Procalcitonin for diagnosis of bacterial pneumonia in critically ill patients during 2009 H1N1 influenza pandemic: a prospective cohort study, systematic review and individual patient data meta-analysis

In 2009, the pandemic influenza A(H1N1) virus infection caused more than 18,000 deaths and was associated with more than 200,000 estimated respiratory deaths worldwide [1]. Although most H1N1 patients in 2009 and 2010 presented with mild febrile illnesses and cough, sore throat and occasional gastrointestinal symptoms, 7% to 25% of patients developed severe respiratory distress requiring admission to ICUs [2‐5]. Bacterial coinfection in ICU patients with 2009 H1N1 influenza infection is reported to be 18% to 34% and is associated with significantly increased mortality [6, 7]. Importantly, the initial clinical features of isolated H1N1 infection cannot be distinguished from additional bacterial coinfection [6]. Antibiotic treatment is not recommended in patients with H1N1 infections, but early antibiotic treatment in patients with community-acquired pneumonia is associated with decreased mortality [8]. A rapid diagnosis at admission of bacterial coinfection or isolated bacterial infection is crucial to early initiation or withdrawal of antibiotic treatment. However, obtaining the results cultures for bacterial pathogens may take several days. As a consequence, researchers in several studies have analyzed the predictive value of biomarkers to distinguish bacterial from isolated viral pneumonia [9, 10].

Procalcitonin (PCT) may be a clinically useful marker for the diagnosis of bacterial infections [11, 12] and may be a useful adjunct for reducing antibiotic usage [13‐15]. However, data on the diagnostic value of PCT, including comparisons with other biomarkers such as C-reactive protein (CRP), in critically ill patients in the setting of 2009 H1N1 influenza infection are rare, and the reported results are inconsistent [16‐20]. Earlier results of studies on PCT in critically ill patients might not be transferable to pandemic H1N1 influenza, because the latter is associated with a remarkable release of proinflammatory cytokines such as interleukin 6 and a defective T-cell response, both of which might affect PCT release and hence discriminatory performance of PCT in diagnosing bacterial infection [21, 22]. Conflicting results of studies of PCT in critically ill patients with suspected 2009 H1N1 influenza are most likely due to the small sample sizes and substantial differences in populations studied. More powerful analyses are needed to address the role of potential confounders of the diagnostic accuracy of PCT, such as antibiotic pretreatment, underlying impairment of the immune system and hospital-acquired infections.

Our aims in the present study were to test the utility of PCT in the diagnosis of bacterial pneumonia in patients admitted to the ICU during the 2009 H1N1 influenza pandemic, including its performance in important clinical subgroups, and to compare its performance with CRP. We performed a prospective cohort study in patients admitted to a 28-bed ICU of a tertiary care hospital with pneumonia during the 2009 and 2010 influenza seasons and then combined our data with individual patient data derived from published studies identified by conducting a systematic literature search and performed a meta-analysis.

By utilizing an individual patient data meta-analytic approach in this study, we show that, among patients admitted to the ICU with pneumonia during the H1N1 pandemics in 2009 and 2010, PCT levels were significantly increased in patients with bacterial pneumonia compared to those with isolated H1N1 pneumonia. The overall discriminatory performance of PCT for identification of bacterial pneumonia was moderate, with an AUC of 0.72, and seemed to be negatively affected by the presence of hospital-acquired infections and immune-compromising disorders. The AUC increased to 0.76 when these conditions were excluded from the analysis, and PCT at a cutoff of 0.5 μg/L had a sensitivity and negative predictive value of 85.5% (95% CI = 73.3 to 93.5) and 82.2% (95% CI = 68.0 to 92.0), respectively.

Patients with 2009 H1N1 infection and potential bacterial coinfection admitted to an ICU are at high risk for mortality [3‐6, 24]. Early differentiation of isolated viral infection from isolated or associated bacterial infection might help in the decision to initiate appropriate antimicrobial treatment or avoid unnecessary treatment. To date, studies evaluating PCT in ICU patients with suspected or confirmed 2009 H1N1 influenza have included between 16 and 52 patients, and, accordingly, they have major inherent limitations [16‐20]. Owing to small sample sizes, differences in inclusion criteria, patient selection and outcome definitions, the discriminatory capacity of PCT appears to be unclear [17, 19], with researchers in some studies reporting good discriminatory performance of PCT [16, 18, 20]. Authors of a recent meta-analysis examined the value of PCT for the detection of bacterial coinfection in patients with influenza [25]. Of note, only two of the six studies included in that meta-analysis exclusively comprised ICU patients. Both of these studies were included in our analyses [16, 18]. Additionally, the authors of that meta-analysis did not use individual patient data, but instead reported only meta-analysis of summary statistics such as AUC, sensitivity and specificity.

Our collaborative approach yielded a sample of 161 patients, which is three times larger than any earlier study on this ICU collective and allowed us to estimate discriminatory performance more robustly than other studies, not only in the overall population but also in important subgroups. We demonstrate improved accuracy when patients with compromised immune systems and hospital-acquired disease are excluded, albeit that this difference did not reach statistical significance, most likely because of the limited sample size. The first finding might be explained by a trend toward overall lower PCT levels in immune-compromised patients [26]. Regarding the latter point, PCT is known to have weaker diagnostic utility in patients with nosocomial pneumonia because hospitalized patients generally have higher exposure to bacterial infections independently of potential pneumonia [27, 28]. Furthermore, we provide a set of cutoffs for PCT for detection or exclusion of bacterial infection in ICU patients with potential 2009 H1N1 influenza. Three of the earlier studies did not report any cutoff, owing to weak overall discrimination of PCT or small sample size, and the two largest studies reported a cutoff of 0.8 μg/L. This cutoff is in between the generally accepted levels in ICU patients with respiratory tract infections [29], and the generalizability might be limited by the small and selected derivation sample. We show that recommended rule-out and rule-in cutoffs (0.25, 0.5 and 1 μg/L) for bacterial infection and, as a consequence, for antibiotic treatment discriminate moderately in patients with suspected 2009 H1N1 influenza infection. In patients with community-acquired disease without immune-compromising disorder, PCT might be helpful in ruling out bacterial infection or coinfection, with a negative predictive value of 82.2% and a negative likelihood ratio of 0.27. However, this negative predictive value still seems too low to use PCT as a stand-alone marker for withholding antibiotic therapy in such critically ill patients. Serial changes in PCT might further improve discrimination of bacterial pneumonia, but we had serial PCT measures in only 40% of our population, which is not sufficient for analysis. Thus, further study is needed to evaluate the clinical benefit of PCT testing and to define its role in an algorithm for antimicrobial therapy in ICU patients with suspected H1N1 influenza pneumonia.

In our patient population, PCT showed better discrimination than CRP for bacterial infection, albeit the difference was of borderline significance because of the small sample size. This finding is in line with a body of evidence supporting the superiority of PCT over CRP for detection of bacterial infection [30]. PCT is upregulated through stimulation by cytokines released in response to bacterial infection, and, in contrast to CRP, is inhibited through interferon γ, which is released in viral infections, thus exhibiting a strong specificity for bacterial infections [31].

Our study has some limitations. Despite the combination of six cohorts, our total sample size is still limited and thus subgroup analyses must be interpreted cautiously. Furthermore, half of the studies were retrospective and the study populations were variably selected, which limit the generalizability of our results and might have introduced bias. For instance, one-third of patients did not have PCT measured within the first 24 hours after admission. However, we show that these patients with missing PCT data were not significantly different from the patients included in the analysis regarding most baseline characteristics. A major limitation, which applies to almost all observational studies evaluating PCT, is that there is no accurate gold standard for the outcome definition of bacterial infection. The intensity of diagnostics have differed across studies and was clinically driven in all studies but ours and in the one by Piacentini et al. [20], who performed systematic microbiological testing on blood, urinary and tracheobronchial secretions in all patients. Nonetheless, microbiological cultures of bacterial pathogens are generally inaccurate, particularly in cases of respiratory infections [32, 33]. Thus, underdiagnosing of bacterial infection might be common and thus might attenuate the resulting accuracy of PCT as a marker. Hence, to definitively assess the value of PCT in patients with possible H1N1 infection, a randomized trial on the clinical outcome of PCT-guided antimicrobial treatment is needed. Furthermore, our findings are not generalizable to other clinical scenarios, such as in the emergency department and cases of interpandemic pneumonia or other viral strains causing pneumonia, because the inflammatory response, including PCT, might differ, depending on clinical severity and the underlying viral pathogen [34]. We do not have information on the time from symptom onset to admission to the ICU, which might affect not only the risk of bacterial coinfection but also absolute PCT levels because the kinetics of PCT differ between viral and bacterial infections [20]. Hence the discriminatory performance of PCT might vary over time since the onset of infection. This factor might lead to underestimation of overall PCT accuracy.

In patients admitted to the ICU with suspected H1N1 pneumonia, PCT is a sensitive marker with good negative predictive value for the identification of bacterial infection and is superior to CRP. The accuracy of PCT was highest in patients with community-acquired disease and in patients who are not immunocompromised. However, a negative predictive value of 82% might not be sufficient to use PCT as a stand-alone marker for withholding antibiotic treatment, and thus further studies are needed to define the clinical benefit of PCT testing in ICU patients during H1N1 epidemics.