Clinical Question

How can I predict whether a patient with cough has pneumonia?

Evidence Summary

Identifying patients with pneumonia (defined as acute respiratory illness with an infiltrate seen on a chest radiograph) is important. Although there is no evidence that antibiotics benefit otherwise healthy patients with acute bronchitis,¹ antibiotics are recommended to treat patients with bacterial pneumonia. History and physical examination findings that increase the likelihood of pneumonia include fever, chills, sweats, myalgia, tachypnea, egophony, decreased breath sounds, dullness to percussion, crackles (rales), immunosuppression, and dementia. Findings that decrease the likelihood of pneumonia include asthma, sore throat, and rhinorrhea.² However, individual signs and symptoms cannot rule pneumonia in or out.

A number of clinical decision rules that combine signs and symptoms to increase the accurate diagnosis of pneumonia and guide the use of chest radiography have been developed and validated. Gennis and colleagues developed a rule that recommends chest radiography for any patient with an abnormal vital sign (i.e., temperature greater than 100° F [37.8° C], respiratory rate greater than 20 breaths per minute, or heart rate greater than 100 beats per minute).³ In the original population of 308 emergency department patients with acute respiratory illness, of whom 38 percent had a definite or equivocal infiltrate on a chest radiograph, the Gennis rule identified 113 out of 117 patients with pneumonia.³

Diehr and colleagues studied 1,819 emergency department patients at an Army medical center who presented with acute cough and who had received chest radiography. Pneumonia was diagnosed in the 2.6 percent of patients who had an infiltrate on a chest radiograph.⁴ Diehr and colleagues developed a clinical decision rule in 1,400 patients and validated it in 483 patients.⁴ The rule assigned −2 points for rhinorrhea; −1 point for sore throat; 1 point each for night sweats, myalgia, and all-day sputum production; and 2 points each for a respiratory rate greater than 25 breaths per minute and a temperature of 100° F or greater. The likelihood of pneumonia ranged from less than 1 percent for patients with −2 or −3 points to 27 percent for patients with 3 to 6 points.⁴

Heckerling and colleagues developed a clinical decision rule in 1,134 patients presenting to three emergency departments with fever or respiratory symptoms and who received chest radiography; it was validated in 302 patients.⁵ The rule identified five key predictors for pneumonia: temperature greater than 100° F, heart rate greater than 100 beats per minute, crackles, decreased breath sounds, and absence of asthma. The likelihood of pneumonia increased with the number of predictors present (Table 1^2–7 ).⁵

Presenting information this way allows calculation of likelihood ratios that can be applied to different pretest probabilities (i.e., the probability of pneumonia in the entire population at risk). These pretest probabilities are generally lower in the primary care setting and higher in the emergency department setting.

Each of the previously mentioned rules,^3–5 plus a multivariate model,⁶ were prospectively validated in a study of 290 patients who received chest radiography after presenting to a hospital's emergency department or outpatient clinic.⁷ All patients had an acute cough plus fever, hemoptysis, or sputum production; 7 percent of patients were diagnosed with pneumonia based on chest radiography findings. The performance of the clinical decision rules was compared with physician judgment. Although physician judgment was more sensitive than the rules, the specificity and overall accuracy of the Gennis and Heckerling rules were better than physician judgement (overall accuracy was 76 percent for the Gennis rule, 68 percent for the Heckerling rule, and 60 percent for physician judgment).⁷

All of these studies were conducted more than 15 years ago and took place in the emergency department setting.^3–7 Hopstaken and colleagues conducted a more recent study of patients presenting to Dutch general physicians.⁸ Patients were adults presenting with acute cough; shortness of breath, wheezing, chest pain, or auscultation abnormalities; and fever, sweats, headache, or myalgia. All of the patients had suspected bronchitis or pneumonia diagnosed by their general physician.

The Hopstaken and colleagues study was used to develop several multivariate models that included signs, symptoms, erythrocyte sedimentation rate, and C-reactive protein (CRP) levels to predict the risk of pneumonia. Overall, 13 percent of patients had pneumonia. The best independent predictors of pneumonia were dry cough, diarrhea, a temperature of 100° F or greater, and a CRP level of 20 mg per dL (200 mg per L) or greater. Only 3 percent of patients with none or one of these predictors and a CRP level less than 20 mg per dL had pneumonia, and only 2 percent of patients with none of these predictors and a CRP level less than 20 mg per dL had pneumonia.⁸

Which, if any, of these clinical decision rules should a primary care physician use in practice? Although the Hopstaken rule has potential for identifying low-risk patients, it has not been prospectively validated and requires a CRP measurement, which is not readily available in the outpatient setting in the United States. The Gennis and Heckerling rules are the simplest and best validated rules. Because the first two predictors in the Heckerling rule are vital signs also found in the Gennis rule, a simple rule of thumb that combines these rules can be used to determine when to order chest radiography (Table 2).

Applying the Evidence

A 56-year-old man presents to a home-less clinic with an acute cough and a temperature of 102° F (38.9° C). His heart rate is 104 beats per minute; his breath sounds are not decreased, but he has crackles; and he has no history of asthma. How likely is it that he has pneumonia?

Answer: Using the Heckerling rule (Table 1^2–7 ), the patient has four out of five key predictors for pneumonia. In the primary care setting, his posttest probability of pneumonia would be 27 percent, and you would order chest radiography to confirm the diagnosis before prescribing antibiotics. However, because homeless patients do not readily seek care, you estimate that his pretest probability for pneumonia is closer to that in the emergency department setting (15 percent) than that in a typical primary care office (5 percent). You determine that this patient has a posttest probability of 56 percent. Because of the relatively high risk of pneumonia and the difficulty of obtaining a radiograph for the uninsured patient, you prescribe antibiotics.