Erythrocyte Sedimentation Rate and C-Reactive Protein in Acute Inflammation

Meta-Analysis of Diagnostic Accuracy Studies

Ivana Lapić, MSc; Andrea Padoan, PhD; Dania Bozzato, MSc; Mario Plebani, MD

Disclosures

Am J Clin Pathol. 2019;153(1):14-29. 

In This Article

Results

Search Results and Characteristics of the Studies Included

A flowchart of study selection is presented in Figure 1. Table 1 summarizes the characteristics of the included studies.[22–50] Overall, 29 studies were finally selected for meta-analysis, all reporting data on diagnostic sensitivity and specificity of ESR and CRP or providing data for their calculation. Five studies[30,31,34,43,45] also provided diagnostic accuracy data when both tests were above the predefined cutoff value and five[30,31,42,43,45] when either of the two tests was elevated. Eleven studies[24,30–33,36,38,39,43,45,49] applied multiple cutoff values.

Figure 1.

Flowchart of the study selection process.

Thirteen studies had a prospective design,[25,26,28,31–35,39,42,43,47,48] 12 were retrospective studies,[24,27,29,30,36–38,40,44,45,49,50] while four were cross-sectional studies.[22,23,41,46] Twenty studies involved an adult population,[22,24,27–34,37–40,42,43,45,47–49] eight focused on a pediatric population,[23,25,35,36,41,44,46,50] and age was not defined for the patient population in one study.[26]

Sixteen studies were categorized within the group of orthopedic infections,[24–28,30,31,36,38–40,42,44,45,48,49] three studies were included in the rheumatic diseases group,[34,47,50] and the remaining 10 studies were categorized as others.[22,23,29,32,33,35,37,41,43,46] Orthopedic infections group included a total number of 2,941 participants, of whom 955 (32.5%) had a diagnosis of a location-confined or systemic inflammation; the "others" group comprised 3,422 participants, of whom 657 (19.2%) were diagnosed with an inflammatory condition; and the group of studies dealing with rheumatic diseases comprised 896 participants, of whom 233 (26.0%) had confirmed inflammation.

Quality Assessment

The results of multiple-choice questions from the QUADAS-2 questionnaire for each domain are summarized in Figure 2.

Figure 2.

Results of the Quality Assessment of the Diagnostic Accuracy Studies–Revised questionnaire.

A total of 14 studies completely fulfilled the QUADAS-2 criteria.[22,23,25,26,31,32,35,39,40,42,43,45,47,48] Overall, most studies presented a low risk of bias and concerns regarding applicability. However, some potential issues should be considered. The risk of bias in patient selection was identified to be high in six studies[24,29,36,37,41,50] because these were not consecutive studies, included inappropriate patient exclusions, or both, while the study by Xiu-Yu et al[50] did not avoid a case-control design. The study by Bosch et al[24] excluded all cases missing inflammatory marker results or outcome data due to incomplete reporting. Similarly, participants in the study by Leli et al[37] were excluded if a sample for one of the laboratory tests (ie, blood culture, real-time polymerase chain reaction, procalcitonin, CRP, ESR, and WBC count) was not obtained. We assume that by applying these nonrandom exclusion criteria, more patients with or without diagnosis could have been excluded or missed, thus affecting diagnostic sensitivity and specificity of the assessed tests. Concern about the eligibility of included patients was high in one study,[29] while this matter was considered unclear in two studies.[33,37] The index test domain was found to be at high risk of bias in four studies[27,28,49,50] because it was not clear whether the results of the index tests were interpreted without the knowledge of the reference standard. Moreover, in more than 50% of all included studies, it was not clear whether the used threshold for either ESR or CRP was prespecified by the authors before performing the study. Risk of bias regarding the reference standard was found to be high in three studies[27,30,50] because in two of these studies, the index test was not interpreted without the knowledge of the reference standard,[27,50] while in the study by Ghanem et al,[30] the review of the medical record was used as the reference standard, which is not a method that can objectively and absolutely identify the target condition. Flow and timing presented a high risk in four studies[28,34,41,46] because either not all patients were included in the study[28,34,46] or not all patients received the same reference standard.[41]

Meta-analysis

Orthopedic Infections. The median cutoff value in this group was 30 mm for ESR and 10.8 mg/L for CRP. Analyses performed with this fixed cutoff included a total of 16 studies,[24–28,30,31,36,38–40,42,44,45,48,49] with a total of 19 diagnostic accuracy results used for the analyses since the study by Piper et al[42] defined four separated outcomes depending on the implant site. Results for ESR showed a pooled sensitivity of 0.78 (95% confidence interval [CI], 0.66–0.86) and specificity of 0.68 (95% CI, 0.55 to 0.79) Figure 3A, with an area under the summary receiver operating characteristic curve (SAUC) of 0.80 (95% CI, 0.76–0.83) Figure 3B. Pooled positive and negative likelihood ratios for ESR were 2.5 (95% CI, 1.8–3.4) and 0.32 (95% CI, 0.22–0.47). Heterogeneity was elevated, with I 2 being equal to 99% (95% CI, 98%-99%), as estimated by the bivariate model, and the proportion of heterogeneity due to different cutoffs was high (0.50). No covariates were found to be associated with sensitivity or specificity by meta-regression. Analysis of publication bias yielded a nonsignificant regression coefficient (bias = –10.2, P = .232).

Figure 3.

A, Forest plot showing diagnostic accuracy data for erythrocyte sedimentation rate (ESR) reported in the studies from the orthopedic infections group, with pooled sensitivity and specificity data. B, Summary receiver operating curve (SROC) of ESR for the orthopedic infections group. Sensitivity = 0.78 (95% confidence interval [CI], 0.66–0.86); specificity = 0.68 (95% CI, 0.55–0.79); area under the curve = 0.80 (95% CI, 0.76–0.83).

Considering CRP, pooled sensitivity and specificity were 0.79 (95% CI, 0.69–0.87) and 0.70 (95% CI, 0.59–0.79) Figure 4A, respectively; SAUC was 0.81 (95% CI, 0.78–0.84) Figure 4B. Pooled positive and negative likelihood ratios for CRP were 2.7 (95% CI, 2.0–3.6) and 0.29 (95% CI, 0.20–0.43). I 2 was 99% (95% CI, 99%-100%), demonstrating that, also for CRP, a high heterogeneity was present, with an elevated effect of different cutoffs (0.29). No covariates were found to be associated with sensitivity or specificity by meta-regression. A nonsignificant regression coefficient (bias = –5.05, P = .614) was observed for publication bias.

Figure 4.

A, Forest plot showing diagnostic accuracy data for C-reactive protein (CRP) reported in the studies from the orthopedic infections group, with pooled sensitivity and specificity data. B, Summary receiver operating curve (SROC) of CRP for the orthopedic infections group. Sensitivity = 0.79 (95% confidence interval [CI], 0.69–0.87); specificity = 0.70 (95% CI, 0.59–0.79); area under the curve = 0.81 (95% CI, 0.78–0.84).

Furthermore, the meta-analysis modeling with multiple thresholds was applied. Considering ESR, the optimal cutoff was 31.2 mm/h with a corresponding estimated pooled sensitivity and specificity of 0.704 (95% CI, 0.493–0.853) and 0.757 (95% CI, 0.599–0.867), respectively. By emphasizing sensitivity over specificity (choosing λw larger than 0.5—namely, 0.6), results showed a slight improvement for sensitivity (0.783; 95% CI, 0.620–0.889), with a reduction of specificity (0.660; 95% CI, 0.504–0.787). SAUC of the model was 0.793. Results for CRP meta-analysis modeling with multiple thresholds with λw equal to 0.5 showed an optimal cutoff of 12.9 mg/L with an estimated pooled sensitivity and specificity of 0.707 (95% CI, 0.523–0.849) and 0.770 (95% CI, 0.643–0.866), respectively. By using a value of λw = 0.6, the optimal cutoff value for CRP was 8.5, with an estimated pooled sensitivity and specificity of 0.802 (95% CI, 0.676–0.893) and of 0.654 (95% CI, 0.536–0.759), respectively, and a SAUC equal to 0.815.

Others. Analyses performed with the fixed median cutoff value of 25 mm for ESR and 20 mg/L for CRP included a total of 10 studies but 11 evaluations since the study by Ayazi et al[23] dealt with two outcomes. Pooled sensitivity and specificity for ESR were 0.77 (95% CI, 0.62–0.88) and 0.59 (95% CI, 0.41–0.75) Figure 5A, respectively, with a SAUC of 0.75 (95% CI, 0.71–0.79) Figure 5B. Pooled positive and negative likelihood ratios were 1.9 (95% CI, 1.2–3.1) and 0.38 (95% CI, 0.19–0.75), respectively. Heterogeneity was elevated, with I 2 being 97% (95% CI, 95%-99%), but only a very small heterogeneity could be attributable to cutoff differences (0.03). No covariates were associated with sensitivity or specificity by meta-regression. Analysis of publication bias yielded a nonsignificant regression coefficient for bias (bias = 11.6, P = .671).

Figure 5.

A, Forest plot showing diagnostic accuracy data for erythrocyte sedimentation rate (ESR) reported in the studies from the "others" group, with pooled sensitivity and specificity data. B, Summary receiver operating curve (SROC) of ESR for the "others" group. Sensitivity = 0.77 (95% confidence interval [CI], 0.62–0.88); specificity = 0.59 (95% CI, 0.41–0.75); area under the curve = 0.75 (95% CI, 0.71–0.79).

Pooled sensitivity and specificity for CRP were 0.86 (95% CI, 0.67–0.95) and 0.67 (95% CI, 0.34–0.89) Figure 6A, respectively, with a SAUC of 0.86 (95% CI, 0.83–0.89) Figure 6B. Pooled positive and negative likelihood ratios were 2.6 (95% CI, 1.1–6.4) and 0.20 (95% CI, 0.08–0.51), respectively. I 2 was equal to 99% (95% CI, 99%-100%), and the heterogeneity attributable to cutoffs was 0.14. No covariates were associated with sensitivity or specificity by meta-regression. Publication bias analysis was observed to have a nonsignificant regression coefficient for bias (bias = –12.5, P = .744).

Figure 6.

A, Forest plot showing diagnostic accuracy data for C-reactive protein (CRP) reported in the studies from the "others" group, with pooled sensitivity and specificity data. B, Summary receiver operating curve (SROC) of CRP for the "others" group. Sensitivity = 0.86 (95% confidence interval [CI], 0.67–0.95); specificity = 0.67 (95% CI, 0.34–0.89); area under the curve = 0.86 (95% CI, 0.83–0.89).

The meta-analysis modeling with multiple thresholds for ESR yielded an optimal cutoff of 51.6 mm with a corresponding estimated pooled sensitivity and specificity of 0.604 (95% CI, 0.259–0.870) and 0.604 (95% CI, 0.306–0.841), respectively. By emphasizing sensitivity over specificity, estimated specificity results by the model were unsatisfactory. SAUC of the respective model was 0.638. The optimal cutoff for CRP was 20.2 mg/L with a corresponding estimated pooled sensitivity and specificity of 0.833 (95% CI, 0.557–0.952) and 0.703 (95% CI, 0.443–0.875), respectively. By emphasizing specificity over sensitivity with λw equal to 0.6, the best cutoff was 20.2 mg/L, with estimated pooled sensitivity similar to that obtained in the first model (0.833; 95% CI, 0.557–0.952) and specificity (0.703; 95% CI, 0.443–0.875). SAUC of this model was 0.830.

Rheumatic Diseases

Since this group constituted only three studies, it was not possible to form a pooled estimate or perform meta-analysis. Therefore, only descriptive statistics was applied. These three studies dealt with different outcomes—diagnosis of GCA,[34] differentiation of erosive from nonerosive RA[47] and diagnosis of Kawasaki disease[50]—and reported different median cutoff values. Diagnostic accuracy data for ESR show that studies report higher sensitivities than specificities, but values are rather variable. On the other hand, studies by Kermani et al[34] and Shovman et al[47] yielded high diagnostic sensitivities (0.913 and 0.864, respectively) with equally low specificities (0.378 and 0.305, respectively) for ESR while a sensitivity of 0.697 and a specificity of 0.735 were obtained by Xiu-Yu et al.[50] The study by Kermani et al[34] reports equal data for ESR and CRP, with the latter having a sensitivity of 0.864 and a specificity of 0.305. Diagnostic sensitivity of CRP in the two remaining studies was 0.913[47] and 0.690,[50] while specificities were 0.378[47] and 0.727.[50] Moreover, Kermani et al[34] showed that nonsignificant changes of diagnostic sensitivity and specificity (0.808 and 0.412, respectively) can be obtained when both index tests are above the cutoff value for diagnosis of GCA.

Combined use of ESR and CRP

Four studies from the orthopedic infections group,[30,31,42,45] of which the study by Piper et al[42] reported diagnostic accuracy data for four different implant sites, reported diagnostic sensitivity and specificity when either of the two tests was above the cutoff. Obtained pooled sensitivity and specificity values were 0.83 (95% CI, 0.75–0.93) and 0.72 (95% CI, 0.66–0.77), respectively. SAUC was 0.78 (95% CI, 0.74–0.82), while positive and negative likelihood ratios were 3.0 (95% CI, 2.4–3.6) and 0.23 (95% CI, 0.11–0.51), respectively. The study by Pongprasobchai et al[43] dealt also with this issue and observed a sensitivity of 1.0 (95% CI, 0.78–1.0) and a specificity of 0.44 (95% CI, 0.28–0.61).

The five studies dealing with diagnostic accuracy of the combined use of ESR and CRP when both tests were above the cutoff value were from different subgroups (ie, three from orthopedic infections[30,31,45] and one each from rheumatic[34] and others[43]), and thus a pooled estimate or meta-analysis was not performed. While the studies by Greidanus et[31] and Ghanem et al[30] reported similar data for diagnostic sensitivity, both being 0.88 (95% CI, 0.81–0.93 and 0.79–0.98, respectively), and specificity (0.93 with 95% CI, 0.88–0.98 and 0.89 with 95% CI, 0.85–0.92, respectively), the third study from the orthopedic infections group[45] showed a significantly lower sensitivity (0.33 with 95% CI, 0.21–0.48) and a similar specificity (0.84 with 95% CI, 0.71–0.93). Kermani et al[34] reported a sensitivity of 0.808 and specificity of 0.412, while Pongprasobchai et al[43] had a sensitivity of 0.43 (95% CI, 0.21–0.67) and a specificity of 1.0 (95% CI, 0.89–1.0). However, the latter study used significantly higher cutoffs than other studies (ie, 60 mm for ESR and 150 mg/L for CRP).

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....