Elsevier

Journal of Infection

Volume 72, Issue 2, February 2016, Pages 143-151
Journal of Infection

Procalcitonin (PCT) levels for ruling-out bacterial coinfection in ICU patients with influenza: A CHAID decision-tree analysis

https://doi.org/10.1016/j.jinf.2015.11.007Get rights and content

Highlights

  • CHAID model illustrates multilevel interactions among risk factors to identify stepwise pathways for detecting bacterial coinfection.

  • PCT was the strongest variable associated with coinfection diagnosis by CHAID model analysis.

  • Low serum levels of PCT (<0.29 ng/mL) in patients without shock were an accurate predictor for ruling out coinfection.

  • PCT was a more accurate marker than CRP for coinfection diagnosis.

Summary

Objectives

To define which variables upon ICU admission could be related to the presence of coinfection using CHAID (Chi-squared Automatic Interaction Detection) analysis.

Methods

A secondary analysis from a prospective, multicentre, observational study (2009–2014) in ICU patients with confirmed A(H1N1)pdm09 infection. We assessed the potential of biomarkers and clinical variables upon admission to the ICU for coinfection diagnosis using CHAID analysis. Performance of cut-off points obtained was determined on the basis of the binominal distributions of the true (+) and true (−) results.

Results

Of the 972 patients included, 196 (20.3%) had coinfection. Procalcitonin (PCT; ng/mL 2.4 vs. 0.5, p < 0.001), but not C-reactive protein (CRP; mg/dL 25 vs. 38.5; p = 0.62) was higher in patients with coinfection. In CHAID analyses, PCT was the most important variable for coinfection. PCT <0.29 ng/mL showed high sensitivity (Se = 88.2%), low Sp (33.2%) and high negative predictive value (NPV = 91.9%). The absence of shock improved classification capacity. Thus, for PCT <0.29 ng/mL, the Se was 84%, the Sp 43% and an NPV of 94% with a post-test probability of coinfection of only 6%.

Conclusion

PCT has a high negative predictive value (94%) and lower PCT levels seems to be a good tool for excluding coinfection, particularly for patients without shock.

Introduction

Community-acquired respiratory coinfection (CARC) in patients with viral pneumonia caused by influenza A(H1N1)pdm09 has been recognised as a major cause of influenza-related death.1, 2 Since 2009, more than 2000 patients have been admitted to intensive care units (ICUs) for cases of severe influenza in Spain.3, 4, 5, 6 To reduce mortality and morbidity in this vulnerable patient population, early administration of antibiotic (AB) treatment is recommended in patients suspected of having CARC. Yet, different studies have found that both clinical signs and symptoms, and commonly used laboratory markers, are unreliable for assessing the risk of CARC in this particular subset of patients admitted to the ICU.7, 8 As a result, in clinical practice, currently most patients receive antibiotics (AB); according to the Spanish Society of Intensive Care Medicine (SEMICYUC) database, 100% of critically ill patients with influenza were treated with empirical AB upon admission to the ICU.3, 4, 5, 6 However, only in 20% of the cases was CARC eventually confirmed.6 More accurate, prompt diagnostic tools to rule out CARC could potentially limit AB overuse and subsequently reduce unnecessarily high costs, potential side-effects and the development of multi-drug resistance infections.

Procalcitonin (PCT) is a biomarker reported in the event of bacterial infection and adequately correlates with severity and outcome of lower respiratory tract infections (LRTI). In addition, AB guidelines, based on PCT levels, have been seen to significantly reduce AB administration in patients with LRTI in both emergency departments and ICUs.9, 10, 11 However, there is a lack of large-scale clinical data demonstrating the utility of PCT as an accurate biomarker for guiding AB use in patients with severe influenza pneumonia with CARC. Four minor studies have suggested that the use of PCT cut-off ranges in patients infected with influenza A(H1N1)pdm09 may estimate the probability of developing CARC.12, 13, 14, 15 Nevertheless, these studies are limited in terms of small sample size (between 16 and 100 patients) and different criteria for patient selection, cut-offs, and outcome. The aim of this study was to evaluate the potential role of PCT in ruling out the presence of CARC in a large, well-defined cohort of influenza A(H1N1)pdm09-infected patients. Our hypothesis was that the PCT algorithms recommended for AB administration9 could be different from those observed in patients with primary viral pneumonia caused by influenza A(H1N1)pdm09. The main objective of our study was, therefore, to define which variables upon admission to the ICU could be related to the presence of CARC using CHAID (Chi-squared Automatic Interaction Detection) decision-tree analysis16, 17, 18 in order to maximise the probability of a correct diagnosis.16, 17, 18, 19

Section snippets

Study design and patient population

This is a secondary analysis from a prospective, observational cohort study conducted across 148 ICUs in Spain between June 2009 and April 2014. Data were obtained from a voluntary register created by the SEMICYUC.3, 4, 5, 6

The study was approved by the Joan XXIII University Hospital Ethics Committee (IRB NEUMAGRIP/11809). Patients remained anonymous. The requirement for informed consent was waived due to the observational nature of the study and the fact that this activity was considered an

Study population characteristics

Two thousand one hundred thirty-two patients with rt-PCR-confirmed influenza A(H1N1)pdm09 virus infection were admitted to the 148 ICUs in the three periods considered. Nine hundred and seventy-two of them (45.6%) had PCT levels measured upon admission to the ICU and were the population of analysis. Of these patients, 581 (59.8%) were men, and the median was 51 years of age. The mean APACHE II and SOFA scores were 16.5 and 6.4 points, respectively. Patients had a high comorbidity burden (see

Discussion

The main findings of our study conducted in patients with confirmed A(H1N1)pdm09 virus infection were threefold. First, the CHAID model illustrates multilevel interactions among risk factors to identify stepwise pathways to detect CARC. The proposed CHAID model distributed serum PCT into the first level of partition above other variables as the strongest variable associated with CARC. Second, low serum levels of PCT in patients without shock were an accurate predictor for excluding the presence

Conclusion

In patients admitted to the ICU with confirmed influenza A(H1N1)pdm09 infection and without shock, low serum levels of PCT might be a good tool for ruling out the presence of CARC. However, while PCT can assist physicians in developing patient-specific therapeutic plans, such as antibiotic prescription, it is important to highlight that biomarkers are tools that should never replace physician decision-making.

Conflicts of interest and source of funding

Dr A. Rodriguez has participated as a speaker at scientific meetings or courses organised and financed by various pharmaceutical companies including Astellas, Pfizer, Novartis, Gilead, Thermo Fisher, bioMèrieux, and Roche. AR is the principal investigator for research grants of Carlos III National Institute of Health.

I. Martin-Loeches has participated in advisory boards organized and financed by Bayer, Clinigen, Pfizer and Johnson & Johnson. IML is the principal investigator for research grants

Contributors

AR, FXAJ, IML, ED, JSV, MR and PS conceived and designed the study. All authors, apart from MR and PS, contributed to acquisition and local preparation of the constituent database.

FXAJ (Master's degree in epidemiology and statistics) has carried out the analysis of the data and the CHAID model.

AR, ST, IML and FXAJ contributed to database creation and standardisation, design of statistical analyses, and data analysis.

AR, FXAJ, IML, JSV, BS, MR, ED, RZ, and PS interpreted the data and wrote the

Acknowledgements

This study was endorsed by the SEMICYUC (Spanish Society of Intensive Care Medicine). We thank the GETGAG (Influenza A/H1N1 Working Group from SEMICYUC) investigators for their contributions to the research. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the SEMICYUC.

References (29)

  • N. Bhat et al.

    Influenza-associated deaths among children in the United States, 2003–2004

    N Engl J Med

    (2005)
  • J. Rello et al.

    Intensive care adult patients with severe respiratory failure caused by Influenza A (H1N1)v in Spain

    Crit Care

    (2009)
  • I. Martin-Loeches et al.

    Pandemic and post-pandemic influenza A (H1N1) infection in critically ill patients

    Crit Care

    (2011)
  • B.A. Cuhna et al.

    Swine influenza (H1N1) pneumo nia in hospitalized adults. Chest film findings

    Heart Lung

    (2011)
  • Cited by (0)

    1

    SEMICYUC/GETGAG Working Group investigators listed in supplemental digital content (Appendix 1).

    View full text