The effect of study design biases on the diagnostic accuracy of magnetic resonance imaging for detecting silicone breast implant ruptures: a meta-analysis

Abstract

Background: The U.S. Food and Drug Administration has recommended that all silicone breast implant recipients undergo serial screening to detect implant rupture with magnetic resonance imaging. The authors performed a systematic review and meta-analysis to examine the effect of study design biases on the estimation of magnetic resonance imaging diagnostic accuracy measures.

Methods: Studies were identified using the MEDLINE, EMBASE, ISI Web of Science, and Cochrane library databases. Two reviewers independently screened potential studies for inclusion and extracted data. Study design biases were assessed using the Quality of Diagnostic Accuracy Studies tool and the Standards for Reporting of Diagnostic Accuracy Studies checklist. Meta-analyses estimated the influence of biases on diagnostic odds ratios.

Results: Among 1175 identified articles, 21 met the inclusion criteria. Most studies using magnetic resonance imaging (10 of 16) and ultrasound (10 of 13) examined symptomatic subjects. Magnetic resonance imaging studies evaluating symptomatic subjects had 14-fold higher diagnostic accuracy estimates compared with studies using an asymptomatic sample (relative diagnostic odds ratio, 13.8; 95 percent confidence interval, 1.83 to 104.6) and 2-fold higher diagnostic accuracy estimates compared with studies using a screening sample (relative diagnostic odds ratio, 1.89; 95 percent confidence interval, 0.05 to 75.7).

Conclusions: Many of the published studies using magnetic resonance imaging or ultrasound to detect silicone breast implant rupture are flawed with methodologic biases. These methodologic shortcomings may result in overestimated magnetic resonance imaging diagnostic accuracy measures and should be interpreted with caution when applying the data to a screening population.

Figure 1

Steps to a Meta-Analysis

Figure 2

Diagnostic Odds Ratio (A) The diagnostic odds ratio (DOR) is a measure of the overall accuracy of a positive test and combines a test’s sensitivity and specificity. It is interpreted as the odds of a positive test given disease, divided by the odds of a positive test given no disease. A large DOR of a test means the test has a high sensitivity and specificity for detecting a disease. (B) The relative diagnostic odds ratio (RDOR) is a ratio of 2 diagnostic odds ratios. ^†Covariates that will be examined are the different types of study or sample characteristics such as sample characteristics (i.e., symptomatic, asymptomatic, screening), partial verification bias, among others (See Figure 6).

Figure 2

Diagnostic Odds Ratio (A) The diagnostic odds ratio (DOR) is a measure of the overall accuracy of a positive test and combines a test’s sensitivity and specificity. It is interpreted as the odds of a positive test given disease, divided by the odds of a positive test given no disease. A large DOR of a test means the test has a high sensitivity and specificity for detecting a disease. (B) The relative diagnostic odds ratio (RDOR) is a ratio of 2 diagnostic odds ratios. ^†Covariates that will be examined are the different types of study or sample characteristics such as sample characteristics (i.e., symptomatic, asymptomatic, screening), partial verification bias, among others (See Figure 6).

Figure 3

Selection of Studies for Meta-Analysis A trial flow diagram shows the number of identified, screened, and included studies.

Figure 4

Quality Assessment of Magnetic Resonance Imaging Studies (A) Details of the QUADAS assessment of MRI studies are shown. (B) Percentages coded yes, no, or unclear are shown.

Figure 4

Quality Assessment of Magnetic Resonance Imaging Studies (A) Details of the QUADAS assessment of MRI studies are shown. (B) Percentages coded yes, no, or unclear are shown.

Figure 5

Quality Assessment of Ultrasound Studies (A) Details of the QUADAS assessment of ultrasound studies are shown. (B) Percentages coded yes, no, or unclear are shown.

Figure 5

Quality Assessment of Ultrasound Studies (A) Details of the QUADAS assessment of ultrasound studies are shown. (B) Percentages coded yes, no, or unclear are shown.

Figure 6

Study Estimates of Sensitivity and Specificity Values of MRI Studies Forest plots of the sensitivity and specificity values are illustrated. Two sources are listed twice because of separate subgroup analyses., Berg et al., (indicated by double-hatched cross) examined 122 single lumen silicone breast implants; however, only 94 implants were used in the 2×2 table because 28 indeterminant implants were not included in the analysis.

Figure 7

Relative Diagnostic Odds Ratios of Study Design Characteristics and Biases Examined with Univariate Regression Analysis In both panels, the black squares indicate the relative diagnostic odds ratios for each individual study. 95% confidence intervals are indicated by the horizontal line.

Figure 8

Funnel Plots to Assess Publication Bias (A) A funnel plot of the 18 MRI studies illustrates an asymmetric distribution of studies suggesting publication bias (Egger’s test, p=0.01) (B) A funnel plot constructed for the 13 ultrasound studies did not reveal a statistically significant distribution (p=0.87).