Dataset Information

How robust are findings of pairwise and network meta-analysis in the presence of missing participant outcome data?

ABSTRACT:

Background

To investigate the prevalence of robust conclusions in systematic reviews addressing missing (participant) outcome data via a novel framework of sensitivity analyses and examine the agreement with the current sensitivity analysis standards.

Methods

We performed an empirical study on systematic reviews with two or more interventions. Pairwise meta-analyses (PMA) and network meta-analyses (NMA) were identified from empirical studies on the reporting and handling of missing outcome data in systematic reviews. PMAs with at least three studies and NMAs with at least three interventions on one primary outcome were considered eligible. We applied Bayesian methods to obtain the summary effect estimates whilst modelling missing outcome data under the missing-at-random assumption and different assumptions about the missingness mechanism in the compared interventions. The odds ratio in the logarithmic scale was considered for the binary outcomes and the standardised mean difference for the continuous outcomes. We calculated the proportion of primary analyses with robust and frail conclusions, quantified by our proposed metric, the robustness index (RI), and current sensitivity analysis standards. Cohen's kappa statistic was used to measure the agreement between the conclusions derived by the RI and the current sensitivity analysis standards.

Results

One hundred eight PMAs and 34 NMAs were considered. When studies with a substantial number of missing outcome data dominated the analyses, the number of frail conclusions increased. The RI indicated that 59% of the analyses failed to demonstrate robustness compared to 39% when the current sensitivity analysis standards were employed. Comparing the RI with the current sensitivity analysis standards revealed that two in five analyses yielded contradictory conclusions concerning the robustness of the primary analysis results.

Conclusions

Compared with the current sensitivity analysis standards, the RI offers an explicit definition of similar results and does not unduly rely on statistical significance. Hence, it may safeguard against possible spurious conclusions regarding the robustness of the primary analysis results.

SUBMITTER: Spineli LM

PROVIDER: S-EPMC8691029 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Study publication bias and outcome reporting bias have been recognised as two threats to the validity of systematic reviews. The purpose of this research was to estimate the proportion of missing participant outcome data from randomised controlled trials (RCTs) due to lack of publication of whole studies and due to outcome data missing within study publications.Data were extracted from protocols of clinical research projects submitted to the research ethics committee of the University of Freiburg (Germany) between 2000 and 2002 and associated fully published articles. The total amount of published and unpublished outcome data from all trial participants was calculated for each trial and the overall proportion of missing data from both unpublished and published trials computed. Full and partially reported outcome data was also taken into consideration. The impact of funding source on missingness was also considered at the trial level. From 308 parallel group trials in the study cohort, 167 were published and 141 were unpublished. Overall, 260,563 participants contributed to a total of 2,618,116 participant outcome data across all trials. About half (47%) of the participant outcome data from the 308 trials was reported in full but at least 81% were partially reported. Of the 19% of participant data that were missing, 4% was attributable to missing data from published trials and 15% from unpublished trials. Commercially funded trials had a higher probability of publication (relative risk 1.20, 95% confidence interval 0.86, 1.67; p = 0.27) but were less likely to fully report all outcomes than non-commercially funded trials (relative risk 0.64, 95% confidence interval 0.30, 1.38; p = 0.26).Missing participant outcome data from both published and unpublished trials is frequent. Clinical trial registration including outcome information not only identifies that clinical trials exist but the systematic examination and monitoring of trial information within a registry can help detect selective reporting of entire studies and of outcome data within studies and possibly prevent it.

Project description:BackgroundMissing participant outcome data (MOD) are ubiquitous in systematic reviews with network meta-analysis (NMA) as they invade from the inclusion of clinical trials with reported participant losses. There are available strategies to address aggregate MOD, and in particular binary MOD, while considering the missing at random (MAR) assumption as a starting point. Little is known about their performance though regarding the meta-analytic parameters of a random-effects model for aggregate binary outcome data as obtained from trial-reports (i.e. the number of events and number of MOD out of the total randomised per arm).MethodsWe used four strategies to handle binary MOD under MAR and we classified these strategies to those modelling versus excluding/imputing MOD and to those accounting for versus ignoring uncertainty about MAR. We investigated the performance of these strategies in terms of core NMA estimates by performing both an empirical and simulation study using random-effects NMA based on electrical network theory. We used Bland-Altman plots to illustrate the agreement between the compared strategies, and we considered the mean bias, coverage probability and width of the confidence interval to be the frequentist measures of performance.ResultsModelling MOD under MAR agreed with exclusion and imputation under MAR in terms of estimated log odds ratios and inconsistency factor, whereas accountability or not of the uncertainty regarding MOD affected intervention hierarchy and precision around the NMA estimates: strategies that ignore uncertainty about MOD led to more precise NMA estimates, and increased between-trial variance. All strategies showed good performance for low MOD (<5%), consistent evidence and low between-trial variance, whereas performance was compromised for large informative MOD (>?20%), inconsistent evidence and substantial between-trial variance, especially for strategies that ignore uncertainty due to MOD.ConclusionsThe analysts should avoid applying strategies that manipulate MOD before analysis (i.e. exclusion and imputation) as they implicate the inferences negatively. Modelling MOD, on the other hand, via a pattern-mixture model to propagate the uncertainty about MAR assumption constitutes both conceptually and statistically proper strategy to address MOD in a systematic review.

Project description:BackgroundA number of strategies have been proposed to handle missing binary outcome data (MOD) in systematic reviews. However, none of these have been evaluated empirically in a series of published systematic reviews.MethodsUsing published systematic reviews with network meta-analysis (NMA) from a wide range of health-related fields, we evaluated comparatively the most frequently described Bayesian modelling strategies for MOD in terms of log odds ratio (log OR), between-trial variance, inconsistency factor (i.e. difference between direct and indirect estimates for a comparison), surface under the cumulative ranking (SUCRA) and rankings. We extended the Bayesian random-effects NMA model to incorporate the informative missingness odds ratio (IMOR) parameter, and applied the node-splitting approach to investigate inconsistency locally. We considered both pattern-mixture and selection models, different structures for prior distribution of log IMOR, and different scenarios for MOD. To illustrate level of agreement between different strategies and scenarios, we used Bland-Altman plots.ResultsAddressing MOD using extreme scenarios and ignoring the uncertainty about the scenarios led to systematically different and more precise log ORs compared to modelling MOD under the missing at random (MAR) assumption. Hierarchical structure of log IMORs led to lower between-trial variance, especially in the case of substantial MOD. Assuming common-within-network or trial-specific log IMORs yielded similar posterior results for all NMA estimates, whereas intervention-specific structure systematically inflated uncertainty around log ORs and SUCRAs. Pattern-mixture model agreed with selection model, particularly under the trial-specific structure; however, selection model systematically reduced precision around log IMORs. Overall, different strategies and scenarios mostly had good agreement in the case of low MOD.ConclusionsAddressing MOD using extreme scenarios and/or ignoring the uncertainty about the scenarios may negatively affect NMA estimates. Modelling MOD via the IMOR parameter can ensure bias-adjusted estimates and offer valuable insights into missingness mechanisms. The researcher should seek an expert opinion in order to decide on the structure of log IMOR that best aligns to the condition and interventions studied and to define a proper prior distribution for log IMOR. Our findings also apply to pairwise meta-analyses.

Project description:BackgroundTrials with binary outcomes can be synthesised using within-trial exact likelihood or approximate normal likelihood in one-stage or two-stage approaches, respectively. The performance of the one-stage and the two-stage approaches has been documented extensively in the literature. However, little is known about how these approaches behave in the presence of missing outcome data (MOD), which are ubiquitous in clinical trials. In this work, we compare the one-stage versus two-stage approach via a pattern-mixture model in the network meta-analysis using Bayesian methods to handle MOD appropriately.MethodsWe used 29 published networks to empirically compare the two approaches concerning the relative treatment effects of several competing interventions and the between-trial variance (?2), while considering the extent and level of balance of MOD in the included trials. We additionally conducted a simulation study to compare the competing approaches regarding the bias and width of the 95% credible interval of the (summary) log odds ratios (OR) and ?2 in the presence of moderate and large MOD.ResultsThe empirical study did not reveal any systematic bias between the compared approaches regarding the log OR, but showed systematically larger uncertainty around the log OR under the one-stage approach for networks with at least one small trial or low event risk and moderate MOD. For these networks, the simulation study revealed that the bias in log OR for comparisons with the reference intervention in the network was relatively higher in the two-stage approach. Contrariwise, the bias in log OR for the remaining comparisons was relatively higher in the one-stage approach. Overall, bias increased for large MOD. For these networks, the empirical results revealed slightly higher ?2 estimates under the one-stage approach irrespective of the extent of MOD. The one-stage approach also led to less precise log OR and ?2 when compared with the two-stage approach for large MOD.ConclusionsDue to considerable bias in the log ORs overall, especially for large MOD, none of the competing approaches was superior. Until a more competent model is developed, the researchers may prefer the one-stage approach to handle MOD, while acknowledging its limitations.

Project description:BackgroundIn non-randomized studies (NRSs) where a continuous outcome variable (e.g., depressive symptoms) is assessed at baseline and follow-up, it is common to observe imbalance of the baseline values between the treatment/exposure group and control group. This may bias the study and consequently a meta-analysis (MA) estimate. These estimates may differ across statistical methods used to deal with this issue. Analysis of individual participant data (IPD) allows standardization of methods across studies. We aimed to identify methods used in published IPD-MAs of NRSs for continuous outcomes, and to compare different methods to account for baseline values of outcome variables in IPD-MA of NRSs using two empirical examples from the Thyroid Studies Collaboration (TSC).MethodsFor the first aim we systematically searched in MEDLINE, EMBASE, and Cochrane from inception to February 2021 to identify published IPD-MAs of NRSs that adjusted for baseline outcome measures in the analysis of continuous outcomes. For the second aim, we applied analysis of covariance (ANCOVA), change score, propensity score and the naïve approach (ignores the baseline outcome data) in IPD-MA from NRSs on the association between subclinical hyperthyroidism and depressive symptoms and renal function. We estimated the study and meta-analytic mean difference (MD) and relative standard error (SE). We used both fixed- and random-effects MA.ResultsTen of 18 (56%) of the included studies used the change score method, seven (39%) studies used ANCOVA and one the propensity score (5%). The study estimates were similar across the methods in studies in which groups were balanced at baseline with regard to outcome variables but differed in studies with baseline imbalance. In our empirical examples, ANCOVA and change score showed study results on the same direction, not the propensity score. In our applications, ANCOVA provided more precise estimates, both at study and meta-analytical level, in comparison to other methods. Heterogeneity was higher when change score was used as outcome, moderate for ANCOVA and null with the propensity score.ConclusionANCOVA provided the most precise estimates at both study and meta-analytic level and thus seems preferable in the meta-analysis of IPD from non-randomized studies. For the studies that were well-balanced between groups, change score, and ANCOVA performed similarly.

Project description:A fundamental aspect of epidemiological studies concerns the estimation of factor-outcome associations to identify risk factors, prognostic factors and potential causal factors. Because reliable estimates for these associations are important, there is a growing interest in methods for combining the results from multiple studies in individual participant data meta-analyses (IPD-MA). When there is substantial heterogeneity across studies, various random-effects meta-analysis models are possible that employ a one-stage or two-stage method. These are generally thought to produce similar results, but empirical comparisons are few.We describe and compare several one- and two-stage random-effects IPD-MA methods for estimating factor-outcome associations from multiple risk-factor or predictor finding studies with a binary outcome. One-stage methods use the IPD of each study and meta-analyse using the exact binomial distribution, whereas two-stage methods reduce evidence to the aggregated level (e.g. odds ratios) and then meta-analyse assuming approximate normality. We compare the methods in an empirical dataset for unadjusted and adjusted risk-factor estimates.Though often similar, on occasion the one-stage and two-stage methods provide different parameter estimates and different conclusions. For example, the effect of erythema and its statistical significance was different for a one-stage (OR?=?1.35, [Formula: see text]) and univariate two-stage (OR?=?1.55, [Formula: see text]). Estimation issues can also arise: two-stage models suffer unstable estimates when zero cell counts occur and one-stage models do not always converge.When planning an IPD-MA, the choice and implementation (e.g. univariate or multivariate) of a one-stage or two-stage method should be prespecified in the protocol as occasionally they lead to different conclusions about which factors are associated with outcome. Though both approaches can suffer from estimation challenges, we recommend employing the one-stage method, as it uses a more exact statistical approach and accounts for parameter correlation.

Project description:ObjectiveProvider profiling of outcome performance has become increasingly common in pay-for-performance programs. For chronic conditions, a substantial proportion of patients eligible for outcome measures may be lost to follow-up, potentially compromising outcome profiling. In the context of primary care depression treatment, we assess the implications of missing data for the accuracy of alternative approaches to provider outcome profiling.DataWe used data from the Improving Mood-Promoting Access to Collaborative Treatment trial and the Depression Improvement across Minnesota, Offering a New Direction initiative to generate parameters for a Monte Carlo simulation experiment.Study designThe patient outcome of interest is the rate of remission of depressive symptoms at 6 months among a panel of patients with major depression at baseline. We considered two alternative approaches to profiling this outcome: (1) a relative, or tournament style threshold, set at the 80th percentile of remission rate among all providers, and (2) an absolute threshold, evaluating whether providers exceed a specified remission rate (30 percent). We performed a Monte Carlo simulation experiment to evaluate the total error rate (proportion of providers who were incorrectly classified) under each profiling approach. The total error rate was partitioned into error from random sampling variability and error resulting from missing data. We then evaluated the accuracy of alternative profiling approaches under different assumptions about the relationship between missing data and depression remission.Principal findingsOver a range of scenarios, relative profiling approaches had total error rates that were approximately 20 percent lower than absolute profiling approaches, and error due to missing data was approximately 50 percent lower for relative profiling. Most of the profiling error in the simulations was a result of random sampling variability, not missing data: between 11 and 21 percent of total error was attributable to missing data for relative profiling, while between 16 and 33 percent of total error was attributable to missing data for absolute profiling. Finally, compared with relative profiling, absolute profiling was much more sensitive to missing data that was correlated with the remission outcome.ConclusionsRelative profiling approaches for pay-for-performance were more accurate and more robust to missing data than absolute profiling approaches.