This study was conducted in Kisarawe district on the eastern coast of Tanzania. The District Medical Officer selected interested general clinicians who were at the professional level of Clinical Officer or above at health centers within the district. Within the Tanzanian health system Clinical Officers have undergone a 3-year (2 years of course work with 1 year of clinical clerkships) training program that leads to a Diploma in Clinical Medicine. Within the health system in Tanzania, the Clinical Officer scope of practice includes family and emergency medicine without immediate supervision in the most peripheral sectors of the health system, namely dispensaries and health centers. Physicians are regularly available at the district hospital, but cardiologists are regularly available only in higher level hospitals in large cities.

An adult learning approach was used and included didactic and practical sessions. All trainees attended the same in-person training over a 5-day period, Monday through Friday approximately 9 a.m. to 5 p.m. daily. Didactic sessions including PowerPoint presentations comprised approximately 20% (7 h total) of the training period and focused on basic cardiac anatomy and physiology, introduction to cardiac ultrasound, the focused cardiac scanning protocol, the normal cardiac ultrasound, and identification of key cardiac abnormalities using ultrasound. Experiential hands-on learning sessions comprised approximately 70% (24 h) of the training period and were supervised by the three trainers (a registered cardiac sonographer with 9.5 years of experience [SM] and two cardiologists specializing in echocardiography [RB and DK]). Initially the trainees performed supervised scans of six healthy adults known to have normal cardiac ultrasound examinations, in two 3-h sessions. Subsequently, they performed supervised scans of eight assenting adults who were drawn from the inpatient wards and outpatient clinics at JKCI and known to have abnormal cardiac ultrasound examinations. Six patients had abnormal LV systolic function, four had some degree of mitral regurgitation, two had aortic insufficiency, six had some degree of tricuspid regurgitation, and 2 had some degree of pericardial effusion.

The ultrasound scanning protocol, in sequence, included four views with seven images: parasternal long axis 2D with and without color-flow Doppler over the mitral and aortic valve simultaneously, apical 4 chamber 2D with and without color-flow Doppler over the mitral valve, apical 4 chamber 2D with color-flow Doppler over the tricuspid valve, apical 5 chamber 2D with and without color-flow Doppler over the aortic valve, and subcostal view of the inferior vena cava with sniff. To facilitate reproducibility trainees were taught to obtain views in the pre-specified sequence during each patient exam. Views were selected to identify a predetermined set of six key pathologies, namely left ventricular (LV) dysfunction, pericardial effusion, aortic valve regurgitation, moderate to severe mitral valve regurgitation, moderate to severe tricuspid valve regurgitation, and mitral valve stenosis. A subjective assessment of LV function was based on visual estimation of an ejection fraction (EF): > 50% indicating normal LV function and < 50% indicating LV dysfunction. The LV functional assessment was determined using the parasternal long axis and apical 4 and 5 chamber views. The presence of pericardial effusion was documented regardless of size. Pericardial effusion was assessed using the parasternal long axis and apical 4 and 5 chamber views. Assessment of valvular dysfunction was done utilizing qualitative assessment. With regard to regurgitation, subjective vena contracta width and jet area occupying the respective atria (left atrium for mitral regurgitation and right atrium for tricuspid regurgitation) were utilized. Mitral valve regurgitation was assessed using the parasternal long axis view with color Doppler and the apical 4 chamber view with color Doppler over the mitral valve. Tricuspid regurgitation was assessed using the apical 4 chamber view with color Doppler over the tricuspid valve. Aortic regurgitation was assessed as being present or absent; grading of aortic regurgitation was not performed. The presence of aortic regurgitation was assessed using the parasternal long view with color Doppler and the apical 5 chamber view with color Doppler over the aortic valve. A subjective assessment for mitral stenosis was used, including the presence of subjective thickening and restricted motion of the mitral leaflets and color Doppler demonstrating flow acceleration across the mitral valve. Mitral stenosis was only assessed as present or absent, and grading of severity was not preformed. Assessment of mitral stenosis utilized the parasternal long axis view with and without color Doppler over the mitral valve and apical 4 chamber with and without color Doppler over the mitral valve. Didactic and practical sessions used both English and Kiswahili languages.

On day 1 immediately prior to initiation of training activities, trainees completed an intake form to gather information about age, sex, professional education, and prior experience performing and interpreting ultrasound images. They were asked to self-assess their knowledge in cardiology and their ability to learn new technical skills as superior, above average, average, or below average. In addition, they took two pre-training knowledge assessments. One knowledge assessment was comprised of ten multiple choice questions focusing on basic cardiac anatomy and physiology (“10Q Anatomy & Physiology” [Additional file 1: Figure S1]). The second knowledge assessment included cardiac ultrasound still images and video clips, with twenty questions focused on identification of major anatomical structures and key pathologies (“20Q Ultrasound Images” [Additional file 2: Figure S2]). On the last day of the in-person training, all trainees underwent a final hands-on observed assessment of their ability to proficiently conduct the scanning protocol and obtain interpretable images, and also independently completed the 10Q Anatomy & Physiology and 20Q Ultrasound Images knowledge assessments.

Immediately following the in-person training, Vscan with Dual Probe handheld ultrasound devices (model H544H1AS, GE Healthcare, Chicago, United States) were provided to trainees for use in their usual clinical practice settings. Trainees received guidance to consider conducting a focused cardiac POCUS examination using the pre-specified scanning protocol on patients presenting with signs and/or symptoms of potential cardiovascular etiology. For each patient scanned, the trainee completed a case report form to document the scanning procedure, patient clinical features (limited to age, sex, height, weight, and presenting symptoms/signs) and their interpretation of the images with regard to the key pathologies. For each of the six key pathologies, trainees were asked to indicate whether overall image quality allowed for interpretation of the scan for that pathology; if interpretation was possible, then the trainee was asked to determine whether the pathology was absent or present. For each of the seven scanning views, trainees were instructed to record a representative video. De-identified case report forms were then submitted to the study coordinator and corresponding de-identified images were uploaded to Tricefy, a cloud-based, secure imaging solution. The coordinator then sent blank case report forms and the listing of relevant scans to a designated study cardiologist at JKCI for review. The JKCI study cardiologist, blinded to trainee identity and scan interpretation, reviewed each scan using the Tricefy platform and independently interpreted the key pathologies. In addition, for each of the echocardiographic scanning views, the study cardiologist assessed image quality on a 5-point Likert scale: 0—image not obtained; 1—image quality too poor to permit meaningful interpretation; 3—suboptimal image quality but basic image interpretation possible; 5—good image quality, meaningful image interpretation easy. As a quality control measure, a random selection equal to 25% of scans was interpreted by a separate cardiologist at the Medical University of South Carolina, using the Tricefy platform.

Categorical data were presented as absolute values and proportions. The JKCI study cardiologists’ findings served as the reference standard for interpretation with regard to absence or presence of key pathologies. For key pathologies, agreement, sensitivity, and specificity were calculated for examinations in which both the trainee and the cardiologist had indicated that overall image quality allowed for interpretation of that pathology. Agreement was calculated using Cohen’s kappa statistic (κ), with the following interpretive categories: κ < 0, no agreement; 0.00–0.20, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; 0.81–1.00, almost perfect agreement [23]. Pre- and post-training scores were compared using the Wilcoxon Signed Rank test. The Pearson correlation (continuous variables) or Spearman correlation (ordinal variables) was used to evaluate for associations between trainee baseline characteristics including training test scores (predictors) and outcome variables of image quality (expressed as % of images with quality grade 3 or higher based on cardiologist assessment) or image interpretation (expressed as mean kappa values across the 6 key pathologies).

The IRBs of the Medical University of South Carolina and the Jakaya Kikwete Cardiac Institute reviewed the protocol and determined that this study was exempt from Human Research Subject Regulations under 45 CFR 46.104(d) for educational practices; written informed consent was not applicable for this exempt educational activity.

Among the eight trainees, six (75%) were male and two (25%) were female (Table 1). The highest level of professional training was Doctor of Medicine (n = 1, 12.5%) followed by Assistant Medical Officer (n = 1, 12.5%); and six (75%) Clinical Officers. None had previously performed a cardiac ultrasound, while two (25%) had previously performed a limited number of antenatal or abdominal ultrasound examinations (Table 1).

On the ten-question multiple choice assessment of basic cardiac anatomy and physiology (10Q Anatomy & Physiology), median number of correct responses pre-training was 7 (range 4 to 9) and post-training was 8.5 (range 8–10, p = 0.018). On the twenty-question assessment that incorporated ultrasound images and video clips (20Q Ultrasound Images), median number of correct responses pre-training was 9 (range 5–17) and post-training was 17 (range 10–20, p = 0.028).

During a consecutive 9-week period, trainees performed focused cardiac ultrasound examinations on a total of 429 patients at their clinical practice settings. Median number of examinations per trainee was 51 (range 22–97). Characteristics and presenting signs/symptoms of 429 patients are shown in Table 1. Median patient age was 55 (range 9–97). The most common presenting symptoms/signs were hypertension (55.5%), chest pain (46.4%), and palpitations (32.6%).

The percentages of images, by scanning view, that were determined by the cardiologists to be of sufficient quality (grade 3 or higher) for interpretation were as follows (Fig. 1a): parasternal long axis 2D 86.9%; parasternal long axis with color Doppler 82.0%; apical 4 chamber 2D 75.0%; apical 4 chamber with color Doppler 77.3%; apical 5 chamber 2D 69.9%; apical 5 chamber with color Doppler 69.4%; subcostal 2D 56.8%. The percentages of scans that were determined by the cardiologists to contain appropriate images of sufficient quality for interpretation as to whether key pathologies were present or absent were as follows: (Fig. 1b): pericardial effusion 89.9%; LV dysfunction 87.3%; aortic regurgitation 85.1%; mitral valve regurgitation 88.9%; tricuspid valve regurgitation 82.4%; presence of mitral valve stenosis 91.8%. By trainee, the overall percentages of images that were determined by cardiologists to be of grade 3 or higher quality ranged from 18.0% (29/161) to 94.2% (356/378), median 76.7%; 5 of 8 trainees achieved 75% or greater interpretable images (Fig. 2a). Score on the post-training 20Q Ultrasound Images assessment (p = 0.004) and number of ultrasound examinations performed prior to study training (p = 0.039) were each positively associated with image quality in univariate analyses (Table 2).

Kappa values, sensitivity, and specificity of the trainee interpretation versus the gold standard cardiologist interpretation, by key pathology, are shown in Table 3. Based on cardiologist interpretation, the frequencies of each of the key pathologies among all scans performed were as follows: pericardial effusion 1.9%, LV dysfunction 8.0%, aortic regurgitation 13.4%, mitral valve regurgitation 5.0%, tricuspid valve regurgitation 6.1%, and mitral stenosis 0.2%.

Agreement between trainee and cardiologist was moderate for interpretation of tricuspid valve regurgitation; fair for aortic regurgitation and mitral valve regurgitation; slight for LV dysfunction; and poor for the presence of mitral stenosis and pericardial effusion. Additional file 3: Figure S3 through Additional file 5: Figure S5 are examples of trainee-obtained images that were of sufficient quality to be interpretable, yet the trainee did not detect a pathology that was present based on the cardiologist’s interpretation. Scans from eight patients showed a pericardial effusion, but none were detected by trainees. Among 34 patients with LV dysfunction, 18 (52.9%) were interpreted as such by trainees, and among 319 patients with normal LV function, 73 (22.9%) were incorrectly interpreted by trainees as having LV dysfunction. By trainee, mean kappa values across the 6 key pathologies ranged from 0 (no agreement) to 0.32 (fair agreement) (Fig. 2b). There was a significant association between score on the post-training 20Q Ultrasound Images assessment and mean kappa value across the 6 key pathologies in univariate analysis (p = 0.004).

There was almost perfect agreement for image interpretation between the JKCI cardiologist and MUSC cardiologist, with kappa 0.84 (95% CI 0.75–0.92).