IDENTIFICATION OF REGRESSION MODELS WITH A MISCLASSIFIED AND ENDOGENOUS BINARY REGRESSOR

We study identification in nonparametric regression models with a misclassified and endogenous binary regressor when an instrument is correlated with misclassification error. We show that the regression function is nonparametrically identified if one binary instrument variable and one binary covariate satisfy the following conditions. The instrumental variable corrects endogeneity; the instrumental variable must be correlated with the unobserved true underlying binary variable, must be uncorrelated with the error term in the outcome equation, but is allowed to be correlated with the misclassification error. The covariate corrects misclassification; this variable can be one of the regressors in the outcome equation, must be correlated with the unobserved true underlying binary variable, and must be uncorrelated with the misclassification error. We also propose a mixture-based framework for modeling unobserved heterogeneous treatment effects with a misclassified and endogenous binary regressor and show that treatment effects can be identified if the true treatment effect is related to an observed regressor and another observable variable.

CATE meets ML

For treatment effects—one of the core issues in modern econometric analysis—prediction and estimation are two sides of the same coin. As it turns out, machine learning methods are the tool for generalized prediction models. Combined with econometric theory, they allow us to estimate not only the average but a personalized treatment effect—the conditional average treatment effect (CATE). In this tutorial, we give an overview of novel methods, explain them in detail, and apply them via Quantlets in real data applications. We study the effect that microcredit availability has on the amount of money borrowed and if 401(k) pension plan eligibility has an impact on net financial assets, as two empirical examples. The presented toolbox of methods contains meta-learners, like the doubly-robust, R-, T- and X-learner, and methods that are specially designed to estimate the CATE like the causal BART and the generalized random forest. In both, the microcredit and 401(k) example, we find a positive treatment effect for all observations but conflicting evidence of treatment effect heterogeneity. An additional simulation study, where the true treatment effect is known, allows us to compare the different methods and to observe patterns and similarities.

Comparing phase 3 “go” decisions (Ph3-GO) between single arm trials with real-world external control (SAT+rwEC) versus randomized phase 2 trials (rPh2).

e13564 Background: Traditional rPh2 trials have limitations that may yield suboptimal Ph3-GO. Compared to a rPh2 of equivalent sample size, SAT+rwEC allows more patients to receive experimental therapies while preserving the ability to compare experimental and control groups. Bias arising from measurement error and confounding in the rwEC, however, poses challenges to statistical inference. Preliminary studies suggest higher response rates are observed in rwEC than randomized controls. We compared Ph3-GO decisions between SAT+rwEC and rPh2. Methods: Ph3-GO probability was compared using simulation studies that resembled the oncology setting with objective response rate (ORR) endpoint. rPh2 simulated parameters were: sample size (60-120) with 1:1 randomization, ORR in rPh2 control (15%-50%), true treatment effect (ΔORR: 0-50). For each rPh2 of a given sample size, we evaluated an SAT+rwEC that re-allocated all rPh2 control patients to the experimental arm (i.e., doubling the sample size of the experimental arm) and added an rwEC. SAT+rwEC were simulated with assumptions for size (rwEC to SAT ratio: 0.5 to 2) and net bias (-10 to +10), which was simulated as a composite representing ORR measurement error plus residual confounding after multivariable adjustment. Positive direction of net bias corresponds to higher ORR in the rwEC. Ph3-GO thresholds varied from 10-30%. Ph3-GO was considered “False-GO” when true treatment effect < threshold, and “True-GO” when true treatment effect ≥ threshold. Results: With positive net bias of +10, SAT+rwEC had lower False-GO and True-GO decisions compared to rPh2. With negative net bias of -10, both False-GO and True-GO probabilities were higher for the SAT+rwEC. When net bias=0, the increased size of SAT+rwEC resulted in observable Ph3-GO improvements with lower False-GO and higher True-GO than corresponding rPh2. Conclusions: An interactive dashboard was developed for users. The magnitude and direction of net bias relative to the decision threshold affect the performance of SAT+rwEC. The relative sample size of rwEC to rPh2 may also impact performance. The dashboard can provide quantitative guidance for Ph3-GO if net bias can be estimated by independent studies. Further work to quantify net bias and refine Ph3-GO criteria can help reduce the currently high False-GO rates while increasing opportunities for patients to receive experimental therapies through the SAT+rwEC design. Ph3-GO probability for rPh2 vs. SAT+rwEC with threshold=15%, baseline ORR=20% (select scenarios).[Table: see text]

ANCOVA versus Change Score for the Analysis of Nonexperimental Two-Wave Data: A Structural Modeling Perspective

There is an ongoing debate on whether the analysis of covariance (ANCOVA) or the change score approach is more appropriate when analyzing nonexperimental pre-post designs. In this article, we use a structural modeling perspective to clarify the different assumptions that are made by the ANCOVA and the change score approaches to identify the causal effect of a treatment variable. We show that the change score approach offers the option of controlling for unobserved confounders but relies on strong assumptions about the effects of these unobserved confounders and does not allow for dynamic causal relationships. By contrast, the ANCOVA approach is based on a selection-on-observables approach and assumes that all relevant confounders are measured. Furthermore, we illustrate conditions under which the two approaches give lower and upper bounds of the true treatment effect, and we discuss the role of measurement error. Implications for the analysis of nonexperimental two-wave data are discussed.

Error propagation in simulated treatment comparisons for multiple myeloma: Results from a simulation.

e20523 Background: Comparing the effectiveness of multiple myeloma treatments presents a challenge due to the limited number of head-to-head trials with which to conduct indirect treatment comparisons. This is particularly true when subgroup analysis is of interest. In comparative effectiveness research Simulated Treatment Comparisons (STCs) are becoming increasingly common in the absence of head-to-head trials. STCs use estimates from limited IPD to adjust for covariate imbalance between trials, however the uncertainty from these estimates is generally ignored when estimating relative treatment effects. This study demonstrates the need to account for this uncertainty when conducting STCs for indications such as multiple myeloma. We introduce an STC method that accounts for the uncertainty due to covariate adjustment, and demonstrate its effectiveness via simulation. Methods: We simulated two single arm studies (N = 300 for both), each containing age and overall survival. We assume study 1 has individual patient data available, and study 2 only has aggregate age data and a digitized Kaplan-Meier curve. We compute a covariate adjustment term based on the mean age difference between the studies and the age coefficients from fitting a parametric survival model to the observed study 1 IPD. We then estimate the variance of this adjustment term via bootstrapping and incorporate this uncertainty into a Bayesian STC model which estimates the relative treatment effect for the two study datasets converted to a digitized Kaplan-Meier format. Results: The proportion of 95% credible intervals (CrI) that captured the true treatment effect was 86.8% without error propagation, whereas 92.0% of CrI’s captured the true treatment with error propagation. 94.9% of CrI’s contained the true treatment effect when using survival regression with the complete IPD. Conclusions: Failing to account for uncertainty from the covariate adjustment when conducting simulated treatment comparisons generally leads to underestimating the uncertainty of the relative treatment effect. This method better captures the uncertainty introduced when conducting an STC and has the potential to yield more reliable estimates of the comparative effectiveness of multiple myeloma treatments.

Performance of informative priors skeptical of large treatment effects in clinical trials: A simulation study

One of the main advantages of Bayesian analyses of clinical trials is their ability to formally incorporate skepticism about large treatment effects through the use of informative priors. We conducted a simulation study to assess the performance of informative normal, Student- t, and beta distributions in estimating relative risk (RR) or odds ratio (OR) for binary outcomes. Simulation scenarios varied the prior standard deviation (SD; level of skepticism of large treatment effects), outcome rate in the control group, true treatment effect, and sample size. We compared the priors with regards to bias, mean squared error (MSE), and coverage of 95% credible intervals. Simulation results show that the prior SD influenced the posterior to a greater degree than the particular distributional form of the prior. For RR, priors with a 95% interval of 0.50–2.0 performed well in terms of bias, MSE, and coverage under most scenarios. For OR, priors with a wider 95% interval of 0.23–4.35 had good performance. We recommend the use of informative priors that exclude implausibly large treatment effects in analyses of clinical trials, particularly for major outcomes such as mortality.

Abstract W P52: Combination Argatroban and TPA for Acute Ischemic Stroke: A Matched Case-Control Analysis

Background and Objective: Thrombin inhibitors such as Argatroban in combination with full-dose IV-tPA are currently being tested in a series of clinical trials for acute ischemic stroke. A single-arm, phase IIa study (ARTSS-1) of 65 patients treated with low-dose Argatroban + IV-tPA in either the 3 or 4.5 hour treatment window demonstrated an overall sICH rate of 4.6% and 40% complete recanalization within 2-hours. The objective of the study was to assess safety in a matched group of IV-tPA alone patients. Methods: pPAIRS © performs a post-hoc Euclidean matching of subjects on factors of choice and excludes outliers when no close match is found. We applied these techniques to compare outcomes of Argatroban+tPA treated subjects from ARTSS-1 with tPA treated subjects from NINDS. Matching factors were NIHSS, age, glucose, race, gender, hemisphere, and subtype. Outcomes were sICH, death and mRS 0-1, 0-2 at day 7 and 90. Results: Of the 65 Argatroban+TPA patients, 55 and 44 had good NINDS matches at day 7 and 90, respectively. Rates of sICH were identical at both day 7 and 90 between ARTSS-1 and NINDS (3.6% and 2.3%). Mortality was not significantly different (5.5% vs. 3.6% at day 7; p=1.00; and 11.4% in each group at day 90). There was a non-significant trend for higher percent of mRS 0-1 (36.4% vs. 32.0%, p=0.814) and 0-2 (47.7% vs. 43.2%, p=0.823) at day 90 in the ARTSS-1 patients that was not present at day 7. The mRS 0-2 trend remained after excluding ARTSS-1 patients treated with tPA between 3-4.5 hours (see table). Conclusion: There is no evidence for significantly increased mortality or symptomatic hemorrhage after addition of low dose Argatroban compared to a NINDS population with similar baseline characteristics particularly in those tPA-treated within 3 hours. A non-significant trend emerged at 90 days for 14% relative improvement in functional outcomes. A larger, Phase IIb, randomized trial (ARTSS-2) is ongoing to determine whether these trends reflect a true treatment effect.

Osteoporosis in Men

The prevalence of osteoporosis is estimated to be 18% in men, but 30% of all fractures occur in men. With age, men experience a gradual decline in testosterone production and bone density. The rate of trabecular bone loss in the lumbar spine in men over age 50 can be double the rate of loss in men under age 50. Endogenous testosterone, estradiol, and their metabolites play a role in maintaining bone health, but their specific effects on bone turnover have been difficult to elucidate. Recently, large cohort studies have provided more detailed information confirming estrogen’s associations and further characterizing the effect of endogenous testosterone and its metabolites on bone mineral density and fractures. Very few clinical trials have assessed the impact of testosterone replacement therapy (TRT) on bone density and fractures in men. The few studies that have been conducted are generally small and not robust enough to show the true treatment effect of TRT and adequately determine its safety. In the absence of data on patient outcomes, it is important for pharmacists to understand the impact of drug therapy on biomarkers and surrogate markers of disease for optimal pharmacotherapy selection and monitoring.