update docs

Author: ehsanx

_book/reference-keys.txt

@@ -1,5 +1,8 @@
+tab:reg2a
 fig:gcomppic
 fig:cvpic
+tab:ipw2psx4adj2
+tab:ipw2psx4adj3
 fig:role
 rhc-data-description
 data-download

docs/TMLEw.epub

@@ -410,17 +410,25 @@ <h3><span class="header-section-number">8.2.1</span> Prediction goal</h3>
 <li>Assuming all covariates are measured, <strong>parametric models</strong> such as linear and logistic regressions are very efficient, but relies on strong assumptions. In real-world scenarios, it is often hard (if not impossible) to guess the correct specification of the right hand side of the regression equation.</li>
 <li>Machine learning (ML) methods are very helpful for prediction goals. They are also helpful in <strong>identifying complex functions</strong> (non-linearities and non-additive terms) of the covariates (again, assuming they are measured).</li>
 <li>There are many ML methods, but the procedures are very different, and they come with their own advantages and disadvantages. In a given real data, it is <strong>hard to apriori predict which is the best ML algorithm</strong> for a given problem.</li>
-<li>That’s where super learner is helpful in <strong>combining strength from various algorithms</strong>, and producing 1 prediction column that has <strong>optimal statistical properties</strong>.</li>
 </ul>
+<div class="rmdcomment">
+<p>
+Super learner is helpful in <strong>combining strength from various algorithms</strong>, and producing 1 prediction column that has <strong>optimal statistical properties</strong>.
+</p>
+</div>
 </div>
 <div id="causal-inference" class="section level3" number="8.2.2">
 <h3><span class="header-section-number">8.2.2</span> Causal inference</h3>
 <p><img src="images/dagci.png" width="242" /></p>
 <ul>
 <li>For causal inference goals (when we have a primary exposure of interest), machine learning methods are often misleading. This is primarily due to the fact that they usually do not have an inherent mechanism of focusing on <strong>primary exposure</strong> (RHC in this example); and treats the primary exposure as any other predictors.</li>
-<li>When using g-computation with ML methods, estimation of variance becomes a difficult problem. Generalized procedures such as <strong>robust SE or bootstrap methods</strong> are not supported by theory.</li>
-<li>That’s where TMLE methods shine, with the help of it’s important <strong>statistical properties (double robustness, finite sample properties)</strong>.</li>
+<li>When using g-computation with ML methods, estimation of variance becomes a difficult problem (with correct coverage). Generalized procedures such as <strong>robust SE or bootstrap methods</strong> are not supported by theory.</li>
 </ul>
+<div class="rmdcomment">
+<p>
+TMLE method shine, with the help of it’s important <strong>statistical properties (double robustness, finite sample properties)</strong>.
+</p>
+</div>
 </div>
 <div id="identifiability-assumptions" class="section level3" number="8.2.3">
 <h3><span class="header-section-number">8.2.3</span> Identifiability assumptions</h3>

docs/TMLEw_files/figure-html/unnamed-chunk-29-1.png

@@ -1718,6 +1718,11 @@ <h4><span class="header-section-number">2.3.1.1</span> Step 1</h4>
 <span id="cb63-2"><a href="g-computation.html#cb63-2" aria-hidden="true" tabindex="-1"></a>                               <span class="fu">paste</span>(baselinevars,</span>
 <span id="cb63-3"><a href="g-computation.html#cb63-3" aria-hidden="true" tabindex="-1"></a>                                     <span class="at">collapse =</span> <span class="st">&quot;+&quot;</span>)))</span>
 <span id="cb63-4"><a href="g-computation.html#cb63-4" aria-hidden="true" tabindex="-1"></a>fit1 <span class="ot">&lt;-</span> <span class="fu">lm</span>(out.formula, <span class="at">data =</span> ObsData)</span></code></pre></div>
+<div class="rmdcomment">
+<p>
+<span class="math inline"><span class="math inline">\(Q(A,L)\)</span></span> is often used to represent the predictions from the G-comp model.
+</p>
+</div>
 </div>
 <div id="step-2" class="section level4" number="2.3.1.2">
 <h4><span class="header-section-number">2.3.1.2</span> Step 2</h4>

docs/final-words.html

@@ -379,7 +379,7 @@ <h1 class="title">R Guide for TMLE in Medical Research</h1>
 <h1>Preface</h1>
 <div id="background" class="section level2 unnumbered">
 <h2>Background</h2>
-<p>In comparative effectiveness studies, researchers typically use propensity score methods. However, propensity score methods have known limitations in real-world scenarios, when the true data generating mechanism is unknown. Targeted maximum likelihood estimation (TMLE) is an alternative estimation method with a number of desirable statistical properties. It is a doubly robust method, making use of both the outcome model and propensity score model to generate an unbiased estimate as long as at least one of the models is correctly specified. TMLE also enables the integration of machine learning approaches. Despite the fact that this method has been shown to perform better than propensity score methods in a variety of scenarios, it is not widely used in medical research as the technical details of this approach are generally not well understood.</p>
+<p>In comparative effectiveness studies, researchers typically use propensity score methods. However, propensity score methods have known limitations in real-world scenarios, when the true data generating mechanism is unknown. <strong>Targeted maximum likelihood estimation</strong> (TMLE) is an alternative estimation method with a number of desirable statistical properties. It is a doubly robust method, making use of both the outcome model and propensity score model to generate an unbiased estimate as long as at least one of the models is correctly specified. TMLE also enables the integration of machine learning approaches. Despite the fact that this method has been shown to perform better than propensity score methods in a variety of scenarios, it is <strong>not widely used in medical research</strong> as the implementation details of this approach are generally not well understood.</p>
 </div>
 <div id="goal" class="section level2 unnumbered">
 <h2>Goal</h2>
@@ -400,7 +400,7 @@ <h2>Goal</h2>
 </div>
 <div id="philosophy" class="section level2 unnumbered">
 <h2>Philosophy</h2>
-<p>Code-first philosophy is adopted for this workshop; demonstrating the analyses through one real data analysis problem used in the literature.</p>
+<p><strong>Code-first</strong> philosophy is adopted for this workshop; demonstrating the <strong>analyses through one real data analysis</strong> problem used in the literature.</p>
 <ul>
 <li>This workshop is not theory-focused, nor utilizes simulated data to explain the ideas. Given the focus on implementation, theory is beyond the scope of this workshop.</li>
 <li>At the end of the workshop, we will provide key references where the theories are well explained.</li>
@@ -419,6 +419,7 @@ <h2>Pre-requisites</h2>
 <h2>Version history</h2>
 <p>The workshop was first developed for <a href="https://r-medicine.org/schedule/">R/Medicine
 Virtual Conference</a> 2021, August 24th; title: `An Introductory R Guide for Targeted Maximum Likelihood Estimation in Medical Research’.</p>
+<p>Feel free to reach out for any comments, corrections, suggestions.</p>
 </div>
 <div id="contributor-list" class="section level2 unnumbered">
 <h2>Contributor list</h2>