Lung Ultrasound for Diagnosis of Pneumonia in Children

Diagnostics and Likelihood Ratios, Explained

Positive Findings (Patient Has This)

Finding Increased Disease Probability (Positive Likelihood Ratio)
Pneumonia diagnosis 13.4
Sensitivity: 94%
Specificity: 94%

Negative Findings (Patient Doesn't Have This)

Finding Decreased Disease Probability (Negative Likelihood Ratio)
Pneumonia diagnosis 0.07


The diagnosis of pneumonia remains a diagnostic challenge without an acceptable gold standard.1 Chest X-ray has poor test performance and only moderate inter-observer reliability,2 while computed tomography involves radiation exposure levels too high for routine use. A growing body of literature shows lung ultrasound (LUS) to be a potentially useful tool for the diagnosis of pneumonia in both adults3 and children,4 with the added benefit of no exposure to ionizing radiation.

Orso and colleagues performed a systematic review and meta-analysis on this topic.5 Their literature search identified 17 studies evaluating diagnostic accuracy of ultrasound for pneumonia in children. The meta-analysis was applied only to LUS for pneumonia and excluded studies of other diagnoses. All but two of the identified studies were prospective. Studies were evaluated for risk of bias by the QUADAS 2 tool.5 The systematic review reports the diagnostic accuracy results as sensitivity, specificity, and area under the receiver operating characteristic (AUC) curve with respective interquartile range (IQR) between 25% and 75% quartiles.

Overall LUS for pneumonia had a sensitivity of 94% (IQR, 89% to 97%) and specificity 94% (IQR, 86% to 98%). These correspond to a Likelihood ratio for a positive test (LR+) of 13.4 and a negative test (LR-) of 0.07. The area under the receiving operating curve was calculated to be 0.98 (IQR, 0.94 to 0.99), indicating a strong diagnostic accuracy.5

The studies had good agreement on sonographic definitions for pneumonia which included consolidation (hypoechoic subpleural area) and may have also included interstitial pattern, focal B-lines, and sonographic air bronchograms. One study did not report their definition. Those who reported scanning technique tended to follow imaging protocols similar to that described by Copetti, 2008.6

The authors identified retrospective study design of two included studies as the only high risk of bias. They assigned “uncertain risk” of bias to reference standard, convenience sampling, and timing of LUS.


The most significant limitation is the lack of uniform reference standard. This is a challenge in pneumonia research as there is no accepted gold standard. The authors did an analysis evaluating the different reference standards, finding highest performance of LUS when compared to clinical diagnosis and adjudication. The concordance between Chest X-ray (CXR) and LUS was weaker, possibly because LUS likely outperforms CXR for identifying consolidation.6, 7 The comparison to CT is difficult to evaluate in this study given small number of subject who had CT. This metaanalysis found the use of CXR as reference standard to be the primary source of attributable heterogeneity amongst the reviewed studies.

The studies were variable in terms of setting and who was performing ultrasounds. In the majority, the sonographer was a pediatrician. Other scans were performed by emergency physicians, radiologists, and pediatric subspecialists. Similarly, the levels of ultrasound training and experience among sonographers were variable, ranging from 1 hour to 25 years. The most recent meta-analysis on this topic by Tsou et al. found significant difference between novice and experienced sonographers.8 While Tsou et al. included more studies due to their emphasis on sonographer training and not limiting to English language, they found similar pooled test characteristics for performance of LUS for pneumonia.

Other limitations include, all published studies used convenience sampling, many had relatively small sample sizes, and none were randomized controlled trials. These limitations contribute to a risk of bias that the systematic review by Orso et. al. may have underestimated. It is important for us to emphasize that imaging findings (i.e. consolidation) may be different than clinical pathology (i.e. pneumonia) and may have variable clinical relevance. Are small (‘sub-centimeter’) lesions clinically significant? Could they suggest viral or bacterial etiology, or be anatomical variants? Are they prognostic? Do they indicate need for antibiotics? These and other questions continue to challenge this field of research.

Globally, it seems likely that LUS performed by experienced sonographers is a diagnostic test that may help identify lung findings among children being evaluated for pneumonia. The authors of the systematic review discussed here5 astutely cautioned against wholeheartedly taking the overly favorable results. Notably, prominent research limitations (e.g. heterogeneity of studies) raise concerns of reliability, and potential biases (e.g. lack of reference standard) challenge the optimistic results.

The original manuscript was published in Academic Emergency Medicine as part of the partnership between and AEM.


Lilly A. Bellman, MD; Yiju T Liu, MD
Supervising Editor: Shahriar Zehtabchi, MD


February 3, 2020

What are Likelihood Ratios?

LR, pretest probability and posttest (or posterior) probability are daunting terms that describe simple concepts that we all intuitively understand.

Let's start with pretest probability: that's just a fancy term for my initial impression, before we perform whatever test it is that we're going to use.

For example, a patient with prior stents comes in sweating and clutching his chest in agony, I have a pretty high suspicion that he's having an MI – let's say, 60%. That is my pretest probability.

He immediately gets an ECG (known here as the "test") showing an obvious STEMI.

Now, I know there are some STEMI mimics, so I'm not quite 100%, but based on my experience I'm 99.5% sure that he's having an MI right now. This is my posttest probability - the new impression I have that the patient has the disease after we did our test.

And likelihood ration? That's just the name for the statistical tool that converted the pretest probability to the posttest probability - it's just a mathematical description of the strength of that test.

Using an online calculator, that means the LR+ that got me from 60% to 99.5% is 145, which is about as high an LR you can get (and the actual LR for an emergency physician who thinks an ECG shows an obvious STEMI).

(Thank you to Seth Trueger, MD for this explanation!)