Original articles
Exploring Heterogeneity in Meta-Analysis: Is the L'Abbé Plot Useful?

https://doi.org/10.1016/S0895-4356(99)00066-9Get rights and content

Abstract

By using a published meta-analysis as an example, this paper discusses the use of L'Abbé plot for investigating the potential sources of heterogeneity in meta-analysis. As compared with other graphic procedures, the L'Abbé plot is useful to identify not only the studies having different results from other studies, but also the study arms that are responsible for such differences. This may be important for determining the focus of heterogeneity investigations. Results of stochastic simulation indicate that, purely because of random variation, studies with event rates of around 50% are more likely to be identified as outliers in a L'Abbé plot. This paper also demonstrates that different methods may identify different trials as “outliers” in meta-analysis.

Section snippets

Example

Dolan-Mullen and colleagues conducted a meta-analysis to assess the effect of prenatal smoking intervention [8]. By pooling results from 11 randomized controlled trials, it was found that intervention increased the rate of smoking cessation during pregnancy (overall rate ratio 2.08; 95% confidence interval 1.74, 2.49), although statistical test showed significant heterogeneity in rate ratio across the trials (χ2 = 35.26, df = 10, P < 0.001). Figure 1, known as a Forrest plot [9], shows the

Stochastic simulation

Stochastic simulation technique was used to explore factors related to the random discrepancies between study points and the overall RR line in the L'Abbé plot. The simulation results reveal that random variation in the distance between a study point and the overall RR line is negatively related to the sample size (Figure 3). It is also observed that the random variation in the distance is greatest when the event rates in the control and the treatment group are 50%.

Therefore, the absolute

Identifying outliers

In Dolan-Mullen et al.'s meta-analysis [8], L'Abbé plot was used to identify “outliers” according to the distance between a trial and the overall RR line.Then the “outliers” were excluded one by one until the statistical test of heterogeneity across trials was no longer significant. Trial-2 should be excluded from the analysis because it is farthest from the overall RR line (Figure 2). However, the standardized distance is greatest for Trial-10 (Table 3). On the other hand, according to the

Discussion

By examining the rates of smoking cessation in the L'Abbé plot, it is suggested that variations in the intensity of interventions, “usual care” controls, and study settings are potential sources of heterogeneity in Dolan-Mullen et al.'s meta-analysis. However, the usefulness of the L'Abbé plot in this example does not necessarily mean that it can reveal important sources of heterogeneity in all meta-analyses, because the potential causes of heterogeneity may be different and data may not be

References (18)

There are more references available in the full text version of this article.

Cited by (67)

  • Smoking-Cessation Interventions in Appalachia: A Systematic Review and Meta-Analysis

    2020, American Journal of Preventive Medicine
    Citation Excerpt :

    Homogeneity was also assessed quantitatively using Cochran's Q statistic and the I2 index. Because both the Q statistic and I2 index have low power to detect heterogeneity among a small number of studies,19 a L'Abbé plot was also used to attempt to graphically evaluate homogeneity.20,21 If there was evidence of a violation of the homogeneity assumption, the pooled effect of smoking-cessation intervention programs on smoking abstinence was estimated from a random effects model.

  • Efficacy of sulbactam for the treatment of Acinetobacter baumannii complex infection: A systematic review and meta-analysis

    2017, Journal of Infection and Chemotherapy
    Citation Excerpt :

    The heterogeneity across studies was assessed using the χ2 based Q statistics and I2 test, and was defined as low (I2 = 25%–49%), moderate (I2 = 50%–75%) and high (I2 ≥ 75%) [15]. In case of substantial inconsistency, a L'Abbe plot was employed to graphically identify the potential sources of heterogeneity in meta-analysis [16]. Subgroup analyses were conducted by type of infection, study design and antimicrobial agents.

  • Antibiotics for acute otitis media in children

    2023, Cochrane Database of Systematic Reviews
View all citing articles on Scopus
View full text