Published ahead of print on May 8, 2003, doi:10.1164/rccm.200212-1449OC
© 2003 American Thoracic Society
Diagnosing Pneumonia during Mechanical VentilationThe Clinical Pulmonary Infection Score RevisitedService de Réanimation Médicale; Antenne de Pneumologie; Service de Microbiologie; and Service de Réanimation Chirurgicale, Hôpital Henri Mondor, Créteil, France Correspondence and requests for reprints should be addressed to Christian Brun-Buisson, M.D., Service de Réanimation Médicale, Hôpital Henri Mondor, Assistance Publique-Hôpitaux de Paris and Université Paris-12, 51, Avenue du Mal de Lattre de Tassigny, 94010 Créteil Cedex, France. E-mail: christian.brun-buisson{at}hmn.ap-hop-paris.fr
The clinical pulmonary infection score—original or modified—has been proposed for the diagnosis and management of ventilator-associated pneumonia. In 79 episodes of suspected pneumonia, we prospectively assessed the diagnostic accuracy of the physicians' clinical assessment of probability and of the modified clinical pulmonary infection score, both measured before (pretest) and after (post-test) incorporating gram stains results, using bronchoalveolar lavage fluid culture as the reference test. The pretest clinical estimate was inaccurate (sensitivity 50%, specificity 58%); the mean clinical pulmonary infection score at baseline was 6.5 ± 1.3 (range, 3–9) and 5.9 ± 1.7 (range, 3–9), respectively, for the 40 confirmed and the 39 nonconfirmed episodes (p = 0.07), and only slightly more accurate (sensitivity 60%, specificity 59%) than the clinical prediction. Incorporating the gram stain results of either directed or blind protected sampling increased the diagnostic accuracy (sensitivity and specificity of 85% and 49% and 78% and 56%, respectively) of the clinical score and increased the likelihood ratio for pneumonia of a score of more than six from 1.46 to 1.67 and 1.77. The clinical pulmonary infection score has low diagnostic accuracy; however, incorporating gram stains results into the score may help clinical decision making in patients with clinically suspected pneumonia.
Key Words: pneumonia • hospital infection • clinical diagnosis • mechanical ventilation The clinical assessment of ventilator-associated pneumonia (VAP) is usually based on the presence of fever (core temperature of more than 38.3°C), blood leukocytosis (more than 10,000 per mm3), or leukopenia (less than 4,000 per mm3), purulent tracheal secretions, and the presence of a new and/or persistent radiographic infiltrate. However, these parameters taken separately have limited diagnostic value. Pugin and colleagues combined body temperature, white blood cells count, volume, and appearance of tracheal secretions, oxygenation (PAO2/FIO2), chest X-ray, and tracheal aspirate cultures into a clinical pulmonary infection score (CPIS) as a diagnostic tool for pneumonia (1). They found that a CPIS of more than six was associated with a high likelihood of pneumonia with a sensitivity of 93% and a specificity of 100%. More recently, Singh and colleagues (2) used a modified CPIS within a clinical management algorithm in an attempt to reduce unnecessary antibiotic use in patients in whom VAP was suspected. In this series of patients, a modified score remaining less than six at baseline and at 3 days safely allowed stopping antibiotics. However, the diagnostic value of the CPIS has yet to be confirmed. In addition, the clinical utility of such a score would be higher if it helped clinicians in their decision to initiate or withhold antibiotic therapy in patients clinically suspected of VAP rather than only to confirm or exclude pneumonia after 2–3 days when tracheal aspirate culture results are available. The ability of the score to help discriminate which patients should be treated might be improved if it took into account direct examination of respiratory tract specimens at the initial assessment. Using bronchoalveolar lavage (BAL) fluid culture as the reference standard (3), we therefore conducted a prospective study to test the diagnostic value of the modified CPIS and to assess the contribution of gram stains of respiratory secretions specimens to its diagnostic accuracy in clinically suspected VAP.
Patients The study was prospectively conducted between May 2000 and September 2001 in the medical and surgical intensive care units of Henri Mondor University hospital. Based on nosocomial infection surveillance results, the incidence of VAP in these units ranges between 20 of 1,000 and 30 of 1,000 ventilator days. Patients having received mechanical ventilation for more than 48 hours and in whom VAP was clinically suspected were eligible. Exclusion criteria were age of less than 18 years, poor oxygenation (PAO2/FIO2 of less than 100 mm Hg), unstable hemodynamic condition, human immunodeficiency virus infection, and cytotoxic chemotherapy-induced neutropenia. For each patient studied, these parameters were recorded: age, sex, organ failure, and severity scores, assessed by the simplified acute physiology score and the sepsis-related organ failure assessment score (4, 5), date of intubation or tracheostomy, duration of mechanical ventilation, and antibiotics received before samplings. The attending physicians and nursing staff notified one of the investigators when a patient was clinically suspected of pneumonia, based on the usual clinical, radiologic, and biologic criteria (6). A patient could be included twice when two successive episodes of suspected VAP occurred at least 5 days apart. The study was approved by the ethics committee of the Société de Réanimation de Langue Française, and informed consent was waived, as the study was observational and procedures used followed recommendations form the French Society of Intensive Care for standard clinical practice in suspected VAP.
Sampling Procedures
BAL.
Microbiologic Tests
Estimation of the Clinical Probability of VAP and CPIS Calculation
A modified CPIS (2) was calculated by one of the investigators (Table 2) independently of the physicians in charge, at the three steps of the estimation of the clinical probability of pneumonia, that is, at baseline (CPIS baseline), after obtaining the results of the gram stains of BAL and blind PTC (CPIS gram), and when culture results were obtained (CPIS culture).
Statistical Analysis
Patient Population Sixty-eight consecutive patients in whom VAP was clinically suspected were included prospectively over a 16-month period. Patients included (47 males) were aged 59 ± 15 years and had a mean simplified acute physiology score on intensive care unit admission of 52 ± 21; 42% had a rapidly or ultimately fatal underlying disease, according to MacCabe and Jackson classification. Eleven patients were included twice with a mean of 14 ± 8 days (range, 5 to 30 days) apart between the two episodes of suspected pneumonia, and 79 episodes were studied altogether. The mean duration of mechanical ventilation before suspicion of pneumonia was 11.3 ± 8.2 days (range, 2–41 days) (Figure E1 in the online supplement). Antibiotics were being administered in 46 (58%) suspected episodes, and 33 (42%) episodes were free of antibiotics for at least 48 hours at the time of clinical suspicion. Twelve episodes (15%) were associated with new signs and symptoms of severe sepsis or shock (Table 3) . Pneumonia was confirmed by BAL fluid culture in 40 of the 79 episodes (51%). The rate of positive BAL fluid quantitative cultures was higher in episodes free of antibiotics than in episodes with current antibiotics, whether these were ongoing for more than 72 hours or recently introduced (23 of 33 episodes [70%] versus 17 of 46 episodes [37%], p = 0.015).
Clinical Assessment of VAP by Physicians and from Modified CPIS The clinical probability of pneumonia was estimated by physicians at the time of sampling, and the potential improvement over this estimate provided by the modified CPIS measured at baseline was assessed. Physicians providing their estimates were either senior intensive care unit staff members (28 episodes [37%]) or intensivists in training in their last year of residency (47 episodes [63%]).
At baseline.
The mean modified CPIS baseline (2) calculated at inclusion, using the first five variables of the score, was 6.2 ± 1.5 (range, 3 to 9); it was 6.5 ± 1.3 (range, 3 to 9) and 5.9 ± 1.7 (range 3 to 9), respectively, for the confirmed and the nonconfirmed episodes (p = 0.07) (Figure 2) . A score of more than six at baseline had a poor sensitivity and specificity (60% and 59%, respectively), and was associated with a likelihood ratio of 1.46, whereas a "negative" CPIS (six or less) was associated with a likelihood ratio of 0.68 (Table 4) . However, the median (IQR) probability of pneumonia derived from the modified CPIS differed significantly (p < 0.001) between episodes with a score of more than six (85%; IQR, 69–100%) and those with a lower score (50%; 31–73%) (Figure 3) . Whether or not antibiotics were being administered did not influence the diagnostic accuracy of the baseline score (Table E1 in the online supplement).
Post-test probability: integrating the respiratory specimens gram stains results. Gram stains of BAL and of PTC specimens were positive in 34 of 40 (85%) and 24 of 40 (60%) confirmed episodes and in 10 of 39 (26%) and 4 of 39 (10%) nonconfirmed episodes, respectively. At this step, the physicians' estimate was available for only 73 episodes. The overall proportion of episodes misclassified by physicians' estimate of clinical probability of VAP (post-test) decreased nonsignificantly as compared with the pretest estimate (25 of 73 [34%] versus 34 of 73 [47%], p = 0.2); this proportion also decreased nonsignificantly within the subgroup of confirmed episodes (12 of 40 [30%] versus 20 of 40 [50%], p = 0.11) (Figure 1B). Again, the accuracy of the physicians' estimate was not related to their experience. As expected, adding the gram stains results to the CPIS baseline significantly increased the score: The mean CPIS gram for BAL increased in all episodes to 7.3 ± 2.0 (range, 3 to 11), and there was a significant difference in the score between confirmed (8.2 ± 1.6; range, 3 to 11) and nonconfirmed episodes (6.4 ± 2; range, 3 to 11, p < 0.001; Figure 2). Similarly, the mean CPIS gram increased to 6.9 ± 1.8 (range, 4 to 10) for PTC, with respective values of 7.7 ± 1.6 (range, 4 to 10) and 6.1 ± 1.6 (range, 4 to 9) for confirmed and nonconfirmed episodes (p < 0.001; Figure 2). Of note, the CPIS gram significantly increased after integration of the gram stain of BAL fluid in both confirmed and nonconfirmed episodes, whereas a significant increase was noted for only confirmed episodes using PTC. As compared with the CPIS baseline, the accuracy of the modified CPIS gram to discriminate between confirmed and nonconfirmed VAP thus improved after inclusion of results of gram stains into the score (Table 4). The median (IQR) probability of pneumonia derived from the pretest clinical probability and the operative characteristics of the CPIS gram (Table 4) were 88% (73–100%) for a score of more than six using PTC and 25% (25–61%) for a lower score and 87% (71–98.5%) for a score of more than six using BAL, and only 18% (16–36%) for a lower score (Figure 3). Detailed comparisons between the clinical probability estimate and the CPIS at baseline and after gram staining of either BAL fluid or PTC samples are shown in Figures E2–E4 in the online supplement. Similar to baseline, administration of antibiotic did not influence the diagnostic accuracy of the CPIS gram (Table E1).
Post-test probability: integrating results of the blinded respiratory specimens cultures to the CPIS. The mean CPIS culture of EAsq was 7.2 ± 2 (range, 3 to 11) in all episodes, with respective values of 8.0 ± 1.6 (range, 5 to 11) and 6.2 ± 1.9 (range, 3 to 11) for the confirmed and the nonconfirmed episodes (p < 0.001). The mean CPIS culture of PTC increased to 7.0 ± 1.9 (range, 3 to 11) overall, with respective values of 7.9 ± 1.6 (range, 5 to 11) and 6.0 ± 1.7 (range, 3 to 9) for the confirmed and the nonconfirmed episodes (p < 0.001). Taking BAL fluid culture as the reference standard, the sensitivity and the specificity of the CPIS culture for semiquantitative cultures of EA and for PTC were very similar, respectively, of 82% and 61% for the former and 80% and 64% for the latter.
In this study, we have evaluated the clinical diagnosis of ventilator associated pneumonia, assessed on either the routine clinical estimation at the bedside or the modified CPIS and the contribution of the respiratory specimens gram stains results to the diagnosis of VAP, taking BAL fluid culture as the reference test. As expected from previous studies, the clinical prediction alone was inaccurate to diagnose pneumonia; however, incorporating gram stains results of respiratory tract secretions specimens into the score calculation improved its diagnostic accuracy. Thus modified, the CPIS might be of some help in the clinical decision-making process for initiating antibiotic therapy in suspected episodes of VAP. The diagnosis of pneumonia in mechanically ventilated patients remains a difficult challenge because the clinical signs and symptoms lack both sensitivity and specificity and the selection of microbiologic diagnostic procedure is still a matter of debate (9). The clinical assessment of VAP is usually based on body temperature, blood leukocytes count, purulent tracheal secretions, and chest X-ray. However, each of these signs or symptoms taken separately has limited diagnostic value (6, 10). In 1972, Johanson and colleagues (11) proposed to define pulmonary infection by four criteria: (1) the radiographic appearance of a new or progressive pulmonary infiltrate, (2) fever, (3) leukocytosis, and (4) purulent tracheobronchial secretions; probable infection was diagnosed by three criteria, including fever, leukocytosis, and either a new or progressive radiographic infiltrate or the presence of purulent secretions. In this study however, respiratory tract secretions cultures were found to be nonspecific and were not taken into consideration. In 1991, Pugin and colleagues (1) proposed to combine the clinical signs recorded on the day of the clinical suspicion of VAP (including temperature, leukocytes, tracheal aspirate appearance and volume, chest X-ray infiltrates, and PAO2/FIO2 ratio) to the tracheal aspirate gram stain and culture obtained later, into a CPIS as a diagnostic tool of pneumonia. The cutoff, range, and weight attributed to variables included in the score were determined empirically (1). The score varied from 0 to 12 points; a CPIS of more than six was associated with a high likelihood of pneumonia and had a sensitivity of 93% and a specificity of 100%, taking the BAL fluid culture "bacterial index" as the reference standard. However, this first study describing the CPIS had some limitations: Only 28 patients and 40 episodes were studied, and the "bacterial index" (the sum of the logarithm of all bacterial species recovered) has not been well accepted as a reference test for pulmonary infection. The diagnostic accuracy of the CPIS was assessed in two subsequent small studies using both histology and lung tissue cultures as the reference test (12, 13) and providing conflicting evidence. In the first study, including 25 patients (12), the sensitivity and the specificity of the clinical criteria (infiltrates on the chest radiograph and two of the following: fever, purulent secretions, leukocytosis) were 69% and 75%, respectively; the CPIS was not more accurate (sensitivity 77%, specificity 42%). In the second study, including 38 patients (13), the score appeared to have a higher diagnostic accuracy (sensitivity of 72% and specificity of 85%); however, the score calculation used the tracheal aspirate culture recorded within the 48–72 hours preceding the investigation, information that may not be available routinely. In terms of clinical decision making in patients in whom VAP is suspected, the major problem with the original CPIS score is that it requires respiratory tract secretions cultures, which implies waiting for at least 24–48 hours after sampling. To overcome in part this problem, Singh and colleagues tested in a randomized trial the hypothesis that when the score remained low (i.e., six or less) both at baseline and at 3 days, pneumonia could be reasonably excluded and empiric antibiotics safely stopped (2). In that study, a modified CPIS was calculated at baseline from the five first clinical variables of the score (without microbiologic information), and the CPIS at 72 hours was based on all seven variables of the score, including progression of radiologic infiltrate and tracheal aspirate culture results. A substantial reduction of overall antibiotic use was demonstrated, and outcome of patients who discontinued therapy was improved. Thus, Singh and colleagues (2) actually used the score to help decide which patients, within the subgroup of those having a low suspicion of pneumonia to start with, would not require continuation of empiric therapy. In that study, all patients having a score of more than six received a full antibiotic therapy course. An even more important input of a score such as the CPIS would be to help clinicians in the decision to treat empirically patients clinically suspected of pneumonia and to avoid whenever possible treating patients who do not need antibiotic therapy for this purpose. In our series, taking BAL fluid culture as the reference standard, the modified CPIS (2) did not differ in patients with and without pneumonia (as confirmed by BAL fluid culture) at the initial pretest assessment. Although slightly more accurately predictive of pneumonia than the physicians' subjective clinical probability estimate (Table 4 and Figure 3), the clinical score was thus of little help for the decision to initiate therapy. It should be noted that 16 of 39 of patients (41%) having a modified CPIS baseline of less than six did not have confirmed pneumonia, which suggests that many such patients studied by Singh and colleagues did not actually have pneumonia and likely explains in part why a short course of therapy in that group was associated with a better outcome (2). In our study, the addition of gram stains results to clinical information improved the diagnostic accuracy of the CPIS thus modified. Whereas the CPIS baseline did not differ between patients with or without confirmed pneumonia, the CPIS gram differed significantly between these two subgroups (see Figures E2–E4 in the online supplement). According to the sample taken, the prediction of pneumonia based on a CPIS gram of more than six had a sensitivity of 78–85% and a specificity of 49–56%; accounting for the pretest clinical probability, the median probability of pneumonia estimated from the score increased from 85–88% (73–100%) for a CPIS of more than six using PTC and to 87% (71–98.5%) using BAL, after obtaining the gram stain results. It should be noted that the improvement in diagnostic accuracy provided by incorporating gram stains results in the score was similar using "invasive" (directed BAL) or "noninvasive" (PTC blind) diagnostic procedures: The proportion of episodes misclassified decreased from pretest to post-test with both techniques from 32 of 79 (41%) to 26 of 79 (33%) (Table 4). Although a substantial proportion of episodes remained misclassified at this step, most of these were false positives (20 of 26 with BAL and 17 of 26 with PTC), which is of less concern than false negatives in terms of selection of patients who should receive empiric therapy, provided that therapy is re-evaluated after culture results are obtained. It should be emphasized that despite the improvement in diagnostic accuracy provided by the CPIS modified by gram-staining results, there remained a substantial proportion of episodes misclassified by the score. For example, Figure 3 shows that although the median probability of VAP with a modified CPIS of more than six was over 80%, the probability of pneumonia was still of 25% or 16% for a "negative" modified score, using respectively PTC or BAL. This suboptimal specificity of the score may be insufficient to allow withholding antibiotic therapy safely in many patients with suspected pneumonia and a CPIS of less than six. Providing the results of gram stains to clinicians increased their diagnosis accuracy, with a decrease in the proportion of the misclassified episodes from 34 of 73 (47%) to 25 of 73 (34%). Timsit and colleagues have also reported that the accuracy of clinical diagnosis was improved by providing clinicians the results of direct examination of BAL fluid specimens (14); in this study, the authors reported inordinately high predictive values at this step, of 90% and 98%, respectively, for positive and negative predictive values. Again, the physicians' probability estimate increased significantly within the subgroup of confirmed episodes with high (more than 60%) pretest probability, suggesting that they—appropriately—gave more consideration to positive results of gram stains when having a strong pretest clinical suspicion than to negative results (see Figure E5 online supplement). Finally, the clinical prediction of VAP (based on a post-test clinical probability of more than 60%) had a sensitivity of 70% and a specificity of 61%. Although the proportion of microbiologically confirmed VAP episodes in our series (50%) is consistent with others (3, 15), the internal and external validity of our results may be questioned. Although the accuracy of physicians' prediction appeared satisfactory when taking a clinical probability estimate of 60% as the cutoff, it should be noted that these results were obtained in a center in which emphasis on accurate diagnosis of VAP and on the prudent use of antibiotics is an established policy. Although no measure of reproducibility of the clinical assessment across physicians was applied in our series, the physicians who estimated the probability of pneumonia were either senior intensive care unit staff members or intensivists in training in their last year of learning, and the accuracy of the clinical estimates did not differ across these two categories. However, our results may not be applicable to other centers, where variability in the estimates provided by different clinicians in different settings may be expected (16). The potential superiority of the modified CPIS over a "subjective" clinical prediction is to provide a single cutoff value using a reproducible method for calculation. For example, in patients subsequently confirmed to have pneumonia by BAL fluid culture, the diagnostic sensitivity increased from 70% with clinical estimation alone to 78% with CPIS gram using PTC and 85% with CPIS gram using BAL, without loss in specificity. The score might thus be especially useful to inform clinical judgment in case of low–intermediate clinical probability of pneumonia. To summarize, our study shows that the modified CPIS based on clinical criteria alone recorded on the day of clinical suspicion of pneumonia results in misclassifying approximately one-half of patients suspected of pneumonia and is only slightly more accurate than the routine "subjective" clinical estimation; therefore, this score cannot be recommended as an aid for deciding which patients should receive empiric therapy. When the clinical suspicion of pneumonia is high, the modified CPIS does not perform better than clinical prediction alone. Incorporating the results of specimens gram stain increases the sensitivity of the score and the physicians' diagnostic accuracy. This might help therapeutic decision-making in cases of intermediate probability of pneumonia. A similar approach has been proposed by Blot and colleagues (17), combining direct examination of EA and of blinded PTC specimens. These authors proposed that when the gram stain from the PTC sample was positive, empiric therapy should be started because of the high specificity of this test, whereas a negative gram stain of EA would suggest withholding therapy because of its high sensitivity. In intermediate cases, where the PTC gram stain is negative and EA is positive, no recommendation could be provided (17). In such cases, the CPIS modified by the addition of gram staining results might be helpful to consider in addition to other factors such as severity of infection and specific host's risk factors (18). Our results also show that directed or blinded specimens have similar accuracy, which simplifies sampling procedures, where and when bronchoscopy is considered impractical or not readily available. Our findings re-emphasize the difficulties of the clinical diagnosis of pneumonia in mechanically ventilated patients suspected of VAP. Because the clinical and radiologic signs alone or combined into a score have poor accuracy, the results of direct examination of specimens should be taken into consideration in the diagnostic approach. However, because of its suboptimal specificity, even after adding gram stains results, the modified CPIS should be used cautiously in clinical practice, and further refinements of the clinical scoring approach, perhaps by incorporating other biological markers of infection, are needed to improve its value in the management of patients with a clinical suspicion of pneumonia.
This work was presented in part at the ATS International Conference, Atlanta, GA, May 2002. This article has an online supplement, which is accessible from this issue's table of contents online at www.atsjournals.org Received in original form December 11, 2002; accepted in final form May 4, 2003
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
METHODS
