The primary objectives of this study are to characterise lung sounds associated with various clinical diagnoses, radiologically confirmed consolidative pneumonia and diffuse interstitial pneumonia, bronchiolitis, asthma and URIs, in a paediatric population, and to determine if these diagnoses can be differentiated from normal through automated lung sounds analysis and compare with modalities of imaging, current WHO algorithm for ALRI case management and microbiological testing. We then aim to develop a clinical protocol pairing electronic auscultation with a CLSA algorithm to aid in pneumonia diagnosis.
Our design will be a cross-sectional study of lung sounds and other diagnostic modalities from children 2 to 60 months of age presenting with a primary respiratory complaint to the Instituto Nacional de Salud del Niño, a tertiary care hospital in Lima, Peru. Informed consent from parents will be obtained in the emergency department (ED), asthma ward or pulmonary ward where all testing will be performed in a single visit. Parents will be asked to fill out a questionnaire, while the physician reports relevant aspects of the physical exam. Electronic auscultation will then be performed, following by imaging and collection of blood, respiratory, urine and stool samples.
During the initial phase, we will record lung sounds from 600 children from 2 to 59 months of age, 100 each with consolidative pneumonia, diffuse interstitial pneumonia, asthma, bronchiolitis, URIs and normal lungs. The second phase will consist of completing our testing set for external validity and comparing CLSA with the current WHO algorithm and other diagnostic tools such as physical exam findings, chest ultrasound and microbiologic testing in order to construct an improved algorithm for pneumonia diagnosis.
Children from 2 to 59 months of age presenting to the ED or in the asthma or pulmonary ward without a history of chronic lung disease, excluding asthma, or significant cardiac disease will be invited to participate in the study. Children with respiratory complaints will be invited to participate as potential cases, while those without respiratory complaints and no acute respiratory illness within 1 month of presentation will be invited to join the study as controls.
Children will be considered eligible if their parents or guardians are able to provide written informed consent, and they themselves do not require airway management or non-invasive ventilation. Children will be considered ineligible if they have chronic lung diseases other than asthma, such as cystic fibrosis, bronchiectasis and chronic lung disease of prematurity, or significant congenital heart disease. Patients will be considered ineligible post-consent if they were found to have more than one active respiratory diagnosis upon further testing. Group classification also may be modified post-consent and further enrolment required depending on chest x-ray (CXR) final readings and microbiological testing for diagnosis.
Outcomes and case definitions
Because there is no gold standard for diagnosis, we aim to compare our results with common case definitions and clinical diagnoses by experienced physicians. Secondary outcomes will incorporate aetiology information from standard culture and molecular techniques; however, these additional data will not serve as the gold standard.
Pneumonia will be initially categorised upon clinical diagnosis by examining paediatricians at El Instituto Nacional de Salud del Niño and further characterised as consolidative or diffuse interstitial pneumonia given final CXR reading by blinded radiologists from the Johns Hopkins University. Asthma will be defined by the presence of wheeze on physical exam, history of asthma and improvement with bronchodilators. Bronchiolitis will be defined as the presence of wheeze and difficulty breathing on physical exam and viral symptoms (cough, rhinorrhoea), no history of asthma and little or no improvement with bronchodilators if attempted. URI will be defined as respiration rate <50 breaths per minute and associated with one or more of the following: clear nasal secretions, sore or red throat or hoarseness.
Our recruitment goal is 600 subjects, 100 each with consolidative pneumonia, diffuse interstitial pneumonia, asthma, bronchiolitis, URI and normal lungs. Sample size was powered to improve specificity of WHO algorithm upon the addition of electronic auscultation. To detect an improvement in diagnostic specificity from 50% (WHO algorithm for paediatric pneumonia) to 80% (CLSA and WHO algorithm) with 95% power and α of 0.05 between pneumonia and non-pneumonia cases, we require 70 patients per group in the training set. We will also recruit a test set consisting of an additional 30 patients per group (30% of total sample) to estimate areas under the curve for our diagnostic algorithm. Each group will be over-enrolled by 20 to account for post-consent ineligibility, for a total of 720.
A.B. PRISMA in Lima, Peru, and Johns Hopkins University in Baltimore, USA, will provide administrative oversight for the study. There will be a research coordinator at a central location in Lima, Peru, who will provide logistical support and management of the study team. Instituto Nacional de Salud del Niño will provide a team of study nurses and physicians to carry out recruitment, physical examination and collection of specimens. We have also established prior training by an experienced ultrasonographer to conduct chest ultrasonography. A multidisciplinary team of clinicians, field epidemiologists, acoustical engineers and biostatisticians from Johns Hopkins University, Tufts University, Cincinnati Children's Hospital and Instituto Nacional de Salud del Niño will be involved with study design and conduct, statistical analysis and reporting of results.
We will ask the parent or guardian about the child's past medical history, environmental exposures, access to healthcare and current respiratory symptoms. We will enquire about demographic information, nutrition and vaccination history. We will ask about co-morbidities, family history and developmental history. Environmental questions pertained to housing, number of children, rural versus urban living, parent occupations, smoke and allergen exposure, and sick contacts. Current respiratory symptoms were asked of the parent or guardian to answer subjectively and included rapid breathing, difficulty breathing, chest indrawing, cyanosis, cough, sputum production, audible breath sounds and subjective fever.
Physical exam and laboratory testing
The initial set of vital signs will be recorded, including pulse oximetry. During the physical exam, a single examining physician from the larger group of study physicians will be responsible for recording findings for a given patient with emphasis on the respiratory exam. Chest retractions, nasal flaring, grunting, stridor and accessory muscle use will be noted and characterised if present, along with any adventitious lung sounds appreciated by physician and study team member on chest auscultation. Degree of improvement after bronchodilators will also be recorded if administered. Additionally, signs of dehydration and malnutrition will be reported if present. Laboratory results will be recorded if evaluated by the ED and include complete blood count, electrolytes and arterial blood gas.
Parents will be allowed to position the patient supine or upright. The study team member will listen to eight auscultation sites using a ThinkLabs ds32a Digital Stethoscope and mp3 recorder, for 10 s at each site, in the following order of placement: front top left and right, fronterolateral bottom right and left, back top right and left and back bottom left and right (). Auscultation will be performed at the participant's normal breathing patterns during recording without being asked to take deep breaths. We will allow only one repeat of auscultation if recording is interrupted for any reason or due to unacceptable signal quality of the first recording.
Order of auscultation by electronic stethoscope. The study team member will listen to each site, starting with ‘A’ for 10 s each.
All participants will receive bilateral lung ultrasonography carried out on SonoSite portable ultrasound machine with HFL38/13–6 MHz and P17/5–1 MHz MicroMaxx®
transducers by a single ultrasound technician who has been trained to the standardised protocol. Patients will be examined in the supine position with each hemithorax divided into six sections: two anterior, two lateral and two posterior. The posterior area is defined from the posterior axillary line to the paravertebral line. Longitudinal and oblique scans will be obtained at each of the chest zones. Longitudinal scans allow visualisation of the ribs with the pleural line under them.23
Representative images from each section will be saved and later transferred to radiologists at an outside institution.
To assess for pneumonia, two of three radiologists must agree on the description of ultrasound images compatible with pneumonia. Consolidation will be determined by (1) the presence of hypo- or anechoic images with loss of distinct pleural lines and (2) an irregular shredded border of the pleural line that is distinct from the lung line, termed the ‘shred sign’. Additional signs to be reported will include punctate hyperechoic images reflecting air bronchograms, decreased lung sliding and homogeneous hypoechoic images in the pleural space corresponding to pleural effusions. Interstitial infiltrates will be determined by the presence of ‘lung rockets’, which correlate to three or more B-lines in a longitudinal view between two ribs. Additional features to be reported include heterogeneous echotexture, air and fluid bronchograms, lung pulse and additional B-lines.23
All case participants will undergo chest radiography. We will attempt postero-anterior and lateral films but will allow an antero-posterior view if not possible. Digital images will be sent to a third party reading group blinded to clinical information. Using WHO standardisation of CXR interpretation for paediatric pneumonia,24
radiologists will comment on quality as adequate, suboptimal or unreadable and on the presence of pathology as consolidative or interstitial with or without pleural effusions. Radiographic evidence of pneumonia will be confirmed by agreement by two of three radiologist reports for a given patient. Additional pathologic findings not previously characterised will also be recorded and reported to the patient's physician for further intervention if necessary.
Blood, urine and nasopharyngeal samples will be collected according to our study design (). Blood samples will be drawn for cultures and sensitivities. Urine will be tested for detection of pneumococcal antigen and stored for PCR. Nasopharyngeal swabs will be tested for respiratory viruses, along with culture and sensitivities. Respiratory pathogens tested by PCR will include Streptococcus pneumoniae, Haemophilus influenzae and respiratory syncytial virus.
To ensure safety, the researcher collecting data is experienced with providing care to children. The researcher will use this experience to minimise any discomfort the children may have. All blood samples will be collected by a skilled nurse or phlebotomist.
We will adhere to hospital procedures for avoiding hospital-acquired infections. We will wash hands with soap and water or alcohol-based hand sanitizer before patient contact. We will wear gloves, gowns and mask when required. We will clean the devices using alcohol swabs before and after each use.
Data quality and management
Prior to data collection, a Manual of Operations will be developed to ensure standardisation and reliability and contain detailed instructions for all study procedures and guidelines for data collection. The manual will be revised as needed and distributed to members of the study team.
All data are recorded first on paper case report forms and subsequently double entered using Microsoft ACCESS. Data sets will be cross validated and errors corrected. Electronic lung recordings will be transferred from the mp3 player to participant-specific files on the study computer at least every other day and backed-up weekly. Digital CXR images will also be uploaded to these files and backed-up similarly.
Analysis of lung sounds and statistics
An important first step in CLSA is using common signal processing techniques to investigate high- and low-frequency information using methods such as the Short-time Fourier Transform, wavelet transforms and P-spline bases. The second step will be extracting signal processing features to train the classifier.
Based on preliminary recordings to date performed by our group in Baltimore (), we anticipate that wheeze can be characterised using features from the Fourier transform, such as the existence and temporal stability of tonal peaks in the 300–1000 Hz range, while crackles could be recognised using features such as amplitude, the presence of broadband energy and the duration of this energy. Features such as the decrease in signal energy with frequency can characterise movement sounds. We have previously used time–frequency descriptors such as Mel-frequency cepstral coefficients as features and anticipate they will be useful for CLSA but may require temporal information as well. We will use the extracted features from signal processing analyses for classification using machine learning algorithms including: nearest neighbour methods, support vector machines, random forests and gradient boosting. Primary analysis will consist of a fivefold cross validation on the training set to calculate expected prediction errors. The training set will additionally be used to estimate areas under the curve for our diagnostic algorithm, including CLSA, WHO algorithm, imaging and physical exam findings. Secondary analysis will include calculating sensitivities and specificities of experimental diagnostic ultrasound for detection of pneumonia when compared with gold standards (clinical diagnosis and CXR reading). Performance will be measured using logistic and multinomial regression, receiver operating characteristic curves and area under the curve.
Figure 3 Preliminary data suggest a difference in spectral analysis between children with and without wheeze. Short-time Fourier Transform analysis was used to visualise spectrograms of a normal control (A) and asthmatic child with active wheeze (B). Representative (more ...)