in the first place while currently observed safety and efficiency results
% help the sponsor judge whether to continue the trial.
% In contrast, the recruitment rate may depend on the previous results about safety
% and efficacy.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% \subsection{Data Summary}
% %% Describe data here
% Since Sep 27th, 2007 those who conduct clinical trials of FDA controlled
% drugs or devices on human subjects must register
% their trial at \url{ClinicalTrials.gov}
% (\cite{anderson_fdadrugapproval_2022}).
% This involves submitting information on the expected enrollment and duration of
% trials, drugs or devices that will be used, treatment protocols and study arms,
% as well as contact information the trial sponsor and treatment sites.
%
% When starting a new trial, the required information must be submitted
% ``\dots not later than 21 calendar days after enrolling the first human subject\dots''.
% After the initial submission, the data is briefly reviewed for quality and
% then the trial record is published and the trial is assigned a
% National Clinical Trial (NCT) identifier.
% (\cite{anderson_fdadrugapproval_2022}).
%
% Each trial's record is updated periodically, including a final update that must occur
% within a year of completing the primary objective, although exceptions are
% available for trials related to drug approvals or for trials with secondary
% objectives that require further observation\footnote{This rule came into effect in 2017}
% (\cite{anderson_fdadrugapproval_2022}).
% Other than the requirements for the first and last submissions, all other
% updates occur at the discresion of the trial sponsor.
% Because the ClinicalTrials.gov website serves as a central point of information
% on which trials are active or recruting for a given condition or drug,
% most trials are updated multiple times during their progression.
%
% There are two primary ways to access data about clinical trials.
% The first is to search individual trials on ClinicalTrials.gov with a web browser.
% This web portal shows the current information about the trial and provides
% access to snapshots of previously submitted information.
% Together, these features fulfill most of the needs of those seeking
% to join a clinical trial.
% For this project I've been able to scrape these historical records to establish
% snapshots of the records provided.
% %include screenshots?
% The second way to access the data is through a normalized database setup by
% the
% \href{https://aact.ctti-clinicaltrials.org/}{Clinical Trials Transformation Initiative}
% called AACT. %TODO: Get CITATION
% The AACT database is available as a PostgreSQL database dump or set of
% flat-files.
% These dumps match a near-current version of the ClinicalTrials.gov database.
% This format is ameniable to large scale analysis, but does not contain
% information about the past state of trials.
% I combined these two sources, using the AACT dataset to select
% trials of interest and then scraping \url{ClinicalTrials.gov} to get
% a timeline of each trial.
%
% %%%%%%%%%%%%%%%%%%%%%%%% Model Outline
%
% The way I use this data is to predict the final status of the trial
% from the snapshots that were taken, in effect asking:
% ``how does the probability of a termination change from the current state
% of the trial if X changes?''
%
%% Return to causal identification
\subsection{Causal Identification}
Because running experiments on companies running clinical trials is not going
to happen anytime soon, causal identification depends on using a
structural causal model.
Because the data generating process for the clinical trials records is rather
straightforward, this is an ideal place to use
\authorcite{pearl_causalitymodelsreasoning_2009}
Do-Calculus.
This process involves describing the data generating process in the form of
a directed acyclic graph, where the nodes represent different variables
within the causal model and the directed edges (arrows) represent
assumptions about which variables influence the other variables.
There are a few algorithms that then tell the researcher which of the
relationships will be confounded, which ones can be statistically estimated,
and provides some hypotheses that can be tested to ensure the model is
reasonably correct.
In \cref{Fig:CausalModel} I diagram the directed acyclic graph that describes
my proposed data generating process,
It revolves around the decisions made by the study sponsor,
who must decide whether to let a trial run to completion
or terminate the trial early.
While receiving updates regarding the status of the trial, they ask questions
such as:
\begin{itemize}
\item Do I need to terminate the trial due to safety incidents?
\item Does it appear that the drug is effective enough to achieve our
goals, justifying continuing the trial?
\item Are we recruiting enough participants to achive the statistical
results we need in the budget we have?
\item Does the current market conditions and expectations about returns on
investment justify the expenditures we are making?
\end{itemize}
When appropriate issues arise, the study sponsor terminates the trial, otherwise
it continues to completion.
\begin{figure}[H] %use [H] to fix the figure here.
\includegraphics[width=\textwidth]{../assets/img/CausalModel.drawio.png}
\caption{Graphical Causal Model}
% \small{Crimson boxes are the variables of interest,
% white boxes are unobserved, while the gray boxes will be controlled for.}
\label{FIG:CausalModel}
\end{figure}
A quick summary of the nodes of the DAG,
which nodes are captured in the data,
the hypothesized relationships in the model,
and the proposed confounding pathways.
\begin{itemize}
\item Items of Interest (Blue boxes and Arrow)
\begin{enumerate}
\item \texttt{Enrollment Level (Enrollment Status)}:
While occasionally a trial will keep the enrollment numbers
up to date, the only regular information on enrollment recieved
is the enrollment status, i.e. whether they have finished
recruiting or not.
\item \texttt{Will it Terminate?}:
This represents whether the trial was terminated or if it
completed successfully.
\item The effect of \texttt{Enrollment Status} on
\texttt{Will it Terminate?}:
How does changing the enrollment status affect the
probability of termination.
\end{enumerate}
\item Observed values (Solid orange boxes)
\begin{enumerate}
\item \texttt{Condition}
(Not drawn in DAG because it impacts everything):
The underlying condition, classified by IDC-10 group.
This impacts every other aspect of the model and is pulled from
the AACT dataset.
\item \texttt{Market Measures}:
Various measures of the number of alternate drugs on the market.
These are either the number of other drugs with the same active ingredient as the trial
(both generic and originators),
and those considered alternatives in various formularies published by the United States Pharmacopeia.
\item \texttt{Population (market size)}:
Multiple measures of the impact the disease.
These are measured by the DALY cost of the disease, and is
separated by the impact on countries with
High, High-Medium, Medium, Medium-Low, and Low
Socio-Demographic Index (SDI) scores.
This data comes from the Institute for Health Metrics' Global Burden of Disease study
\cite{vos_globalburden369_2020}.
\item \texttt{Elapsed Duration}:
A normalized measure of the time elapsed in the trial.
Comes from the original estimate of the trial's primary completion date and the registered start date.
I take the difference in days between these, and get the percentage of that time that has elapsed.
This calculation is based on data from the snapshots and the
AACT final results.
\item \texttt{Decision to Proceed with Phase III}:
If the compound development has progressed to Phase III.
This is included in the analysis by only including
Phase III trials registered in the AACT dataset.
\end{enumerate}
\item Unobserved (Green Boxes with squiggle hatch marks)
\begin{enumerate}
\item \texttt{Fundamental Efficacy and Safety}:
The underlying safety of the compound.
Cannot be observed, only estimated through scientific study.
\item \texttt{Previously observed Efficacy and Safety}:
The information gathered in previous studies.
This is not available in my dataset because I don't
have links to prior studies.
\item \texttt{Currently observed Efficiency and Safety}:
The information gathered during this study.
This is only partially available, and so is
treated as unavailable.
After a study is over, the investigators are
often publish information about adverse events, but only
those that meet a certain threshold.
As this information doesn't appear to be provided to
participants, we don't consider it.
\end{enumerate}
% \end{itemize}
%
% %
%
% \begin{itemize}
% \item Relationships of interest
% \begin{enumerate}
% \item \texttt{Enrollment Status} $\rightarrow$ \texttt{Will Terminate?}:
% This is the primary effect of interest.
% \item \texttt{Market Measures} $\rightarrow$ \texttt{Will Terminate?}:
% This is the secondary effect of interest.
% \end{enumerate}
\item Jointly determined variables
\begin{enumerate}
\item
\texttt{Enrollment Level (Enrollment Status)}
$\leftrightarrow$
\texttt{Elapsed Duration}:
Because I only observe enrollment status and have not good estimate of
the enrollment process, there is a potential for confounding between
the elapsed duration of a trial and the enrollment status.
The proposed mechansims are through the partially observed levels of
enrollment.
First, as a trial progresses, the enrollment levels should grow until
it matches the planned enrollment and the trial ends.
Thus under good circumstances, elapsed duration drives
enrollment levels.
Under bad circumstances though, low enrollment levels may cause the
duration to extend, as study sponsors spend more resources
to complete the trial successfully.
This is an issue because the only complete measure of enrollment
that we currently have is the enrollment status, and thus I cannot
control for this effect.
\item
\texttt{Market Conditions}
$\leftrightarrow$
\texttt{Population}:
There exists an endogenous dynamic between
between the treatments available for a disease and the
market size/population with that disease.
\authorcite{cerda_endogenousinnovationspharmaceutical_2007}
proposes two mechanisms
that link the drugs on the market and market size.
The first is that a larger population increases the potential
profitability, trying to get more treatments allowed.
The inverse is that for many chronic diseases with high mortality rates,
more drugs cause better survivability, increasing the size of those markets.
\end{enumerate}
\item Confounding Pathways
\begin{enumerate}
\item
\texttt{Condition} (Not drawn in figure \ref{FIG:CausalModel}):
Interacts with everything.
\item Backdoor Pathway
between \texttt{Will Terminate?} and
\texttt{Enrollment Status} through
\texttt{Fundamental Safety and Efficacy}.
The concern is that since previously learned information
and current information are driven by the same underlying
physical reality, the enrollment process and
termination decisions may be correlated.
Controlling for the decision to proceed with the trial is the
best adjustment available to block this confounding pathway.
Below I describe the exact pathways.
\begin{enumerate}
\item
\texttt{Will Terminate?}
$\leftarrow$
\texttt{Currently Observed Efficacy and Safety}
$\leftarrow$
\texttt{Fundamental Efficacy and Safety}
$\rightarrow$
\texttt{Previously Observed Efficacy and Safety}
$\rightarrow$
\texttt{Is likely safe and effective (Decision to proceed with Phase III trial)}
$\rightarrow$
\texttt{Enrollment Process Parameters}
$\rightarrow$
\texttt{Enrollment Levels (Enrollment Status)}
\end{enumerate}
\item
Backdoor Pathways through \texttt{Population} and
\texttt{Market Conditions}
The concern with this pathway is that the rate of enrollment, and
thus the enrollment status, is affected by the Population with
the disease and the market condition.
\begin{enumerate}
\item
\texttt{Will Terminate?}
$\leftarrow$
\texttt{Market Conditions}
$\rightarrow$
\texttt{Enrollment Process Parameters}
$\rightarrow$
\texttt{Enrollment Levels (Enrollment Status)}
\item
\texttt{Will Terminate?}
$\leftarrow$
\texttt{Market Conditions}
$\leftrightarrow$
\texttt{Population}
$\rightarrow$
\texttt{Enrollment Process Parameters}
$\rightarrow$
\texttt{Enrollment Levels (Enrollment Status)}
\end{enumerate}
\item Backdoor Pathway through
\texttt{Elapsed Duration}.
\begin{enumerate}
\item
\texttt{Will Terminate?}
$\leftarrow$
\texttt{Elapsed Duration}
$\leftrightarrow$
\texttt{Enrollment Levels (Enrollment Status)}
\end{enumerate}
\end{enumerate}
\end{itemize}
\end{document}