Ventilator-associated pneumonia (VAP) is a serious problem in critically ill patients. Despite great efforts to prevent VAP, it remains very common and is regarded as one of the main quality of care indicators, since it accounts for up to 25% of all nosocomial infection episodes seen in Departments of Intensive Care Medicine.1 No less important is the high risk of recurrence (in the order of 25%), particularly associated to the development of septic shock or adult respiratory distress syndrome,2 and its impact upon patient mortality. Indeed, VAP has been associated with a mortality risk of 35%, which proves even higher when associated among other factors to inappropriate treatment, the development of severe sepsis or bacteremia, or when VAP is late in manifesting or causes respiratory failure.3

The diagnosis of VAP presents still unresolved problems that complicate early and adequate antibiotic treatment capable of reducing the risk of life-threatening situations in the Department of Intensive Care Medicine. The diagnosis is fundamentally clinical, combining radiological criteria (appearance of new infiltrates or progression of already existing infiltrates), evidence of respiratory deterioration (PO2/FiO2), and signs of both local (purulent bronchorrhea) and systemic inflammatory reaction (fever/hypothermia, leukopenia/leukocytosis).4 An adequate diagnosis requires microbiological confirmation, however, and this may take 3–5 days. From the microbiological perspective, gram staining of the respiratory secretions and anticipative rapid cultures can help establish both the diagnosis and the treatment strategy–thereby shortening the response times.5

Procalcitonin (PCT) is a soluble protein composed of 116 amino acids with a sequence identical to that of the calcitonin prohormone, produced under normal conditions by the C cells of the thyroid gland secondary to internal PCT proteolysis. The basal circulating levels are very low (<0.05ng/ml).6 Situations of sepsis, bacterial infections or severe inflammatory reactions increase expression of the CALC-1 gene, favoring the production of PCT in all tissues (including lung, liver, kidney, adipocytes and muscle), and in all differentiated cells of the body. The levels increase quickly 2–3h after the triggering stimulus,7 remaining stable both in vivo and in vitro, and laboratory assay based on immunoanalytical methods allows reliable determination of the plasma concentrations.

Procalcitonin is used in the diagnosis of sepsis, for differentiating bacterial infections from other causes of systemic inflammatory reaction, as a severity marker, and as a prognostic element in the evaluation of the impact of antibiotic treatment and in certain organic infections.8–10 The role of PCT in the diagnosis of VAP has been investigated in a limited number of studies, with conflicting results.11 The joint utilization of clinical criteria, preliminary microbiological data and plasma PCT levels could allow early diagnosis and treatment, thereby contributing to lessen the impact of VAP.

The present study was carried out to determine the usefulness of the plasma PCT levels in diagnosing VAP and to identify possible factors capable of modifying the diagnostic performance of the molecule.

A systematic review and metaanalysis was carried out in accordance with the published guides referred to diagnostic tests,12,13 with the purpose of determining the true usefulness of the plasma PCT levels in diagnosing VAP in adult critical patients.

The initial literature search yielded 1700 articles, which were reduced to 167 publications on considering only prospective clinical studies carried out over the last two decades and involving adult critical patients. Following detailed evaluation, these studies were further reduced to 11 articles, of which 7 were finally selected and included in the metaanalysis.24–30 Of the four excluded articles, two were rejected because valid results for analysis could not be drawn; a third was rejected because it evaluated PCT in the differential diagnosis of pneumonia versus aspiration pneumonitis; and the fourth was excluded because it assessed the role of PCT as a risk variable in the development of VAP among polytraumatized patients within the context of multivariate models–not as a diagnostic tool31–34 (Fig. 1).

Figure 1.

Study selection flow chart.

(0.31MB).

Quality of the studies and publication bias

The QUADAS-2 questionnaire, designed to evaluate the methodological quality of diagnostic test studies, includes four domains that analyze the patients included in the studies, the diagnostic tests evaluated, the references tests, patent flow within the study and the time of the final diagnosis (flow and timing). Each evaluated aspect in turn is classified as presenting low risk of bias, high risk of bias, or indeterminate bias (in the event of insufficient data in the publication to answer the corresponding question).

In the studies included in our review, all the patients were consecutively enrolled, case-control designs were avoided, and there were no inappropriate exclusions. In all the studies the threshold values (PCT cutoff points) were previously specified, and in 5 of the 7 articles the study test was evaluated without knowing the result of the reference test (in the other 2 studies this aspect was not specified). In all of the studies the reference test was interpreted independently of the PCT levels, and in 6 of the 7 publications we considered that the reference test correctly classified the patients. The time interval between conduction of the diagnostic test and the reference test was appropriate, and all patients were subjected to reference testing and included in the subsequent analyses.

On the other hand, in over 85% of the studies (6 out of 7) the patient profile was representative of the population subjected to testing in routine practice; the screening criteria were clearly described; both the study and reference tests were applied to the entire patient sample (avoiding differential and partial verification bias, respectively); both tests (blood PCT assay and reference test) were reproducible; and the same clinical information was available for interpreting the test results as that available in routine clinical practice. Fig. 2 graphically summarizes the risks of bias in each of the four domains of the questionnaire, as well as the application problems that may be found on the basis of the results obtained.

Figure 2.

Graphic summary of the results obtained (%) with the different items of the QUADAS-2 questionnaire.

(0.17MB).

The concordance between the two investigators in assessing methodological quality was high, with a kappa index of 0.86 (95% confidence interval (95%CI) 0.66–1.04). We found no evidence of publication bias either graphically (Fig. 3A) or on applying the Begg (Z: 1.501, p=0.133) or Egger statistics (t: 1.657, df 5, p=0.158).

Figure 3.

(A) Funnel plot for the detection of publication bias. (B) Galbraith plot for the analysis of heterogeneity among studies.

(0.17MB).

Usefulness of procalcitonin in the diagnosis of ventilator-associated pneumonia

The patients with evident clinical manifestations and high plasma PCT levels were at an increased risk of suffering VAP (69% versus 8.8%, OR 8.39, 95%CI 5.4–12.6). The Spearman correlation coefficient between sensitivity and specificity, 0.667 (p=0.102), was not indicative of a threshold effect, thus allowing calculation of the average of the different diagnostic indexes. In this regard, we respectively recorded sensitivity 76.7% (95%CI 69–82), specificity 79.9% (95%CI 74–84), LR+ 4.35 (95%CI 2.48–7.62), LR− 0.26 (95%CI 0.15–0.46) and DOR 17.9 (95%CI 10.1–31.7) (Fig. 4).

Figure 4.

Area under the curve (A) for the diagnosis of VAP. Summary of the values referred to sensitivity (B), specificity (C), positive (LR+) (D) and negative likelihood ratio (LR−) (E) and diagnostic odds ratio (F) of plasma PCT in VAP.

AUC: area under the curve. SE: standard error. Q*: Q* point as global test performance parameter.

(0.76MB).

No heterogeneity was noted either with the Q statistic (chi-squared 5.50, p=0.48) or graphically (Fig. 3B). The graphic representation of the pooled values of the different diagnostic indexes showed some inconsistency, with the exception of DOR–this leading us to perform the analysis of subgroups. We examined diagnostic performance according to the different plasma PCT cutoff points and in patients without exposure to antibiotic treatment and in situations of late VAP. Table 2 shows the results obtained.

Table 2.

Pooled results of the different diagnostic indexes.

	CS (no.)	S (%)	Sp (%)	LR+	LR−	DOR
Total studies	7 (434)	76 (69–82)	79 (74–84)	4.35 (2.48–7.62)	0.26 (0.15–0.46)	17.9 (10–31.7)	Q*=0.808; AuROC=0.87
PCT>0.5	4 (230)	83 (75–89)	60 (52–68)	2.5 (1.11–5.66)	0.28 (0.1–0.76)	9.24 (1.6–53.1)	Q*=0.88; AuROC=0.94
PCT>1	4 (262)	62 (50–74)	79 (73–85)	3.85 (1.8–8.3)	0.48 (0.34–0.66)	10.5 (4.9–22.2)	Q*=0.75; AuROC=0.81
PCT>3	2 (116)	48 (31–66)	96 (89–99)	13.5 (4.2–42.9)	0.55 (0.4–0.75)	28.7 (7.1–116)	Q*=0.81; AuROC=0.88
Without ATB	4 (255)	74 (65–81)	89 (83–94)	6.1 (3.2–11.4)	0.25 (0.10–0.64)	30.6 (14–56)	NA
Late	4 (208)	86 (78–92)	74 (66–81)	3.94 (1.9–8.1)	0.18 (0.11–0.31)	20.25 (9.4–43.5)	Q*=0.852; AuROC=0.91

ATB: antibiotic treatment; AuROC: area under the ROC curve; Sp: specificity; CS: clinical studies; LR+: positive likelihood ratio; LR−: negative likelihood ratio; no.: number of patients; NA: not applicable; DOR: diagnostic odds ratio; PCT: procalcitonin; S: sensitivity.

By generating a regression (metaregression) model, we can analyze the impact of certain cofactors, included in the model as independent variables, upon diagnostic validity (DOR), which constitutes the dependent variable in the model. The exponential transformation of the coefficients of each independent variable indicates the change in the diagnostic performance of the test, and the result is expressed as relative DOR (rDOR). In our study the diagnostic performance of plasma PCT was lowered by previous exposure to antibiotics (rDOR 0.11; 95%CI 0.02–0.069). In contrast, no such effects were recorded for the type of critical patients (rDOR 1.08, 95%CI 0.25–4.74), the methodological quality of the studies (rDOR 1.04; 95%CI 0.59–1.83) or the timing of the development of VAP (rDOR 2.91; 95%CI 0.37–2.28). These results confirm those shown in Table 2 and have a greater explanatory and hypothesis-generating potential than the demonstration of statistical evidence alone.17,21

The rapid and precise diagnosis of VAP continues to pose a challenge, as reported by Rea-Neto et al.35 in a recent systematic review in which important between-observer variability was observed in the use of clinical scales, with diagnostic accuracy that did not increase with the addition of microbiological data. The authors drew these conclusions from studies taking histopathological tests as reference, and considered quantitative microbiological cultures as the diagnostic equivalent to be used in clinical practice.36 This approach, although little used in our setting where qualitative cultures predominate, should be evaluated with a view to incorporating it to routine practice in our Units.5,37 Blood cultures show low performance (in the range of 25%),5 and in this respect the results of cytological tests and the utilization of biomarkers appear promising.11 Specifically, the usefulness of PCT as a prognostic marker and as a guide to treatment appears clear, while its use in the diagnosis of localized infections remains to be defined.

Following the GRADE recommendations,38 diagnostic tests are useful only if they improve patient outcome. The accuracy of these tests is merely an indirect indicator of such clinical relevance. Deducing that increased diagnostic precision is equivalent to improved clinical outcome implies the existence of effective treatment for VAP episodes.

In our study we recorded adequate diagnostic behavior of the PCT levels in VAP in terms of sensitivity, specificity and DOR. All patients with evident clinical manifestations presented a 17-fold increased risk of suffering VAP in the presence of elevated plasma PCT levels versus patients in which the PCT levels were normal or low. These results improve upon those published by Shan et al.39 in a recent metaanalysis of the role of the Pugin (Clinical Pulmonary Infection Score), and are comparable to those obtained by O’Horo et al.40 in another recent study on the role of gram staining in bronchial secretions, referred to the diagnosis of VAP (see summary in Table 3). Fig. 5 offers a graphic comparison of the diagnostic tests according to the results obtained by the mentioned metaanalyses.

Figure 5.

Representation of the sensitivity and specificity of the Clinical Pulmonary Infection Score (CPIS), PCT and gram staining. The lines traced from coordinates (0.0) and (1.1) generate four spaces that allow the inter-comparison of tests. Area 1: better performance; area 2: better only if negative result; area 3: always poorer; area 4: only better if positive result.

(0.09MB).

Despite failure of the statistical tests to detect a threshold effect, diagnostic variation was noted according to the plasma PCT cutoff point. Since these were patients with suspected VAP, and therefore with relatively high pre-test probabilities (≥40%), the use of low cutoff points (PCT<0.5) is associated to a non-negligible number of false positive results and, therefore, to a risk of over-treatment, toxicity and the development of resistances–in turn balanced by a greater proportion of true positive results that lessen the risk of a poor outcome or recurrences. High cutoff points (PCT>3), always in situations of high pre-test probability, imply a larger proportion of false negative results and the risk of late diagnosis, septic complications and poorer outcome.

The diagnostic performance of the plasma PCT levels proved better in patients without previous antibiotic treatment. Their true usefulness in early pneumonias remains to be established, though on the basis of the findings referred to late pneumonia, the performance of the plasma PCT levels in patients with prolonged stays does not appear to decrease. We found no evidence of publication bias capable of contaminating the results obtained, though a certain heterogeneity was noted in relation to the type of critical patient involved.

It is well known that elevated plasma PCT levels are not specific of organic infection, and may be due to infections in other locations. However, the fact that the studies involved patients with a strong suspicion of respiratory infection and mean PCT levels appreciably lower than those published in other critical care contexts makes this possibility less likely.

Knowledge of the values obtained for the different diagnostic indexes, and particularly of the likelihood ratios (LR+/LR−), also allows us to determine (adopting a probabilistic approach) the probability of suffering a VAP episode according to whether the test proves positive or negative, as shown in Fig. 6.

Figure 6.

Pre- and post-test probability according to the results of the metaanalysis of PCT in patients without previous antibiotic treatment.

(0.14MB).

However, the true usefulness of plasma PCT levels in VAP should be established following their incorporation to the diagnostic algorithms and after evaluating their impact upon the management and especially the outcome of critical patients.

Among the limitations of our study, mention must be made of the few published articles, which implies a relatively small number of included patients, and on the other hand causes the statistical tests to be scantly consistent. This fact does not allow us to firmly establish the true impact of factors such as antibiotic treatment, the role of PCT in early or late pneumonias, or the ideal cutoff point. On the other hand, the assessment of publication bias has not been fully standardized in the context of metaanalyses of diagnostic tests–though it appears to be clear that there is a relationship between elevated PCT levels and the risk of suffering VAP in patients with suggestive clinical data. This makes the existence of significant unpublished studies capable of changing the sense of the association rather unlikely. Lastly, it must be mentioned that the type of patients included did not cover the full spectrum of critical patients, and this consequently questions the validness of generalizing the results obtained.

In conclusion, assuming that a rapid diagnosis of VAP episodes must be based on the sum of clinical, radiological and laboratory test findings, together with anticipative microbiological information (e.g., gram staining), plasma PCT levels offer real-time information allowing us to establish the probability of suffering VAP and to start empirical antibiotic treatment capable of palliating its impact upon critical patient outcome. These PCT levels represent a complement to the known use of PCT in guiding antibiotic therapy and as a prognostic marker. It is necessary to clarify the true value of these plasma levels in early pneumonia, the potential impact of previous antibiotic treatments and immune modulating therapies upon diagnostic performance, and their behavior in the different types of critical patients.

This study has received no financial support.

The authors declare that they have no conflicts of interest.