Biased Estimates Of Treatment Effects Research Articles

Instrumental variable based estimation under the semiparametric accelerated failure time model.

Randomized controlled trials are the gold standard for estimating causal effects of treatments or interventions, but in many cases are too costly, too difficult, or even unethical to conduct. Hence, many pressing medical questions can only be investigated using observational studies. However, direct statistical modeling of observational data can result in biased estimates of treatment effects due to unmeasured confounding. In certain cases, instrumental variable based techniques can be used to remove such biases. These techniques are indeed widely studied and used in econometrics under parametric outcome models, however limited works have focused on the utilization of instrumental variables in survival analysis, where semiparametric models are often necessary. The additional challenge in analyzing survival data is the presence of censoring. In this paper, we introduce an instrumental variable method that relaxes the strong assumptions of previous works and provides consistent estimation of the causal effect of a treatment on a survival outcome. We demonstrate the efficacy of our method in various simulated settings and an analysis of Medicare enrollment data comparing two prevalent surgical procedures for abdominal aortic aneurysm from an observational study.

Regression discontinuity designs with unknown state-dependent discontinuity points: estimation and testing

Abstract This paper extends Regression discontinuity designs with unknown discontinuity points developed by (Porter, J., and P. Yu. 2015. “Regression Discontinuity Designs with Unknown Discontinuity Points: Testing and Estimation.” Journal of Econometrics 189: 132–147.) to allow for state-dependent discontinuity points. We discuss the estimation of the model, and propose test statistics for treatment effect and state dependency in the discontinuity points. We conduct Monte Carlo simulations to compare the proposed estimator with these based on the constant discontinuity RDD and the classic fuzzy RDD, and find that overlooking the state dependency can lead to biased estimates of treatment effects, while the proposed estimator works well and is robust when applied to constant discontinuity RDDs. Monte Carlo experiments also point out that the sizes and powers of the proposed test statistics are generally satisfactory. The model is illustrated with an empirical application.

Sensitivity analysis for non-monotone missing binary data in longitudinal studies: Application to the NIDA collaborative cocaine treatment study.

Conventional approaches for handling missingness in substance use disorder trials commonly rely upon a single deterministic "worst value" imputation that posits a perfect relationship between missingness and drug use ("missing value = presumed drug use"); this yields biased estimates of treatment effects and their standard errors. Instead, deterministic imputations should be replaced by probabilistic versions that encode researchers prior beliefs that those with missing data are more likely to be using drugs at those occasions. Motivated by this problem, we present a method for handling non-monotone missing binary data in longitudinal studies. Specifically, we consider a joint model that combines a not missing at random (NMAR) selection model with a generalized linear mixed model for longitudinal binary data. The selection model links the distribution of a missing outcome to the corresponding distribution of the outcome for those observed at that occasion via a fixed and known sensitivity parameter. The mixed model for longitudinal binary data assumes the random effects have bridge distributions; the latter yields regression parameters that have both subject-specific and marginal interpretations. This approach is completely transparent about what is being assumed about missing data and can be used as the basis for sensitivity analysis.

Impact of Non-Adherence and Flare Resolution on the Cost-Effectiveness of Treatments for Gout: Application of a Linked Pharmacometric/Pharmacoeconomic Model.

Dual urate-lowering therapy (ULT) with lesinurad in combination with either allopurinol or febuxostat is an option for patients with gout unsuccessfully treated on either monotherapy. Treatment failure is often a result of poor medication adherence. Imperfect adherence in clinical trials may lead to biased estimates of treatment effect and confound the results of cost-effectiveness analyses. To estimate the impact of varying medication adherence on the cost effectiveness of lesinurad dual therapy and estimate the value-based price of lesinurad at which the incremental cost-effectiveness ratio is equalto £20,000 per quality-adjusted life-year (QALY). Treatment effect was simulated using published pharmacokinetic-pharmacodynamic models and scenarios representing adherence in clinical trials, routine practice, and perfect use. The subsequent cost and health impacts, over the lifetime of a patient cohort, were estimated using a bespoke pharmacoeconomic model. The base-case incremental cost-effectiveness ratios comparing lesinurad dual ULT with monotherapy ranged from £39,184 to £78,350/QALY gained using allopurinol and £31,901 to £124,212/QALY gained using febuxostat, depending on the assumed medication adherence. Results assuming perfect medication adherence imply a per-quarter value-based price of lesinurad of £45.14 when used in dual ULT compared with allopurinol alone and £57.75 compared with febuxostat alone, falling to £25.41 and £3.49, respectively, in simulations of worsening medication adherence. The estimated value-based prices of lesinurad only exceeded that which has been proposed in the United Kingdom when assuming both perfect drug adherence and the eradication of gout flares in sustained treatment responders.

Causal Inference with Networked Treatment Diffusion

Treatment interference (i.e., one unit’s potential outcomes depend on other units’ treatment) is prevalent in social settings. Ignoring treatment interference can lead to biased estimates of treatment effects and incorrect statistical inferences. Some recent studies have started to incorporate treatment interference into causal inference. But treatment interference is often assumed to follow a simple structure (e.g., treatment interference exists only within groups) or measured in a simplistic way (e.g., only based on the number of treated friends). In this paper, I highlight the importance of collecting data on actual treatment diffusion in order to more accurately measure treatment interference. Furthermore, I show that with accurate measures of treatment interference, we can identify and estimate a series of causal effects that are previously unavailable, including the direct treatment effect, treatment interference effect, and treatment effect on interference. I illustrate the methods through a case study of a social network–based smoking prevention intervention.

PRM116 - Do Models With Better Goodness Of Fit Ensure Less Biased Treatment Effect Estimates?

There is widespread misconception that goodness-of-fit statistics are critical in determining whether treatment effects estimated from a model are unbiased. We aimed to examine whether models with better goodness-of-fit ensure less biased treatment effect estimates. We constructed a simulated dataset of 10,000 patients to compare two treatments on a continuous outcome of a prognostic score (mean: 0.82; range:-6.2to8.0) wherein a higher score indicates a higher risk of mortality within one year. By design in our simulation, the new treatment would lower the prognostic score by 0.1 compared to the traditional treatment. We then generated two study designs using this dataset: 1) a randomized controlled trial (RCT) with randomly assigned treatment; 2) an observational study design wherein patients receiving the new treatment tended to be sicker on observed (e.g. older age, higher comorbidity score) and unobserved covariates. Treatment effect estimates were obtained from the observational study design using: 1) multivariable linear regression accounting for observed covariates, and 2) instrumental variable two-stage least squares (2SLS) regression accounting for observed and unobserved confounders and were compared with 3) the regression results from the RCT. Bootstrapping with 1000 resamples and standardized errors (SE) were used to estimate out-of-sample prediction for each of the three models. Across the three models, multivariable linear regression had the highest goodness-of-fit (adjusted R-squared 0.215) and best out-of-sample prediction (SE: 2.15), but very biased treatment effect estimates due to unmeasured confounders (bias: 1.119, 95%CI: 1.101to1.137). The RCT regression had the lowest R-squared (0.001) and 2SLS had a R-squared of 0.122. Both RCT (bias: 0.003, 95%CI: -0.017to0.024) and 2SLS (bias: 0.004, 95%CI: -0.039to0.046) had unbiased estimates. Models with better goodness-of-fit would have better predictive power but do not ensure less biased treatment effect estimates. Correct study design and model specification are important for unbiased treatment effect estimates.

An experimental study in distinguishing an authentic herbal substance from sham herbal substances

ObjectivesAn unblinded randomised trial can result in biased treatment effect estimates and lead to erroneous conclusions on the efficacy of the therapeutic intervention. Unlike pharmaceutical substances, Chinese herbal medicines have special characteristics including texture, colour, odour and taste as the origin of these constituents are different. In addition, its distinctive odour makes blinding of Chinese herbal medicine RCTs very difficult, as the placebo substance needs to match the special characteristics of the herbal substance being investigated. For these reasons, two studies were undertaken to evaluate whether a participant could distinguish a herbal intervention capsule (Ganopoly combination) when compared to three types of capsules containing culinary materials following a visual, odour and taste evaluation. DesignStudy One, was a pilot involving eleven participants (n = 11) while Study Two, involved eighty one participants (n = 81) and was conducted to make improvements on Study One. For both studies, participants were asked to identify which of four capsules were a herbal substance following a visual, smell and taste evaluation. ResultsFor study One it was found that for both odour (p = .484) and visual appearance (p = .077) the number of participants selecting the herbal substance was not significantly different from what may have been selected by chance. This was not the case for taste where significantly more participants selected capsule B as the herbal substance (p = .004). For Study Two test it was found that all three evaluations for odour, visual appearance and taste significantly more participants selected the herbal substance (p < .0001 in all cases). This indicates that the participant guesses were not evenly distributed across the four choices and suggests a failure to blind. ConclusionThe failure to blind participants highlights the difficulties in preparing sham herbal substances that look, smell and taste like the real herbal substance.

2SLS versus 2SRI: Appropriate methods for rare outcomes and/or rare exposures.

This study used Monte Carlo simulations to examine the ability of the two-stage least squares (2SLS) estimator and two-stage residual inclusion (2SRI) estimators with varying forms of residuals to estimate the local average and population average treatment effect parameters in models with binary outcome, endogenous binary treatment, and single binary instrument. The rarity of the outcome and the treatment was varied across simulation scenarios. Results showed that 2SLS generated consistent estimates of the local average treatment effects (LATE) and biased estimates of the average treatment effects (ATE) across all scenarios. 2SRI approaches, in general, produced biased estimates of both LATE and ATE under all scenarios. 2SRI using generalized residuals minimized the bias in ATE estimates. Use of 2SLS and 2SRI is illustrated in an empirical application estimating the effects of long-term care insurance on a variety of binary health care utilization outcomes among the near-elderly using the Health and Retirement Study.

Regularization and Confounding in Linear Regression for Treatment Effect Estimation

This paper investigates the use of regularization priors in the context of treatment effect estimation using observational data where the number of control variables is large relative to the number of observations. First, the phenomenon of “regularization-induced confounding” is introduced, which refers to the tendency of regularization priors to adversely bias treatment effect estimates by over-shrinking control variable regression coefficients. Then, a simultaneous regression model is presented which permits regularization priors to be specified in a way that avoids this unintentional “re-confounding”. The new model is illustrated on synthetic and empirical data.

WITHDRAWN: Chemoimmunotherapy versus chemotherapy for metastatic malignant melanoma.

Malignant melanoma, one of the most aggressive of all skin cancers, is increasing in incidence throughout the world. Surgery remains the cornerstone of curative treatment in earlier stages. Metastatic disease is incurable in most affected people, because melanoma does not respond to most systemic treatments. A number of novel approaches are under evaluation and have shown promising results, but they are usually associated with increased toxicity and cost. The combination of chemotherapy and immunotherapy has been reported to improve treatment results, but it is still unclear whether evidence exists to support this choice, compared with chemotherapy alone. No language restrictions were imposed. To compare the effects of therapy with chemotherapy and immunotherapy (chemoimmunotherapy) versus chemotherapy alone in people with metastatic malignant melanoma. We searched the Cochrane Skin Group Specialised Register (14 February 2006), the Cochrane Central Register of Controlled Trials (The Cochrane Library Issue 3, 2005), MEDLINE (2003 to 30 January 2006 ), EMBASE (2003 to 20 July 2005) and LILACS (1982 to 20 February 2006). References, conference proceedings, and databases of ongoing trials were also used to locate trials. All randomised controlled trials that compared the use of chemotherapy versus chemoimmunotherapy on people of any age, diagnosed with metastatic melanoma. Two authors independently assessed each study to determine whether it met the pre-defined selection criteria, with differences being resolved through discussion with the review team. Two authors independently extracted the data from the articles using data extraction forms. Quality assessment included an evaluation of various components associated with biased estimates of treatment effect. Whenever possible, a meta-analysis was performed on the extracted data, in order to calculate a weighed treatment effect across trials. Eighteen studies met our criteria and were included in the meta-analysis, with a total of 2625 participants. We found evidence of an increase of objective response rates in people treated with chemoimmunotherapy, in comparison with people treated with chemotherapy. Nevertheless, the impact of these increased response rates was not translated into a survival benefit. We found no difference in survival to support the addition of immunotherapy to chemotherapy in the systemic treatment of metastatic melanoma, with a hazard ratio of 0.89 (95% CI 0.72 to 1.11, P = 0.31). Additionally, we found increased hematological and non-hematological toxicities in people treated with chemoimmunotherapy. We failed to find any clear evidence that the addition of immunotherapy to chemotherapy increases survival of people with metastatic melanoma. Further use of combined immunotherapy and chemotherapy should only be done in the context of clinical trials.

A comparison of meta-analytic methods for synthesizing evidence from explanatory and pragmatic trials

BackgroundThe pragmatic–explanatory continuum indicator summary version 2 (PRECIS-2) tool has recently been developed to classify randomized clinical trials (RCTs) as pragmatic or explanatory based on their design characteristics. Given that treatment effects in explanatory trials may be greater than those obtained in pragmatic trials, conventional meta-analytic approaches may not accurately account for the heterogeneity among the studies and may result in biased treatment effect estimates. This study investigates if the incorporation of PRECIS-2 classification of published trials can improve the estimation of overall intervention effects in meta-analysis.MethodsUsing data from 31 published trials of intervention aimed at reducing obesity in children, we evaluated the utility of incorporating PRECIS-2 ratings of published trials into meta-analysis of intervention effects in clinical trials. Specifically, we compared random-effects meta-analysis, stratified meta-analysis, random-effects meta-regression, and mixture random-effects meta-regression methods for estimating overall pooled intervention effects.ResultsOur analyses revealed that mixture meta-regression models that incorporate PRECIS-2 classification as covariate resulted in a larger pooled effect size (ES) estimate (ES = − 1.01, 95%CI = [− 1.52, − 0.43]) than conventional random-effects meta-analysis (ES = − 0.15, 95%CI = [− 0.23, − 0.08]).ConclusionsIn addition to the original intent of PRECIS-2 tool of aiding researchers in their choice of trial design, PRECIS-2 tool is useful for explaining between study variations in systematic review and meta-analysis of published trials. We recommend that researchers adopt mixture meta-regression methods when synthesizing evidence from explanatory and pragmatic trials.

Assessment of quality of reporting of Helicobacter pylori related randomized controlled trials: a focus on highly ranked gastroenterology journals

&lt;p class="abstract"&gt;&lt;strong&gt;Background:&lt;/strong&gt; Randomized controlled trials are often considered as the gold standard for measuring the effectiveness of an intervention. However, inappropriate or poor reporting in randomized controlled trials can produce biased estimates of treatment effects.&lt;strong&gt; &lt;/strong&gt;Clinical trials that do not use the CONSORT statement for reporting their findings will have limited value to the clinicians and researchers due to the risk of bias in their results. This review aims to assess the quality of reporting of randomized controlled trials in &lt;em&gt;Helicobacter pylori&lt;/em&gt; associated infections by using the CONSORT 2010 checklist.&lt;/p&gt;&lt;p class="abstract"&gt;&lt;strong&gt;Methods:&lt;/strong&gt; All issues of 20 highly ranked gastroenterology journals published from Jan 2011 up to November 2017 were searched. Searches were conducted in November 2017. Randomized controlled trials reporting on &lt;em&gt;Helicobacter pylori&lt;/em&gt; associated infections were included in the review.&lt;/p&gt;&lt;p class="abstract"&gt;&lt;strong&gt;Results:&lt;/strong&gt; 21 randomized controlled trials published in gastroenterology journals were included in the study. All included studies adequately reported (100%) on items including description of interventions, outcomes assessed, total number of participants analysed, baseline characteristics and results of outcome assessed. However, items including blinding and mechanism of allocation concealment were reported in only 12 randomized controlled trials (50%).&lt;strong&gt; &lt;/strong&gt;The maximum and minimum scores and percentage of compliance of included randomised controlled trials were 24 (100%) and 15 (62.5%) respectively.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusions: &lt;/strong&gt;The finding of this review suggests that the overall quality of reporting in the included randomized controlled trials was adequate. However, items including trial design, trial registration and protocol and sample size calculations should be reported adequately in the future randomized controlled trials to improve the quality of reporting and replicability of clinical trials.&lt;/p&gt;

Near-Far Matching in R: The nearfar Package.

Estimating the causal treatment effect of an intervention using observational data is difficult due to unmeasured confounders. Many analysts use instrumental variables (IVs) to introduce a randomizing element to observational data analysis, potentially reducing bias created by unobserved confounders. Several persistent problems in the field have served as limitations to IV analyses, particularly the prevalence of "weak" IVs, or instrumental variables that do not effectively randomize individuals to the intervention or control group (leading to biased and unstable treatment effect estimates), as well as IV-based estimates being highly model dependent, requiring parametric adjustment for measured confounders, and often having high mean squared errors in the estimated causal effects. To overcome these problems, the study design method of "near-far matching" has been devised, which "filters" data from a cohort by simultaneously matching individuals within the cohort to be "near" (similar) on measured confounders and "far" (different) on levels of an IV. To facilitate the application of near-far matching to analytical problems, we introduce the R package nearfar and illustrate its application to both a classical example and a simulated dataset. We illustrate how the package can be used to "strengthen" a weak IV by adjusting the "near-ness" and "far-ness" of a match, reduce model dependency, enable nonparametric adjustment for measured confounders, and lower mean squared error in estimated causal effects. We additionally illustrate how to utilize the nearfar package when analyzing either continuous or binary treatments, how to prioritize variables in the match, and how to calculate F statistics of IV strength with or without adjustment for measured confounders.

Methods for time-varying exposure related problems in pharmacoepidemiology: An overview.

PurposeLack of control for time‐varying exposures can lead to substantial bias in estimates of treatment effects. The aim of this study is to provide an overview and guidance on some of the available methodologies used to address problems related to time‐varying exposure and confounding in pharmacoepidemiology and other observational studies. The methods are explored from a conceptual rather than an analytical perspective.MethodsThe methods described in this study have been identified exploring the literature concerning to the time‐varying exposure concept and basing the search on four fundamental pharmacoepidemiological problems, construction of treatment episodes, time‐varying confounders, cumulative exposure and latency, and treatment switching.ResultsA correct treatment episodes construction is fundamental to avoid bias in treatment effect estimates. Several methods exist to address time‐varying covariates, but the complexity of the most advanced approaches—eg, marginal structural models or structural nested failure time models—and the lack of user‐friendly statistical packages have prevented broader adoption of these methods. Consequently, simpler methods are most commonly used, including, for example, methods without any adjustment strategy and models with time‐varying covariates. The magnitude of exposure needs to be considered and properly modelled.ConclusionsFurther research on the application and implementation of the most complex methods is needed. Because different methods can lead to substantial differences in the treatment effect estimates, the application of several methods and comparison of the results is recommended. Treatment episodes estimation and exposure quantification are key parts in the estimation of treatment effects or associations of interest.

The impact of response-guided baseline phase extensions on treatment effect estimates

BackgroundWhen developmental disabilities researchers use multiple-baseline designs they are encouraged to delay the start of an intervention until the baseline stabilizes or until preceding cases have responded to intervention. Using ongoing visual analyses to guide the timing of the start of the intervention can help to resolve potential ambiguities in the graphical display; however, these forms of response-guided experimentation have been criticized as a potential source of bias in treatment effect estimation and inference. Aims and methodsMonte Carlo simulations were used to examine the bias and precision of average treatment effect estimates obtained from multilevel models of four-case multiple-baseline studies with series lengths that varied from 19 to 49 observations per case. We varied the size of the average treatment effect, the factors used to guide intervention decisions (baseline stability, response to intervention, both, or neither), and whether the ongoing analysis was masked or not. ResultsNone of the methods of responding to the data led to appreciable bias in the treatment effect estimates. Furthermore, as timing-of-intervention decisions became responsive to more factors, baselines became longer and treatment effect estimates became more precise. ConclusionsAlthough the study was conducted under limited conditions, the response-guided practices did not lead to substantial bias. By extending baseline phases they reduced estimation error and thus improved the treatment effect estimates obtained from multilevel models.

A DAG-based comparison of interventional effect underestimation between composite endpoint and multi-state analysis in cardiovascular trials

BackgroundComposite endpoints comprising hospital admissions and death are the primary outcome in many cardiovascular clinical trials. For statistical analysis, a Cox proportional hazards model for the time to first event is commonly applied. There is an ongoing debate on whether multiple episodes per individual should be incorporated into the primary analysis. While the advantages in terms of power are readily apparent, potential biases have been mostly overlooked so far.MethodsMotivated by a randomized controlled clinical trial in heart failure patients, we use directed acyclic graphs (DAG) to investigate potential sources of bias in treatment effect estimates, depending on whether only the first or multiple episodes are considered. The biases first are explained in simplified examples and then more thoroughly investigated in simulation studies that mimic realistic patterns.ResultsParticularly the Cox model is prone to potentially severe selection bias and direct effect bias, resulting in underestimation when restricting the analysis to first events. We find that both kinds of bias can simultaneously be reduced by adequately incorporating recurrent events into the analysis model. Correspondingly, we point out appropriate proportional hazards-based multi-state models for decreasing bias and increasing power when analyzing multiple-episode composite endpoints in randomized clinical trials.ConclusionsIncorporating multiple episodes per individual into the primary analysis can reduce the bias of a treatment’s total effect estimate. Our findings will help to move beyond the paradigm of considering first events only for approaches that use more information from the trial and augment interpretability, as has been called for in cardiovascular research.

A comparison of approaches for adjudicating outcomes in clinical trials

BackgroundIncorrect classification of outcomes in clinical trials can lead to biased estimates of treatment effect and reduced power. Ensuring appropriate adjudication methods to minimize outcome misclassification is therefore essential. While there are many reported adjudication approaches, there is little consensus over which approach is best.MethodsUnder the assumption of non-differential assessment (i.e. that misclassification rates are the same in each treatment arm, as would typically be the case when outcome assessors are blinded), we use simulation and theoretical results to address four different questions about outcome adjudication: (a) How many assessors should be used? (b) When is it better to use onsite or central assessment? (c) Should central assessors adjudicate all outcomes, or only suspected events? (d) Should central assessment with multiple assessors be done independently or through group consensus?ResultsNo one adjudication approach performs optimally in all settings. The optimal approach depends on the misclassification rates of site and central assessors, and the correlation between assessors. We found: (a) there will generally be little incremental benefit to using more than three assessors and, for outcomes with very high correlation between assessors, using one assessor is sufficient; (b) when choosing between site and central assessors, the assessor with the smallest misclassification rate should be chosen; when these rates are unknown, a combination of one site assessor and two central assessors will provide good results across a range of scenarios; (c) having central assessors adjudicate only suspected events will typically increase bias, and should be avoided, unless the threshold for sending outcomes for central assessment is extremely low; (d) central assessors can adjudicate either independently or in a group, and the preferred option should be dictated by whichever is expected to have the lowest misclassification rate.ConclusionsOutcome adjudication is of critical importance to ensure validity of trial results, although no one approach is optimal across all settings. Investigators should choose the best strategy based on the specific characteristics of their trial. Regardless of the adjudication strategy chosen, assessors should be qualified and receive appropriate training.

Combining Propensity Score and Random Coefficient Modelling as an Approach to Analyse Complex Longitudinal Data

ABSTRACT&#x0D; ObjectivesWe looked for an approach to analyze/visualize a set of repeated measures of renal laboratory data (eGFR [estimated Glomerular Filtration Rate] from an observational population-based data set) as safety parameters in a longitudinal design and calculate annual changes in different sub-cohorts. Previous meta-analyses had struggled to address this problem (due to poor data quality and strong heterogeneity in underlying historical studies) and previous large population-based observational studies had only looked into binary outcomes.&#x0D; Particular challenges lay (1) in the complexity of the data set with irregularly spaced observation points, (2) in the observational character of the data with associated bias and confounding by indication and co-medication and (3) in the change of lab method during the observation period.&#x0D; ResultsOut of a population base of 400.000 we analysed linked longitudinal data of more than 1000 eligible patients with over 1000.000 prescription records of index drug or co-prescription drugs. Data were provided by the Dundee University Health Informatics Centre (HIC).&#x0D; We addressed the differences in covariates (which typically can lead to biased estimates of treatment effects in observational studies) via individual propensity scores (to reduce this bias by balancing the covariates in the two groups) and a hierarchical modelling approach.&#x0D; A Random Coefficient Model (via proc mixed in SAS 9.3) proved a much more powerful statistical tool than analysis of covariance of the summary measure during follow-up, particularly as the latter approach is less efficient when applied to longitudinal data with missing data points and irregularly spaced repeated measures.Visualization was achieved with the SAS 9.3 GPLOT procedure combined with a spline function. The historical change in lab method was addressed via a conversion of lab results to an internationally recognized standard (IDMS aligned method).&#x0D; We were able to achieve plausible and more precise estimates of the annual decline in eGFR in the patient group of interest than previous attempts from other research publications. This led to a publication in a high profile journal.&#x0D; ConclusionOur approach of combining a Propensity Score and Random Coefficient Modelling was successful to answer a question in drug safety using repeated measurement data from a longitudinal observational population based data set. This approach may be useful for other research questions in Drug Safety or in Comparative Clinical Effectiveness Research for continuous outcome measures.

Assessment of Confounders in Comparative Effectiveness Studies From Secondary Databases.

Secondary clinical databases are an important and growing source of data for comparative effectiveness research (CER) studies. However, measurement of confounders, such as biomarker values or patient-reported health status, in secondary clinical databases may not align with the initiation of a new treatment. In many published CER analyses of registry data, investigators assessed confounders based on the first questionnaire in which the new exposure was recorded. However, it is known that adjustment for confounders measured after the start of exposure can lead to biased treatment effect estimates. In the present study, we conducted simulations to compare assessment strategies for a dynamic clinical confounder in a registry-based comparative effectiveness study of 2 therapies. As expected, we found that adjustment for the confounder value at the time of the first questionnaire after the start of exposure creates a biased estimate the total effect of exposure choice on outcome when the confounder mediates part of the effect. However, adjustment for the prior value can also be badly biased when measured long before exposure initiation. Thus, investigators should carefully consider the timing of confounder measurements relative to exposure initiation and the rate of change in the confounder in order to choose the most relevant measure for each patient.

Heterogeneous populations driven by clinical practice: A case example comparing enzalutamide (ENZA) and bicalutamide (BIC).

e571 Background: Comparative real-world insurance claims data (RWD) analysis can be used to compare homogeneous cohorts (to avoid selection bias). Since the National Comprehensive Cancer Network guidelines recommend ENZA for metastatic castration-resistant prostate cancer (PC) and BIC for localized PC, it was of interest to evaluate these 2 heterogeneous cohorts as the ENZA cohort, being further advanced in the disease, has a worse prognosis than the BIC cohort. Here, we describe the results of a post hoc analysis on the ENZA and BIC cohorts to confirm the heterogeneity of demographics presented in a published study (Behl A et al. J Clin Oncol. 2016;34:235). Methods: The RWD captured enrollees with PC, claims for ENZA or BIC from 09/01/2012–12/31/2014, and at least 6 mo of continuous enrollment prior to the index date. After replicating methods from Behl et al using Truvan, we explored the balance of demographics in baseline periods greater than 6 mo. Differences in demographics and the post hoc analysis were compared. Results: Behl et al reported balanced age between the ENZA (73.1±10.0 y) and BIC (71.4±10.5 y) cohorts. However, 90% (531/592) of the ENZA cohort were metastatic vs 35% (1917/5524) of the BIC cohort. The Quan-Charlson comorbidity index was higher for the ENZA (5.6±2.1 y) than BIC (3.2±2.1 y) cohort. In the ENZA cohort, abiraterone acetate (ABI) or BIC utilization in the baseline period was higher (53%) than in the BIC cohort (0%). When the baseline period was greater than 6 mo, per our post hoc analysis, prior use of ABI or BIC increased to 68% for the ENZA cohort and remained unchanged in the BIC cohort. Conclusions: Despite results from the Behl analysis, the ENZA and BIC cohorts were dissimilar; ENZA patients had more advanced disease (metastatic disease; received ≥ 1 therapy) than BIC patients. Comparison of such heterogeneous populations can lead to biased estimates of treatment effects. Analyses lacking appropriate measures of prognostic variables that guide treatment decisions should be cautiously interpreted. Further analysis is required to identify best practices for using RWD and to address confounding driven by clinical practice.