Does This Patient Have Community-Acquired Pneumonia? Diagnosing Pneumonia by History and Physical Examination Joshua P. Metlay, MD, PhD; Wishwa N. Kapoor, MD, MPH; Michael J. Fine, MD, MSc

Community-acquired pneumonia is an important cause of acute respiratory chitis, upper respiratory tract infection, symptoms (eg, cough) in the ambulatory care setting. Distinguishing pneumo- asthma, and sinusitis.1 Though pneumo¬ nia from other causes of respiratory illnesses, such as acute bronchitis and up- nia may represent a small proportion of all the accu¬ per respiratory tract infections, has important therapeutic and prognostic impli- acute respiratory illnesses, cations. The reference standard for is chest rate identification ofthis subgroup is im¬ diagnosing pneumonia radiography, because of the distinct but it is on portant very likely that many physicians rely the patient's history and their physi- therapeutic and prognostic features of cal examination to diagnose or exclude this disease. A review of published this illness. studies of are no in- patients suspected of having pneumonia reveals that there In the preantibiotic era, mortality dividual clinical findings, or combinations of findings, that can rule in the diag- from pneumococcal pneumonia was con¬ nosis of pneumonia for a patient suspected of having this illness. However, sistently higher than 20% for all cases, some studies have shown that the absence of any vital sign abnormalities or any rising to more than 60% for bacteremic abnormalities on chest auscultation substantially reduces the likelihood of cases.2 Since the introduction of antibi¬ pneumonia to a point where further diagnostic evaluation may be unnecessary. otics no one has reported results from studies antibiotic This article reviews the literature on the appropriate use of the history and large-scale comparing physical examination in diagnosing community-acquired pneumonia. therapy to nonantibiotic therapy for pa¬ JAMA. tients with pneumonia. As a result, such 1997;278:1440-1445 therapy is universally recommended and has become a standard of care for all CLINICAL SCENARIO—DOES last 2 days, she has noted increasing spu¬ patients with pneumonia. No such stan¬ THIS PATIENT HAVE PNEUMONIA? tum production with her cough and dard exists for alternative respiratory worsening fatigue. She has felt warm but infections such as bronchitis3 or the com¬ A woman comes to has documented fever or 53-year-old your not any night mon cold.4 Moreover, inappropriate use office with a cough ofmore than 1 week's sweats. On physical examination, her of antibiotics for these alternative res¬ duration. She was in excellent health un¬ oral is 38.3°C her temperature (101°F), piratory infections may be an important til 7 days ago when she developed a non¬ heart rate is 110 beats per minute, and determinant of the rise in antibiotic re¬ mild sore and auscultation of her chest productive cough, throat, reveals inspi- sistance among common respiratory myalgia. She recalls no past history of ratory crackles on the left side. pathogens.5·6 asthma or chronic obstructive pulmo¬ In terms of prognosis, patients with and she does not smoke. nary disease, WHY IS THIS AN IMPORTANT continue to have an overall home from work forthe pneumonia Despite staying QUESTION TO ANSWER WITH high mortality from this illness, ranging A CLINICAL EXAMINATION? from as low 5% in studies of hospitalized and ambulatory patients to as high as Physicians commonly encounter pa¬ 37% in studies of ad¬ From the General Internal Medicine Unit, Department tients with simi¬ patients requiring of General respiratory complaints mission intensive care units.7 This Medicine, Massachusetts Hospital and Har- lar those in the clinical scenario. In to per¬ vard Medical School, Boston (Dr Metlay), and the Divi- to sistently high mortality underscores the sion of General Internal Medicine, Department of Medi- 1994, there were over 10 million visits to Center cine, for Research on Health Care, University of primary care physicians by adults with a Pittsburgh, Pittsburgh, Pa (Drs Kapoor and Fine). Dr chief of Metlay is now with the Division of General Internal complaint cough, representing Medicine, Department of Medicine, University of Penn- over 4% of all visits to physicians that The Rational Clinical Examination section editors: sylvania Medical Center, Philadelphia. year. Pneumonia 5% David L. Simel, MD, MHS, Durham Veterans Affairs Reprints: Michael J. Fine, MD, MSc, Montefiore Uni- represented only Medical Center and Duke Medical Cen- of all causes for these visits and was University versity Hospital, 8 East Room 824, 200 Lothrop St, ter, Durham, NC; Drummond Rennie, MD, Deputy Pittsburgh, PA 15213 (e-mail: [email protected]). the fifth leading diagnosis, after bron- Editor (West), JAMA. Downloaded from jama.ama-assn.org at HINARI on August 27, 2011 need for physicians to choose carefully HOW TO ELICIT THESE SYMPTOMS can yield abnormal auscultatory findings between home or hospital therapy for all AND SIGNS in as many as 50% of normal subjects.16 patients with pneumonia.8 For these rea¬ Finally, both percussion and ausculta¬ sons, physicians need to know how opti¬ Patients with community-acquired tion of the chest should proceed in a sys¬ mally to use their clinical examination to pneumonia present with a large number tematic fashion, with an examination of identify patients at suitable risk for of possible symptoms. In a study of more symmetric areas on both the anterior pneumonia to require further, definitive than 1800 patients with community- and posterior chest wall. diagnostic testing. acquired pneumonia, these presenting Chest is the reference radiography symptoms ranged from typical respira¬ METHODS standard for diagnosing community- tory complaints, including productive acquired pneumonia and provides addi¬ cough, dyspnea, and pleuritic chest pain, Literature Search tional information on the prognosis of pa¬ to predominately systemic complaints We searched English-language medi¬ tients with this illness,9 as well as the of fatigue, anorexia, and myalgias. More¬ cal literature to determine the precision presence of coexisting conditions such as over, the pattern ofpresenting symptoms of the clinical examination in patients bronchial obstruction or pleural effu¬ varied considerably among patients, par¬ with community-acquired pneumonia sions.10 Moreover, chest radiography is ticularly among elderly patients with and the accuracy of the examination in highly reliable,11 safe, generally avail¬ pneumonia who less frequently reported diagnosing patients suspected of having able, and relatively inexpensive, so that a wide range of symptoms.14 As a result, this illness. We searched MEDLINE it is a standard part of the evaluation of careful history taking in a patient sus¬ from 1966 through October 1995 based any patient with suspected pneumonia. pected of having community-acquired on an initial search strategy similar to It is possible that some physicians con¬ pneumonia should consider a broad that used by other authors in this series. tinue to diagnose and manage patients range of possible symptoms, including (The search strategy is available on re¬ with pneumonia without the aid of chest both respiratory and nonrespiratory quest.) The initial retrieval oftitles (n=7 radiography, while other physicians rou¬ symptoms. for precision, n=140 for diagnostic accu¬ tinely obtain chest radiographs for all pa¬ In contrast, the examination of the racy) was reviewed by 2 of us (J.P.M., tients suspected ofhaving pneumonia. We chest in patients with suspected pneu¬ M.J.F.). Articles that focused on hospi¬ do not know the proportion of physi¬ monia is traditionally carried out in a tal-acquired pneumonia, pediatrie pneu¬ cians who choose these alternative strat¬ structured manner, proceeding through monia, or acquired immunodeficiency egies. Therefore, the aims of this article the 4 steps of inspection, palpation, per¬ syndrome-related pneumonia were ex¬ are both to assess the validity of the cussion, and auscultation. The chest is cluded. The remaining articles were re¬ former approach (diagnosing pneumo¬ inspected for signs of asymmetric chest trieved, as well as any potentially eli¬ nia without chest radiography, using his¬ expansion, defined as a visible difference gible articles identified through review tory and physical examination alone) and in excursion between the 2 sides of the ofthe article reference lists (n=7 for pre¬ to identify elements of the clinical ex¬ chest. The chest wall is palpated while cision, =52 for diagnostic accuracy). amination that might improve the effi¬ the patient speaks in order to assess the A set of explicit inclusion and exclu¬ ciency of the latter approach (ordering transmission of sound, or tactile fremi¬ sion criteria were applied to each re¬ chest radiographs for all patients with tus. Percussion over symmetric areas of trieved article. Inclusion criteria re¬ suspected pneumonia). the anterior and posterior chest wall de¬ quired that the study be an original tects diminution in the resonance of the of the or of the PATHOPHYSIOLOGY OF study accuracy precision percussion note, or dullness to percus¬ history and/or physical examination in COMMUNITY-ACQUIRED sion. auscultation ofthe as¬ the of commu¬ PNEUMONIA Finally, lung determining diagnosis sesses the intensity of normal breath nity-acquired pneumonia. Exclusion cri¬ In patients with community-acquired sounds, the transmission of spoken teria consisted of studies of (1) patients pneumonia, the site of infection can in¬ words, and the presence of adventitious younger than 16 years, (2) patients with volve the pulmonary interstitium, alveoli, sounds. Auscultation in the peripheral known immunosuppression, or (3) pa¬ or both. This provides the physiologic ba¬ lung fields may detect the replacement tients with nosocomial infections. In ad¬ sis for the principle chest examination of the normal vesicular breath sounds dition, case series (<10 observations) or findings in pneumonia, which include dull¬ with tubular or bronchial breath sounds, review articles without original data ness to percussion, changes in the inten¬ which are normally heard only over were excluded. oftactile fremitus and breath the trachea. Increased transmission of sity sounds, Review of Articles and inspiratory crackles. Dullness to per¬ speech may be detected as the increased Quality cussion and local changes in the intensity clarity of whispered phrases, known as The remaining eligible articles were of tactile fremitus and breath sounds are whispered pectoriloquy, or as the change each evaluated by one of us (J.P.M.) the result of diffuse replacement of the in timbre of vowel sounds in the form of based on a méthodologie quality filter pulmonary parenchyma with inflamma¬ "e" to "a," known as egophony.12 The that assigned a level of evidence from I tory tissue leading to pulmonary consoli¬ principal abnormal sounds in commu¬ to V based on the internal validity of the dation or the presence of pleural effu¬ nity-acquired pneumonia are known as study. Level I evidence refers to a pri¬ sions.12 In a patient with pneumonia, crackles, which are nonmusical, discon¬ mary, prospective study of the accuracy crackles (formerly called "rales") are tinuous sounds and should be detected or precision ofthe clinical examination in caused by the delayed opening of alveoli with the patient in the upright position. community-acquired pneumonia. For in deflated regions of pathologically in¬ It has been suggested that auscultation studies dealing with accuracy, this re¬ flamed lung.13 Of note, crackles refer to of each lung in the lateral dependent po¬ quires independent, blind comparisons any discontinuous adventitious lung sition is a more sensitive technique for of clinical findings with a criterion stan¬ sounds and can therefore be heard in a crackles, but this has not been indepen¬ dard (or "gold standard") of diagnosis or variety of pulmonary diseases that cause dently validated.15 Auscultation should etiology among a large number (>50) of lung stiffening, including congestive occur with the patient breathing at nor¬ consecutive patients suspected of hav¬ heart failure, pulmonary fibrosis, and ob¬ mal tidal volumes, since inspiration from ing community-acquired pneumonia. structive lung disease.12 lower lung volumes (ie, residual volume) For studies dealing with precision, this Downloaded from jama.ama-assn.org at HINARI on August 27, 2011 Table 1.—Precision of Physical Examination Find¬ monia has not been directly examined. Accuracy of the Clinical ings in Examination of the Chest* However, analogous work in assessing History in the Diagnosis of in Pneumonia Physical symptom prevalence large-scale epi¬ Community-Acquired Examination Finding Agreement, %t Value demiologie studies has revealed consid¬ Tachypnea 63 0.25 erable interobserver variation in the re¬ For this review, 4 studies were Reduced chest movement 70 0.38 of This has led to to have level I evidence on the Increased tactile fremitus 85 0.01 cording symptoms.20,21 judged the ofstandardized test characteristics of individual items Dullness to percussion 77 0.52 adoption respiratory Decreased breath sounds 4 0.43 questionnaires in epidemiologie studies in the clinical history in the diagnosis Wheezes ..79 0.51 of chronic illnesses. How¬ of Crackles 72 0.41 respiratory community-acquired pneumonia.26'29 no such standardized In each of these the reference Bronchial breath sounds 4 0.32 ever, question¬ studies,

Whispered pectoriloquy .. 4 0.11 naires exist for recording symptoms in standard for the diagnosis of pneumo¬ . . patients with acute respiratory infec¬ nia was a new infiltrate on a chest *Adapted from Spiteri et al.23 tions.22 Table 2 summarizes the fCalculated based on data provided in Table 1 of radiograph. Spiteri et al.23 It has also been appreciated for some value of findings from the history, in¬ ÍMean pair agreement rates were not calculated for time that the examination ofthe nonres- the signs for which 2 or more physicians in a group failed physical cluding respiratory symptoms, to report the presence or absence of the sign. chest is hampered by a high degree of piratory symptoms, and information on interobserver error. While no study has past medical history. specifically addressed the reliability of Though all 4 studies were based in or more blinded requires 2 independent the physical examination in patients emergency departments, variations in raters of symptoms or signs in a large with community-acquired pneumonia, the patterns of the results reflect, number of patients suspected of having et al23 measured in variation in the selection crite¬ Level Spiteri reliability part, community-acquired pneumonia. among 24 physicians in the examination ria for each study. For example, in the II studies were to level I stud¬ analogous of 24 with a variety of respira¬ study by Diehr et al,26 chest radio¬ ies but with smaller numbers of patients patients tory conditions, 4 of whom had radio- graphs were obtained for all patients (10-50), widening the confidence limits graphic evidence of pneumonia. Table 1 presenting with acute cough, while the of the resulting calculations. Level III presents the calculated interobserver other studies obtained chest radio¬ studies were based on a retrospective the for sev¬ when the clinical determined reliability among physicians graphs only primary physi¬ design (ie, findings eral chest The results are cian previously determined a need for chart Level IV studies in¬ signs. present¬ by review). ed in the form ofboth mean pair observer them, often to confirm or exclude a cluded nonconsecutive patients, gener¬ agreement rates and values, which ac¬ diagnosis of pneumonia. The selected because of their definitive suspected ally count for rates of chance agreement latter approach provides a more highly results for the under or a findings study, from when is no selected population of patients with nonblinded of clinical find¬ ranging 0, agreement comparison better than chance, to 1, when there is acute respiratory complaints that may ings with a gold standard. Level V stud¬ perfect agreement. In fact, 2 ofthe most alter the measured test characteristics ies included studies with an uncertain reliable findings, dullness to percussion of individual clinical findings. This se¬ standard or a defined gold poorly study and wheezes on auscultation, had only lection bias is reflected in the fact that not even have com¬ population (ie, may fair to good values of 0.52 and 0.51, the prevalence of pneumonia in the For the munity-acquired pneumonia). to rates of study populations ranged from as low of this studies of corresponding agreement purposes study, only 77% and 79%, respectively. Crackles had as 2.6%26 to as high as 38.3%.27 level I also called A evi¬ quality, grade a value of 0.41 (agreement rate of 72%) Still, certain patterns emerge. For dence, were considered for the main and several such as there are no individual items and tables. Summaries of rel¬ findings whispered example, analyses pectoriloquy, and increased tactile from the clinical history whose pres¬ evant level II V studies are through pro¬ fremitus had values indicating poor ence or absence would reduce the odds vided in the text. agreement (range, 0.01-0.11), in part ex¬ of disease sufficiently to exclude pneu¬ Data Analysis plained by the rarity of these findings monia and eliminate the need to obtain overall. a chest radiograph. The 1 exception to Likelihood ratios were calculated for Similarly poor interobserver reliabil¬ this is the presence of a medical his¬ the presence (positive likelihood ratio ity has been noted in the chest examina¬ tory of asthma, which reduces the odds or absence likelihood [LR+]) (negative tion ofother respiratory disease. For ex¬ of pneumonia by a factor of 0.1, though ratio of individual clinical find¬ [LR-]) ample, Schilling et al24 noted an agree¬ this has been demonstrated in only 1 ings.1718 Only those findings significantly ment rate of 76% for abnormal chest study.29 associated with the or absence presence sounds in the examination of 187 men Similarly, the presence of no single of in least 1 based pneumonia at study, with interstitial lung disease and 88 con¬ item in the clinical history raises the on a 2-tailed or Fisher exact test with 2 trols; this yields a value of0.25. Smyllie odds of pneumonia high enough to con¬ were included in the results. P<.05, et al25 measured agreement rates among firm the without a chest ra¬ the actual value of diagnosis However, diagnostic 9 who examined 20 the presence of find¬ still de¬ physicians patients diograph. Though statistically significant findings with a variety of chronic lung diseases. ings with a LR+ ranging from 2 (fever on both the of pends prior probability rates were mid¬ or to 3 (history of and how much the likelihood Agreement generally immunosuppression) pneumonia way between chance and perfect agree¬ dementia) may be helpful, they are not ratio moves the posterior probability ment for a number of chest examination confirmatory, particularly given the from the prior probability.19 findings, including diminished breath typically low prevalence of pneumonia RESULTS sounds, decreased percussion note, and in the study populations. For example, crackles. the basis for the rela¬ in the Diehr et al, the pres¬ Precision of and Though study by Symptoms Signs tively low interobserver reliability in ence of subjective fever (LR+=2.1, of Community-Acquired Pneumonia chest examination is unknown, at least 1 95% confidence interval [CI], 1.4-2.9) Interobserver variation in the record¬ group has suggested that deficiencies in had a positive predictive value of only ing of the presence of symptoms in pa¬ the teaching of the clinical examination 5.5%, reflecting the low prevalence of tients with community-acquired pneu- are to blame.23 pneumonia in the population.26 Downloaded from jama.ama-assn.org at HINARI on August 27, 2011 Table 2.—Likelihood Ratios for Pneumonia Given the Presence or Absence of Individual History Findings* Positive Likelihood Ratiof Negative Likelihood Ration Diehr Gennis Singal Heckerling Diehr Gennis Singal et al,26 1984 et al,271988 et al,281989 et al,291990 et al,261984 et al,271988 et al,281989 Respiratory symptoms Cough NS NS NS 0.31 NS Dyspnea NS NS 0.67 NS NS Sputum production 1.3 NS NS 0.55 NS NS Nonrespiratory symptoms Fever 2.1 NS 1.7 0.71 NS 0.59 Chills 1.6 1.3 1.7 0.85 0.72 0.70 Night sweats 1.7 0.83 Myalgias 1.3 NS 0.58 NS Sore throat 0.78 NS 1.6 NS Rhinorrhea 0.78 NS 2.4 NS Past medical history Asthma 0.10 3.8 Immunosuppression 2.2 0.85 Dementia 3.4 0.94

*Only those findings significantly associated with the presence or absence of pneumonia in at least 1 study were included (P<.05 in a 2-tailed 2 or Fisher exact test). Ellipses indicate the result is not available; NS, result not significant. tPositive likelihood ratio for pneumonia when symptom present (sensitivity/1-specificity). ^Negative likelihood ratio for pneumonia when symptom absent (1-sensitivity/specificity).

Table 3.—Likelihood Ratios for Pneumonia Given the Presence or Absence of Physical Examination Findings* Positive Likelihood Ratiof Negative Likelihood Ration I I I I Diehr Gennis Singal Heckerling Diehr Gennis Singal Heckerling et al,261984 et al,271988 et al,281989 et al,291990 et al,261984 et al,271988 et al,281989 et al,291990 Vital signs Respiratory rate, breaths/min >20 1.2 0.66 >25 3.4 NS§ 0.78 NS 0.82 >30 2.6 0.80 Heart rate, beats/min >100 NS 1.6 NS§ 2.3 NS 0.73 NS 0.49 >120 1.9 Temperature >37.8°C (100°F) 4.4 1.4 2.4 2.4 0.63 0.68 0.58 Any abnormal vital sign 1.2 0.1E Chest examination Asymmetric respiration 0.96 Dullness to percussion NS 2.2 NS 0.93 0.79 Decreased breath sounds 2.3 2.5 0.78 0.64 Crackles 2.7 1.6 2.6 0.87 0.83 0.78 0.62 Bronchial breath sounds 3.5 0.90 Rhonchi NS 1.5 NS 0.85 0.76 Egophony 8.6 2.0 5.3 0.96 0.96 0.76 Any chest finding 1.3 0.57

*Only those findings that were significantly associated with the presence or absence of pneumonia in at least 1 study were included (P<.05 In a 2-tailed 2 or Fisher exact test). Ellipses indicate result is not available; NS, result not significant. tPositive likelihood ratio for pneumonia when finding present (sensitivity/1 —specificity). ^Negative likelihood ratio for pneumonia when finding absent (1-sensitivity/specificity). §Actual cut points not specified in this study.

Accuracy of Physical Examination these abnormalities did not have a sub¬ patient without any vital sign abnormali¬ Findings in the Diagnosis of stantial impact on the calculated LRs.27 ties would have a predicted probability Community-Acquired Pneumonia Similarly, LRs for the absence of any in¬ of pneumonia of less than 1%. dividual vital sign abnormality (LR-) The presence of several individual Table 3 summarizes the accuracy of 10 ranged from 0.5 to 0.8. However, Gennis findings on chest examination signifi¬ different potential findings (3 vital signs et al27 demonstrated an LR- of0.18 (95% cantly raised the likelihood of pneumo¬ and 7 abnormal findings on chest exami¬ CI, 0.07-0.46) for the diagnosis of pneu¬ nia. For example, in 1 study the presence nation) from the physical examination monia based on the absence of all 3 vital of asymmetric respirations essentially in patients with suspected pneumonia sign abnormalities (ie, respiratory rate guaranteed the diagnosis of pneumonia based on results from the 4 previously <30 breaths per minute, heart rate < 100 (LR+=oc,95%CI,3.2-^).26Howeverthe identified studies. Likelihood ratios for beats per minute, and temperature usefulness ofthis finding was limited by the presence of any individual vital sign <37.8°C [100°F]). Based on this finding, the fact that only 4% of patients with abnormality (LR+), including tachy- if the baseline prevalence of pneumonia pneumonia were noted to have asym¬ pnea, tachycardia, or fever, ranged from among ambulatory patients with respi¬ metric respirations. The presence of 2 to 4. Moreover, various cut points for ratory illnesses is assumed to be 5%, a other findings, including egophony and Downloaded from jama.ama-assn.org at HINARI on August 27, 2011 Table 4.—Predictive Rules for Pneumonia Diag¬ mal auscultatory finding yielded an LR- The 3 scores summarized in Table 4, nosed by Chest Radiography* of only 0.68 (95% CI, 0.44-0.89), both of along with the decision rule suggested Diehr et al26 which would translate into very small by Gennis et al (ie, only obtaining chest Add points when presents effects on the probability ofpneumonia.32 radiographs for patients suspected of Rhinorrhea In contrast, another study found that with at least 1 vital Sore throat having pneumonia the absence of on chest were for Night sweats any abnormality sign abnormality),27 compared Myalgias auscultation resulted in an LR- of 0.13 their ability to predict correctly the re¬ Sputum all day (95% CI, 0.07-0.24),31 which might sub¬ sults ofchest in an indepen¬ >25 radiography Respiratory rate breaths/min reduce the ofpneu¬ dent Emerman et al.33 Patients Temperature 237.8°C (100°F) stantially probability study by monia. However, this result has not presenting to an emergency department Singal et al28 been replicated in prospective studies, or outpatient medical clinic with a com¬ Probability=1/(1+e-Y)4 which would be to less bias in of were enrolled Y=-3.095+1.214(cough) subject plaint cough prospec- +1.007 (fever) the recording of physical examination tively, and chest radiographs were ob¬ +0.823 (crackles) findings. tained for all patients regardless of the Each variablen if present primary physician's clinical impression. Heckerling et al29 Evaluating Algorithms Overall, the of Determine the number of prevalence pneumonia findings to Predict Pneumonia the was 7%. In the present§: among study patients Absence of asthma Because the accuracy of individual absence of an explicit guideline, physi¬ Temperature >37.8°C (100°F) symptoms or signs for predicting pneu¬ cian judgment that the patient did not Heart rate >100 beats/min monia is several studies have at¬ need chest reduced the Decreased breath sounds low, radiography Crackles tempted to build prediction rules that probability of pneumonia to just less incorporate the presence or absence of than 2% (LR-=0.25, 95% CI, 0.09-0.61), *Adapted from Emerman et al.33 several or examination which exceeded all 4 prediction rules. tFor example, a threshold score of -1 (ie, all patients history physical with scores a-1 are considered to have pneumonia), findings. Table4summarizesthefeatures In contrast, physician judgment that yields a positive likelihood ratio (LR+)=1.5 and negative of 3 such rules. Though initially designed the patient needed chest radiography likelihood ratio a threshold score of +1 (LR-)=0.22, as aids in the of chest radio¬ to increased yields a LR+=5.0 and LR-=0.47, and a threshold score ordering diagnose pneumonia only of +3 yields a LR+=14.0 and LR-=0.82, based on the graphs for patients with suspected pneu¬ the probability of pneumonia to 13% data.26 original study are considered as which meant JFIrst calculate Y and then calculate the predicted monia, they reasonably (LR+=2.0,95% CI, 1.5-2.4), probability of pneumonia. prediction rules for the diagnosis ofpneu¬ that reliance on implicit physician judg¬ §For example, based on a prevalence of pneumonia monia in these and ment alone would have led to many un¬ of 5%, the presence of 0,1, 2, 3, 4, or 5 findings yields patients yield prob¬ probabilities of pneumonia of <1%, 1%,3%, 10%, 25%, abilities ofpneumonia after completion of necessary chest radiographs. and 50%, respectively, based on a nomogram provided the clinical examination. For the rule of In comparison, the simple decision by Heckerling et al.29 Diehr et al, points are assigned for each rule of Gennis et al—ordering chest ra¬ clinical finding and summed to yield a dis¬ diographs only for patients with abnor¬ dullness to percussion, significantly in¬ criminant score. For example, a thresh¬ mal vital signs—yielded the highest creased the likelihood of pneumonia. old score of -1 (ie, all patients with scores overall LR+ for predicting pneumonia, However, given the low prevalence of >-l are considered to have pneumonia) but the LR+ was a modest 2.6 (95% CI, pneumonia in the overall study popula¬ yields an LR+ of 1.5 and an LR- of 0.22, a 1.6-3.7). Using this rule, 40% fewer ra¬ tions, the impact ofobserving these find¬ threshold score of+1 yields an LR+ of 5.0 diographs would have been ordered ings on estimating the probability of and an LR- of0.47, and a threshold score compared with unaided physician judg¬ pneumonia was only modest. For ex¬ of+3 yields an LR+ of 14.0 and an LR- of ment. However, excluding pneumonia ample, the presence of egophony had a 0.82, based on the original study data.26 on the basis of the absence of any vital positive predictive value ranging from The rule of Singal et al28 is a logistic func¬ sign abnormalities would have missed as low as 20%26 to no higher than 56%.27 tion that can yield probabilities of pneu¬ 38% of patients subsequently shown to Finally, all 4 studies support the con¬ monia ranging from 4% (no findings pres¬ have pneumonia on chest radiography clusion that the presence or absence of ent) to 49% (all 3 findings present).28 (LR-=0.50 [95% CI, 0.27-0.78], com¬ crackles on examination would not be The final prediction rule, by Hecker¬ pared with LR-=0.18 in the original sufficient to rule in or rule out the diag¬ ling et al,29 is based on the presence or study of Gennis et al27). The clinical sig¬ nosis. For example, with a prevalence of absence of 5 clinical findings. The perfor¬ nificance of this finding remains un¬ pneumonia of 5%, the absence of crack¬ mance of this prediction rule depends on known. les reduces the probability to 3%, at the the pretest probability of pneumonia in It should be emphasized that an algo¬ lowest, and the presence of crackles the population. In most ambulatory care rithm that is less than perfect, ie, not all raises the probability to 10%, at the high¬ settings, this probability will be rela¬ ordered chest radiographs demonstrate est. Moreover, the absence ofany abnor¬ tively low. For example, as noted earlier, a new infiltrate, will still be acceptable mality on chest examination yielded an in a national survey, only 5% of all pa¬ given the relatively low cost and risk as¬ LR- of 0.57 (95% CI, 0.39-0.83),27 which tients visitingprimary care physicians for sociated with this test. Ultimately, opti¬ is too close to the indeterminate LR cough were diagnosed as having pneumo¬ mum yields for chest radiography in the value of 1.0 to substantially reduce the nia.1 In this setting, the presence of2,3, or evaluation of patients with suspected probability of pneumonia. 4 predictors would result in predicted pneumonia will need to be determined, The low accuracy of individual find¬ probabilities ofpneumonia of3%, 10%, or balancing the costs of the test with the ings on chest examination for detecting 25%, respectively, based on a nomogram costs of missed diagnoses. Additional pneumonia has also been supported by provided by Heckerling et al.29 The rule factors, such as illness severity and pa¬ studies that relied on retrospective data would yield a maximum probability of tient preferences, will also play a role in gathering30·31 or incomplete application pneumonia of 50% if all 5 of its clinical determining the appropriate threshold of chest radiography to all study pa¬ predictors were present. These findings for ordering chest radiographs in pa¬ tients.32 In 1 study, the absence of crack¬ emphasize the inaccuracy in diagnosing tients with acute respiratory illnesses. les yielded an LR- of only 0.71 (95% CI, pneumonia clinically, in the absence of For example, thresholds may be lower 0.47-0.90) and the absence of any abnor- confirmatory chest radiography. for patients who appear severely ill or Downloaded from jama.ama-assn.org at HINARI on August 27, 2011 who express strong desires to have a de¬ data) and a calculated probability of firm the diagnosis ofpneumonia. If diag¬ finitive diagnosis. We suggest that an pneumonia of 42%. Finally, the patient nostic certainty is required in the algorithm that yields less than a 100% has 3 of 3 of the criteria of Singal et al, management of a patient with sus¬ negative predictive value may still be yielding a probability of pneumonia of pected pneumonia, then chest radiogra¬ acceptable assuming that the missed 49%, based on their logistic formula.28 phy should be performed. cases of pneumonia continue to have We conclude that none of these combi¬ 3. Future research should examine good clinical outcomes. However, this nations offindings can be said to "rule in" ways to improve the precision of the hypothesis will need to be tested. the diagnosis, yet the possibility ofpneu¬ clinical examination in patients with monia remains high that further as well as to de¬ RETURN TO THE CLINICAL enough suspected pneumonia, diagnostic testing, in particular chest ra¬ termine the accuracy of the clinical ex¬ SCENARIO diography, is warranted. amination in these patients in settings The with outside the emergency In patient presents typical THE BOTTOM LINE department. symptoms ofcommunity-acquired pneu¬ addition, studies should address appro¬ monia, including a productive cough and 1. Physicians frequently disagree priate thresholds for obtaining chest ra¬ fever. Physical examination reveals fe¬ about the presence or absence of indi¬ diographs and treating accordingly vs ver and crackles on chest auscultation. vidual findings on chest examinations of empirical antimicrobial therapy vs clini¬ In particular, the patient has 4 of 5 ofthe patients with respiratory illnesses, in¬ cal observation in the management of clinical pneumonia predictors identified cluding community-acquired pneumonia. patients with suspected community- by Heckerling et al (absence of asthma, 2. Individual symptoms and signs have acquired pneumonia. presence of fever, tachycardia, and inadequate test characteristics to rule in crackles). Using the nomogram of Heck¬ or rule out the diagnosis of pneumonia. Dr Metlay was supported by a General Medicine erling et al, a 5% prevalence of pneumo¬ Decision rules that use the presence or Research Fellowship, National Research Service Award 5 T32 PE11001-08. Dr Fine was as nia a 25% absence of several and supported among outpatients yields prob¬ symptoms signs a Robert Wood Johnson Foundation Generalist ability ofpneumonia.29 Similarly, the pa¬ to modify the probability of pneumonia Physician Faculty Scholar. tient is at the threshold score of+3 points are available, the simplest of which re¬ The authors gratefully acknowledge the contri¬ on the prediction rule of Diehr et al26 quires the absence of any vital sign ab¬ bution of Melanie E. Smith, MPIA, to the develop¬ ment of the MEDLINE search and initial of sore normalities to exclude the strategy (presence throat, sputum, myal¬ diagnosis. retrieval ofarticles. We also thank Daniel E. Singer, gias, and fever), yielding an LR for pneu¬ There are no combinations ofhistory and MD, for editorial suggestions on an earlier version monia of 14.0 (based on the original study physical examination findings that con- of the manuscript. References 1. Metlay JP, Stafford RS, Singer DE. National 11. Albaum MN, Hill LC, Murphy M, et al. Interob- I. Disagreement between observers in an epidemio- trends in the management of acute cough by pri- server reliability ofthe chest radiograph in commu- logical study of respiratory disease. BMJ. 1955;1: mary care physicians. J Gen Intern Med. 1997;12 nity-acquired pneumonia. Chest. 1996;110:343-350. 65-68. (suppl):77. 12. Sapira JD. The Art and Science of Bedside Di- 25. Smyllie HC, Blendis LM, Armitage P. Observer 2. Bullowa JGM,Wilcox C. Incidence of bacteremia agnosis. Baltimore, Md: Williams & Wilkins; 1990. disagreement in physical signs of the respiratory in the pneumonias and its relation to mortality. Arch 13. Forgacs P. Lung sounds. Br J Dis Chest. 1969; system. Lancet. 1966;2:412-413. Intern Med. 1935;55:558-573. 63:1-12. 26. Diehr P, Wood RW, Bushyhead J, Krueger L, 3. Orr PH, Scherer K, Macdonald A, Moffatt MEK. 14. Metlay JP, Schulz R, Li YH, et al. Influence of Wolcott B, Tompkins RK. Prediction of pneumonia Randomized placebo-controlled trials of antibiotics age on symptoms at presentation in community- in outpatients with acute cough: a statistical ap- for acute bronchitis: a critical review of the litera- acquired pneumonia. Arch Intern Med. 1997;157: proach. J Chronic Dis. 1984;37:215-225. ture. J Fam Pract. 1993;36:507-512. 1453-1459. 27. Gennis P, Gallagher J, Falvo C, Baker S, Than 4. Mainous AG, Hueston WJ, Clark JR. Antibiotics 15. Gilbert VE. Detection of pneumonia by auscul- W. Clinical criteria for the detection ofpneumonia in and upper respiratory infection: do some folks think tation of the lungs in the lateral decubitus positions. adults: guidelines for ordering chest roentgeno- there is a cure for the common cold? J Fam Pract. Am Rev Respir Dis. 1989;140:1012-1016. grams in the emergency department. J EmergMed. 1996;42:357-361. 16. Thacker RE, Kraman SS. The prevalence ofaus- 1989;7:263-268. 5. Jernigan DB, Cetron MS, Breiman RF. Minimiz- cultatory crackles in subjects without lung disease. 28. Singal BM, Hedges JR, Radack KL. Decision ing the impact ofdrug-resistant Streptococcus pneu- Chest. 1982;81:672-674. rules and clinical prediction of pneumonia: evalua- moniae (DRSP): a strategy from the DRSP Work- 17. Koopman PAR. Confidence intervals for the ra- tion of low-yield criteria. Ann Emerg Med. 1989;18: ing Group. JAMA. 1996;275:206-209. tio of two binomial proportions. Biometrics. 1984; 13-20. 6. Gonzales R, Steiner JF, Sande M. Antibiotic pre- 40:513-517. 29. Heckerling PS, Tape TG, Wigton RS, et al. Clini- scribing for adults with colds, upper respiratory 18. Centor RM. Estimating confidence intervals of cal prediction rule for pulmonary infiltrates. Ann tract infections, and bronchitis by ambulatory care likelihood ratios. Med Decis Making. 1992;12:229\x=req-\ Intern Med. 1990;113:664-670. physicians. JAMA. 1997;278:901-904. 233. 30. Osmer JC, Cole BK. The stethoscope and roent- 7. Fine MJ, Smith MA, Carson CA, et al. Prognosis 19. Sackett DL. A primer on the precision and ac- genogram in acute pneumonia. South Med J. 1966; and outcomes of patients with community-acquired curacy ofthe clinical examination. JAMA. 1992;267: 59:75-77. pneumonia: a meta-analysis. JAMA. 1996;275:134-141. 2638-2644. 31. Heckerling PS. The need for chest roentgeno- 8. Fine MJ, Auble TE, Yealy DM, et al. A prediction 20. Cochrane AL, Chapman PJ, Oldham PD. Ob- grams in adults with acute respiratory illness: clini- rule to identify low-risk patients with community- servers' errors in taking medical histories. Lancet. cal predictors. Arch Intern Med. 1986;146:1321\x=req-\ acquired pneumonia. N Engl J Med. 1997;336:243\x=req-\ 1951;1:1007-1008. 1324. 250. 21. Fletcher CM. The problem of observer varia- 32. Melbye H, Straume B, Aasebo U, Dale K. Diag- 9. Hasley PB, Albaum MN, Li YH, et al. Do pulmo- tion in medical diagnosis with special reference to nosis ofpneumonia in adults in general practice: rela- nary radiographic findings at presentation predict chest diseases. Method Inform Med. 1964;3:98-103. tive importance of typical symptoms and abnormal mortality in patients with community-acquired 22. Graham NMH. The epidemiology of acute res- chest signs evaluated against a radiographic refer- pneumonia? Arch Intern Med. 1996;156:2206-2212. piratory infections in children and adults: a global ence standard. Scand J Prim Health Care. 1992;10: 10. American Thoracic Society. Guidelines for the perspective. Epidemiol Rev. 1990;12:149-178. 226-233. initial management of adults with community-ac- 23. Spiteri MA, Cook DG, Clarke SW. Reliability of 33. Emerman CL, Dawson N, SperoffT, et al. Com- quired pneumonia: diagnosis, assessment of sever- eliciting physical signs in examination of the chest. parison of physician judgment and decision aids for ity, and initial antimicrobial therapy. Am Rev Respir Lancet. 1988;1:873-875. ordering chest radiographs for pneumonia in outpa- Dis 1993;148:1418-1426. 24. SchillingRSF, HughesJPW,Dingwall-Fordyce tients. Ann Emerg Med. 1991;20:1215-1219.

Downloaded from jama.ama-assn.org at HINARI on August 27, 2011