Professional Documents
Culture Documents
tumors (mostly pineoblastoma).2,3 Magnetic resonance imaging (MRI) also can aid retinoblastoma diagnosis in case of
an unclear ocular medium. Treatment strategies are focused
on survival, preserving vision, and nally on avoiding
enucleation. Although the gold standard for diagnosis and
local staging relies on pathologic analysis, the treatment
strategy frequently is based on clinical and radiologic ndings only, notably in eyes treated conservatively. In patients
treated by primary enucleation, extraocular extension rst
1109
Methods
We performed this study according to the Preferred Reporting
Items for Systematic Reviews and Meta-Analyses (PRISMA)
statement.18,19
Search Strategy
We searched Medline (PubMed) and Embase for English, Dutch,
German, and Spanish literature published through April 2013
evaluating tumor extent of retinoblastoma by MRI or CT. We also
included studies found in alternative ways (e.g., through checking
references in included studies). When necessary, we contacted
authors for additional data. The search included keywords corresponding to the index test MRI and CT, the target condition
(retinoblastoma), and diagnostic performance (Appendix A,
available at www.aaojournal.org).
Study Selection
Article titles and abstracts were reviewed independently for eligibility by 2 authors (M.C.J. and D.P.N.), and discrepancies were
1110
Data Extraction
Two authors (M.C.J. and D.P.N.) independently extracted study
data. Discrepancies were resolved by consensus. When studies
reported multiple sets of sensitivities and specicities from multiple readers separately, the set with the highest diagnostic odds ratio
(DOR) was used for the gures and meta-analysis. We did this to
prevent data from the same patient population from being used
twice. This could have resulted in an overestimation of the summary estimates; therefore, we also reported the overall summary
estimates including sensitivity and specicity from the reader with
the lowest DOR.
De Jong et al
Results
Medline and Embase searches yielded 426 unique studies. We
excluded 376 articles based on title and abstract (Fig 1). We
excluded 32 studies based on the full text; see Figure 1 for
Figure 1. Flowchart showing systematic literature search. CT computed tomography; FN false-negative results; FP false-positive results; MRI
magnetic resonance imaging; TN true-negative results; TP true-positive results.
1111
Summary Estimates
Compared with histopathologic analysis, the meta-analysis of MRI
for the different tumor-extent parameters gave us a sensitivity of
73% (95% CI, 45%e90%), a specicity of 96% (95% CI, 64%e
100%), and a DOR of 59.7 (95% CI, 4.36e818) for prelaminar
ONI from 8 studies (Table 5; Fig 2, available at www.
aaojournal.org). For intralaminar ONI, MRI showed a sensitivity
of 38% (95% CI, 0%e100%), a specicity of 89% (95% CI,
47%e99%), and a DOR of 4.84 (95% CI, 0.00e3.14104) from
4 studies. For postlaminar ONI, MRI demonstrated a sensitivity
of 59% (95% CI, 37%e78%), a specicity of 94% (95% CI,
84%e98%), and a DOR of 20.9 (95% CI, 4.75e92.2) from 10
studies. For choroidal invasion, MRI showed a sensitivity of
74% (95% CI, 52%e88%), a specicity of 72% (95% CI, 31%e
94%), and a DOR of 7.38 (95% CI, 1.21e45.0) from 6 studies.
For scleral invasion, MRI demonstrated a sensitivity of 88%
(95% CI, 20%e100%), a specicity of 99% (95% CI, 86%e
100%), and a DOR of 503 (95% CI, 24.9e1.02104) from 6
studies. The ROC plots show the summary estimates of
specicity on the x-axis and sensitivity on the y-axis for these 5
tumor-extent parameters with their respective 95% condence
areas (Fig 3).
1112
studies for scleral invasion explicitly reported data based on primary enucleations only. Results based on this subset gave lower
DORs for most tumor-extent paramaters; only the DOR of
choroidal invasion did not show much of a difference, and for
scleral invasion, we could not calculate summary estimates
(Table 5).
We also assessed the effect of including data from the readers of
the MRI scans with the lowest DORdfrom studies that report
multiple readersdon the summary estimates. Three of 8 studies
assessing prelaminar ONI, 1 of 4 studies assessing intralaminar
ONI, and 2 of 10 studies assessing postlaminar ONI reported 2 sets
of sensitivity and specicity from 2 readers. When data from the
worst reader (i.e., with the lowest DOR) were included in the
analysis, the results did not change much (Table 5).
For prelaminar ONI and choroidal invasion, the results from
metaregression analysis showed higher DORs for studies with a
higher image quality. The results for postlaminar ONI showed a
higher DOR only after removal of a potential outlier. However,
none of these differences were statistically signicant (Table 6;
Fig 2, available at www.aaojournal.org).
Discussion
We were able to perform a meta-analysis of the diagnostic
accuracy of MRI for detection of optic nerve, choroidal, and
scleral invasion in retinoblastoma. Study data of extrascleral
invasion, AES involvement, and ciliary body invasion
proved insufcient for meta-analysis. Studies reporting the
De Jong et al
Figure 3. Receiver operator characteristic spaces with 95% condence areas. The sizes of the rectangles correlate with the size of the study.
1113
TP
FP
TN
FN
8
4
8
47 (157/334)
70 (119/169)
47 (157/334)
342
564
329
23
55
24
507
241
506
128
140
141
4
3
4
17 (36/217)
19 (35/181)
17 (36/217)
62
31
64
92
128
94
742
679
740
104
162
102
59 (37e78) 94 (84e98)
56 (27e81) 93 (77e98)
58 (31e81) 94 (85e97)
20.9 (4.75e92.2)
16.0 (2.26e113)
19.9 (4.64e85.4)
10
7
10
17 (69/418)
18 (59/331)
17 (69/418)
97
100
95
53
61
53
781
761
782
68
78
70
74 (52e88) 72 (31e94)
69 (26e93) 78 (6e100)
7.38 (1.21e45.0)
7.73 (0.07e854)
6
4
42 (105/253)
44 (81/184)
309
302
165
123
420
437
107
139
6
2
4 (12/281)
5 (6/133)
38
IC
14
IC
944
IC
5
IC
1114
Limitations
Because the incidence of retinoblastoma is low, published
studies usually have small patient populations, resulting in
De Jong et al
Figure 4. Graph showing the risk of bias and applicability concerns: review of authors judgements about each domain, presented as percentages across
included studies.
happen is unattainable in practice because of ethical considerations. In this case, enucleation depends on previous
clinical tests, but also on the index test. Verication bias can
lead to overestimation of sensitivity and underestimation of
specicity but generally does not inuence the DOR.20,43
Verication bias is likely to be present in all included
studies to varying degrees. Studies report varying prevalences of tumor-extent parameters (Tables 2e4, available at
www.aaojournal.org). Prevalence depends on more than
verication bias alone, but a higher than normal prevalence
may be attributed in part to verication bias. Inherently, the
results presented in this meta-analysis are based on patients
with severe retinoblastoma because only these patients undergo enucleation, allowing for a histopathologic evaluation.
To what extent these results are also generalizable to the
entire retinoblastoma patient population is difcult to say.
Studies have not always reported explicitly that the interpretation of the index test or histopathologic results was performed in a blinded manner. Not blinding the interpretation
can result in a slight overestimation of the DOR.20 Most
studies were unclear regarding whether their patients were
enrolled consecutively or randomly (yes, 4 studies; unclear,
13 studies; no, 1 study), and most studies either reported a
retrospective study design or did not mention this in their
article (retrospective, 9 studies; not specied, 7 studies;
prospective, 2 studies); however, Lijmer et al20 showed that
these study design aspects do not impact the DOR of a
diagnostic test. Some studies included both affected eyes of
1 patient, which might have caused an overestimation of
diagnostic accuracy; this issue was not part of the QUADAS
checklist. See Table 1 (available at www.aaojournal.org) for
studies that included more eyes than patients.
In meta-analyses, there is a risk that publication bias affects
the results. Publication bias usually occurs when small negative
studies (i.e., studies with poor diagnostic accuracy and a low
number of patients) are deemed uninteresting and are not published. Missing these studies could lead to an overestimation of
the summary estimates. We have chosen not to look at publication bias formally because of the lack of a good test.44
Heterogeneity across studies is a common limitation of
meta-analyses. The studies we included differ in terms of
imaging technique and quality, prevalence of tumor-extent
parameters, inclusion criteria (some also included secondary enucleations), and time between index and reference tests.
Sensitivity analysis by excluding studies that did not explicitly report only including primary enucleations showed
slightly lower diagnostic accuracy for most tumor-extent
1115
Sensitivity*
Specicity*
No. of Studies
Disease Prevalencey
57 (27e83)
83 (58e95)
0.12
97 (55e100)
93 (38e100)
0.59
47.0 (1.51e1.47103)
67.2 (2.74e1.65103)
0.85
4
4
43 (58/135)
50 (99/199)
59 (32e82)
58 (27e84)
0.94
76 (45e93)
0.26
94 (78e98)
94 (77e98)
0.98
96 (76e100)
0.63
21.1 (3.01e148)
20.5 (2.41e175)
0.98
84.8 (7.72e932)
0.30
5
5
21 (44/210)
12 (25/208)
11 (16/148)
71 (7e99)
77 (52e92)
0.80
IC
66 (16e95)
76 (21e97)
0.72
IC
4.84 (0.10e246)
10.9 (0.93e128)
0.62
IC
3
3
42 (64/151)
40 (41/102)
36 (24/66)
1116
References
1. Moll AC, Kuik DJ, Bouter LM, et al. Incidence and survival of
retinoblastoma in The Netherlands: a register based study
1862e1995. Br J Ophthalmol 1997;81:55962.
De Jong et al
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
1117
1118
Financial Disclosure(s):
The author(s) have no proprietary or commercial interest in any materials
discussed in this article.
Supported by the ODAS Foundation, Delft, The Netherlands (M.C.J.). The
European Retinoblastoma Imaging Collaboration (ERIC) is nancially
supported by a grant from the ODAS Foundation. The authors work was
independent of the funding organizations. The funding organizations had no
involvement in the design or conduct of this study, data management and
analysis, or manuscript preparation and review or authorization for
submission.
Correspondence:
Marcus C. De Jong, MD, MSc, Department of Radiology, VU University
Medical Center, P.O. Box 7057, 1007 MB Amsterdam, The Netherlands.
E-mail: mc.dejong@vumc.nl.