Guido Schwarzer

Recommendations for examining and interpreting funnel plot asymmetry in meta-analyses of randomised controlled trials.

Funnel plots, and tests for funnel plot asymmetry, have been widely used to examine bias in the results of meta-analyses. Funnel plot asymmetry should not be equated with publication bias, because it has a number of other possible causes. This article describes how to interpret funnel plot asymmetry, recommends appropriate tests, and explains the implications for choice of meta-analysis model

Recombinant human erythropoiesis-stimulating agents and mortality in patients with cancer: a meta-analysis of randomised trials.

BACKGROUND Erythropoiesis-stimulating agents reduce anaemia in patients with cancer and could improve their quality of life, but these drugs might increase mortality. We therefore did a meta-analysis of randomised controlled trials in which these drugs plus red blood cell transfusions were compared with transfusion alone for prophylaxis or treatment of anaemia in patients with cancer. METHODS Data for patients treated with epoetin alfa, epoetin beta, or darbepoetin alfa were obtained and analysed by independent statisticians using fixed-effects and random-effects meta-analysis. Analyses were by intention to treat. Primary endpoints were mortality during the active study period and overall survival during the longest available follow-up, irrespective of anticancer treatment, and in patients given chemotherapy. Tests for interactions were used to identify differences in effects of erythropoiesis-stimulating agents on mortality across prespecified subgroups. FINDINGS Data from a total of 13 933 patients with cancer in 53 trials were analysed. 1530 patients died during the active study period and 4993 overall. Erythropoiesis-stimulating agents increased mortality during the active study period (combined hazard ratio [cHR] 1.17, 95% CI 1.06-1.30) and worsened overall survival (1.06, 1.00-1.12), with little heterogeneity between trials (I(2) 0%, p=0.87 for mortality during the active study period, and I(2) 7.1%, p=0.33 for overall survival). 10 441 patients on chemotherapy were enrolled in 38 trials. The cHR for mortality during the active study period was 1.10 (0.98-1.24), and 1.04 (0.97-1.11) for overall survival. There was little evidence for a difference between trials of patients given different anticancer treatments (p for interaction=0.42). INTERPRETATION Treatment with erythropoiesis-stimulating agents in patients with cancer increased mortality during active study periods and worsened overall survival. The increased risk of death associated with treatment with these drugs should be balanced against their benefits. FUNDING German Federal Ministry of Education and Research, Medical Faculty of University of Cologne, and Oncosuisse (Switzerland).

On the misuses of artificial neural networks for prognostic and diagnostic classification in oncology.

The application of artificial neural networks (ANNs) for prognostic and diagnostic classification in clinical medicine has become very popular. In particular, feed-forward neural networks have been used extensively, often accompanied by exaggerated statements of their potential. In this paper, the essentials of feed-forward neural networks and their statistical counterparts (that is, logistic regression models) are reviewed. We point out that the uncritical use of ANNs may lead to serious problems, such as the fitting of implausible functions to describe the probability of class membership and the underestimation of misclassification probabilities. In applications of ANNs to survival data, further difficulties arise. Finally, the results of a search in the medical literature from 1991 to 1995 on applications of ANNs in oncology and some important common mistakes are reported. It is concluded that there is no evidence so far that application of ANNs represents real progress in the field of diagnosis and prognosis in oncology.

Treatment of ocular hypertension and open angle glaucoma: meta-analysis of randomised controlled trials

Abstract Objective Open angle glaucoma is one of the most common causes of blindness in industrialised nations. Treatments to lower ocular pressure are widely used in glaucoma prevention and treatment, despite conflicting evidence. Design We performed meta-analyses to reassess the effectiveness of pressure lowering treatment to delay the development of glaucoma in ocular hypertension, as well as progression of manifest open angle glaucoma. Data sources Medline, Embase, and the Cochrane Library. Selection of studies Eligible studies were randomised controlled trials with a concurrent untreated control group and information on time to glaucomatous changes to visual field and optic disc. Trial reports were reviewed independently by two investigators in an unblinded standardised manner. Results Meta-analysis of trials in ocular hypertension showed a significant preventive effect of reducing intraocular pressure on progression to glaucoma (hazard ratio 0.56, 95% confidence interval 0.39 to 0.81, P = 0.01; number needed to treat 12). Pooled data of studies in manifest glaucoma showed a significant delay of visual field deterioration (0.65, 0.49 to 0.87, P = 0.003; NNT = 7), with subgroup analysis showing a larger effect in patients with raised pressure and a reduced effect in normal tension glaucoma (subgroup comparison: not significant). Conclusions Lowering intraocular pressure in patients with ocular hypertension or manifest glaucoma is beneficial in reducing the risk of visual field loss in the long term.

Why add anything to nothing? The arcsine difference as a measure of treatment effect in meta-analysis with zero cells.

For clinical trials with binary endpoints there are a variety of effect measures, for example risk difference, risk ratio and odds ratio (OR). The choice of metric is not always straightforward and should reflect the clinical question. Additional issues arise if the event of interest is rare. In systematic reviews, trials with zero events in both arms are encountered and often excluded from the meta-analysis.The arcsine difference (AS) is a measure which is rarely considered in the medical literature. It appears to have considerable promise, because it handles zeros naturally, and its asymptotic variance does not depend on the event probability.This paper investigates the pros and cons of using the AS as a measure of intervention effect. We give a pictorial representation of its meaning and explore its properties in relation to other measures. Based on analytical calculation of the variance of the arcsine transformation, a more conservative variance estimate for the rare event setting is proposed. Motivated by a published meta-analysis in cardiac surgery, we examine the statistical properties of the various metrics in the rare event setting.We find the variance estimate of the AS to be more stable than that of the log-OR, even if events are rare. However, parameter estimation is biased if the groups are markedly unbalanced. Though, from a theoretical viewpoint, the AS is a natural choice, its practical use is likely to continue to be limited by its less direct interpretation.

Seroprevalence of Chronic Hepatitis B Virus Infection and Prior Immunity in Immigrants and Refugees: A Systematic Review and Meta-Analysis

Background International migrants experience increased mortality from hepatocellular carcinoma compared to host populations, largely due to undetected chronic hepatitis B infection (HBV). We conducted a systematic review of the seroprevalence of chronic HBV and prior immunity in migrants arriving in low HBV prevalence countries to identify those at highest risk in order to guide disease prevention and control strategies. Methods and Findings Medline, Medline In-Process, EMBASE and the Cochrane Database of Systematic Reviews were searched. Studies that reported HBV surface antigen or surface antibodies in migrants were included. The seroprevalence of chronic HBV and prior immunity were pooled by region of origin and immigrant class, using a random-effects model. A random-effects logistic regression was performed to explore heterogeneity. The number of chronically infected migrants in each immigrant-receiving country was estimated using the pooled HBV seroprevalences and country-specific census data. A total of 110 studies, representing 209,822 immigrants and refugees were included. The overall pooled seroprevalence of infection was 7.2% (95% CI: 6.3%–8.2%) and the seroprevalence of prior immunity was 39.7% (95% CI: 35.7%–43.9%). HBV seroprevalence differed significantly by region of origin. Migrants from East Asia and Sub-Saharan Africa were at highest risk and migrants from Eastern Europe were at an intermediate risk of infection. Region of origin, refugee status and decade of study were independently associated with infection in the adjusted random-effects logistic model. Almost 3.5 million migrants (95% CI: 2.8–4.5 million) are estimated to be chronically infected with HBV. Conclusions The seroprevalence of chronic HBV infection is high in migrants from most world regions, particularly among those from East Asia, Sub-Saharan Africa and Eastern Europe, and more than 50% were found to be susceptible to HBV. Targeted screening and vaccination of international migrants can become an important component of HBV disease control efforts in immigrant-receiving countries.

Ranking treatments in frequentist network meta-analysis works without resampling methods

BackgroundNetwork meta-analysis is used to compare three or more treatments for the same condition. Within a Bayesian framework, for each treatment the probability of being best, or, more general, the probability that it has a certain rank can be derived from the posterior distributions of all treatments. The treatments can then be ranked by the surface under the cumulative ranking curve (SUCRA). For comparing treatments in a network meta-analysis, we propose a frequentist analogue to SUCRA which we call P-score that works without resampling.MethodsP-scores are based solely on the point estimates and standard errors of the frequentist network meta-analysis estimates under normality assumption and can easily be calculated as means of one-sided p-values. They measure the mean extent of certainty that a treatment is better than the competing treatments.ResultsUsing case studies of network meta-analysis in diabetes and depression, we demonstrate that the numerical values of SUCRA and P-Score are nearly identical.ConclusionsRanking treatments in frequentist network meta-analysis works without resampling. Like the SUCRA values, P-scores induce a ranking of all treatments that mostly follows that of the point estimates, but takes precision into account. However, neither SUCRA nor P-score offer a major advantage compared to looking at credible or confidence intervals.

Treatment-effect estimates adjusted for small-study effects via a limit meta-analysis

Statistical heterogeneity and small-study effects are 2 major issues affecting the validity of meta-analysis. In this article, we introduce the concept of a limit meta-analysis, which leads to shrunken, empirical Bayes estimates of study effects after allowing for small-study effects. This in turn leads to 3 model-based adjusted pooled treatment-effect estimators and associated confidence intervals. We show how visualizing our estimators using the radial plot indicates how they can be calculated using existing software. The concept of limit meta-analysis also gives rise to a new measure of heterogeneity, termed G(2), for heterogeneity that remains after small-study effects are accounted for. In a simulation study with binary data and small-study effects, we compared our proposed estimators with those currently used together with a recent proposal by Moreno and others. Our criteria were bias, mean squared error (MSE), variance, and coverage of 95% confidence intervals. Only the estimators arising from the limit meta-analysis produced approximately unbiased treatment-effect estimates in the presence of small-study effects, while the MSE was acceptably small, provided that the number of studies in the meta-analysis was not less than 10. These limit meta-analysis estimators were also relatively robust against heterogeneity and one of them had a relatively small coverage error.

Extent of Non-Publication in Cohorts of Studies Approved by Research Ethics Committees or Included in Trial Registries

Background The synthesis of published research in systematic reviews is essential when providing evidence to inform clinical and health policy decision-making. However, the validity of systematic reviews is threatened if journal publications represent a biased selection of all studies that have been conducted (dissemination bias). To investigate the extent of dissemination bias we conducted a systematic review that determined the proportion of studies published as peer-reviewed journal articles and investigated factors associated with full publication in cohorts of studies (i) approved by research ethics committees (RECs) or (ii) included in trial registries. Methods and Findings Four bibliographic databases were searched for methodological research projects (MRPs) without limitations for publication year, language or study location. The searches were supplemented by handsearching the references of included MRPs. We estimated the proportion of studies published using prediction intervals (PI) and a random effects meta-analysis. Pooled odds ratios (OR) were used to express associations between study characteristics and journal publication. Seventeen MRPs (23 publications) evaluated cohorts of studies approved by RECs; the proportion of published studies had a PI between 22% and 72% and the weighted pooled proportion when combining estimates would be 46.2% (95% CI 40.2%–52.4%, I2 = 94.4%). Twenty-two MRPs (22 publications) evaluated cohorts of studies included in trial registries; the PI of the proportion published ranged from 13% to 90% and the weighted pooled proportion would be 54.2% (95% CI 42.0%–65.9%, I2 = 98.9%). REC-approved studies with statistically significant results (compared with those without statistically significant results) were more likely to be published (pooled OR 2.8; 95% CI 2.2–3.5). Phase-III trials were also more likely to be published than phase II trials (pooled OR 2.0; 95% CI 1.6–2.5). The probability of publication within two years after study completion ranged from 7% to 30%. Conclusions A substantial part of the studies approved by RECs or included in trial registries remains unpublished. Due to the large heterogeneity a prediction of the publication probability for a future study is very uncertain. Non-publication of research is not a random process, e.g., it is associated with the direction of study findings. Our findings suggest that the dissemination of research findings is biased.

Cerebrovascular reactivity predicts stroke in high-grade carotid artery disease

Objective: To assess the usefulness of transcranial Doppler CO2 reactivity (CO2R) for prediction of ipsilateral ischemic stroke in carotid artery stenosis and occlusion with a meta-analysis of prospective studies based on individual patient data. Methods: We searched Medline, Biosis Previews, Science Citation Index, The Cochrane Library, and EMBASE for studies in which patients with severe carotid artery stenosis or occlusion underwent Doppler CO2R testing (inhalation of CO2 or breath-holding) and were prospectively followed for ipsilateral ischemic stroke. Individual data from 754 patients from 9 studies were included. We used percentage cerebral blood flow velocity increase (pCi) during hypercapnia as the primary CO2R measure, and defined impaired reactivity as pCi <20% increase. Results: In a multiple regression model, impaired CO2R was independently associated with an increased risk of ipsilateral ischemic stroke (hazard ratio [HR] 3.69; confidence interval [CI] 2.01, 6.77; p < 0.0001). Risk prediction was similar for recently symptomatic vs asymptomatic patients. Using continuous values of pCi, a significant association between decreasing pCi and increasing risk of ipsilateral stroke was found: HR of 1.64 (95% CI 1.33, 2.02; p < 0.0001) per 10% decrease in pCi. For patients with asymptomatic internal carotid artery stenosis only (n = 330), a comparable stroke risk prediction was found: increasing HR 1.95 (95% CI 1.26, 3.04; p = 0.003) per 10% decrease in pCi. Conclusions: This analysis supports the usefulness of CO2R in risk prediction for patients with severe carotid artery stenosis or occlusion, both in recently symptomatic and asymptomatic patients. Further studies should evaluate whether treatment strategies in asymptomatic patients based on CO2R could improve patient outcomes.