Current issues of ACP Journal Club are published in Annals of Internal Medicine


Prognosis

Pneumonia in adults was predicted by fever, high pulse rates, rales, decreased breath sounds, and asthma

ACP J Club. 1991 Mar-Apr;114:58. doi:10.7326/ACPJC-1991-114-2-058


Source Citation

Heckerling PS, Tape TG, Wigton RS, et al. Clinical prediction rule for pulmonary infiltrates. Ann Intern Med. 1990;113:664-70.


Abstract

Objective

To derive and validate a clinical prediction rule for radiographic pneumonia in patients over 16 years old presenting to emergency departments with acute respiratory symptoms or fever.

Design

Inception cohorts of consecutive patients, assembled to derive and validate the rule.

Setting

3 university-associated emergency departments.

Patients

The derivation cohort included 1134 patients seen in Chicago, who presented with fever or respiratory symptoms and for whom the examining physician ordered a chest roentgenogram. 16 patients were excluded because a roentgenogram was not found. The validation samples were 150 and 152 patients seen in Omaha, Nebraska, and Richmond, Virginia, respectively.

Assessment of prognostic factors

Data forms that included demographics, symptoms, signs, and coexisting medical conditions were completed within 24 hours of inception.

Main outcome measures

"Definite" or "probable" pneumonia as decided by 2 independent reviewers judging the hospital radiologist's report. (Their unweighted agreement was 92% corrected for chance).

Main results

Significant demographic differences between the derivation and validation cohorts in racial mix, age, and coexisting diseases occurred. The prevalence of radiographic pneumonia was 12.4% (Illinois), 21.5% (Virginia), and 30.0% (Nebraska) in the 3 groups. In the validation cohorts, the 5 factors independently predicting outcome by step-wise logistic regression were temperature > 37.8°C (odds ratio [OR] 2.69, 95% CI 1.73 to 4.17), pulse > 100 beats/min (OR 2.35 CI 1.52 to 3.65), rales (OR 3.73 CI 2.43 to 5.72), decreased breath sounds (OR 3.58, CI 2.33 to 5.50), and absence of asthma (OR 3.98, CI 1.89 to 8.42). The area under the receiver operating curve (area under the ROC) was 0.82 (CI 0.78 to 0.86). When applied to the validation cohorts and adjusted for prevalence, the area under the ROCs were 0.82 (CI 0.74 to 0.90, for Nebraska) and 0.76 (CI 0.66 to 0.86, for Virginia) and were not significantly different from the derivation value (P = 0.93). A nomogram is provided in the article for estimating the probability of pneumonia in a new patient.

Conclusions

Predictors of radiographic pneumonia in patients > 16 years old, with fever or acute respiratory symptoms included temperature > 37.8°C, pulse > 100 beats/min, rales, decreased breath sounds, and the absence of asthma. A clinical prediction rule and nomogram for estimating the probability of radiologic pneumonia after clinical examination were derived and validated. The latter was done in populations differing in demographic features and pneumonia prevalence from the derivation cohort.

Source of funding: Not stated.

Address for reprints: Dr. P.S. Heckerling, Department of Medicine, University of Illinois, Box 6998, Chicago, IL 60680, USA.


Commentary

Although the study by Heckerling and colleagues addresses an important question, it has several drawbacks. First, the clinical utility and effect of the prediction rule were not assessed: The authors did not estimate the number of roentgenograms that might have been eliminated if their rule had been used. My estimate, using the 5% cut-off as indication of no pneumonia, suggests that the effect may be small. Second, the classification used for determining the presence or absence of pulmonary infiltrates, which was based on review of the roentgenogram report, is also of concern. Inter- and intra-rater reliability for these readings were not assessed but are important, especially because 12.7% of cases in the derivation cohort had "possible" pneumonia, and this classification has a substantial effect on the prediction rule and its accuracy. Third, patients were included if the physician evaluating them decided to order a chest roentgenogram. As the authors point out, this may alter the value of some predictor variables and might diminish the accuracy of the rule when applied to all patients with fever or respiratory symptoms in an emergency department setting. This may explain why "absence of asthma" was identified as a predictor variable.

Unfortunately, although this study examines an important question, the findings are too preliminary to be useful in clinical decision making. Individual practitioners, out of interest, might wish to informally trace the rule's accuracy when they evaluate patients with fever and respiratory symptoms.

Edgar R. Black, MD
University of Rochester Medical CenterRochester, New York, USA


Authors' Response

The reviewer raises concerns that inter- and intra-observer variation in roentgenogram readings and clinical measurements might diminish the accuracy of our prediction rule. However, we collected data using standard roentgenogram readings and clinical measurements specifically to enhance the applicability of the rule to actual practice. The degree of explanatory power of a prediction rule reflects the robustness of the measurement of its predictor and outcome variables. That our rule was validated in 2 different settings shows that the symptoms, signs, and radiologic classifications that we used were sufficiently robust.

Paul S. Heckerling, MD
Thomas G. Tape, MDRobert S. Wigton, MD