Tests for publication bias are unreliable in case of heteroscedasticity

被引：7

作者：

Almalik, Osama ^{[1
]}

Zhan, Zhuozhao ^{[1
]}

van den Heuvel, Edwin R. ^{[1
,2
]}

机构：

[1] Eindhoven Univ Technol, Dept Math & Comp Sci, POB 513, NL-5600 MB Eindhoven, Netherlands

[2] Boston Univ, Sch Med, Dept Prevent Med & Epidemiol, Boston, MA 02118 USA

来源：

CONTEMPORARY CLINICAL TRIALS COMMUNICATIONS | 2021年 / 22卷

关键词：

Heteroscedastic mixed effects model; Aggregated data meta-analysis; Mean difference treatment effect sizes; METAANALYSIS; VARIANCE; VARIABILITY; ESTIMATORS; REGRESSION;

D O I：

10.1016/j.conctc.2021.100781

中图分类号：

R-3 [医学研究方法]; R3 [基础医学];

学科分类号：

1001 ;

摘要：

Regression based methods for the detection of publication bias in meta-analysis have been extensively evaluated in literature. When dealing with continuous outcomes, specific hidden factors (e.g., heteroscedasticity) may interfere with the test statistics. In this paper we investigate the influence of residual heteroscedasticity on the performance of four tests for publication bias: the Egger test, the Begg-Mazumdar test and two tests based on weighted regression. In the presence of heteroscedasticity, the Egger test and the weighted regression tests highly inflate the Type I error rate, while the Begg-Mazumdar test deflates the Type I error rate. Although all three tests already have low statistical power, heteroscedasticity typically reduces it further. Our results in combination with earlier discussions on publication bias tests lead us to conclude that application of these tests on continuous treatment effects is not warranted.

引用

页数：6

共 33 条

[1] OPERATING CHARACTERISTICS OF A BANK CORRELATION TEST FOR PUBLICATION BIAS [J].

BEGG, CB ;

MAZUMDAR, M .

BIOMETRICS, 1994, 50 (04) :1088-1101

[2] A comparison of statistical methods for meta-analysis [J].

Brockwell, SE ;

Gordon, IR .

STATISTICS IN MEDICINE, 2001, 20 (06) :825-840

[3] Random-Effects Meta-analysis of Inconsistent Effects: A Time for Change [J].

Cornell, John E. ;

Mulrow, Cynthia D. ;

Localio, Russell ;

Stack, Catharine B. ;

Meibohm, Anne R. ;

Guallar, Eliseo ;

Goodman, Steven N. .

ANNALS OF INTERNAL MEDICINE, 2014, 160 (04) :267-270

[4] VARIANCE FUNCTION ESTIMATION [J].

DAVIDIAN, M ;

CARROLL, RJ .

JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION, 1987, 82 (400) :1079-1091

[5] SOME SIMPLE METHODS FOR ESTIMATING INTRAINDIVIDUAL VARIABILITY IN NONLINEAR MIXED EFFECTS MODELS [J].

DAVIDIAN, M ;

GILTINAN, DM .

BIOMETRICS, 1993, 49 (01) :59-73

[6] The performance of tests of publication bias and other sample size effects in systematic reviews of diagnostic test accuracy was assessed [J].

Deeks, JJ ;

Macaskill, P ;

Irwig, L .

JOURNAL OF CLINICAL EPIDEMIOLOGY, 2005, 58 (09) :882-893

[7] METAANALYSIS IN CLINICAL-TRIALS [J].

DERSIMONIAN, R ;

LAIRD, N .

CONTROLLED CLINICAL TRIALS, 1986, 7 (03) :177-188

[8] DISTRIBUTION OF FORCED EXPIRATORY VOLUME IN ONE 2ND AND FORCED VITAL CAPACITY IN HEALTHY, WHITE, ADULT NEVER-SMOKERS IN 6 UNITED-STATES CITIES [J].

DOCKERY, DW ;

WARE, JH ;

FERRIS, BG ;

GLICKSBERG, DS ;

FAY, ME ;

SPIRO, A ;

SPEIZER, FE .

AMERICAN REVIEW OF RESPIRATORY DISEASE, 1985, 131 (04) :511-520

[9] A nonparametric "trim and fill" method of accounting for publication bias in meta-analysis [J].

Duval, S ;

Tweedie, R .

JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION, 2000, 95 (449) :89-98

[10] Bias in meta-analysis detected by a simple, graphical test [J].

Egger, M ;

Smith, GD ;

Schneider, M ;

Minder, C .

BMJ-BRITISH MEDICAL JOURNAL, 1997, 315 (7109) :629-634

← 1 2 3 4 →