A total of 2,644 studies were screened. Twenty studies met the eligibility criteria and were included[10–12, 33–49] in the quality assessment and meta-analysis (Fig. 1 and S-Table 1).
Study characteristics
Data from 49,043 patients, including 8,300 patients (16.9%) with pelvic fracture, were included in the analysis (Table 1). The median prevalence of pelvic fracture was 10.5% (IQR: 5.1–16.5). Sixteen studies[10, 12, 33–36, 38–44, 47–49] were prospective, and four studies[11, 37, 45, 46] were retrospective. Most studies were set at a trauma center or emergency department of a university hospital. Three studies[37, 39, 46] included children (< 18 years). Seven studies[33, 35, 36, 38, 41, 44, 49] included patients who were either alert or had minor impairments to consciousness (Glasgow Coma Scale [GCS] score ≥ 13); other studies included patients with GCS score ≤ 13. Patient characteristics, index test definitions, and reference standards used in each study are summarized in Tables 2 and 3. Physical examination included inquiries about pelvic pain, inspection and palpation of the pelvis, assessment of the stability of the pelvis, and other procedures. Physical examinations were performed by an emergency physician in a trauma bay, emergency department, or surgical department. The reference standards were x-ray[33–36, 38, 40, 44, 46, 49], unclear[11], or x-ray or CT[10, 12, 37, 39, 41–43, 45, 47, 48]. Findings were interpreted by a radiologist[33–35, 39, 41, 43, 45, 46, 49], surgeon[12, 33, 35, 36, 42], or an unreported specialist.[10, 11, 37, 38, 40, 44, 47, 48] One study[48] focused on an unstable pelvic fracture.
Table 1
Summary of primary study characteristics
Author
|
Year
|
Country
|
Design
|
Setting
|
N
|
Pelvic Fx
Prevalence
|
Fund
|
COI
|
Civil, et al
|
1988
|
USA
|
Pro
|
Trauma C
|
133
|
8 (6.0%)
|
Unclear
|
Unclear
|
Grant
|
1990
|
UK
|
Pro
|
ED
|
36
|
22 (61.1%)
|
Unclear
|
Unclear
|
Salvino, et al
|
1992
|
USA
|
Pro
|
Trauma C
|
810
|
39 (4.8%)
|
Unclear
|
Unclear
|
Yugueros, et al
|
1995
|
USA
|
Pro
|
ED
|
608
|
59 (9.7%)
|
Unclear
|
Unclear
|
SD. John, et al
|
1996
|
USA
|
Retro
|
Pediatric ED
|
292
|
6 (2.1%)
|
Unclear
|
Unclear
|
Heath, et al
|
1997
|
USA
|
Pro
|
ED
|
82
|
9 (11%)
|
Unclear
|
Unclear
|
Junkins, et al
|
2001
|
USA
|
Pro
|
Pediatric Trauma C
|
140
|
16 (11.4%)
|
Unclear
|
Unclear
|
Duane, et al
|
2002
|
USA
|
Pro
|
Trauma C
|
247
|
45 (18.2%)
|
Unclear
|
Unclear
|
Gonzalez, et al
|
2002
|
USA
|
Pro
|
Trauma C
|
2176
|
97 (4.5%)
|
Unclear
|
Unclear
|
Pehle, et al
|
2003
|
Germany
|
Pro
|
ED
|
979
|
111 (11.3%)
|
Unclear
|
Unclear
|
Waydhas, et al
|
2007
|
Germany
|
Pro
|
Trauma C
|
784
|
93 (11.9%)
|
Unclear
|
Unclear
|
Duane, et al
|
2008
|
USA
|
Pro
|
Trauma C
|
1388
|
168 (12.1%)
|
Unclear
|
Unclear
|
Duane, et al
|
2009
|
USA
|
Pro
|
Trauma C
|
197
|
8 (4.1%)
|
Unclear
|
Unclear
|
Shlamovitz, et al
|
2009
|
USA
|
Retro
|
Trauma C
|
1316
|
109 (8.3%)
|
Unclear
|
Unclear
|
Lagisetty, et al
|
2012
|
USA
|
Retro
|
Pediatric Trauma C
|
504
|
19 (3.8%)
|
Unclear
|
Unclear
|
Lustenberger, et al
|
2016
|
Germany
|
Retro
|
Trauma registry
|
35490
|
7201 (20.3%)
|
Unclear
|
Unclear
|
Majidinejad, et al
|
2018
|
Iran
|
Pro
|
ED
|
3527
|
224 (6.4%)
|
Declared
|
Declared
|
Schweigkofler, et al
|
2017
|
Germany
|
Pro
|
Trauma C
|
147
|
57 (38.8%)
|
Unclear
|
Declared
|
Leent, et al
|
2019
|
Netherlands
|
Pro
|
Trauma C
|
54
|
11 (20.3%)
|
Unclear
|
Unclear
|
Moosa, et al
|
2019
|
Pakistan
|
Pro
|
ED
|
133
|
16 (12.0%)
|
Declared
|
Declared
|
Pro: Prospective study, Retro: Retrospective study, Trauma C: Trauma center, ED: Emergency department, N: The number of total patients included in analysis, Fx: Fracture, COI: Conflict of interest |
Table 2
Summary of primary study characteristics, continued
Author
|
Year
|
Inclusion
|
Index test
|
Reference standard
|
|
|
Age(y)
|
GCS
|
Setting
|
Ask
|
Inspection
|
Palpation
|
Stability
|
Other
|
Modality
|
Radiologist
|
Blind
|
Civil, et al
|
1988
|
-
|
15
|
In-hos
|
+
|
+
|
+
|
+
|
+
|
Xp
|
+
|
?
|
Grant
|
1990
|
-
|
-
|
In-hos
|
-
|
-
|
-
|
+
|
-
|
Xp
|
+
|
?
|
Salvino, et al
|
1992
|
≥ 12
|
≥ 13
|
In-hos
|
+
|
+
|
+
|
+
|
+
|
Xp
|
+
|
-
|
Yugueros, et al
|
1995
|
> 13
|
≥ 14
|
In-hos
|
-
|
-
|
-
|
+
|
-
|
Xp
|
-
|
+
|
SD. John, et al
|
1996
|
< 18
|
-
|
In-hos
|
+
|
?
|
?
|
?
|
?
|
Xp or CT
|
-
|
?
|
Heath, et al
|
1997
|
≥ 18
|
≥ 14
|
In-hos
|
?
|
?
|
?
|
?
|
?
|
Xp
|
-
|
?
|
Junkins, et al
|
2001
|
< 18
|
-
|
In-hos
|
-
|
+
|
+
|
+
|
+
|
Xp or CT
|
+
|
?
|
Duane, et al
|
2002
|
-
|
-
|
In-hos
|
+
|
+
|
+
|
+
|
-
|
Xp
|
?
|
?
|
Gonzalez, et al
|
2002
|
> 14
|
≥ 14
|
In-hos
|
+
|
+
|
-
|
+
|
+
|
Xp or CT
|
+
|
+
|
Pehle, et al
|
2003
|
-
|
-
|
In-hos
|
-
|
+
|
+
|
+
|
+
|
Xp or CT
|
-
|
?
|
Waydhas, et al
|
2007
|
-
|
≤ 13
|
In-hos
|
-
|
-
|
-
|
+
|
-
|
Xp or CT
|
-
|
+
|
Duane, et al
|
2008
|
> 16
|
-
|
In-hos
|
+
|
+
|
-
|
+
|
-
|
Xp or CT
|
+
|
?
|
Duane, et al
|
2009
|
> 16
|
≥ 13
|
In-hos
|
+
|
+
|
-
|
+
|
+
|
Xp
|
?
|
?
|
Shlamovitz, et al
|
2009
|
-
|
-
|
In-hos
|
+
|
+
|
+
|
+
|
-
|
Xp or CT
|
+
|
?
|
Lagisetty, et al
|
2012
|
< 18
|
-
|
In-hos
|
+
|
+
|
+
|
+
|
-
|
Xp
|
+
|
+
|
Lustenberger, et al
|
2016
|
-
|
-
|
Pre-hos
|
?
|
?
|
?
|
?
|
?
|
?
|
?
|
?
|
Majidinejad, et al
|
2018
|
5–64
|
-
|
In-hos
|
+
|
-
|
+
|
-
|
-
|
Xp or CT
|
?
|
?
|
Schweigkofler, et al
|
2017
|
-
|
-
|
Both
|
?
|
?
|
?
|
?
|
?
|
Xp or CT
|
?
|
?
|
Leent, et al
|
2019
|
≥ 18
|
-
|
Pre-hos
|
-
|
-
|
-
|
+
|
-
|
Xp or CT
|
?
|
?
|
Moosa, et al
|
2019
|
≥ 16
|
15
|
In-hos
|
-
|
-
|
+
|
-
|
-
|
Xp
|
+
|
?
|
GCS: Glasgow coma scale, In-hos: In-hospital, Pre-hos: Pre-hospital, Xp: X-ray picture, CT: Computed tomography, ?: Unclear |
Table 3
Demographic and clinical characteristics of patients included in the primary studies
Author
|
Year
|
Age (years)
|
Men
|
Mechanism
|
GCS
|
Severity
|
Civil, et al
|
1988
|
PE + 31/ PE- 34
|
-
|
Fx+: MVA100%
|
15: 100%
|
ISS PE + 11.7/ PE- 8.6
|
Grant
|
1990
|
46, Range (9–95)
|
47%
|
TA 61%
|
≥ 13: 94%
|
TS 15–16: 92%
|
Salvino, et al
|
1992
|
33, Range (12–78)
|
66%
|
MVA 58%
|
-
|
ISS 11
|
Yugueros, et al
|
1995
|
Median 33, Range (14–90)
|
73%
|
-
|
15: 74%
|
median ISS 8
|
SD. John, et al
|
1996
|
10, Range (5 m − 17)
|
55%
|
MVA 55%
|
-
|
-
|
Heath, et al
|
1997
|
Range (18–81)
|
-
|
MVA 79%
|
≥ 14:100%
|
-
|
Junkins, et al
|
2001
|
Fx + 9.8/ Fx-7.8
|
51%
|
MVA Fx+:69%/ Fx-: 54%
|
-
|
Median ISS Fx+:9/ Fx-: 8
|
Duane, et al
|
2002
|
Fx+:36(17)/ Fx-:34(19)
|
-
|
-
|
Fx + 14.4(2.1)/ Fx- 14.4(2.1)
|
ISS Fx+:11.5 (7.4)/ Fx-:5.9 (6.6)
|
Gonzalez, et al
|
2002
|
36, Range (14–93)
|
62%
|
MVA73%
|
≥ 14:100%
|
-
|
Pehle, et al
|
2003
|
PE+: 40(22)/ PE-: 44(20)
|
71%
|
-
|
PE + 10.7(4.8)/ PE-:10.8(4.6)
|
ISS PE + 42.3(19.6)/ PE- 19.9(15.6)
|
Waydhas, et al
|
2007
|
-
|
71%
|
-
|
9.8 (4.7)
|
ISS 23.3 (17.4)
|
Duane, et al
|
2008
|
Fx+: 41(18)/ Fx-: 39 (17)
|
-
|
-
|
Fx+: 12.3 (4.6)/ Fx-: 13.9 (3.1)
|
-
|
Duane, et al
|
2009
|
34(12)
|
80%
|
MVA 76%
|
12.8(4.1)
|
-
|
Shlamovitz, et al
|
2009
|
36 (20)
|
68%
|
MVA 44%
|
14 (2.6)
|
RTS 10.6 (1.5)
|
Lagisetty, et al
|
2012
|
-
|
-
|
MVA68%
|
< 14: 11.3%
|
-
|
Lustenberger, et al
|
2016
|
43(20) in True-positive
|
66% in TP
|
-
|
11.8(4.4) in TP
|
29.6(14.6) in TP
|
Majidinejad, et al
|
2018
|
32(14)
|
76%
|
-
|
-
|
-
|
Schweigkofler, et al
|
2017
|
46
|
69%
|
TA 51%
|
-
|
-
|
Leent, et al
|
2019
|
49(20)
|
70%
|
TA 63%
|
≤ 13: 46%
|
-
|
Moosa, et al
|
2019
|
37(14)
|
92%
|
-
|
-
|
-
|
GCS: Glasgow coma scale, Fx: Pelvic fracture, PE: Physical examination, MVA: Motor vehicle accident, TA: Traffic accident, TS: Trauma score, ISS: Injury severity score, RTS: Revised trauma score. Data is described as mean (standard deviation) or proportion (%) unless otherwise noted. |
Risk of bias assessment
For patient selection, we evaluated 11 studies[10, 11, 34, 37, 38, 40, 43–45, 47, 48] as having high risk or high concern in applicability (Fig. 2) because of poorly described inclusion criteria, nonreproducible methodology, inappropriate patient selection, or poor exclusion criteria, such as the selective exclusion of patients who did not have a reference standard (x-ray or CT) or complete physical examination data. In two studies,[10, 11] it was not clear when the physical examination was performed. For the index test, we evaluated six studies[11, 37, 38, 45, 46, 48] as having high risk or high concern because the physical examination findings were retrospectively collected or because the index test was poorly described. For the reference standard, we evaluated nine studies[12, 34–36, 39, 41, 43, 45, 46, 49] as having low risk of bias because the readers of the imaging scans were blinded to the physical examination findings or because the reference was based on the radiologist’s findings; otherwise, studies were considered to have high risk of bias and high concern in applicability. Moreover, we evaluated five studies[34–36, 46, 49] as having high concern in applicability because of the reference standard being x-ray only despite CT scan being the current gold standard in trauma diagnosis. In patient flow assessment, we deemed nine studies[10, 34, 37, 38, 40, 43–46] to have high risk of bias because these studies excluded a certain number of patients from analysis without proper reporting. The overall quality of the included studies was low. The details of the assessment are shown in the supplementary file (S-Table 2).
Results of meta-analysis
The summary of the diagnostic accuracy and hierarchical ROC of physical examination for each study is presented in Fig. 3. DOR was 76.8 (95% CI: 37.3–157.9). The calculated pooled sensitivity using HSROC parameters was 0.859 (95% CI: 0.761–0.952) at a given specificity of 0.920 (median value among included studies). The positive and negative likelihood ratios were 10.7 (95% CI: 9.5–11.9) and 0.153 (95% CI: 0.05–0.26), respectively. Given a sensitivity of 0.859, the pooled specificity was 0.923 (95% CI: 0.839–0.988). In a population of 1000 patients with a given pelvic fracture prevalence of 10%, the following was detected: 86 patients (95% CI: 76–95) with true TP, 14 patients (95% CI: 5–24) with FN, 831 patients (95% CI: 755–889) with TN, and 69 patients (95% CI: 11–145) with FP. Findings for different prevalence estimates (5% and 15%) are presented in Table 4.
Table 4
The number of TP, TN, FM, FP patients by physical examination in 1,000 patients
Prevalence
|
5%
|
10%
|
15%
|
TP
|
43 (38–48)
|
86 (76–95)
|
129 (114–143)
|
FN
|
7 (2–12)
|
14 (5–24)
|
21 (7–36)
|
TN
|
877 (797–939)
|
831 (755–889)
|
785 (713–840)
|
FP
|
73 (11–153)
|
69 (11–145)
|
65 (10–137)
|
Sensitivity: 0.859 [95%CI: 0.761–0.952] at fixed specificity as 0.920 |
Specificity: 0.923[95%CI: 0.839–0.988] at fixed sensitivity as 0.859 |
TP: True positive, FN: False negative, TN: True negative, FP: False positive |
The number of patients and 95%CI of TP, FN, TN, FP among the 1,000 trauma patients. |
Net benefit and decision curve analysis
Findings from the decision curve analysis at a fixed specificity of 0.92 are shown in Fig. 4 (A). When the threshold probabilities were set at < 0.008, 0.017, and 0.026 with a 5%, 10%, and 15% prevalence, respectively, the net benefit of imaging was higher than that of physical examination. Otherwise, the net benefit of physical examination was higher than that of any imaging tests.
Subgroup analysis
Subgroup analysis could not be performed owing to a lack of adequate data or the small number of studies. We assessed the heterogeneity of covariates, such as level of consciousness (GCS ≥ 13 only or including GCS ≤ 13). The level of consciousness subgroup analysis revealed that the overall risk of bias and applicability were respectively low and of low concern in the subgroup without patients who have impaired consciousness compared with the subgroup with patients who have impaired consciousness (Fig. 4 and supplementary file). The HSROC parameters captured the between-group heterogeneity (Fig. 4). The DORs were 342.8 (70.8–1659.9) and 43.4 (20.4–92.0) for patients with GCS ≥ 13 (13 studies) and GCS ≤ 13 (7 studies), respectively. The RDOR for subgroup comparisons was 7.9 (1.4–44.6), and the p-value was 0.027 in the likelihood ratio test. The pooled sensitivity for patients with GCS ≥ 13 based on HSROC parameters was 0.933 (0.847–0.998) at a given specificity of 0.920; for patients with GCS ≤ 13, the corresponding value was 0.761 (0.560–0.932), suggesting that sensitivity among patients with GCS ≥ 13 was higher than that among patients with GCS ≤ 13.
For patients with GCS ≥ 13; threshold probabilities of < 0.003, 0.008, and 0.013; and prevalence of 5%, 10%, and 15%, the net benefit of the imaging tests was higher than that of physical examination (Fig. 4B). Otherwise, the net benefit of physical examination was higher than that of imaging tests. For patients with GCS ≤ 13; threshold probabilities of < 0.014, 0.028, and 0.044; and prevalence of 5%, 10%, and 15%, the net benefit of imaging tests was higher than that of physical examination.
Sensitivity analysis
In sensitivity analysis, we excluded the following studies to assess the robustness of the results. For the index test, we excluded studies in which the examination type was not relevant to this review, such as neurogenic examination or rectal examination. Furthermore, we excluded studies in which the reference standard was defined as only x-ray or unclear. From post-hoc sensitivity analysis, we excluded studies that were assessed as having high risk of bias in at least one domain. The relevant findings are shown in a supplementary file (S-Tables 1, 2, and 3). Excluding studies with high risk of bias marginally improved diagnostic accuracy, but other exclusions did not affect any estimates.