Relevance of Crackles to Clinician Diagnosis of Pneumonia Open Access

The diagnosis of pneumonia in children is often based on clinical findings such as fever, focal crackles, and hypoxemia. National guidelines recommend against the use of chest radiography (CXR) to confirm the diagnosis of pneumonia in ambulatory children, based on its poor ability to discern viral from bacterial etiologies, cost, and radiation exposure.¹ Without CXR, clinicians are left to judge the risk of pneumonia by clinical features alone, many of which have been shown in large studies to have poor predictive value for the presence of radiographic pneumonia.^2,3 Focal crackles are often considered the quintessential examination finding among children with pneumonia. It is unknown the extent to which the presence of crackles on examination influences clinicians’ determination of pneumonia and subsequent antibiotic use. The objective of this study was to evaluate the association of crackles with the clinical diagnosis of pneumonia and to compare this with the actual association between crackles and radiographic pneumonia. We hypothesized that physicians would overvalue the association between crackles and pneumonia as compared with the gold standard of CXR.

This was single-center prospective cohort study of children aged 3 months through 18 years undergoing CXR for suspected pneumonia in an urban, tertiary-care emergency department (ED) between May 2015 and February 2024. Our site maintains ongoing enrollment of patients undergoing CXR to evaluate for pneumonia in an effort to promote judicious use of antibiotics among children with respiratory illness. The electronic ED tracking board was monitored by trained research coordinators (RCs) while on shift (generally from 7 am to 11 pm 7 days a week) to identify patients with respiratory complaints in whom CXR was ordered. We excluded children with complex chronic medical conditions predisposing to pneumonia (eg, chronic lung disease, complex congenital cardiac disease, immunodeficiency, cancer, chronic aspiration), those already on antibiotics at the time of enrollment, those who had already undergone CXR earlier in the illness episode, children whose caregivers did not speak English, children already enrolled in another study during the ED visit, and children deemed by their treating clinician to be unfit to participate, due to either acuity or psychosocial factors.⁴ 

Following patient consent and enrollment, RCs administered a survey to the patient’s treating physician. Physicians were asked to indicate physical examination findings and were asked a series of questions relating to their patient’s likelihood of radiographic pneumonia. Physicians were also asked to indicate whether they had already seen the chest radiograph or reviewed the radiologist’s report. For this study, we only included data from children for whom the physicians had no knowledge of CXR findings prior to being surveyed. RCs queried the electronic medical record to ascertain demographic information, vital sign information, and the radiograph report.

The outcome of clinician-diagnosed pneumonia was ascertained by a physician response of “yes” to following question: “If you did not perform an x-ray, would you prescribe antibiotics for pneumonia?” We used this question in order to best approximate the physician’s binary determination of pneumonia in the patient prior to knowledge of radiography results and thus whether pneumonia would have been diagnosed if CXR had not been performed.

The outcome of radiographic pneumonia was ascertained by CXR reports written by radiologists at the time of the patients’ ED visit. These CXR reports were independently reviewed by 3 board-certified pediatric emergency medicine physicians and study investigators (A.H., S.L., and M.N.) and categorized based on the radiologist’s final impression as definite pneumonia, probable pneumonia, equivocal for pneumonia, unlikely to represent pneumonia, or negative for pneumonia, according to a previously described method. Radiographs with classic viral features such as peribronchial cuffing, perihilar infiltrates, and hyperinflation without focal opacities/consolidation were considered negative for pneumonia.^4,5 In this study, chest radiographs categorized as definite or probable pneumonia were considered positive for the outcome of radiographic pneumonia.

The primary exposure was rales/crackles auscultated on lung examination. We included both terms as they are used synonymously in the literature.⁶ The presence or absence of auscultatory findings was determined by the physician survey at the time of the patient encounter. If the physician indicated that the patient had rales/crackles, wheeze, or decreased breath sounds, they were asked to specify if each finding was focal (limited to 1 lung region) or diffuse (present throughout all lung fields). Physicians also specified whether retractions and/or respiratory distress were present. Caregivers indicated whether children had fever and/or cough at home. We defined a child as having fever if they had a reported history of fever, a temperature greater than or equal to 38 °C in the ED, or both. Tachypnea was based on the triage respiratory rate (RR) recorded and was defined as follows: RR greater than or equal to 50/min in age younger than 1 year, RR greater than or equal to 40 in age 1 to younger than 5 years, RR greater than or equal to 30 in age 5 to younger than 10 years, and RR greater than or equal to 25 in age 10 to 18 years.⁴ Hypoxemia was defined as an oxygen saturation of 96% or below⁷; the first oxygen saturation recorded during the patient’s visit was used in analysis.

We summarized characteristics of the cohort using frequencies and percentages for categorical variables and medians with IQRs for continuous variables. Among children with focal crackles, we determined the proportions of children with clinician-diagnosed pneumonia and radiographic pneumonia. To compare these proportions, we estimated 2 intercept-only generalized linear models (GLMs) with the binomial family and identity link, where the dependent variables were clinician-diagnosed pneumonia and radiographic pneumonia, respectively.⁸ Using the suest command in Stata, we combined the estimation results of these 2 GLMs into a single parameter vector and simultaneous sandwich/robust (co)variance matrix and conducted comparisons of the proportions of children with clinician-diagnosed vs radiographic pneumonia. Statistical significance was defined by α = 0.05. We repeated this analysis for children with diffuse crackles and for children with crackles in tandem with other specified combinations of signs (fever, hypoxemia, focally decreased breath sounds, absence of wheeze). For patients with focal crackles, we calculated test characteristics with 95% CIs of clinician diagnosis against CXR as the gold standard.

Using multivariable logistic regression, we constructed 2 models, using (1) clinician-diagnosed pneumonia and (2) radiographic pneumonia as the dependent variables. The independent variable in each model was a 3-level variable for crackles on lung auscultation: no crackles (referent group), diffuse crackles, and focal crackles. Models were adjusted for other factors that may be associated with a clinical or radiographic diagnosis of pneumonia: fever at home or in the ED, cough, first oxygen saturation recorded, wheeze, focally decreased breath sounds, retractions, respiratory distress, and age-adjusted tachypnea. Patients without an oxygen saturation measurement were excluded from the main regression model; sensitivity analyses were performed with additional models imputing the missing oxygen saturations at 90%, 95%, and 100%. No other data points were missing from the data set. Considering the knowledge that the majority of respiratory illness in children younger than 2 years is viral in origin,¹ we performed another sensitivity analysis restricting our regression analysis to children 2 years and older. Additionally, as our study dates spanned those of the global COVID-19 pandemic, we performed a sensitivity analysis excluding children enrolled during the early phase of the pandemic (March 15, 2020, to December 31, 2022).⁹ 

This study was approved by the institutional review board at the study institution. All statistical analyses were performed using Stata version 17 (StataCorp).

Among 2902 children approached for inclusion, 2115 consented to participate; the study cohort consisted of 1594 children for whom the physician had not yet reviewed the chest radiograph or radiologist interpretation prior to completion of the study survey. The median age was 3.3 years (IQR, 1.5–6.1). The majority of patients (92%) had a report of cough and 78% had a report of fever. On physical examination, 394 children (24.7%) had crackles noted, 66% of which were focal. Two hundred forty-one patients (15.1%) had radiographic pneumonia (Table 1).

The prevalence of radiographic pneumonia varied by the presence and type of crackles. Among the 1200 patients without crackles, 162 (13.5%) had pneumonia. Among the 134 patients with diffuse crackles, 20 (14.9%) had pneumonia. Among the 260 patients with focal crackles, 59 (22.7%) had pneumonia (P = .001). Of the 260 children with focal crackles on examination, 94 (36.2%) had clinician-diagnosed pneumonia (95% CI, 30.3%-42.3%), whereas 59 (22.7%) children (95% CI, 17.7%-28.3%) had radiographic pneumonia (P < .001). Among subgroups of children with focal crackles in combination with other clinical findings commonly associated with pneumonia, this difference persisted (Figure 1A). Among children with diffuse crackles alone or combined with other findings, there was no significant difference between the proportion with clinician-diagnosed pneumonia and the proportion with radiographic pneumonia (Figure 1B).

Among children with focal crackles, substantial discordance was found between the children with clinician-diagnosed pneumonia and those with radiographic pneumonia. Specifically, among the 94 children with focal crackles judged clinically to have pneumonia, only 29 (30.9%) had radiographic pneumonia. Conversely, of the 166 children with focal crackles judged clinically not to have pneumonia, 30 (18.1%) had radiographic pneumonia. Using radiographic pneumonia as the gold standard, the sensitivity of physician judgment in children with focal crackles was 49.2% (95% CI, 36.1–62.4) and the specificity was 67.7% (95% CI, 60.7–74.0). Across subgroups of children defined by the presence of focal crackles with other signs of pneumonia, sensitivity of physician judgment ranged from 49.2% to 76.9% and specificity ranged from 45.5% to 67.7% (Table 2).

In our multivariable model adjusted for other historical and examination features (excluding 31 patients with missing oxygen saturation), those with diffuse crackles were more likely to have clinician-diagnosed pneumonia (adjusted odds ratio [aOR], 2.4; 95% CI, 1.3–4.2) compared with children without crackles. However, in the model predicting radiographic pneumonia, no association was found with the presence of diffuse crackles on examination (aOR, 1.3; 95% CI, 0.7–2.1). Children with focal crackles had 6.8 (95% CI, 4.8–9.8) times the odds of clinician-diagnosed pneumonia compared with children without crackles. The presence of focal crackles was significantly associated with a higher odds of radiographic pneumonia (aOR, 1.7; 95% CI, 1.2–2.4) (Table 3). Sensitivity analyses imputing oxygen saturation values of 90%, 95%, and 100% for the patients with missing oxygen saturation information did not materially differ from the main analysis (Supplemental Table 1), nor did the sensitivity analysis restricting the model to children 2 years and older (Supplemental Table 2). In the sensitivity analysis excluding children enrolled during the early phase of the COVID-19 pandemic, we found similar results for clinician-diagnosed pneumonia (aOR, 6.3; 95% CI, 4.2–9.3 in the COVID-exclusion cohort and aOR, 6.8; 95% CI, 4.8–9.8 in the entire cohort); other results also did not materially differ from the main analysis (Supplemental Table 3).

Among children with signs and symptoms of pneumonia, we observed that although radiographic pneumonia was more common in children with focal crackles than in those without, physicians overestimated the value of focal crackles in determining the presence of pneumonia. To a lesser extent, diffuse crackles were associated with clinician-diagnosed pneumonia, but they were not independently associated with radiographic pneumonia. When focal crackles were combined with additional auscultatory findings, physicians overestimated the proportion of patients with pneumonia. We also found significant discordance between clinician-diagnosed pneumonia and radiographic pneumonia on the individual patient level. Taken together, our findings suggest that crackles are generally overvalued as a sign of pneumonia.

Focal crackles may be appreciated when pneumonia is localized to one region or lobe of the lung, which typically occurs with bacterial pneumonia. Conversely, atypical or viral pneumonia often presents with more diffuse findings on both physical examination and radiography. This explains the lower observed rates of both clinician-diagnosed and radiographic pneumonia among children with diffuse crackles as compared with those with focal crackles.

Although several single-center studies have observed an association between crackles and radiographic pneumonia in children,^10–13 data from 2 systematic review/meta-analyses summarizing data from thousands of children with suspected pneumonia observed that auscultatory findings were not highly associated with the presence of radiographic pneumonia.^2,3 Furthermore, crackles and decreased breath sounds have poor inter-rater reliability, potentially limiting their utility.¹⁴ Despite this, our findings suggest that physicians rely heavily on findings heard with a stethoscope when making decisions about which children with respiratory illness would benefit from antibiotic treatment. Our findings are supported by those from a recent survey of clinicians using artificially generated patient cases, in which focal crackles were associated with a 4-fold increase in CXR ordering and 3.5-fold increase in antibiotic prescribing.¹⁵ 

Based on the prevalence of radiographic pneumonia in children with focal crackles (22.7% for focal crackles alone and up to 37.1% in combination with other signs of pneumonia), clinicians would need to treat 3 to 5 children in these subgroups with antibiotics in order to treat 1 child with radiographic pneumonia. The significance of this, and its impact on clinical care, likely depends on practice location. For example, in settings where CXR is readily available, clinicians may choose to obtain CXR prior to making decisions about antibiotics. A negative chest radiograph effectively rules out pneumonia and so could be used as a tool to decrease antibiotic use, especially in children with focal crackles on examination.¹⁰ Although positive radiographs may still lead to unnecessary antibiotic use for viral etiologies, CXR is currently the standard for pneumonia diagnosis in children. On the other hand, clinicians practicing in a setting where CXR is not readily available may accept a number needed to treat of 3 to 5 as reasonable and prescribe empirical antibiotics based on physical examination findings.

On a more granular level, among patients with focal crackles, we observed substantial discordance between the clinical and radiographic diagnosis of pneumonia in individual patients. This suggests not only that physicians overvalue the importance of focal crackles in pneumonia diagnosis but also that they may also be clinically diagnosing pneumonia in the wrong patients, that is, providing antibiotics to patients who are unlikely to benefit from them and withholding them from patients who may benefit.

The observations in this study are consistent with those from other studies suggesting that health care providers generally overestimate the pretest probability of disease.^16,17 One potential way to mitigate this phenomenon is through the use of rigorously derived and validated clinical risk scores.^4,18–20 Clinical risk scores have the advantage of quantifying the effect of multiple predictors in an objective way, thus minimizing the “overvaluation” effect that we observed in this study. The Pneumonia Risk Score (PRS), for example, includes crackles as a predictor of radiographic pneumonia but accounts for their modest effect in the final score.⁴ A score such as the PRS could aid clinicians in more accurately identifying children with respiratory illness who are more likely to benefit from antibiotic treatment, with a focus on minimizing unnecessary CXR performance and antibiotic use.

As with most studies on the diagnosis of community-acquired pneumonia in children, ours is limited by the use of CXR as a gold standard for pneumonia. Bacterial pathogens are infrequently identified in children with pneumonia, even in those with radiographic infiltrates.²¹ This is likely partially due to technological limitations in isolating bacterial pathogens but suggests that even radiographic pneumonia may be purely viral in origin. Our question to clinicians about their intent to prescribe antibiotics should CXR not have been performed assumes a bacterial etiology for any pneumonia identified. Although this could in theory lead to “false” discordance (radiographic pneumonia in a patient without clinician intent to treat), analysis of our data shows that more than 90% of children with “definite” or “probable” pneumonia on CXR are treated with antibiotics. Thus, we believe clinicians are generally equating radiographic pneumonia with bacterial pneumonia, at least in terms of treatment decisions.

This study has several other limitations. First, this was a secondary analysis of a single-center study, and due to funding constraints, we were only able to enroll English-speaking families in this study. This may limit the generalizability of our findings, particularly to non-ED settings and in non–English-speaking patients. Second, clinician-diagnosed pneumonia was based on a physician’s answer to a binary question asked prior to knowledge of CXR findings. This is not synonymous with what would have happened had a chest radiograph indeed not been obtained, and it is possible that physicians’ real-life decisions may have differed from their hypothetical decisions. However, because all children in this study did undergo CXR, this question was the only way to approximate a clinical judgment of pneumonia. Third, although rales/crackles are specifically defined in the literature and used interchangeably,⁶ clinician interpretation of examination findings may vary widely. One examiner may use the strict definition of crackles, whereas another may label findings such as transmitted upper airway sounds or coarse rhonchi as crackles. Children with fine inspiratory crackles may be more likely to have radiographic pneumonia than children with coarse rhonchi, but given the difficulties in auscultation of the pediatric chest, it is unlikely that clinicians are able to consistently discern this type of crackles from others. Fourth, clinicians are often taught that the radiographic findings of pneumonia may lag behind clinical signs; thus, it is conceivable that some patients with focal crackles may have had falsely negative chest radiographs. However, prior work has demonstrated that this rarely occurs in children¹⁰ and would not materially impact our findings. Fifth, by enrolling only children who underwent CXR, we would not have included children at very low and very high risk of pneumonia who were managed without the use of CXR. However, CXR is used in more than 80% of children diagnosed with pneumonia at US children’s hospitals nationally^22,23; additionally, our division has a clinical pathway in place recommending the use of CXR in most cases of suspected pneumonia, mitigating this concern. The presence of a clinical pathway also mitigates the risk of Hawthorne bias as it relates to clinicians’ decisions to order CXR.

In summary, in this prospective study of children undergoing CXR for suspected pneumonia, we observed that the presence of focal crackles is strongly associated with clinician-diagnosed pneumonia and only modestly associated with the presence of radiographic pneumonia. These findings underscore the need for integration of validated clinical risk scores^4,18–20 that can more accurately risk stratify children and aid clinicians in making decisions about CXR and antibiotic usage.

aOR

adjusted odds ratio

CXR

chest radiography

ED

emergency department

GLM

generalized linear model

PRS

Pneumonia Risk Score

RC

research coordinator

RR

respiratory rate