Focused Ultrasound of the Heart and Lungs in Patients With Respiratory Failure: A Randomized Controlled Multicenter Trial.


 Background Ultrasound is a focus oriented tool for differentiating among cardiopulmonary diseases. Its value in the hands of emergency physicians, with various ultrasound experience, remains uncertain. We tested the hypothesis that, in emergency department patients with signs of respiratory failure, a focused cardiopulmonary ultrasound along with standard clinical examination, performed by emergency physicians with various ultrasound experience would increase the proportion of patients with correct diagnoses at 4 hours after admission compared to standard clinical examination alone. Methods In this prospective multicenter superiority trial in Danish emergency departments we randomly assigned patients presenting with acute signs of respiratory failure to intervention or control in a 1:1 ratio by block randomization. Patients received a focused cardiopulmonary ultrasound examination within 4 hours from admission. Ultrasound results were unblinded for the treating emergency physician in the intervention group. Final diagnoses and treatment were determined by blinded review of the medical record after the patients´ discharge.Results From October 9, 2015 to April 5, 2017, we randomized 218 patients and included 211 in the final analyses. At 4 hours we found; no change in the proportion of patients with correct presumptive diagnoses; intervention 79·25% (95% CI 70·3-86·0), control 77·1% (95% CI 68·0-84·3), an increased proportion of correct treatment prescribed; intervention 79·3% (95% CI 70·3-86·0), control 65·7% (95% CI 56·0-74·3) and of patients who spent less than 1 day in hospital; intervention n=42 (39·6%; 25·8 38·4), control n=25 (23·8%; 16·5-33·0). No adverse events were reported. Conclusions Focused cardiopulmonary ultrasound added to standard clinical examination in patients with signs of respiratory failure had no impact on the diagnostic accuracy, but significantly increased the proportion of correct treatment prescribed and the proportion who spent less than 1 day in hospital.Trial registration https://clinicaltrials.gov/, number NCT 02550184.

The amount of diagnostic accuracy studies is vast and we are short of studies of the F-us´ impact on patient related outcomes. Results from diagnostic accuracy studies are rarely directly translational to clinical practice hence we need randomized pragmatic studies to investigate the potential impact of F-us in daily clinical practice.
We investigated the hypothesis that adding a cardiopulmonary F-us to standard clinical examination of patients admitted to the ED with acute signs of respiratory failure could increase the proportion of patients with a correct presumptive diagnosis at 4 hours (4 h) after admission when compared to standard clinical examination alone. 15

Study design
This prospective randomized semi-blinded, multicenter, superiority trial with a parallel group design and allocation ratio of 1:1 was undertaken in 10 of the 21 EDs in Danish community Hospitals. Three of these EDs were located at university hospitals and seven in secondary or tertiary hospitals. Patients were enrolled from October 9, 2015 to April 5, 2017.
Acute admissions to Danish EDs are to public hospitals and established by the general practitioner or by emergency call.
Exceptions are patients suspected of having heart disease who are admitted directly to the cardiology department.
A presumptive diagnosis at admission and at 4h as well as a treatment plan must ideally be journalized by the EP (emergency physician) within 4h from the patient´s admission to the ED.

Inclusion and exclusion
We screened patients admitted to the ED for inclusion. Patients ≥18 years with a primary sign or symptom of respiratory failure of; cough, dyspnea, chest pain, respiratory frequency >20 or peripheral oxygen saturation < 95% or any combination of these were included upon written informed consent. Exclusion criteria were inability to give written informed consent, Fus of the lungs or heart already performed by others than the investigator in relation to the primary examination, or inability to randomize or perform the F-us within 4h from the patient´s admission to the ED.

Randomization and masking
Randomization numbers were created by a block randomization database (REDCap, OPEN) using permuted blocks of random numbers to ensure equal numbers of patients in both trial arms at each center. 16 The allocation sequence of randomization numbers was generated by a data manager from REDCap OPEN and the project manager, MR, using an online random number service. Randomization numbers were paired with the REDCap OPEN database developed for this trial by MR.
The investigators screened the patients for inclusion at alternating shifts including day and night shifts on all weekdays. Once a patient signed the informed consent, the investigator registered the patient in the trial database that randomly allocated the patient and assigned the patient a unique computerized randomization number. Patients were aware of their group assignment.
The investigator then noted the basic clinical values upon the patient´s admission and performed the F-us unaware of the primary presumptive diagnoses. All F-us results were entered in the project database. F-us results from patients in the control group remained blinded to the EP, whereas once the investigator had received the primary presumptive diagnoses, he unblinded the intervention groups F-us results to the EP both orally and in writing on a paper record marked with the patients ID number and stored it in an accessible briefcase in the ED. The EP was then free to re-evaluate his presumptive diagnoses and treatment according to the ED´s clinical guidelines. The EP was instructed to leave all information of randomization and F-us results out of the medical records to ensure blinding of the medical record audit.
In the medical record the investigator only noted the projects identi cation number and localization of the paper records containing the F-us results.

Procedures
In all patients admitted to the ED the EP assessed the primary presumptive diagnoses using standard methods of diagnostic examination (e.g., clinical examination, blood samples, ECG, chest x-ray) as soon as possible after the patient´s arrival.
Standard diagnostic tests were available within 4h. Supplementary imaging examinations such as computerized tomography (CT), ultrasonography and echocardiography performed by specialists were available if necessary.
The investigator performed the F-us blinded to the EP in both groups. Subsequently, the EP announced his primary presumptive diagnoses to the investigator who then unblinded the F-us ndings in the intervention group.
The F-us was performed within 4h from the patients´ admission to the ED and consisted of F-us of the heart and lungs. The F-us protocol was de ned as follows: The F-us of the lungs was a modi cation of the ultrasound protocol used by Laursen et al. originally modi ed from the principles of lung ultrasound by Volpicelli and Lichtenstein. 12,17,18 It was performed as follows: The anterior and lateral part of thorax was divided into a superior and inferior quadrant. Each quadrant represented a zone in which the probe was placed centrally to create a transverse picture of the costae and pleurae. We looked for pleural effusion, interstitial syndrome/pulmonary edema and pneumothorax.
The focused cardiac ultrasound was performed according to the principles described in the international evidence based guideline. 19 The views used were the 4-chamber picture of the heart achieved either from a sub-xiphoid or an apical window. We looked for pericardial effusion, altered left ventricular ejection fraction and right ventricular overload.
There were no regulations in the choice of ultrasound machines or probes used for inclusion as long as the F-us was performed with an image quality deemed su cient for evaluation by the operator.
All F-us examinations were performed by investigators who were either specialists or in specialist training and who received patients in the ED on a regular basis. All investigators used F-us on a daily basis in their clinical practice but had varying degrees of F-us experience. Prior to become an investigator they all received an educational program regarding data collection and F-us examination, which was composed by MR and consisted of an e-learning presentation with instructional videos. 20 Then, MR made a 4h on site presentation comprising the project, the collection and registration of data and an introduction to the project´s F-us protocol. Hereafter, MR evaluated each investigator´s F-us skills and handling of the project database by hands on and by multiple choice questionnaire tests to ensure that the investigators were familiar with their investigator tasks prior to initiation of inclusion. During the period of inclusion, investigators could take daily contact to the project manager for questions.
The diagnostic criteria for the F-us examination are provided in Appendix 1.

Diagnostic examinations
The primary presumptive diagnoses were registered in the medical record. The EP was free to reassess the presumptive diagnoses, diagnostic tests, and treatment.
The 4h presumptive diagnosis was the last registered clinical examination made by the EP within 4h from the patientsá dmission to the ED and was assessed by blinded audit of the medical record after the patients discharge from hospital. New diagnostic tests and treatments could be prescribed after the 4h examination but would then be part of the nal diagnosis.
The nal diagnosis was assessed by blinded audit of the medical record and included electronic journal data (e.g. clinical, microbiological and biochemical data, and imaging results) and was performed after the patients´ discharge from hospital by two auditors, who independently of each other, set the nal diagnosis. In case of discrepancy a third auditor set the nal diagnosis. A prede ned audit protocol with diagnostic criteria of the most common diagnoses was used (Appendix 2).

Outcomes
The primary outcome was to assess the percentage of patients with a correct presumptive diagnosis at 4h after the patient s admission to the ED. The 4h cut off was set because various Danish hospitals request that examination, presumptive diagnoses and plan for further treatment are assessed within 4h from admission to the ED. 21 The secondary outcomes were as follows: 1. Diagnostic accuracy of the primary presumptive diagnoses made upon arrival.
2. The proportion of patients receiving a correct primary presumptive diagnosis made upon arrival.
3. The proportion of patients who, within 4h after admission to the ED, is given the correct treatment.
4. Time spent in the ED (hours).

Time spent in hospital (days).
6. The proportion of patients being transferred from the ED to the intensive care unit.
7. The proportion of patients being transferred from the ED to a hospital ward.
8. The proportion of patients being discharged from the ED and directly to their home. 9. The proportion of patients being readmitted to hospital within 30 days from discharge.
As a negative F-us can be found in various pulmonary diseases (e.g. chronic obstructive pulmonary disease (COPD), asthma) we could not compare F-us ndings directly with nal diagnoses and were omitted from the secondary endpoint The proportion of F-us examinations with a correct diagnostic examination´ as written in the published protocol. 15 Unblinding For safety and ethical reasons the F-us ndings in the control group were unblinded to the EP in charge of the patient if the F-us raised suspicion of a life-threatening condition (e.g. pulmonary edema, pneumothorax, pericardial effusion, or heart failure). 15 Statistical analysis The sample size estimate was based on the results from a previous similar trial where about 65% of the patients had a correct diagnosis at 4h after admission to the ED when F-us was not used. 11 A clinically signi cant improvement of the diagnosis by using F-us in a multicenter trial was set to be 15%.
To detect a 15% increase in the number of correct tentative diagnoses, from 65% in the control group to 80% in the intervention group, with an 80% chance for detection, a level of signi cance of 5% and with an estimated drop out of 6% we had to include 288 patients, 144 patients in each group (intervention/control). Sample size calculations were made with the online database for clinical trials; Sealed Envelope. 22 We used the intention-to-treat method as main comparative analysis on all participants. Descriptive statistics were handled as follows: Categorical data by number and percentages of patients with 95% con dence intervals (CI), continuous data by number of patients (n), mean, standard deviation, median, minimum and maximum.
Missing data in the baseline characteristics were handled as simple imputation when represented as dichotomous data.
Other missing data were evaluated to be missing at random and was handled by multiple imputations using the Markov Chain Monte-Carlo method including auxiliary variables in the model. We plotted the mean value of the imputed variables in the spot of missing data. The rule of three was used to nd 95% CI in categories without events. Categorical endpoints were summarized by numbers and percentages with 95% con dence intervals. 23 We used the Chi 2 test and the Fischer exact test for comparison of proportions expressed as percentages and for the comparison of continuous endpoints we used the Student t test (means) and the Mann Whitney test (medians). A twosided signi cance level of 5% was applied to all tests.
We used the diagnoses from blinded audit as reference test for determining the diagnostic accuracy of the presumptive diagnoses established at admission and at 4h and their 95% con dence interval based at binominal distribution. To assess the interrater reliability of the nal diagnoses we used the Cohens kappa coe cient. Data analyses were conducted using STATA Release V. 15.0 (Stata Corp) including professional statistical advice. To assess the time a patient spent in the ED or hospital and the mortality we withdrew data from the Danish National Patient Register (LPR) administered by the Danish Health Data Authority. No data monitoring committee oversaw the trial.

Role of the funding source
Funders of this trial had no role in the trial design, data collection, data analysis, data interpretation or writing of the report. The corresponding author had full access to all the trial data and had nal responsibility for the decision to submit for publication.

Results
We included 220 patients from October 9, 2015 to April 5, 2017 in 10 hospitals counting three university hospitals and seven tertiary hospitals with a total of 21 investigators. The trial was ended before we reached the prede ned sample size as investigators travelled to other departments and the inclusion stagnated. We initially excluded 9 patients: two patient contacts due to double inclusions (their latest inclusions were excluded), one patient who withdrew informed consent, one patient who had no randomization or F-us within 4 h from admission and seven patients due to loss to follow up.

Figure 1 Trial pro le (appended)
There are no deviations of the F-us examination in either group due to adverse events or complications related to the trial.
For patient baseline characteristics see Table 1. For ultrasound ndings see Appendix 3.  Final diagnoses were assessed by the two auditors with an overall agreement of 93·18% (kappa 0·58) ranging from 81·04-100% (kappa − 0·02-1·0) within the speci c diagnoses. A detailed description of the interrater reliability of the audit is provided in Appendix 4. The most common nal diagnoses in the intervention and control group were; pneumonia (25%/ 32%), exacerbation of COPD (25%/27%) and systolic heart failure (16%/21%). The proportion of correctly assessed diagnoses at 4 h in the intervention group is increased compared to the control group for the diagnoses of exacerbation of COPD (89%/75%), pulmonary edema (53%/33%) and para-pneumonic effusion (77%/38%) but with overlapping CIs. The proportion of correctly diagnosed pneumonia is equal among the groups (88%/88%) (  Intention to treat population. Abbreviations: n (number of patients). COPD (Chronic obstructive pulmonary disease). ILD (interstitial lung disease). *Too few ratings to perform diagnostic accuracy calculations.
Moreover, a signi cantly increased proportion of patients in the intervention group spent less than 1 day in hospital, n = 42 (39·6%; 25·8-38·4) compared to the control group n = 25 (23·8%; 16·5-33·0) (p = 0·01) clari ed by an absolute increase of 15·8%. Aside from that we found no signi cant changes in the proportion of time spent in the ED or in hospital, in the patients´ itinerary after discharge from the ED or in the proportion of readmissions and mortality ( Table 4).
The diagnostic accuracy of the 4 h presumptive diagnoses compared to nal diagnoses are listed in Table 2 and added contingency tables in Appendix 5. For `Other diagnoses´ than nal diagnoses, see Appendix 6. Causes of death in patients who died within 30 days from admission are summarized in Appendix 7. For per protocol analysis, see Appendix 8.
No adverse events related to the F-us were observed.

Discussion
In this trial we found that adding F-us to standard clinical assessment in adult patients with acute signs of respiratory failure led to a non-signi cant, absolute increase of 2·1% (p = 0·71) in the proportion of patients with correct diagnoses 4 h after admission to the ED. We observed a signi cant, absolute increase of 13·5% (p = 0·03) in the proportion of patients with appropriate treatment prescribed within 4 h from admission and of patients who stayed less than 1 day in hospital (Table 4).
To our knowledge, no similar randomized multicenter trial has been performed in adult ED patients admitted with signs of acute respiratory failure looking at both the diagnostic accuracy and the proportion of correct treatment prescribed and length of stay when adding F-us to standard clinical assessment.
Apart from a slight imbalance in the distribution of gender, the demographic and clinical characteristics were well matched and in combination with the vital signs and medical history upon admission, our patient population was comparable to similar studies. 11,24 The high 4 h diagnostic accuracy found in the control group is surprising when compared to a similar single center trial, which reported an absolute increase of correct diagnoses of 24·3% (from 63·7% to 88·0%, 95% CI 15·0-33·1) and of appropriate treatment prescribed of 21·2% (from 56·7-78%, 95% CI 10·8-30·9) at 4 h after the patients admission (p < 0·0001). They added cardiopulmonary F-us to standard clinical assessment performed by a single physician specialized in point-of care ultrasound. 11 However, this does not explain the high 4 h diagnostic accuracy we found in both groups nor the signi cant increase of `correct treatment prescribed´ that we found in the intervention group. 11 Instead, the high diagnostic accuracy at 4 h might be explained by the augmented focus on getting more specialist doctors in front in the rst critical hours from the patients´ admission to the ED which may have increased the diagnostic accuracy in general.
The signi cant increase of 13.5% in the proportion of patients in the intervention group who, at 4 h from admission, had an appropriate treatment prescribed despite an insigni cant difference in diagnostic accuracy was similar to the single center study by Laursen et al. who found an absolute increase of 21·2% (Table 4). 11 Laursen et al. found that the diagnostic tests prescribed within the rst 4 h in the intervention group had a higher proportion of tests in which the suspected diagnoses were con rmed, but that the proportion of diagnostic tests prescribed evened out when compared throughout the entire hospital stay. 11 These results indicate that the implementation of F-us leads to faster execution of diagnostic tests earlier in the hospital stay to con rm or reject a presumptive diagnosis and could be a contributory cause to the signi cant increase in the proportion of correctly prescribed treatment at 4 h in the intervention group. Finally, combined with the signi cant increase in the proportion of intervention patients who spent less than 1 day in hospital, these results indicate that the F-us has a positive impact on several clinical outcome parameters.
Apart from the above mentioned, we found no difference in the patients´ itinerary, time spent in hospital or in the rate of readmissions or mortality as found by Laursen et al. 11 The length of time spent in the ED or in hospital is often fairly short hence, larger studies are needed to assess whether F-us has a bene t on these parameters.
Finally, in the control group, the investigator revealed the F-us ndings to the EP in two patients and in four patients the investigator became the treating physician within 6 h from the patients admission (Fig. 1).
In our study population we nd a low number of severely ill patients (Table 1) and surprisingly few with pulmonary edema, pulmonary embolism and acute myocardial infarction despite a greater proportion of patients with systolic heart failure than found in similar trials. 11,24,25 We cannot rule out that a selection bias has taken place where severely ill patients have been deselected due to the more cumbersome process of including these patients in a study trial.
We may have reduced the number of nal diagnoses of pulmonary edema (Table 2) as the investigators´ F-us nding of pulmonary edema did not overrule a negative chest x-ray in the diagnostic criteria despite that F-us of the lungs has shown higher diagnostic accuracy. And as we did not implement pro-brain natriuretic peptide (BNP) in the diagnostic criteria, we may have reduced the number of nal diagnoses of cardiac pathology. 9 The proportion of patients with pulmonary embolism or AMI might be low due to the Danish emergency department setting that bypasses the ED and directs these patients straight to the department of cardiology or to outpatient clinics and because the prevalence of cardiopulmonary diseases is known to vary across studies despite similar inclusion criteria (Table 2). 11 24 26 Hence, we nd no reason to believe that the manner of evaluating the nal diagnoses plays a signi cant role in the results as it has been tested in similar studies. 11,27,28 This study indicates that cardiopulmonary F-us is in the hands of non-specialist sonographers can impact patient treatment, underlined by studies of steep learning curves which demonstrates that novice sonographers can nd pathology comparable to gold standards. 13,14 The pragmatic multicenter trial design increases the generalizability of the results.
We limited the subjective bias from the blinded audit of the nal diagnoses by prede ned diagnostic criteria based on internationally accepted guidelines. However, the kappa agreement between the two auditors was only weak to moderate according to Cohen´s guidelines, probably because the auditors had to agree on all diagnoses to obtain agreement or because the kappa statistic values tend to underestimate the agreement in situations with high inter-observer agreement (Appendix 4). However, the inter-observer variability re ects daily clinical dilemmas where a nal clear diagnosis is not always possible to establish.
The study has limitations. Firstly, we never reached sample size (inclusion: 220 versus sample size: 288) which might have caused the lack of statistically signi cant results. Second, enrolment only took place when the investigators were present in the ED which might have caused bias. Third, the broad inclusion criteria prompted the inclusion of a large number of patients with diseases others than cardiopulmonary. Compared to other studies we ended up with several patients who had pathology that the F-us has no diagnostic impact on which may have caused a reduction in the absolute effect of our outcome parameters. 11 Fourth, our diagnostic criteria may have been too strict as we omitted from the BTS criteria indicating that pneumonia could be diagnosed if there is "no other explanation for the illness, which is treated as community acquired pneumonia (CAP) with antibiotics" (Appendix 2) and from the investigators positive F-us ndings of pneumonia, despite lung F-us has proven to be superior to chest x-ray in diagnosing pneumonia: This is emphasized by a higher proportion of diagnoses of pneumonia at 4 h than at nal diagnoses in both groups and a lack of increase in the diagnostic accuracy (Table 3). 8 29 Fifth, the general absence of a CT-thorax or complete echocardiography in the assessment of the nal diagnoses is a weakness in the standard criteria of the nal cardiopulmonary diagnoses and may have reduced the amount of signi cant ndings. But, as it applies to all included patients we believe it to be of less importance.
Our study was not designed robust enough to enlighten the actual impact that F-us might have on acute as well as on hard clinical outcomes as patient morbidity, mortality and time spent in hospital. Larger studies are needed to specify the recommendations and the selection of patients whom would bene t the most from this diagnostic test as it has substantial healthcare and socioeconomic potential.  protocol has previously been published. 15 All patients provided signed informed consent prior to participation.

Consent for publication
No individual person´s data are presented in this manuscript.

Availability of data and material
The datasets analyzed during the current study are available from the corresponding author on reasonable request.