Portal de Periódicos da CAPES

Clinical prediction rules for pulmonary embolism: a systematic review and meta-analysis

2010; Elsevier BV; Volume: 8; Issue: 5 Linguagem: Inglês

10.1111/j.1538-7836.2010.03801.x

1538-7933

Elisa Ceriani, Christophe Combescure, Grégoire Le Gal, Mathieu Nendaz, Thomas Perneger, H Bounameaux, Arnaud Perrier, Marc Righini,

Acute Ischemic Stroke Management

Journal of Thrombosis and HaemostasisVolume 8, Issue 5 p. 957-970 Free Access Clinical prediction rules for pulmonary embolism: a systematic review and meta-analysis E. CERIANI, E. CERIANI Division of Angiology and Hemostasis, Geneva University Hospital and Faculty of Medicine, GenevaSearch for more papers by this authorC. COMBESCURE, C. COMBESCURE Division of Clinical Epidemiology, Geneva University Hospital and Faculty of Medicine, Geneva, SwitzerlandSearch for more papers by this authorG. LE GAL, G. LE GAL Department of Internal Medicine and Chest Diseases, Brest University Hospital, Brest, FranceSearch for more papers by this authorM. NENDAZ, M. NENDAZ General Internal Medicine, Geneva University Hospital and Faculty of Medicine, Geneva, SwitzerlandSearch for more papers by this authorT. PERNEGER, T. PERNEGER Division of Clinical Epidemiology, Geneva University Hospital and Faculty of Medicine, Geneva, SwitzerlandSearch for more papers by this authorH. BOUNAMEAUX, H. BOUNAMEAUX Division of Angiology and Hemostasis, Geneva University Hospital and Faculty of Medicine, GenevaSearch for more papers by this authorA. PERRIER, A. PERRIER General Internal Medicine, Geneva University Hospital and Faculty of Medicine, Geneva, SwitzerlandSearch for more papers by this authorM. RIGHINI, M. RIGHINI Division of Angiology and Hemostasis, Geneva University Hospital and Faculty of Medicine, GenevaSearch for more papers by this author E. CERIANI, E. CERIANI Division of Angiology and Hemostasis, Geneva University Hospital and Faculty of Medicine, GenevaSearch for more papers by this authorC. COMBESCURE, C. COMBESCURE Division of Clinical Epidemiology, Geneva University Hospital and Faculty of Medicine, Geneva, SwitzerlandSearch for more papers by this authorG. LE GAL, G. LE GAL Department of Internal Medicine and Chest Diseases, Brest University Hospital, Brest, FranceSearch for more papers by this authorM. NENDAZ, M. NENDAZ General Internal Medicine, Geneva University Hospital and Faculty of Medicine, Geneva, SwitzerlandSearch for more papers by this authorT. PERNEGER, T. PERNEGER Division of Clinical Epidemiology, Geneva University Hospital and Faculty of Medicine, Geneva, SwitzerlandSearch for more papers by this authorH. BOUNAMEAUX, H. BOUNAMEAUX Division of Angiology and Hemostasis, Geneva University Hospital and Faculty of Medicine, GenevaSearch for more papers by this authorA. PERRIER, A. PERRIER General Internal Medicine, Geneva University Hospital and Faculty of Medicine, Geneva, SwitzerlandSearch for more papers by this authorM. RIGHINI, M. RIGHINI Division of Angiology and Hemostasis, Geneva University Hospital and Faculty of Medicine, GenevaSearch for more papers by this author First published: 04 May 2010 https://doi.org/10.1111/j.1538-7836.2010.03801.xCitations: 52 Marc Righini, Division of Angiology and Hemostasis, Department of Internal Medicine, Geneva University Hospital and Faculty of Medicine, 24 rue Micheli-du-Crest, CH-1211 Geneva 14, Switzerland.Tel.: +41 22 372 92 94; fax: +41 22 372 92 99.E-mail: Marc.Righini@hcuge.ch AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinked InRedditWechat Abstract Summary. Background: Pretest probability assessment is necessary to identify patients in whom pulmonary embolism (PE) can be safely ruled out by a negative D-dimer without further investigations. Objective: Review and compare the performance of available clinical prediction rules (CPRs) for PE probability assessment. Patients/methods: We identified studies that evaluated a CPR in patients with suspected PE from Embase, Medline and the Cochrane database. We determined the 95% confidence intervals (CIs) of prevalence of PE in the various clinical probability categories of each CPR. Statistical heterogeneity was tested. Results: We identified 9 CPR and included 29 studies representing 31215 patients. Pooled prevalence of PE for three-level scores (low, intermediate or high clinical probability) was: low, 6% (95% CI, 4–8), intermediate, 23% (95% CI, 18–28) and high, 49% (95% CI, 43–56) for the Wells score; low, 13% (95% CI, 8–19), intermediate, 35% (95% CI, 31–38) and high, 71% (95% CI, 50–89) for the Geneva score; low, 9% (95% CI, 8–11), intermediate, 26% (95% CI, 24–28) and high, 76% (95% CI, 69–82) for the revised Geneva score. Pooled prevalence for two-level scores (PE likely or PE unlikely) was 8% (95% CI,6–11) and 34% (95% CI,29–40) for the Wells score, and 6% (95% CI, 3–9) and 23% (95% CI, 11–36) for the Charlotte rule. Conclusion: Available CPR for assessing clinical probability of PE show similar accuracy. Existing scores are, however, not equivalent and the choice among various prediction rules and classification schemes (three- versus two-level) must be guided by local prevalence of PE, type of patients considered (outpatients or inpatients) and type of D-dimer assay applied. Introduction Clinical assessment of the probability of pulmonary embolism (PE) is a crucial step in contemporary diagnostic strategies because the correct interpretation of test results depends on it. For example, the association of a low or intermediate clinical probability of PE with a normal D-dimer ELISA (enzyme-linked immunosorbent assay) test confidently rules out PE, reducing the need for further testing and costs [1]. However, in the presence of a high clinical probability, most clinicians feel that additional tests are necessary because the posttest probability of PE is still high (between 10 and 20%) despite a normal D-dimer result [2]. A normal computed tomographic pulmonary angiography (CTPA) may also safely rule out PE if clinical pretest probability is low or intermediate [2, 3]. The negative predictive value of a negative CTPA is, however, low in patients with a high clinical probability. Therefore, clinical probability assessment is of utmost importance in the diagnostic approach of PE. Until recently, grouping patients into low, intermediate and high probability of PE was done implicitly using global clinical judgment. In the Prospective Investigation of Pulmonary Embolism Diagnosis (PIOPED) study [4], the prevalence of PE in the low, intermediate and high clinical probability categories was 9%, 30% and 68%, respectively. Although several studies confirmed the fair accuracy of implicit evaluation, it has been criticized, mainly because it is not standardized. Therefore, attempts have been made to standardize and render explicit the evaluation of clinical probability using statistically derived scores or clinical prediction rules (CPRs) that are able to provide estimates of the probability of PE based on clinical information. As these scores tend to be tailored to the characteristics of the patients used in the derivation model, they may be less accurate when applied to different sets of patients and must be externally validated. Ideally, outcome studies should demonstrate that patients may be safely managed on the basis of the assessment of the clinical probability they provide [5, 6]. The first prediction rule for PE was reported by Hoellerich et al. in 1986 [7]: this eight-item prediction rule was initially derived and locally validated in a small sample but the following steps of validation were never performed. Within the last two decades, many different rules have been described. The most widely used CPR is the Wells rule, which includes the physician’s judgement of whether an alternative diagnosis is more likely than PE. This criterion, which carries a major weight in the score, is subjective and cannot be standardized. Therefore, several efforts were made to develop entirely objective scores such as the Geneva, Charlotte and Miniati rules. The first fully objective scores required diagnostic tests that are not always available, such as chest X-ray or blood gas analysis on room air. More recent rules are based only on clinical elements. Furthermore, in the last decade the most used rules were modified to increase their usefulness and acceptability for clinicians. Cut-off scores of the three-level rules that stratified patients into three levels of clinical probability (low, intermediate or high) were modified to obtain two-level rules classifying patients in only two categories (‘PE likely/unlikely’ for the Wells score or ‘safe/unsafe’ for the Charlotte rule). Existing scores were simplified by assigning one point to each item instead of the variable number of points per item in the initial rules, to simplify the memorization and computation of the score. This variety makes a rational choice among available scores for assessing clinical probability of PE difficult. The aim of this meta-analysis is to compare the accuracy of the principal CPRs for PE pretest probability estimation and review their level of validation. Methods Search strategy and study selection We systematically searched Medline, Embase and the Cochrane Controlled Trials registry, using the following key words: pulmonary embolism AND (decision tree OR clinical prediction rule OR clinical prediction score OR clinical decision rule OR clinical decision score OR management studies OR outcome studies OR D-dimer). The search was performed for English, French, Italian, Spanish and German language, and completed on 15 July 2008. To ensure a comprehensive literature search, we examined reference lists from retrieved articles and reference literature (guidelines and systematic reviews) and questioned experts in PE diagnostic strategies for possible missing studies. Eligible studies were studies on the diagnosis of PE that used a CPR in the diagnostic work-up of patients with suspected PE. External validation in at least one study was required for inclusion of a given prediction rule in this systematic review. We accepted derivation and validation studies and retrospective and prospective studies. Studies in which the pretest probability was evaluated implicitly were excluded. If key data were missing, we contacted the study authors to request relevant data. Two investigators independently evaluated studies for possible inclusion (E.C., M.R.). For each study, two investigators who were blinded to the study authors and journal in which the studies were published assessed study quality independently and extracted the data on study design and patients characteristics. Disagreements about extracted data were resolved by consensus or by discussion with a third reviewer. Quality assessment and data extraction Methodological quality of included studies was assessed independently by two observers (E.C. and M.R.) using the QUADAS tool (Table 1), which is a quality assessment tool specifically developed for systematic reviews of diagnostic accuracy studies [8]. Regarding the patient spectrum criterion, we considered as representative cohorts that included consecutive patients with a PE prevalence between 15 and 25%. The QUADAS standards also include clarity of the methodological description for the diagnostic tool evaluated (in our case a CPR), an explicit and valid reference criterion standard and blinded interpretation of both CPR and the criterion standard. We considered appropriate the following reference standard tests: negative D-dimer in the presence of low or intermediate clinical probability, ventilation-perfusion lung scan when used and interpreted according to well-accepted criteria, helical computed tomography and pulmonary angiography [9]. The description of the reference standard was considered appropriate if explained with sufficient detail to permit replication of the test. The time between reference standard and index test (CRP calculation) was considered acceptable if < 24 h. To assess the influence of study quality on the results of our analysis, we calculated a global QUADAS score for each study, consisting of the sum of the scores for each individual criterion (score of 1 for criterion met, score of 0 for each criterion not met or if it was unclear whether the criterion was met). Studies with a global score above the median score were considered high quality and those with a score below the median low quality. Table 1. Summary of retrieved studies and their quality evaluation using the QUADAS tool [8] Author, year Score 1.Representative spectrum of patients 2.Selection criteria clearly described 3.Reference standard likely to correctly classify the target condition 4.Time between reference standard and index test short enough to be reasonably sure that the target condition did not change 5.Whole sample or a random selection of the sample, receive verification using a reference standard of diagnosis 6.Patients receive the same reference standard regardless of the index test result 7.Reference standard independent of the index test (i.e. the index test did not form part of the reference standard) 8.Score assessment described in sufficient detail to permit its replication 9.Reference standard described in sufficient detail to permit its replication 10.Pretest probability scored without knowledge of the results of the reference standard 11.Reference standard results interpreted without knowledge of PP 12.Same clinical data available when test results were interpreted as would be available when the test is used in practice 13.Uninterpretable/intermediate test results reported 14.Withdrawals from the study explained Total n/14 Righini et al., 2008 [25] GENEVA RV 1 1 1 1 1 0 1 1 1 1 1 1 1 1 13 Klok et al., 2008 [39] GVA RV. SIMP. 1 1 1 1 1 0 0 1 1 1 1 1 1 1 12 Klok et al., 2008 [26] GENEVA RV; WELLS, 3-level 1 1 1 0 1 0 0 1 1 1 1 1 0 0 9 Gibson et al., 2008 [40] WELLS SIMP. 1 1 1 0 1 0 0 1 1 1 1 1 1 1 11 Goekoop et al., 2007 [41] WELLS, 2-level 1 1 1 0 1 0 0 1 0 1 0 1 0 1 8 Penaloza et al., 2007 [18] WELLS, 3-level; WELLS, 2-level 1 1 1 1 1 0 0 1 1 1 0 1 1 1 11 Christ Study Inv, 2006 [20] WELLS, 2-level 1 1 1 1 1 0 1 1 1 1 1 1 1 1 13 Kearon et al., 2006 [42] WELLS, 3-level 1 1 1 0 1 0 1 1 1 1 0 1 1 1 11 Kline et al., 2006 [15] WELLS, 3-level; CHARLOTTE 0 1 1 0 1 0 0 1 1 1 1 1 0 1 9 Le Gal et al., 2006 [24] GENEVA RV 1 1 1 1 1 0 1 1 1 0 0 1 1 1 11 Ollengerg et al., 2006 [23] WELLS, 3-level; GENEVA 0 0 1 0 1 1 1 1 1 0 0 1 0 0 7 Kline et al., 2006 [43] WELLS, 3-level 0 1 1 0 1 1 1 1 1 1 1 1 0 0 10 Miniati et al., 2005 [16] MINIATI, 3-level, WELLS, 3-level, GENEVA 0 1 1 1 1 1 1 0 1 1 0 1 0 1 10 Perrier et al., 2005 [21] GENEVA 1 1 1 1 1 0 1 1 1 1 1 1 1 1 13 Steeghs et al., 2005 [44] WELLS, 2-level 1 1 1 0 1 0 0 1 1 1 0 1 1 1 10 Runyon et al., 2005 [45] WELLS, 3-level, CHARLOTTE 0 0 1 0 1 1 1 1 0 1 1 1 0 0 8 Bosson et al., 2005 [28] WELLS, 3-level 0 1 1 0 1 0 0 1 0 1 1 1 1 1 9 Kabrhel et al., 2005 [17] WELLS, 2-level, WELLS, 3-level 0 0 1 0 1 1 1 1 0 1 0 1 0 0 7 Wolf et al., 2004 [19] WELLS, 2-level, WELLS, 3-level 0 0 1 0 1 0 1 1 0 1 0 1 1 1 8 Kline et al., 2004 [46] CHARLOTTE 1 0 1 0 1 0 0 1 0 1 0 1 1 1 8 Perrier et al., 2004 [1] GENEVA 1 1 1 1 1 0 1 1 1 1 1 1 1 1 13 Miniati et al., 2003 [47] MINIATI, 4-level 0 1 0 1 1 0 1 1 0 1 1 1 1 1 10 Miniati et al., 2003 [48] MINIATI, 4-level 0 0 1 0 1 0 1 0 0 1 0 1 0 1 6 Aujesky et al., 2003 [22] GENEVA 1 1 1 0 1 0 0 1 1 1 0 1 1 1 10 Kline et al., 2002 v CHARLOTTE 0 1 1 0 1 0 0 1 0 1 0 1 1 0 7 Wells et al., 2001 [13] WELLS, 3-level 1 1 1 1 1 0 0 1 1 1 1 1 1 1 12 Wicki et al., 2001 [30] GENEVA 1 1 1 1 1 0 1 1 1 1 0 1 1 0 11 Sanson et al., 2000 [14] WELLS, 3-level 1 1 1 1 1 1 1 1 1 1 0 1 0 0 11 Wells et al.0, 2000 [29] WELLS, 3-level; WELLS, 2-level 1 1 1 1 1 0 1 1 1 1 0 1 1 0 11 PP, pretest probability. We collected the following data for each study: year of publication, type of CPR evaluated, data collection methods (retrospective or prospective), setting (outpatient, inpatient or both), geographic location, demographic characteristics of studied population (mean age, percentage of women), type of reference standard applied, prevalence of chronic obstructive lung disease, heart failure and cancer, follow-up duration, prevalence of PE in the study population, distribution of patients in each group of pretest probability and prevalence of PE in each category of pretest probability. We considered as acceptable a PE definition which was in line with well-established diagnostic criteria reported in the literature. Venous thromboembolisms (VTE) detected during follow-up in patients in whom anticoagulation was withheld were included in the pooled proportion. As most studies used a 90-day follow-up (mean follow-up period was 106 days, median value was 90 days) we did not perform an adjustment for the difference in follow-up length. Primary outcome of the study was to estimate the pooled prevalence of PE in each group of pretest probability for retrieved CPR. Data analysis We determined the 95% confidence intervals (CIs) of the prevalence of PE in the various clinical probability categories using the exact method (see Appendix S1).The prevalence of VTE in each level of clinical probability was separately assessed using the method of the inverse variance on the arcsine transformed proportions [10]. Heterogeneity was tested with the Cochran Q statistic and also quantified by the indicator I2 (ranging from 0% for perfect homogeneity to 100% for extreme heterogeneity) [11]. In case of heterogeneity (Cochran Q test with a P-value < 0.10 or I2 > 50%) a random effect model was used [12]. The heterogeneity factors were explored graphically by plotting the prevalence in each level against the factors. For the most important heterogeneity factors, a sub-group analysis was performed. The pooled results were given for each sub-group. Heterogeneity factors were explored only if the number of studies was greater than five (two- and three-level Wells score and Geneva score). Potential factors were: prospective/retrospective design, multicenter or single center study, blinded or unblinded readers, consecutive patients vs. other selection mechanism, location (North America or Europe), type of reference standard (imaging, follow-up or both), mean age, inpatients and outpatients/outpatients only recruitment and overall prevalence of PE. For the last point a cut off of 20% was chosen because it is the mean and median value of PE prevalence among selected studies. Moreover, PE prevalence among European studies is 20–25%, whereas in American and Canadian studies it is usually between 10 and 20%; thus this cut-off of 20% could highlight different local management of suspected PE. Publication bias was explored by funnel plots and Egger’s test. Potential bias concerned the low risk group, so we plotted the logarithm of the odds ratio of PE (between low group and the other groups) against its standard error. Results Selection The literature search yielded 3752 articles, of which 130 were potentially relevant for our purpose and, therefore, screened in detail. Of these 130 articles, 104 were discarded for various reasons (Fig. 1), and 26 were included. Three additional articles were manually retrieved by asking experts in VTE or by reviewing the reference list of retrieved articles. Twenty-nine studies were finally included in this systematic review (Fig. 1). Figure 1Open in figure viewerPowerPoint Study flow diagram. *score that was derived in a single study, but never validated Some studies evaluated more than one CPR. Nine different CPRs were identified: the Wells score, three-level (low, intermediate and high clinical probability) or two-level (‘PE likely’, ‘PE unlikely’), the Wells Simplified score, the Geneva score, the Revised Geneva score, the Simplified Revised Geneva score, the Miniati score at three or four levels of probability and the Charlotte rule. The three-level Wells score was examined in 14 studies, of which 11 were prospective and three were retrospective. The two-level Wells score was tested in seven studies, of which six were prospective and one was retrospective. The Geneva score was examined in six studies, of which four were prospective and two were retrospective. The number of studies assessing the performance of the other rules was as follows: Charlotte rule, four; Revised Geneva score, three; four-level Miniati score, two; three-level Miniati score, one; Simplified Revised Geneva score, one; and Simplified Wells score, one. Table 2 displays in detail the retrieved CPRs, while Tables 1 and 3 outline the characteristics of the studies in which these rules were evaluated. Pooled results of PE prevalence in each pretest probability for different clinical prediction models are displayed in Table 4. Table 2. Clinical scores for predicting the pretest probability of pulmonary embolism Table 3. Summary of the study characteristics Studies Characteristics* N Mean age, years Sex (%F) COPD (%) HF (%) Cancer (%) Setting Follow up (d) Prevalence (%) Study In each level, % (95%CI) WELLS, 3-level Low Intermediate High Wells et al., a [29] P; D 972 NA NA NA NA NA In-out 90 17 4 (2–6) 21 (17–24) 67 (54–78) Wells et al., b [29] P; V 247 NA NA NA NA NA In-out 90 16 2 (0–7) 19 (12–27) 50 (27–73) Wells et al. [13] P; V 930 50.5 62.7 NA NA 7.2 Out 90 9 1 (0–3) 16 (13–21) 38 (26–51) Wolf et al. [19] P; V 134 58† 54 NA NA NA Out 90 12 2 (0–9) 15 (7–26) 43 (18–71) Kabrhel et al. [17] P; V 607 47.9 74 NA NA NA Out 90 10 4 (2–7) 14 (10–19) 33 (20–48) Bosson et al. [28] P; V 1528 67 54.1 NA NA 13 In-out 0 20 5 (4–7) 28 (24–31) 47 (39–55) Runyon et al. [45] R; V 2477 45 70 NA NA 8 Out 45 6 3 (2–4) 12 (9–15) 33 (20–48) Kline et al. [43] P; V 178 48 NA NA NA NA Out 90 14 3 (1–8) 24 (13–37) 62 (32–86) Penaloza et al. [18] P; V 185 56 58.9 NA NA 11 Out 90 18 3 (1–9) 31 (21–43) 100 (47–100) Kearon et al. [42] P; V 1126 57 65 NA NA 14 In-out 180 15 5 (4–7) 26 (22–31) 55 (43–67) Kline et al. [15] P; V 2302 44.7 69 13 7 8 Out 90 5 3 (2–4) 9 (6–11) 26 (13–42) Sanson et al. [14] P; V 414 51 58 NA NA 10 In-out 0 29 28 (21–36) 30 (25–36) 38 (9–76) Miniati et al. [16] P; V 215 70 64 NA NA 17 NA 365 43 13 (6–23) 54 (45–63) 64 (45–80) Ollengerg et al. [23] R; V 1359 55 55 12 18 15 In-out 365 29 17 (15–21) 36 (33–40) 55 (46–65) Gibson et al. [40] R; V 3298 NA NA NA NA NA In-out 90 21 7 (6–9) 26 (24–27) 58 (50–65) WELLS, 2-level Unlikely Likely Wells et al., a [29] P; D 964 NA NA NA NA NA In-out 90 18 8 (6–10) 41 (35–47) Wells et al., b [29] P; V 247 NA NA NA NA NA In-out 90 15 5 (2–9) 39 (28–52) Wolf et al. [19] P; V 134 58† 54 NA NA NA Out 90 12 3 (1–10) 28 (16–44) Kabrhel et al. [17] P; V 607 47.9 74 NA NA NA Out 90 10 6 (4–8) 23 (17–31) Steeghs et al. [44] P; V 331 51 61.9 13.3 3 0.9 Out 90 14 11 (7–15) 31 (19–45) Christopher Inv. [20] P; V 3306 53 57.4 10.3 7.4 14 In-out 90 20 12 (11–14) 37 (34–40) Penaloza et al. [18] R; V 185 56 58.9 NA NA 11 Out 90 18 9 (5–15) 49 (33–65) Goekoop et al. [41] P; V 879 51 62.6 13 20 1.4 Out 90 13 11 (8–13) 29 (20–39) WELLS SIMP. Gibson et al. [40] R; V 3298 NA NA NA NA NA In-out 90 21 11 (10–12) 36 (33–38) GENEVA Low Intermediate High Wicki et al. [30] P; D 986 62† 55 12 NA 13 Out 90 27 10 (8–13) 38 (33–43) 81 (69–90) Aujesky et al. [22] P; V 259 63† 58 16 10 13 Out 90 30 9 (4–15) 41 (32–52) 59 (43–74) Perrier et al. [1] P; V 965 61 58 10 10 9 Out 90 23 7 (5–9) 34 (29–39) 85 (75–92) Perrier et al. [21] P; V 756 60 60 10 8 10 Out 90 26 7 (5–10) 30 (25–36) 96 (90–99) Miniati et al. [16] R; V 215 70 64 NA NA 17 NA 365 43 50 (30–70) 39 (31–48) 49 (36–62) Ollengerger et al. [23] R; V 998 55 55 12 18 15 In-out 365 29 18 (14–23) 31 (27-35) 44 (36-51) GENEVA REVISED Low Intermediate High Le Gal et al., a [24] P; D 956 60.6 58.2 10.3 9.8 9.2 Out 90 23 9 (6-13) 28 (24-31) 72 (58-83) Le Gal et al., b [24] P; V 749 NA NA NA NA NA Out 90 26 8 (5-12) 29 (24-33) 74 (60-85) Klok et al. [26] R; V 300 NA NA NA NA NA In-out 90 16 8 (5-14) 23 (16-31) 71 (29-96) Righini et al. [25] P; V 1693 59.3 55,3 NA NA NA Out 90 18.2 9 (7-12) 25 (22-28) 84 (71-93) GENEVA REVISED SIMPLIFIED Low Intermediate High Klok et al. [39] R;V 1049 NA NA NA NA NA In-out 90 23 8 (5-11) 29 (26-33) 64 (48-78) CHARLOTTE Unsafe Safe Kline et al. [27] P; D 934 52 68 21 NA 17 Out 180 19 13 (11–16) 42 (35–49) Kline et al. [46] P; V 1339 44 69 NA NA NA Out 90 6 4 (3–6) 22 (15–30) Kline et al. [15] P; V 2302 44.7 69 13 7 8 Out 90 5 4 (3–5) 12 (8–17) Runyon et al. [45] P; V 2477 45 70 NA NA 8 Out 45 6 4 (3–5) 17 (13–22) MINIATI, 4-level Low Intermediate Moderately-high High Miniati et al. [47] P; V 390 68.5 58 NA NA 14 In-out 365 41 3,1 (1–7,1) 24 (16–34) 83 (66–93) 100 (96–100) Miniati et al. [48] P; D 1100 68† 55 NA NA 14.7 In-out 180 40 4 (3–7) 22 (17–27) 74 (62–83) 98 (95–99) MINIATI, 3-level Low Intermediate High Miniati et al. [16] P; V 215 70 64 NA NA 17 NA 365 43 5 (1–12) 42 (31–53) 98 (91–100) NA, not specifically recorded; HF, heart failure; COPD, chronic obstructive pulmonary disease. *Study characteristics refers to study design (P, prospective; R, retrospective; V, validation; D, derivation). †Median age. Table 4. Pooled prevalence of PE in each pretest probability level for the different retrieved CPR Scores Studies Levels Wells 3-level Low I 2 (%) Intermediate I 2 (%) High I 2 (%) All (n = 14) 5.7 (3.7–8.2) 94 23.2 (18.3–28.4) 95 49.3 (42.6–56.0) 71 With prev < 0.20 3.4 (2.7–4.3) 57 18.8 (14.3–23.8) 92 46.8 (38.3–55.5) 73 With prev > 0.20 15.6 (7.6–25.7) 96 35.6 (26.5–45.3) 90 56.9 (51.5–62.2)* 0 In-Out patients 8.5 (4.6–13.4) 95 26.6 (22.8–30.6) 87 54.4 (50.5–58.4)* 40 Out patients 2.9 (2.4–3.5)* 19 15.8 (11.6–20.4) 84 40.8 (30.4–51.6) 61 Wells 2-level Unlikely Likely All (N = 8) 8.4 (6.4–10.6) 81 34.4 (29.4–39.7) 70 North American studies 6.5 (5.3–7.9)* 30 32.9 (23.2–43.3) 82 European studies 11.5 (10.5–12.6)* 49 36.7 (34.0–39.3)* 0 Geneva Low Intermediate High All (N = 6) 12.8 (7.9–18.7) 91 34.7 (31.3–38.2) 55 71.1 (49.6–88.5) 96 Prospective studies 8.0 (6.7–9.4)* 37 35.2 (31.0–39.6) 57 82.1 (65.8–94.0) 90 Retrospective studies 32.1 (7.1–64.8) 91 34.2 (26.6–42.2) 67 45.1 (38.9–51.5)* 0 In-out patients 18.4 (14.4-23.0) 30.9 (26.8–35.2) 43.7 (36.2–51.4) Out patients 8.0 (6.7–9.4)* 37 35.2 (31.0–39.6) 57 82.1 (65.8–94.0) 90 Revised Geneva Low Intermediate High All (N = 6) 9.0 (7.6–10.6)* 0 26.2 (24.4–28.0)* 15 75.7 (69.0–81.8)* 0 Revised Geneva Simplified Low Intermediate High All (N = 1) 7.7 (5.2–10.8) 29.3 (25.7–33.0) 64.3 (48.0–78.4) Charlotte Safe Unsafe All (N = 4) 5.9 (3.3–9.3) 96 22.5 (11.4–36.2) 95 *Heterogeneity was not detected and a fixed effects model was used. The other results were derived from a random effect model (resulting in an increased confidence interval to take into account the heterogeneity). Three-level Wells score The prevalence of PE ranged from 1.3% [13] to 27.9% [14] in the low clinical probability category, from 8.6% [15] to 54.2% [16] in the intermediate probability category and from 33.3% [17] to 100% [18] in the high probability category. Pooled prevalence was 5.7% in the low, 23.2% in the intermediate and 49.3% in the high clinical probability groups. Random effects models were used because heterogeneity was detected in each of the three levels (P < 0.001 for each level, Cochran’s test). Two-level Wells score The prevalence of PE ranged from 3.4% [19] to 12.1% [20] in the ‘PE unlikely’ group and from 22.8% [17] to 48.8% [18] in the ‘PE likely’ group. The pooled prevalence was 8.4% in the ‘PE unlikely’ group and 34.4% in the ‘PE likely’ one. Random effects models were used because heterogeneity was detected (P < 0.001 in both levels). Geneva score The prevalence of PE ranged from 6.5% [1] to 50.0% [16] in the low probability group, from 30.0% [21] to 41.4% [22] in the intermediate group and from 43.7% [23] to 96.3% [21] in the high probability group. The prevalence of PE in the low probability subgroup was exceptionally high in the study from Miniati (50.0%). In the other studies, this prevalence ranged from 6.5% [1] to 18.4%[23]. The pooled prevalence of PE (including the study from Miniati) was 12.8% in the low probability group, 34.7% in the moderate and 71.1% in the high probability group. Random effects models were also used because heterogeneity was detected in the three categories (P < 0.001 in low and high and 0.05 in intermediate, Cochran’s test). Revised Geneva score The prevalence of PE ranged from 7.9% [24] to 9.4% [25] in the low probability group, from 22.8% [26] to 28.5% [24] in the intermediate level and from 71.4% [26] to 84.0% [25] in the high probability category. The pooled prevalence of PE was 9.0% in the low probability level, 26.2% in the intermediate one and 75.7% in the high one. Heterogeneity was not detected in the intermediate level (P < 0.001) nor in the low and high levels (respectively P = 0.92, P = 0.31 and P = 0.43 in the low, intermediate and high classes respectively). Charlotte Rule The prevalence ranged from 3.9% [1