Use of C-reactive protein as a predictor of chorioamnionitis in preterm prelabour rupture of membranes: a systematic review

Authors


Dr RD Trochez-Martinez, Level 4 Maternity Unit, Torbay Hospital, Lawes Bridge, Torquay, Devon TQ2 7AA, UK. Email ruben.trochez-martinez@nhs.net

Abstract

Background  Studies examining the use of C-reactive protein (CRP) as a predictor of chorioamnionitis in preterm prelabour rupture of membranes (PPROM) report highly conflicting results. Despite this, CRP is commonly used for the early diagnosis of chorioamnionitis.

Objective  To determine the diagnostic accuracy of CRP in the detection of chorioamnionitis in women with PPROM.

Design  Systematic review.

Search strategy  Studies were identified from MEDLINE (1966–2006), EMBASE (1974–2006), PubMed and the Cochrane Library (2005) and from reference lists from primary studies and reviews.

Selection criteria  Only studies of good methodological quality that evaluated the diagnostic performance of CRP in chorioamnionitis in women with PPROM were selected.

Data collection and analysis  Positive and negative likelihood ratios (LR) and diagnostic odds ratios (DOR) were calculated. An attempt was made at pooling data for meta-analysis, but this was considered inappropriate due to the significant unexplained heterogeneity between studies.

Main results  There were eight primary studies comprising 610 cases which met the inclusion criteria. There was wide variation in the positive and negative LR and DOR between the studies, with significant lack of precision demonstrated by wide confidence intervals. Three of the studies concluded that CRP was a useful diagnostic tool for chorioamnionitis (DOR ranging from 4.2 to 191.6), although one of them suggested a higher CRP threshold. The other five studies concluded the opposite (DOR ranging from 1.4 to 17.7).

Author’s conclusions  There is no clear evidence to support the use of CRP for the early diagnosis of chorioamnionitis. Further research is required to address the contradictory findings of diagnostic accuracy.

Introduction

The aetiology of preterm birth (PTB) is multifactorial, although infection is thought to play an important role in a high proportion of cases.1 Preterm prelabour rupture of membranes (PPROM) has been strongly associated with both histological chorioamnionitis and prematurity.2

Balancing the benefits of prolonging pregnancy to allow for maturation of the fetus against the risks of infection in cases of PPROM represents a major challenge for obstetricians. Early detection of infection would allow PTB to be expedited. Early infection is not reliably predicted by commonly used laboratory variables such as erythrocyte sedimentation rate, white blood cell count, neutrophil count or vaginal bacterial culture. Clinical signs such as fever and fetomaternal tachycardia usually appear late.3 Amniotic fluid markers, such as bacterial culture, have been proposed but require amniocentesis, which carries logistical difficulties following PPROM and may require repeat procedures.4

C-reactive protein (CRP) is an acute-phase protein secreted by the liver in response to inflammation. It is not specific for infection but is a marker used for the diagnosis of many inflammatory, infective and malignant conditions. CRP has been widely used in obstetrics despite conflicting reports of its benefit, particularly in the early diagnosis of chorioamnionitis in the absence of clinical signs of infection. Existing reviews on the use of CRP as a predictor of intra-amniotic infection are nonsystematic, fail to assess study quality,1,3 accept different reference standards and do not focus specifically on CRP.4

The aim of this systematic review is to evaluate the diagnostic accuracy of CRP in the prediction of chorioamnionitis following PPROM and summarise the current evidence with respect to its clinical use.

Methods

This review was conducted by one reviewer using a prospective protocol based on recommended methods.5–8

The following databases were accessed in search of relevant citations: MEDLINE (1966–2006), EMBASE (1974–2006), PubMed and the Cochrane Library (2005). In MEDLINE, the combination of medical subject headings ‘C-reactive protein’, ‘chorioamnionitis’, ‘chorionitis’, ‘amnionitis’, ‘premature rupture of membranes’, ‘prelabour/prelabor rupture of membranes’ and ‘amniorrhexis’ were combined using the Boolean operators ‘OR’ and ‘AND’ as appropriate, without language restrictions, generating a subset of citations relevant to the research question. This search strategy was adapted as necessary for use in the other electronic databases. Relevant studies were also sought manually in the reference lists of primary papers and reviews. Selected papers in languages other than English were translated by medically qualified professionals unaffiliated to this study.

Full manuscripts of studies evaluating the performance of CRP for the diagnosis of chorioamnionitis following PPROM were scrutinised and subsequently selected for the review if they fulfilled the following inclusion criteria: a) recruited women before 37 completed weeks of gestation with rupture of membranes prior to labour, b) compared directly CRP to histological chorioamnionitis as the reference test and c) had a clearly predefined cutoff for CRP with a description of the kit used.

The methodological quality of the studies was assessed using recognised methods and checklists.9–12 Studies that had a prospective design, used consecutive recruitment and verified the test result with the reference standard in all subjects were considered of good quality. Since placental histology is only available after delivery, blinding was not considered important for quality assessment.

Accuracy results were extracted from all selected articles to construct 2 × 2 tables of positive and negative CRP results (above or below the predefined cutoff, respectively) versus presence or absence of histological chorioamnionitis as defined in the primary study. Accuracy was determined by likelihood ratios (LR) which express how much more frequent the respective test result (positive or negative) is among subjects with disease than among subjects without disease. The diagnostic odds ratio (DOR) is another measure of test accuracy which combines both LR (DOR = LR+/LR−) and expresses how much greater are the odds of having the disease for subjects with a positive test result compared with those subjects with a negative test result.

The heterogeneity of LR and DOR was assessed graphically using forest plots and statistically using χ2 test. Since there was significant graphical and statistical heterogeneity, a random effects model meta-analysis was used, and possible sources of heterogeneity were sought using meta-regression analysis. The following independent explanatory variables defined a priori were used: gestational age (<37 completed weeks or other), study design (prospective or retrospective), recruitment (consecutive or other), test description (adequate or inadequate) and confirmation of diagnosis with reference standard in all subjects (yes or no). The relative diagnostic odds ratio (RDOR), which represents the change in the diagnostic performance of the test per change in the respective co-variable, was calculated for each predefined variable. In addition, the effect of heterogeneity was quantified by calculating inconsistency (I2), which describes the percentage of total variation across studies that is due to heterogeneity rather than chance.

All statistical analyses were performed using Meta-Disc for Windows package.13

Results

Figure 1 shows the search strategy. There were eight primary studies14–21 comprising 610 cases which met the inclusion criteria for this review. Table 1 summarises the salient characteristics of these studies.

Figure 1.

Search strategy.

Table 1.  Features of studies of CRP in the prediction of chorioamnionitis
StudyNumber of womenCRP threshold (mg/l)Gestation (weeks)DesignRecruitmentTest descriptionHistology in all womenExclusions accounted for
Bankowska et al.14671524–36UnknownConsecutiveAdequateYesNo
Farb et al.15242020–36ProspectiveUnknownAdequateYesYes
Fisk et al.165120, 30, 35 and 4026–36ProspectiveConsecutiveAdequateYesYes
Hawrylyshyn et al.175212.520–34ProspectiveConsecutiveAdequateYesNo
Kurki et al.1814712<37UnknownConsecutiveAdequateYesYes
Nowak et al.198012 and 20<35ProspectiveConsecutiveInadequateYesNo
Sereepapong et al.201265>28UnknownConsecutiveAdequateYesYes
Yoon et al.2163723–41ProspectiveConsecutiveAdequateNoNo

There was significant graphical and statistical heterogeneity and inconsistency of both LR and DOR between the studies (LR+χ2= 31.51, P= 0.001, inconsistency I2= 65.1%; LR−χ2= 59.19, P= 0.000, inconsistency I2= 81.4%; DOR χ2=34.03, P= 0.000, inconsistency I2= 67.7%). Table 2 shows the LR+ and LR− of the studies. There seems to be a tendency for the LR+ to increase and the LR− to decrease as the cutoff for CRP increases, but there is wide variation between studies with wide confidence intervals. Table 3 shows the DOR: it shows a tendency for the DOR to increase with increasing CRP threshold, but again, variation between studies is high with lack of precision demonstrated by wide confidence intervals.

Table 2.  Positive and negative LRs (random effects model)
StudyCRP cutoff (mg/l)LR+95% CILR−95% CI
  1. Heterogeneity (χ2) for LR+ is 31.51, P= 0.001; heterogeneity (χ2) for LR− is 59.19, P= 0.000.

  2. Inconsistency (I2) for LR+ is 65.1%; inconsistency (I2) for LR− is 81.4%.

Bankowska et al.14151.350.55–3.290.910.70–1.19
Farb et al.15202.531.14–5.600.290.05–1.73
Fisk et al.16202.621.01–6.800.620.41–0.93
Fisk et al.16304.901.24–19.330.590.41–0.85
Fisk et al.16358.401.18–59.770.630.46–0.86
Fisk et al.164016.321.01–262.630.640.49–0.85
Hawrylyshyn et al.1712.523.003.35–157.970.120.04–0.35
Kurki et al.18121.881.53–2.300.120.03–0.47
Nowak et al.19123.561.64–7.700.200.10–0.38
Nowak et al.19202.141.29–3.520.150.06–0.37
Sereepapong et al.2051.330.93–1.900.760.53–1.10
Yoon et al.2173.801.46–9.890.530.36–0.79
Table 3.  DOR (random effects model)
StudyCRP cutoff (mg/l)DOR95% CI
  1. Heterogeneity (χ2) is 34.03, P= 0.000.

  2. Inconsistency (I2) is 67.7%.

Bankowska et al.14151.480.47–4.69
Farb et al.15208.670.79–95.09
Fisk et al.16204.251.15–15.65
Fisk et al.16308.311.64–42.17
Fisk et al.163513.331.57–113.00
Fisk et al.164025.361.40–459.50
Hawrylyshyn et al.1712.5191.6718.59–1975.80
Kurki et al.181215.503.54–67.84
Nowak et al.191217.785.20–60.80
Nowak et al.192014.404.09–50.74
Sereepapong et al.2051.750.85–3.60
Yoon et al.2177.122.04–24.87

Analysis of the variation in DOR in relation to diagnostic threshold revealed that there was some variation (Moses’ regression model coefficient b= 0.019), but this was nonsignificant (P= 0.92), yielding an asymmetrical receiver operating characteristic curve (Figure 2) and suggesting that the observed heterogeneity may not be attributed to threshold effect.

Figure 2.

Asymmetric SROC curve. This summarises graphically the results of the studies. The ideal test will have the most convex part of the curve closest to the top left corner, where sensitivity = specificity, represented by the Q* index. SROC, summary receiver operating characteristic curve. AUC, area under the curve; SE, standard error.

By excluding those studies which accepted more than one CRP threshold16,19 or by excluding those with extreme cutoff values of CRP,16,20,21 no significant difference was found in the results.

The meta-regression analysis showed that test accuracy was not affected by gestational age (P= 0.30, RDOR = 0.26, 95% CI 0.01–5.16), study design (P= 0.09, RDOR = 0.15, 95% CI 0.01–1.58), recruitment (P= 0.63, RDOR = 1.91, 95% CI 0.07–51.57), adequacy of test description (P= 0.64, RDOR = 2.29, 95% CI 0.03–177.81) or whether the reference test was performed in all subjects (P= 0.84, RDOR = 1.50, 95% CI 0.01–270.32).

Discussion

There is poor agreement among clinicians as to the gold standard for diagnosing chorioamnionitis whether clinical or histological. To our knowledge, no other review of CRP in women with chorioamnionitis and PPROM has been conducted which is systematic, addresses study quality in good detail and focuses specifically on CRP across different reference standards. Amniorrhexis relation to chorioamnionitis has been addressed previously, but this was an indirect comparison relating CRP with fetal bacteraemia and positive amniotic fluid culture22 and subsequently relating these to histological chorioamnionitis,23 and for this reason, these studies were not included in this review. We chose histological chorioamnionitis because it is strongly related to infection of the chorioamnion and to PTB.2 In addition, clinical chorioamnionitis is usually easier to diagnose and management decisions might be delayed by further testing.

The variation in the reported value of CRP for the diagnosis of chorioamnionitis is wide. There are several laboratory techniques to measure CRP including nephelometry, turbidimetry and radial immunodiffusion or electroimmunodiffusion. The studies included in this review that described the CRP method used either nephelometry or turbidimetry, which seem to correlate well.24 Although some interlaboratory variation may be expected, this is unlikely to be of clinical significance. High-sensitivity CRP (hs-CRP) test is a novel type of CRP, which has significantly lower limits of detection (range 0.5–10 mg/l) and is most often used to help predict a healthy person’s risk of cardiovascular disease.25 The values of hs-CRP test have been shown to increase earlier than that of conventional CRP in women with PPROM,26 but its correlation with chorioamnionitis or indeed its use in obstetrics has been little investigated.

Most studies in this review had sensitivities of 50–80% for a false-positive rate of 10–30% (Figure 2). How useful is the LR+ depends on a number of factors: a) how it compares with other tests; b) how easy it is to combine with other clinical or laboratory variables; c) what are the risks of intervention in response to a false-positive test, in this case, unnecessary elective PTB and d) what are the risks of no intervention due to a false-negative test, in this case, fetomaternal infectious morbidity and mortality. Therefore, while the sole use of CRP may be unhelpful, its relatively high LR+ means that it is still an adjunct to diagnosis.

It would be inappropriate to pool data for meta-analysis because of the statistically significant heterogeneity between studies, which remained unexplained. We did not test for publication bias, but given the wide variation between published results, this seems unlikely.

The quality of systematic reviews is dependent upon the quality of studies included. The strict selection criteria used for this review ensured that the quality of included studies was good. An important weakness in the studies examined was the wide variation in test thresholds as well as the small sample sizes, with wide confidence intervals.

Fisk et al.16 was the only study which fulfilled all six methodological quality criteria, particularly accounting for exclusions, which might have been a source of bias in other studies14,17,19,21 (Table 1). Fisk et al.16 concluded that there was a significant association between an elevated CRP level and histological chorioamnionitis but acknowledged that the usefulness of CRP as a diagnostic test was limited due to overlap between infected and noninfected cases, particularly when generally accepted normal levels of CRP such as 20 mg/l or lower were used. They proposed using higher CRP thresholds of 30, 35 and 40 mg/l for single estimations to improve specificity but also accepted serial estimations of CRP of >20 mg/l as highly predictive of infection. Two other studies supported the use of CRP as a predictor of infection.17,19 Nowak et al.19 used two different CRP thresholds (12 mg/l and 20 mg/l) and considered an increasing value above 12 mg/l on two consecutive estimations as highly reliable.

Five studies14,15,18,20,21 concluded that the overall performance of CRP in the diagnosis of chorioamnionitis was poor. All these studies fulfilled at least four of the six methodological quality inclusion criteria, and there were no significant differences in the populations studied or in the overall quality of the studies.

There is disagreement on what should be considered the normal CRP values in pregnancy and whether the normal concentration of CRP increases as pregnancy progresses.3 This may reflect the different thresholds chosen arbitrarily by different investigators and might play a role in the difficulty in determining the usefulness of CRP for the diagnosis of chorioamnionitis.

Conclusions and recommendations

There is no clear evidence to support the use of CRP as an early diagnostic test of chorioamnionitis following PPROM. As with other diagnostic aids, its predictive value has to be weighted in the context of individual clinical situations together with other clinical predictors, bearing in mind the significant risks of intervention or no intervention in response to a false-positive or false-negative result, respectively. Although there is an association between an elevated CRP level and histological chorioamnionitis, the use of commonly accepted CRP thresholds might be misleading. If single measurements were to be used, it would seem appropriate to consider higher cutoff levels of at least 30 mg/l. For serial CRP estimations, levels of 20 mg/l or above seem predictive of infection. The use of serial CRP measurements seems promising but requires further investigation.

An agreement on the normal values of CRP at different gestations in pregnancy is overdue. From there, further research with larger populations is needed to determine the usefulness of CRP for the diagnosis of chorioamnionitis.

Acknowledgements

The authors thank Barrie Undy for his help with statistics and Beata Klepacka and Tilo Asmussen for their help with translation.

Ancillary