Covariate Adjustment Methods Research Articles

MD3 RISK ESTIMATION BY BOOSTED DOUBLY ROBUST METHOD

The choice of a covariate adjustment method may affect risk estimates in observational studies. We developed a method, the boosted doubly robust (BDR), that can effectively handle numerous correlated covariates and reduce bias and variance in the effect estimates. The method may be applied to many outcome analyses including costs.

Invited Commentary: Theoretical Considerations and Real-World Challenges for Research on Proxy Exposures and Ovarian Reserve.

In this issue of the Journal, Upson et al. (Am J Epidemiol. 2021;190(1):116-124) assess urinary cadmium level as a potential environmental influence on ovarian reserve, as measured using serum follicle-stimulating hormone, in data from 1,681 US women (1988-1994). They compare 3 methods for modeling urinary proxy exposures-standardization, covariate adjustment, and covariate-adjusted standardization. Observing positive associations with all 3 approaches but higher-magnitude estimates using covariate adjustment as compared with standardization and covariate-adjusted standardization-proposed to be the result of collider-stratification bias-the authors conclude that cadmium may affect ovarian aging, and they recommend careful consideration of modeling approach. Comparisons of methodology in practice using real data are not straightforward, and additional complication arises from using a proxy outcome-serum follicle-stimulating hormone level to represent diminished ovarian reserve. In this commentary, I describe the theoretical basis for approaches for modeling urinary proxy exposures; consider potential explanations for why the approaches may yield different results in practice and describe why measurement error may play a larger role than collider-stratification bias; discuss challenges related to studies of ovarian reserve; and emphasize the importance of addressing both theoretical concerns and real-world challenges in methodological research and epidemiologic studies of ovarian reserve.

Evaluation of the Diabetes Screening Component of a National Cardiovascular Risk Assessment Programme in England: a Retrospective Cohort Study

Type 2 Diabetes (T2D) is increasing but the effectiveness of large-scale diabetes screening programmes is debated. We assessed associations between coverage of a national cardiovascular and diabetes risk assessment programme in England (NHS Health Check) and detection and management of incident cases of non-diabetic hyperglycaemia (NDH) and T2D. Retrospective analysis employing propensity score covariate adjustment method of prospectively collected data of 348,987 individuals aged 40–74 years and registered with 455 general practices in England (January 2009-May 2016). We examined differences in diagnosis of NDH and T2D, and changes in blood glucose levels and cardiovascular risk score between individuals registered with general practices with different levels (tertiles) of programme coverage. Over the study period 7,126 cases of NDH and 12,171 cases of T2D were detected. Compared with low coverage practices, incidence rate of detection in medium and high coverage practices were 15% and 19% higher for NDH and 10% and 9% higher for T2D, respectively. Individuals with NDH in high coverage practices had 0.2 mmol/L lower mean fasting plasma glucose and 0.9% lower cardiovascular risk score at follow-up. General practices actively participating in the programme had higher detection of NDH and T2D and improved management of blood glucose and cardiovascular risk factors.

Identifying Risk of Depressive Symptoms in Adults With Physical Disabilities Receiving Rehabilitation Services: Propensity Score Approaches.

ObjectiveTo compare the risk of depressive symptoms in adults with physical disabilities between rehabilitation-receivers and non-receivers.MethodsA total of 3,568 adults with physical disabilities were retrieved from the 2014 Korean National Survey on People with Disabilities database. Four covariate adjustment methods (a multivariable regression model, inverse probability of treatment weighting [IPTW] adjusted for normalized weight, IPTW with stabilized weight, and greedy algorithm with 1:1 propensity score matching) were used to estimate the odds of having depressive symptoms. The dependent variable was depressive symptoms and the independent variable was the use of rehabilitation services. Baseline covariates were 19 demographic variables and 10 chronic condition variables.ResultsThe four covariate adjustment methods revealed that adults with physical disabilities receiving rehabilitation services had a higher risk of depressive symptoms than those who did not receive these services (adjusted odds ratio, 1.191–1.294).ConclusionOur findings suggest that adults with physical disabilities receiving rehabilitation services have higher risk of developing depressive symptoms. Therefore, rehabilitation professionals need to pay attention to depressive symptoms and establish therapeutic strategies that can reduce such risk in rehabilitation settings.

Controlling for confounding via propensity score methods can result in biased estimation of the conditional AUC: A simulation study.

In the medical literature, there has been an increased interest in evaluating association between exposure and outcomes using nonrandomized observational studies. However, because assignments to exposure are not random in observational studies, comparisons of outcomes between exposed and nonexposed subjects must account for the effect of confounders. Propensity score methods have been widely used to control for confounding, when estimating exposure effect. Previous studies have shown that conditioning on the propensity score results in biased estimation of conditional odds ratio and hazard ratio. However, research is lacking on the performance of propensity score methods for covariate adjustment when estimating the area under the ROC curve (AUC). In this paper, AUC is proposed as measure of effect when outcomes are continuous. The AUC is interpreted as the probability that a randomly selected nonexposed subject has a better response than a randomly selected exposed subject. A series of simulations has been conducted to examine the performance of propensity score methods when association between exposure and outcomes is quantified by AUC; this includes determining the optimal choice of variables for the propensity score models. Additionally, the propensity score approach is compared with that of the conventional regression approach to adjust for covariates with the AUC. The choice of the best estimator depends on bias, relative bias, and root mean squared error. Finally, an example looking at the relationship of depression/anxiety and pain intensity in people with sickle cell disease is used to illustrate the estimation of the adjusted AUC using the proposed approaches.

Effects of HIV viremia on the gastrointestinal microbiome of young MSM.

We employed a high-dimensional covariate adjustment method in microbiome analysis to better control for behavioural and clinical confounders, and in doing so examine the effects of HIV on the rectal microbiome. Three hundred and eighty-three MSM were grouped into four HIV viremia categories: HIV negative (n = 200), HIV-positive undetectable (HIV RNA < 20 copies/ml; n = 66), HIV-positive suppressed (RNA 20-200 copies/ml; n = 72) and HIV-positive viremic (RNA > 200 copies/ml; n = 45). We performed 16S rRNA gene sequencing on rectal swab samples and used inverse probability of treatment-weighted marginal structural models to examine differences in microbial composition by HIV viremia category. HIV viremia explained a significant amount of variability in microbial composition in both unadjusted and covariate-adjusted analyses (R = 0.011, P = 0.02). Alterations in bacterial taxa were more apparent with increasing viremia. Relative to the HIV-negative group, HIV-positive undetectable participants showed depletions in Brachyspira, Campylobacter and Parasutterella, while suppressed participants demonstrated depletions in Barnesiella, Brachyspira and Helicobacter. The microbial signature of viremic men was most distinct, showing enrichment in inflammatory genera Peptoniphilus, Porphyromonas and Prevotella and depletion of Bacteroides, Brachyspira and Faecalibacterium, among others. Our study shows that, after accounting for the influence of multiple confounding factors, HIV is associated with dysbiosis in the gastrointestinal microbiome in a dose-dependent manner. This analytic approach may allow for better identification of true microbial associations by limiting the effects of confounding, and thus improve comparability across future studies.

On Binscatter

\textit{Binscatter}, or a binned scatter plot, is a very popular tool in applied microeconomics. It provides a flexible, yet parsimonious way of visualizing and summarizing mean, quantile, and other nonparametric regression functions in large data sets. It is also often used for informal evaluation of substantive hypotheses such as linearity or monotonicity of the unknown function. This paper presents a foundational econometric analysis of binscatter, offering an array of theoretical and practical results that aid both understanding current practices (i.e., their validity or lack thereof) as well as guiding future applications. In particular, we highlight important methodological problems related to covariate adjustment methods used in current practice, and provide a simple, valid approach. Our results include a principled choice for the number of bins, confidence intervals and bands, hypothesis tests for parametric and shape restrictions for mean, quantile, and other functions of interest, among other new methods, all applicable to canonical binscatter as well as to nonlinear, higher-order polynomial, smoothness-restricted and covariate-adjusted extensions thereof. Companion general-purpose software packages for \texttt{Python}, \texttt{R}, and \texttt{Stata} are provided. From a technical perspective, we present novel theoretical results for possibly nonlinear semi-parametric partitioning-based series estimation with random partitions that are of independent interest.

Are Rehabilitation Services Effective for Depressive Symptoms? Propensity Score Matching Approaches

Abstract Date Presented 4/19/2018 We examined people with physical disabilities who received rehabilitation to see whether they had a higher risk of depressive symptoms than those who did not receive rehabilitation using four covariate adjustment methods. Our results revealed that rehabilitation receivers had a higher risk of depressive symptoms than nonreceivers. Primary Author and Speaker: Yejin Lee Additional Authors and Speakers: Ickpyo Hong, Mi Jung Lee, Hae Yean Park

The LOOP Estimator: Adjusting for Covariates in Randomized Experiments.

When conducting a randomized controlled trial, it is common to specify in advance the statistical analyses that will be used to analyze the data. Typically, these analyses will involve adjusting for small imbalances in baseline covariates. However, this poses a dilemma, as adjusting for too many covariates can hurt precision more than it helps, and it is often unclear which covariates are predictive of outcome prior to conducting the experiment. This article aims to produce a covariate adjustment method that allows for automatic variable selection, so that practitioners need not commit to any specific set of covariates prior to seeing the data. In this article, we propose the "leave-one-out potential outcomes" estimator. We leave out each observation and then impute that observation's treatment and control potential outcomes using a prediction algorithm such as a random forest. In addition to allowing for automatic variable selection, this estimator is unbiased under the Neyman-Rubin model, generally performs at least as well as the unadjusted estimator, and the experimental randomization largely justifies the statistical assumptions made.

TREATMENT OF STAGE 1, LOW RISK HYPERTENSION IS ASSOCIATED WITH REDUCTION IN CARDIOVASCULAR MORTALITY

Objective: There is lack of evidence for the benefit of treatment in uncomplicated, low risk grade I hypertension. As such, some of the major guidelines recommend treatment for grade I hypertensives who have underlying cardiovascular disease, or are at high risk. Design and method: From National Health Insurance Service (NHIS) Health Examination Database, subjects with grade I hypertension between 2005 and 2006 were selected and followed-up until December, 2015. The subjects had a SBP of 140–159 mmHg and/or DBP of 90–99 mmHg and were not undergoing treatment at baseline. Subjects were grouped into controlled (<140/90 mmHg; n = 99,301) and uncontrolled group (> = 140/90 mmHg; n = 49,460) according to mean of the BP recorded during follow-up health examination. All-cause death and cardiovascular outcomes including myocardial infarction (MI), ischemic stroke, hemorrhagic stroke, and end stage renal disease (ESRD) were examined by using Cox proportional hazard models with covariate adjustment method using the propensity scores. Results: Median follow-up duration was 124 months (IQR: 116–131). Compared to subjects with uncontrolled BP, controlled group had significantly lower risk of all-cause death (HR, 0.50; 95% CI, 0.48–0.52; p < 0.0001). For non-fatal events, controlled BP was associated with the lower risk of all stroke (HR, 0.88; 95% CI, 0.82–0.93; p < 0.0001), hemorrhagic stroke (HR, 0.75; 95% CI, 0.66–0.85; p < 0.0001), ischemic stroke (HR, 0.91; 95% CI, 0.84–0.98; p = 0.0083), and ESRD (HR, 0.42; 95% CI, 0.30–0.59; p < 0.0001). There was no significant difference between the two groups for non-fatal MI. Importantly, the benefit was evident in young aged hypertensives below the age of 50. The optimal level of BP associated with the lowest risk of all-cause mortality was 120 to < 130 mmHg for SBP and 70 to < 80 mmHg for DBP. However, there was increased risk of MI for subjects with SBP of < 120 mmHg and DBP of 70 to < 80 mmHg. Conclusions: In a large cohort of treated low risk, grade I hypertensive subjects, BP below 140/90 mmHg was associated with significant reduction in the risk of mortality, stroke and ESRD, with the lowest risk of mortality in the range of 120 to < 130 mmHg and 70 to < 80 mmHg.

OP68 Methods For The Economic Evaluation Of Precision Medicine

Introduction:Methods that accommodate heterogeneity in outcomes are not widely used in economic evaluation. With the growth of precision medicine (PM), where choice of treatment is informed by the molecular characteristics of the patient or disease, we expect to see greater heterogeneity in effectiveness and cost of interventions. Our objective was to compare analytical frameworks for valuing heterogeneity in economic evaluation, and consider their strengths and weaknesses for applications in PM.Methods:We conducted a literature review to identify papers that proposed an analytical framework for economic evaluation of a health intervention, and that placed a value on heterogeneous effects. We compared the frameworks considering the purpose of the analysis, including where in the product lifecycle the framework could be used, the types of PM interventions where the framework could be applied, and its ability to address methodological challenges of evaluating PM.Results:Five analytical frameworks were identified: covariate adjustment methods, value of stratification, value of heterogeneity (VoH), expected value of individualized care (EVIC), and loss with respect to efficient diffusion (LED) metrics. Each framework addresses a slightly different research question, and is suited to different settings and interventions. With the exception of covariate adjustment, all focus on maximizing net benefit within certain constraints and quantify the opportunity cost of ignoring heterogeneity. Only VoH considers the relationship between heterogeneity and uncertainty, and no framework explicitly includes the cost or uncertainty associated with identifying subgroups.Conclusions:The ability to value heterogeneity is a critical component of economic evaluations of PM. The choice of an appropriate analytical framework will help strengthen the quality of economic evidence available to support health technology assessment of PM technologies, informing PM adoption decisions, and supporting efficient allocation of health care resources.

Pomalidomide with Dexamethasone for Treating Relapsed and Refractory Multiple Myeloma Previously Treated with Lenalidomide and Bortezomib: An Evidence Review Group Perspective of an NICE Single Technology Appraisal.

The National Institute for Health and Care Excellence (NICE), as part of the institute’s single technology appraisal (STA) process, invited the manufacturer of pomalidomide (POM; Imnovid®, Celgene) to submit evidence regarding the clinical and cost effectiveness of the drug in combination with dexamethasone (POM + LoDEX) for the treatment of relapsed and refractory multiple myeloma (RRMM) after at least two regimens including lenalidomide (LEN) and bortezomib (BOR). Kleijnen Systematic Reviews Ltd (KSR) and Erasmus University Rotterdam were commissioned as the Evidence Review Group (ERG) for this submission. The ERG reviewed the evidence submitted by the manufacturer, validated the manufacturer’s decision analytic model, and conducted exploratory analyses in order to assess the robustness and validity of the presented clinical and cost-effectiveness results. This paper describes the company submission, the ERG assessment, and NICE’s subsequent decisions. The company conducted a systematic review to identify studies comparing POM with comparators outlined in the NICE scope: panobinostat with bortezomib and dexamethasone (PANO + BOR + DEX), bendamustine with thalidomide and dexamethasone (BTD) and conventional chemotherapy (CC). The main clinical effectiveness evidence was obtained from MM-003, a randomized controlled trial (RCT) comparing POM + LoDEX with high-dose dexamethasone (HiDEX; used as a proxy for CC). Additional data from other studies were also used as nonrandomized observational data sources for the indirect treatment comparison of POM + LoDEX with BTD and PANO + BOR + DEX. Covariate or treatment switching adjustment methods were used for each comparison. The model developed in Microsoft® Excel 2010 using a semi-Markov partitioned survival structure, submitted in the original submission to NICE for TA338, was adapted for the present assessment of the cost effectiveness of POM + LoDEX. Updated evidence from the clinical-effectiveness part was used for the survival modelling of progression-free survival and overall survival. For POM + LoDEX, the patient access scheme (PAS) discount was applied to the POM price. Three separate comparisons were conducted for each comparator, each comparison using a different dataset and adjustment methods. The ERG identified and corrected some errors, and the corrected incremental cost-effectiveness ratios (ICERs) for POM + LoDEX versus each comparator were presented: approximately £45,000 per quality-adjusted life-year (QALY) gained versus BTD, savings of approximately £143,000 per QALY lost versus PANO + BOR + DEX, and approximately £49,000 per QALY gained versus CC. The ERG also conducted full incremental analyses, which revealed that CC, POM + LoDEX and PANO + BOR + DEX were on the cost-effectiveness frontier. The committee’s decision on the technology under analysis deemed that POM + LoDEX should be recommended as an option for treating multiple myeloma in adults at third or subsequent relapse of treatments including both LEN and BOR, contingent on the company providing POM with the discount agreed in the PAS.

Methodological approaches in analysing observational data: A practical example on how to address clustering and selection bias

BackgroundBecause not every scientific question on effectiveness can be answered with randomised controlled trials, research methods that minimise bias in observational studies are required. Two major concerns influence the internal validity of effect estimates: selection bias and clustering. Hence, to reduce the bias of the effect estimates, more sophisticated statistical methods are needed. AimTo introduce statistical approaches such as propensity score matching and mixed models into representative real-world analysis and to conduct the implementation in statistical software R to reproduce the results. Additionally, the implementation in R is presented to allow the results to be reproduced. MethodWe perform a two-level analytic strategy to address the problems of bias and clustering: (i) generalised models with different abilities to adjust for dependencies are used to analyse binary data and (ii) the genetic matching and covariate adjustment methods are used to adjust for selection bias. Hence, we analyse the data from two population samples, the sample produced by the matching method and the full sample. ResultsThe different analysis methods in this article present different results but still point in the same direction. In our example, the estimate of the probability of receiving a case conference is higher in the treatment group than in the control group. Both strategies, genetic matching and covariate adjustment, have their limitations but complement each other to provide the whole picture. ConclusionThe statistical approaches were feasible for reducing bias but were nevertheless limited by the sample used. For each study and obtained sample, the pros and cons of the different methods have to be weighted.

Logistic Regression Likelihood Ratio Test Analysis for Detecting Signals of Adverse Events in Post-market Safety Surveillance

ABSTRACTThe Vaccine Adverse Event Reporting System (VAERS) and other product surveillance systems compile reports of product-associated adverse events (AEs), and these reports may include a wide range of information including age, gender, and concomitant vaccines. Controlling for possible confounding variables such as these is an important task when utilizing surveillance systems to monitor post-market product safety. A common method for handling possible confounders is to compare observed product–AE combinations with adjusted baseline frequencies where the adjustments are made by stratifying on observable characteristics. Though approaches such as these have proven to be useful, in this article we propose a more flexible logistic regression approach which allows for covariates of all types rather than relying solely on stratification. Indeed, a main advantage of our approach is that the general regression framework provides flexibility to incorporate additional information such as demographic factors and concomitant vaccines. As part of our covariate-adjusted method, we outline a procedure for signal detection that accounts for multiple comparisons and controls the overall Type 1 error rate. To demonstrate the effectiveness of our approach, we illustrate our method with an example involving febrile convulsion, and we further evaluate its performance in a series of simulation studies.

Combining Double Sampling and Bounds to Address Nonignorable Missing Outcomes in Randomized Experiments

Missing outcome data plague many randomized experiments. Common solutions rely on ignorability assumptions that may not be credible in all applications. We propose a method for confronting missing outcome data that makes fairly weak assumptions but can still yield informative bounds on the average treatment effect. Our approach is based on a combination of the double sampling design and nonparametric worst-case bounds. We derive a worst-case bounds estimator under double sampling and provide analytic expressions for variance estimators and confidence intervals. We also propose a method for covariate adjustment using poststratification and a sensitivity analysis for nonignorable missingness. Finally, we illustrate the utility of our approach using Monte Carlo simulations and a placebo-controlled randomized field experiment on the effects of persuasion on social attitudes with survey-based outcome measures.

A Simulation-Based Comparison of Covariate Adjustment Methods for the Analysis of Randomized Controlled Trials.

Covariate adjustment methods are frequently used when baseline covariate information is available for randomized controlled trials. Using a simulation study, we compared the analysis of covariance (ANCOVA) with three nonparametric covariate adjustment methods with respect to point and interval estimation for the difference between means. The three alternative methods were based on important members of the generalized empirical likelihood (GEL) family, specifically on the empirical likelihood (EL) method, the exponential tilting (ET) method, and the continuous updated estimator (CUE) method. Two criteria were considered for the comparison of the four statistical methods: the root mean squared error and the empirical coverage of the nominal 95% confidence intervals for the difference between means. Based on the results of the simulation study, for sensitivity analysis purposes, we recommend the use of ANCOVA (with robust standard errors when heteroscedasticity is present) together with the CUE-based covariate adjustment method.

Covariate-adjusted measures of discrimination for survival data.

MotivationDiscrimination statistics describe the ability of a survival model to assign higher risks to individuals who experience earlier events: examples are Harrell’s C-index and Royston and Sauerbrei’s D, which we call the D-index. Prognostic covariates whose distributions are controlled by the study design (e.g. age and sex) influence discrimination and can make it difficult to compare model discrimination between studies. Although covariate adjustment is a standard procedure for quantifying disease-risk factor associations, there are no covariate adjustment methods for discrimination statistics in censored survival data.ObjectiveTo develop extensions of the C-index and D-index that describe the prognostic ability of a model adjusted for one or more covariate(s).MethodWe define a covariate-adjusted C-index and D-index for censored survival data, propose several estimators, and investigate their performance in simulation studies and in data from a large individual participant data meta-analysis, the Emerging Risk Factors Collaboration.ResultsThe proposed methods perform well in simulations. In the Emerging Risk Factors Collaboration data, the age-adjusted C-index and D-index were substantially smaller than unadjusted values. The study-specific standard deviation of baseline age was strongly associated with the unadjusted C-index and D-index but not significantly associated with the age-adjusted indices.ConclusionsThe proposed estimators improve meta-analysis comparisons, are easy to implement and give a more meaningful clinical interpretation.

Inferring model‐based probability of occurrence from preferentially sampled data with uncertain absences using expert knowledge

Summary For several data bases in species distribution modelling (SDM), presences are known but absences are uncertain and can be non‐detections. The environment influences both the non‐detections and the species occurrences (preferential sampling), affecting model predictions. The objective is to estimate model‐based probability of occurrence when absences are uncertain and sampling is preferential. Our Bayesian image restoration (BIR) model has an environmental, spatial and non‐detection component. The latter is based on the number of ‘control species’ (selected by expert knowledge to have wide geographic range and occupy several habitats) observed. Control species are assumed present in all grid cells so an observed absence of the control species is a non‐detection. The observed absence of a focal species is likely a non‐detection when the environment is conducive, the focal species was observed in the neighbourhood, and the number of control species is small. Simulating an artificial species (true prevalence 0·31) and preferential sampling (observed prevalence 0·18), the BIR modelled prevalence was 0·32 (0·30–0·34). The estimation was robust to failure of assumptions with a mis‐specified environmental model. The restored map was close to the true map, and the true realized niche of the species was inferred (with uncertainty). The link between the non‐detection and control species is crucial, and a sensitivity analysis is recommended. BIR restored the simulated atlas map more accurately than a covariate adjustment method. When applied to the vascular plant Asarum europaeum L., BIR changed the interpretation. The uncertainty maps assisted decision‐making. A BIR implementation new to the SDM field is presented. BIR using control species is most fruitfully applied at coarse scale and intermediate extent, with limited environmental extremes and sufficient expert knowledge (e.g. many Central European countries). BIR is a framework, and we studied one implementation. Any proxy (road density, distance to city, direct expert knowledge, detectability traits, etc.) associated with non‐detection can be used so BIR can be applied in diverse settings. Including non‐detections reduces model indeterminacy but often requires expert knowledge. However, the benefits of using a priori knowledge and Bayesian estimation are expected to outweigh the risks of model‐based inference.

Empirical comparison of four baseline covariate adjustment methods in analysis of continuous outcomes in randomized controlled trials.

BackgroundAlthough seemingly straightforward, the statistical comparison of a continuous variable in a randomized controlled trial that has both a pre- and posttreatment score presents an interesting challenge for trialists. We present here empirical application of four statistical methods (posttreatment scores with analysis of variance, analysis of covariance, change in scores, and percent change in scores), using data from a randomized controlled trial of postoperative pain in patients following total joint arthroplasty (the Morphine COnsumption in Joint Replacement Patients, With and Without GaBapentin Treatment, a RandomIzed ControlLEd Study [MOBILE] trials).MethodsAnalysis of covariance (ANCOVA) was used to adjust for baseline measures and to provide an unbiased estimate of the mean group difference of the 1-year postoperative knee flexion scores in knee arthroplasty patients. Robustness tests were done by comparing ANCOVA with three comparative methods: the posttreatment scores, change in scores, and percentage change from baseline.ResultsAll four methods showed similar direction of effect; however, ANCOVA (−3.9; 95% confidence interval [CI]: −9.5, 1.6; P=0.15) and the posttreatment score (−4.3; 95% CI: −9.8, 1.2; P=0.12) method provided the highest precision of estimate compared with the change score (−3.0; 95% CI: −9.9, 3.8; P=0.38) and percent change (−0.019; 95% CI: −0.087, 0.050; P=0.58).ConclusionANCOVA, through both simulation and empirical studies, provides the best statistical estimation for analyzing continuous outcomes requiring covariate adjustment. Our empirical findings support the use of ANCOVA as an optimal method in both design and analysis of trials with a continuous primary outcome.

A Note on Partial Covariate-Adjustment and Design Considerations in Noninferiority Trials When Patient-Level Data are not Available

The traditional fixed margin approach to evaluating an experimental treatment through an active-controlled noninferiority trial is simple and straightforward. However, its utility relies heavily on the constancy assumption of the experimental data. The recently developed covariate-adjustment method permits more flexibility and improved discriminatory capacity compared to the fixed margin approach. However, one major limitation of this covariate-adjustment methodology is its adherence on the patient-level data, which may not be accessible to investigators in practice. In this article, under some assumptions, we examine the feasibility of a partial covariate-adjustment approach based on data typically available from journal publications or other public data when the patient-level data are unavailable. We illustrate the usefulness of this approach through two real examples. We also provide design considerations on the efficiency of the partial covariate-adjustment approach.