Can serum procalcitonin levels help interpret indeterminate chest radiographs in patients hospitalized with acute respiratory illness?

Subjects

The study was performed in a 520‐bed, general medicalsurgical hospital, using subjects participating in a study of the relationship between biomarkers and microbiologic diagnoses (to be described in a separate publication). Adults 21 years of age admitted with a diagnosis compatible with respiratory tract infection (pneumonia, acute exacerbations of chronic obstructive pulmonary disease [AECOPD], acute bronchitis, asthma, upper respiratory infection, viral syndrome, respiratory failure, and congestive heart failure [CHF] with signs of infection) were recruited during winters of 20082009 and 20092010. Patients were screened within 24 hours of admission; immunosupression, lung abscess, witnessed aspiration, or previous antibiotic use were exclusion criteria. Blood for ProCT measurement was collected in the emergency room prior to antibiotics. Subjects or legal guardians provided written informed consent, and institutional review boards approved the study.

Illness Evaluation

At enrollment, demographic, clinical, and laboratory data were collected. To provide consistency of data, 1 investigator (a clinical pulmonary specialist), after interviewing, examining each subject, and considering the radiographic findings, assigned a primary and secondary clinical admitting diagnosis: pneumonia, AECOPD, asthma exacerbation, CHF, acute bronchitis, viral syndrome or influenza, other respiratory illness, or non‐respiratory illness. These diagnoses were used for the analysis but were not available to treating physicians. Discharge diagnoses, based on International Classification of Diseases, Ninth Revision (ICD‐9) codes, and official results of follow‐up chest radiographs were also recorded.

Radiographic Interpretations

All official admission chest radiographic reports were reviewed by one of the investigators. Radiographic descriptions were noted with attention to the following comments: clear or no acute change from prior radiographs, infiltrates without consolidation, consolidation, edema, atelectasis, and pleural effusion. For our analysis, chest radiographs were categorized into 4 groups: (1) No acute disease (NAD) (ie, clear or no change from prior radiographs); (2) Other definitive radiographic abnormality (atelectasis, edema, or pleural effusion); (3) Indeterminate (findings consistent with pneumonia, but also other processes including atelectasis, edema, scarring but with no favored interpretation [often including the phrase pneumonia cannot be excluded]); and (4) infiltrate (ie, an infiltrate or consolidation most consistent with pneumonia). The pulmonary specialist independently categorized CXRs and provided a definitive radiographic interpretation whenever possible.

Microbiology

Blood cultures, sputum Gram stain and culture were performed by the clinical laboratory. Pneumococcal‐specific urinary antigen and serology were performed as previously described.[22]

Procalcitonin Measurement

Serum ProCT was measured using resolved amplified cryptate emission technology (Kryptor PCT, Brahms, Henningsdorf, Germany). Functional sensitivity is 0.06 ng/mL (normal serum levels are 0.033 0.003 ng/mL).[23, 24, 25] ProCT results were not available to investigators at the time of clinical or radiographic assessments.

Statistical Analysis

Categorical and continuous distributions were compared using Fisher's exact and Wilcoxon tests, respectively, and summarized by proportions and medians, and means and standard deviations. Log‐axis boxplots were used to graphically display the skewed distributions of ProCT. Stacked bar charts were used to depict discrete distributions of diagnoses by ProCT group. Receiver operating characteristic (ROC) curves relating ProCT to chest radiographs were computed, and sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated for ProCT thresholds. Logistic regression was used to estimate odds ratios (OR), and associated P values and confidence intervals, to quantify associations between infiltrate/non‐infiltrate and dichotomized or log‐transformed ProCT. SAS 9.2 (SAS Institute, Inc, Cary, NC) was used for analyses, using 2‐sided 0.05 level tests.

Study Population and Clinical Diagnoses

During 2 winters, 532 subjects admitted for 556 respiratory illnesses were recruited. Two did not have chest radiographs, 16 illnesses were non‐pulmonary, and 10 did not have admission ProCT levels, leaving 528 illnesses for analysis (Table 1). Subjects averaged 65 years of age, predominately lived at home, and a high percentage had underlying diseases. The leading primary admission diagnoses assigned by the pulmonologist were pneumonia and AECOPD at 31% and 27%, respectively. Of the 163 illnesses assigned a primary or secondary admitting diagnosis of pneumonia, infiltrates were identified on the admission radiograph by the pulmonologist in 156 (96%).

Radiographic Classifications

Based on radiology reports, 213 (40%) were classified as NAD, 76 (14%) as other definitive findings, 75 (14%) as infiltrates, and 164 (31%) as indeterminate. The pulmonologist concurred with the radiology report for most NAD (199/213) and infiltrate (67/75) classifications, but only 4 as indeterminate, assigning approximately half the radiologist's indeterminate CXRs to infiltrate and one‐quarter each to NAD and other categories.

Relationship of ProCT With Clinical and Radiographic Features

The relationship between clinical admitting diagnosis and ProCT is shown in Figure 1A. Subjects with pneumonia had a median ProCT of 0.27 ng/ml (interquartile range [IQR] 1.3), significantly greater than subjects with AECOPD, asthma, acute bronchitis, and viral/emnfluenza. Subjects assigned to the other diagnoses category (ie, skin or urinary infections, empyema) generally had higher ProCT values (median 0.70 ng/ml, IQR 4.6).

Figure 1

The relationship between the radiologist's classification and ProCT is shown in Figure 1B. Median ProCT levels were significantly lower in subjects with radiographs classified as NAD compared to those showing infiltrates (P < 0.0001). Those classified as indeterminate in radiology reports had a median ProCT value (0.13 ng/ml) midway between those with NAD (0.08 ng/ml) and infiltrates (0.21 ng/ml). Similarly, illnesses with radiographs classified by the pulmonologist as infiltrates also had significantly higher median ProCT levels (0.33 ng/ml, IQR 1.33) than those classified as NAD (0.08 ng/ml, IQR 0.06; P < 0.0001), and other definitive findings (0.11 ng/ml, IQR 0.17). Notably, duration of symptoms prior to evaluation did not affect the relationship between ProCT level and radiographic classification (data not shown). All subjects with infiltrates according to the radiologist's report received early antibiotics, as did 97% and 80% with indeterminate radiographs and NAD, respectively.

Predictive Value of ProCT for Interpreting Indeterminate Radiographs

We were specifically interested to learn if ProCT could help clinicians interpret the large number of indeterminate radiographs. As a first step, we evaluated the diagnostic accuracy of ProCT for radiographic infiltrates by calculating the ROC curve (Figure 2A), using cases in which radiologist and pulmonologist concurred on the classification as infiltrate (n = 67) or no infiltrate (ie, NAD or other definitive finding; n = 273). For these cases, the ROC had an area under curve (AUC) of 0.80 (P < 0.0001), indicating moderate predictive accuracy of ProCT for the presence of an infiltrate. The OR for an infiltrate increased steadily with higher ProCT thresholds, quadrupling from 3.9 to 15.6 as the threshold increased from 0.05 to 0.50 ng/ml (Table 2). Using the commonly defined 0.25 ng/ml ProCT threshold for which antibiotics have been recommended for respiratory infections, the PPV for an infiltrate was 61% and the NPV 89%.

Figure 2

Next, we analyzed the 164 illnesses with indeterminate radiographs. Of these, the pulmonologist classified 79 as infiltrate (median ProCT 0.29 ng/ml), 40 as NAD (median ProCT 0.08 ng/ml), 41 as other definitive findings (median ProCT 0.10 ng/ml), and 4 as indeterminate. The admitting diagnosis and corresponding median ProCT for these subjects was pneumonia in 78 (0.28 ng/ml), AECOPD in 26 (0.09 ng/ml), acute bronchitis in 30 (0.09 ng/ml), asthma in 11 (0.08 ng/ml), and CHF in 14 (0.10 ng/ml). The ROC AUC for these cases (excluding 4 considered indeterminate by the pulmonologist) was slightly lower (0.72, P < 0.0001) than the prior analysis (Figure 2B), but ProCT retained moderate predictive value for the presence of infiltrates as diagnosed by the pulmonologist on indeterminate radiographs (Table 3).

Since clinical determinations by a single pulmonologist may be subjective, we sought to objectively assess the accuracy of assigned admission diagnoses by analyzing additional data, including bacterial tests, follow‐up radiographs, and final discharge diagnoses. As a surrogate of invasive bacterial infection, we used identification of S. pneumoniae as the outcome, since multiple complementary diagnostic assays were used, thus minimizing the uncertainty of sputum cultures. Overall, we identified 58 pneumococcal infections (Table 4). The pulmonologist's admitting diagnosis of pneumonia captured a higher proportion (52% for any test positive) of S. pneumoniae cases than identification of a radiographic infiltrate by the radiologist (26%), and equivalent to the pulmonologist's radiographic reading (57%) or elevated ProCT alone (57%). There was an absolute 14% increase (27% relative increase) in the detection of S. pneumoniae infections by addition of ProCT to the pulmonologist's diagnosis. A similar analysis limited to the 164 subjects with indeterminate radiographs found that 13 of 19 (68%) of S. pneumoniae diagnoses had been assigned a clinical diagnosis of pneumonia by the pulmonologist.

We also determined how often subsequent radiographs showed evolution to a definitive infiltrate in indeterminate cases that the pulmonologist had classified as either NAD or other finding. Only 3/20 with repeat films showed a definitive infiltrate, none of whom had evidence of bacterial infection, and all 3 had ProCT <0.25 ng/ml. In contrast, 10 of 21 subjects with indeterminate CXRs, that the pulmonologist had classified as an infiltrate, developed definitive infiltrates on follow‐up studies (P = 0.04), and 8 remained indeterminate.

Finally, 93% of those assigned an admitting pneumonia diagnosis by the pulmonologist had primary (71%) or secondary (22%) discharge diagnosis of pneumonia. Of the additional 28 subjects with a discharge diagnosis of pneumonia, for whom the pulmonologist assigned non‐pneumonia admitting diagnoses, none had positive blood cultures or evidence of S. pneumoniae infection, and only one had a follow‐up radiograph showing a definitive infiltrate.