Search Results
Following deduplication, the literature search yielded 1010 abstracts. Following abstract screening, 96 full-text articles were retrieved and assessed for eligibility. Data was extracted from 33 articles, seven of which did not report sufficient information to allow for calculation of sensitivity and specificity values, and therefore were excluded. A total of 26 studies totalling 1145 patients were included in the systematic review (see Table 1).
Table 1
– Details of included studies.
Method
|
Study
|
Sample size
|
PsP/TTP
|
M/F
|
Mean age (yrs)
|
WHO Grade
|
Radiation therapy
|
DSC-pMRI
|
Baek et al. 2012[61]
|
79
|
37/42
|
46/33
|
51
|
IV
|
CRT-TMZ
|
Cha et al. 2014[62]
|
35
|
24/11
|
18/17
|
49
|
IV
|
CRT-TMZ
|
Kerkhof et al. 2017[31]
|
58
|
26/32
|
41/17
|
60
|
IV
|
CRT-TMZ
|
Kong et al. 2011[63]
|
59
|
26/33
|
49/10
|
50
|
IV
|
CRT-TMZ
|
Mangla et al. 2010[64]
|
19
|
7/12
|
NS
|
63
|
IV
|
RT-TMZ
|
Martínez-Martínez et al. 2014[34]
|
34
|
17/17
|
14/20
|
48
|
III-IV
|
CRT-TMZ
|
Mihailović et al. 2017[65]
|
40
|
8/32
|
37/3
|
51
|
IV
|
RT-TMZ
|
DSC-pMRI, DWI
|
Kim et al. 2019[24]
|
34
|
20/14
|
25/9
|
62
|
IV
|
CRT-TMZ
|
Prager et al. 2015[25]
|
51
|
8/43
|
NS
|
55
|
IV
|
CRT-TMZ
|
DSC-pMRI, ASL
|
Jovanovic et al. 2017[26]
|
31
|
11/20
|
21/10
|
49
|
IV
|
RT-TMZ
|
DSC-pMRI, ASL,
Combination of
DSC-pMRI and ASL
|
Choi et al. 2013[27]
|
62
|
28/34
|
37/25
|
49
|
IV
|
CRT-TMZ
|
Combination of
DSC-pMRI and DCE-pMRI
|
Elshafeey et al. 2019[37]
|
98
|
22/76
|
57/41
|
50
|
IV
|
CRT-TMZ
|
DWI
|
Bulik et al. 2015[66]
|
24
|
6/18
|
17/7
|
50
|
IV
|
CRT-TMZ
|
Chu et al. 2013[67]
|
30
|
15/15
|
16/14
|
51
|
IV
|
CRT-TMZ
|
Kazda et al. 2016[68]
|
39
|
10/29
|
28/11
|
51
|
IV
|
CRT-TMZ
|
Lee et al. 2012[69]
|
22
|
12/10
|
NS
|
49
|
IV
|
CRT-TMZ
|
Reimer et al. 2017[70]
|
35
|
7/28
|
26/9
|
58
|
III-IV
|
CRT-TMZ
|
Song et al. 2013[71]
|
20
|
10/10
|
10/10
|
51
|
IV
|
CRT-TMZ
|
Yoo et al. 2015[72]
|
42
|
18/24
|
27/15
|
61
|
IV
|
CRT-TMZ
|
DCE-pMRI
|
Nam et al. 2017[73]
|
37
|
22/15
|
26/11
|
58
|
IV
|
CRT-TMZ
|
Suh et al. 2013[74]
|
79
|
37/42
|
36/43
|
49
|
IV
|
CRT
|
Thomas et al. 2015[75]
|
37
|
13/24
|
25/12
|
63
|
IV
|
CRT-TMZ
|
Yun et al. 2014[76]
|
33
|
16/17
|
22/11
|
55
|
IV
|
CRT-TMZ
|
18F-FET PET
|
Galldiks et al. 2015[28]
|
22
|
11/11
|
14/8
|
56
|
IV
|
CRT-TMZ
|
APTw-MRI
|
Ma et al. 2016[29]
|
32
|
12/20
|
21/11
|
56
|
III-IV
|
CRT-TMZ
|
Conventional MRI
|
Young et al. 2011[30]
|
93
|
30/63
|
58/35
|
59
|
IV
|
CRT-TMZ
|
APTw-MRI = amide proton transfer-weighted MRI; ASL = arterial spin labelling; CRT-TMZ = chemoradiotherapy with adjuvant temozolomide; NS = not specified, TTP = true progression. |
Of the 1145 patients, 453 (39.6%) cases were confirmed to have PsP following radiotherapy, according to histological examination and/or examination according the RANO criteria. The mean age across the studies was 54, with a range of study mean age of 48 to 62 years. Based on 23 studies with relevant reporting, the male/female ratio was 1.8/1.
Seven distinct methods for differentiating PsP from TTP were identified: DSC-pMRI, DWI, DCE-pMRI, ASL, APTw-MRI, 18F-FET PET, and conventional MRI. Two studies compared DSC-pMRI and DWI,[24],[25] one compared DSC-pMRI and ASL,[26] and one compared DSC-pMRI, ASL, and a combination of the two.[27]
Data on seven distinct methods and two combinations of methods were included in the non-subgroup analysis (Figure 2A). The highest performing singular methods were FET PET (100% sensitivity and 91% specificity) and APTw-MRI (95% sensitivity and 92% specificity), although these subgroups were based on single, low-powered studies.[28],[29] Three studies (two using DSC-pMRI and one using DWI) reported 100% sensitivity and 100% specificity. The lowest sensitivity (38%) for identifying tumour progression was reported by Young and colleagues,[30] who examined visual signs such as enhancement on conventional MRI across 93 patients. The lowest specificity for true progression (23%) was reported by Kerkhof and colleagues,[31] which used visual interpretation of relative cerebral blood volume (rCBV) maps from DSC-pMRI.
Subgroup Analysis
Twenty-two studies assessing three distinct methods reported sufficient data to include in a separate set of subgroup analyses: DSC-pMRI (n = 11), DCE-pMRI (n = 4), and DWI (n = 9). Subgroups included 484, 186, and 309 patients, respectively. Studies by Kim and colleagues[24] and Prager and colleagues,[25] measured DSC-pMRI and DWI separately and thus, these studies appear twice in the forest plot (Figure 2B).,
All three subgroups had a high diagnostic accuracy for differentiating PsP from TTP, supported by high diagnostic odds ratios (Figure 3, Table 2). The DWI subgroup showed the highest diagnostic accuracy (sensitivity = 0.91 [0.84 – 0.95], specificity = 0.87 [0.67 – 0.96], AUC = 0.95 [0.92 – 0.96]), followed by DSC-pMRI (sensitivity = 0.89 [0.81 – 0.94], specificity = 0.89 [0.68 – 0.97], AUC = 0.94 [0.91 – 0.95]). The sensitivity of DCE-pMRI and DWI to the identification of true tumour progression was higher than the specificity of both methods.
Table 2 - Summary statistics comparing DSC-pMRI, DCE-pMRI, and DWI.
|
pooled
|
Likelihood ratio
|
DOR
|
SROC AUC
|
|
SENSITIVITY
|
SPECIFICITY
|
POSITIVE
|
NEGATIVE
|
DSC-pMRI
|
0.89
[0.81 - 0.94]
|
0.89
[0.68 - 0.97]
|
8.2 [2.4 – 27.8]
|
0.13 [0.07 - 0.24]
|
65 [11 - 373]
|
0.94
[0.91 - 0.95]
|
DCE-pMRI
|
0.88
[0.79 - 0.93]
|
0.77
[0.66 - 0.86]
|
3.8 [2.4 - 6.1]
|
0.16 [0.09 - 0.29]
|
24 [9 - 60]
|
0.90
[0.87 - 0.93]
|
DWI
|
0.91
[0.84 - 0.95]
|
0.87
[0.67 - 0.96]
|
6.9 [2.4 - 19.7]
|
0.10 [0.05 - 0.20]
|
68 [14 - 330]
|
0.95
[0.92 - 0.96]
|
DOR = diagnostic odds ratio; LR = likelihood ratio; SROC AUC = Summary receiver operator characteristics area under curve.
Heterogeneity for both DCE-pMRI and DWI was calculated as I2 = 0%, but high heterogeneity was reported in the DSC-pMRI subgroup (I2 = 79%). This heterogeneity was more predominant in the reported specificity (I2 = 85%). The DSC-pMRI subgroup had the highest number of studies and the greatest variation in methodology. However, true heterogeneity is unlikely to be zero in the DCE-pMRI and DWI subgroups, and the small sample size may have led to an underestimation.[32]
Quality Assessment
Twelve of the included 26 studies were determined to have a high risk of bias. Nearly all included studies had a high risk of bias in the index test section. The parameter cut-off values were not pre-specified and instead defined post-hoc. High risk of bias was also apparent in patient selection. This was largely due to some studies including patients who received steroids with standard chemoradiotherapy. Details of patient enrolment and inclusion/exclusion were unclear in some studies, and nearly 40% of included patients presented with PsP, which is higher than previously reported.[8] There were low applicability concerns observed in the included studies. The full risk of bias table and a more detailed summary of quality assessment across all studies is detailed in Supplementary Material B1 & B2.