Task shifting to clinical officer-led echocardiography screening for detecting rheumatic heart disease in Malawi, Africa

The Use of Artificial Intelligence Guidance for Rheumatic Heart Disease Screening by Novices

INTRODUCTION

A novel technology utilizing artificial intelligence (AI) to provide real-time image-acquisition guidance, enabling novices to obtain diagnostic echocardiographic images, holds promise to expand the reach of echo screening for rheumatic heart disease (RHD). We evaluated the ability of nonexperts to obtain diagnostic-quality images in patients with RHD using AI guidance with color Doppler.

METHODS

Novice providers without prior ultrasound experience underwent a 1-day training curriculum to complete a 7-view screening protocol using AI guidance in Kampala, Uganda. All trainees then scanned 8 to 10 volunteer patients using AI guidance, half RHD and half normal. The same patients were scanned by 2 expert sonographers without the use of AI guidance. Images were evaluated by expert blinded cardiologists to assess (1) diagnostic quality to determine presence/absence of RHD and (2) valvular function and (3) to assign an American College of Emergency Physicians score of 1 to 5 for each view.

RESULTS

Thirty-six novice participants scanned a total of 50 patients, resulting in a total of 462 echocardiogram studies, 362 obtained by nonexperts using AI guidance and 100 obtained by expert sonographers without AI guidance. Novice images enabled diagnostic interpretation in >90% of studies for presence/absence of RHD, abnormal MV morphology, and mitral regurgitation (vs 99% by experts, P ≤ .001). Images were less diagnostic for aortic valve disease (79% for aortic regurgitation, 50% for aortic stenosis, vs 99% and 91% by experts, P < .001). The American College of Emergency Physicians scores of nonexpert images were highest in the parasternal long-axis images (mean, 3.45; 81% ≥ 3) compared with lower scores for apical 4-chamber (mean, 3.20; 74% ≥ 3) and apical 5-chamber images (mean, 2.43; 38% ≥ 3).

CONCLUSIONS

Artificial intelligence guidance with color Doppler is feasible to enable RHD screening by nonexperts, performing significantly better for assessment of the mitral than aortic valve. Further refinement is needed to optimize acquisition of color Doppler apical views.

Machine Learning for Pediatric Echocardiographic Mitral Regurgitation Detection

BACKGROUND

Echocardiography-based screening for valvular disease in at-risk asymptomatic children can result in early diagnosis. These screening programs, however, are resource intensive and may not be feasible in many resource-limited settings. Automated echocardiographic diagnosis may enable more widespread echocardiographic screening, early diagnosis, and improved outcomes. In this feasibility study, the authors sought to build a machine learning model capable of identifying mitral regurgitation (MR) on echocardiography.

METHODS

Echocardiograms were labeled by clip for view and by frame for the presence of MR. The labeled data were used to build two convolutional neural networks to perform the stepwise tasks of classifying the clips (1) by view and (2) by the presence of any MR, including physiologic, in parasternal long-axis color Doppler views. The view classification model was developed using 66,330 frames, and model performance was evaluated using a hold-out testing data set with 45 echocardiograms (11,730 frames). The MR detection model was developed using 938 frames, and model performance was evaluated using a hold-out testing data set with 42 echocardiograms (182 frames). Metrics to evaluate model performance included accuracy, precision, recall, F1 score (average of precision and recall, ranging from 0 to 1, with 1 suggesting perfect precision and recall), and receiver operating characteristic analysis.

RESULTS

For the parasternal long-axis view with color Doppler, the view classification convolutional neural network achieved an F1 score of 0.97. The MR detection convolutional neural network achieved testing accuracy of 0.86 and an area under the receiver operating characteristic curve of 0.91.

CONCLUSIONS

A machine learning model is capable of discerning MR on transthoracic echocardiography. This is an encouraging step toward machine learning-based diagnosis of valvular heart disease on pediatric echocardiography.

Task-sharing to support paediatric and child health service delivery in low- and middle-income countries: current practice and a scoping review of emerging opportunities

BACKGROUND

Demographic and epidemiological changes have prompted thinking on the need to broaden the child health agenda to include care for complex and chronic conditions in the 0-19 years (paediatric) age range. Providing such services will be undermined by general and skilled paediatric workforce shortages especially in low- and middle-income countries (LMICs). In this paper, we aim to understand existing, sanctioned forms of task-sharing to support the delivery of care for more complex and chronic paediatric and child health conditions in LMICs and emerging opportunities for task-sharing. We specifically focus on conditions other than acute infectious diseases and malnutrition that are historically shifted.

METHODS

We (1) reviewed the Global Burden of Diseases study to understand which conditions may need to be prioritized; (2) investigated training opportunities and national policies related to task-sharing (current practice) in five purposefully selected African countries (Kenya, Uganda, Tanzania, Malawi and South Africa); and (3) summarized reported experience of task-sharing and paediatric and child health service delivery through a scoping review of research literature in LMICs published between 1990 and 2019 using MEDLINE, Embase, Global Health, PsycINFO, CINAHL and the Cochrane Library.

RESULTS

We found that while some training opportunities nominally support emerging roles for non-physician clinicians and nurses, formal scopes of practices often remain rather restricted and neither training nor policy seems well aligned with probable needs from high-burden complex and chronic conditions. From 83 studies in 24 LMICs, and aside from the historically shifted conditions, we found some evidence examining task-sharing for a small set of specific conditions (circumcision, some complex surgery, rheumatic heart diseases, epilepsy, mental health).

CONCLUSION

As child health strategies are further redesigned to address the previously unmet needs careful strategic thinking on the development of an appropriate paediatric workforce is needed. To achieve coverage at scale countries may need to transform their paediatric workforce including possible new roles for non-physician cadres to support safe, accessible and high-quality care.

Feasibility of point-of-care cardiac ultrasound performed by clinicians at health centers in Tanzania

BACKGROUND

Point-of-care cardiac ultrasound (cardiac POCUS) has potential to become a useful tool for improving cardiovascular care in Tanzania. We conducted a pilot program to train clinicians at peripheral health centers to obtain and interpret focused cardiac POCUS examinations using a hand-held portable device.

METHODS

Over a 5-day period, didactic and experiential methods were used to train clinicians to conduct a pre-specified scanning protocol and recognize key pathologies. Pre- and post-training knowledge and post-training image acquisition competency were assessed. In their usual clinical practices, trainees then scanned patients with cardiovascular signs/symptoms, recorded a pre-specified set of images for each scan, and documented their interpretation as to presence or absence of key pathologies on a case report form. A cardiologist subsequently reviewed all images, graded them for image quality, and then documented their interpretation of key pathologies in a blinded fashion; the cardiologist interpretation was considered the gold standard.

RESULTS

8 trainees (6 Clinical Officers, 1 Assistant Medical Officer, and 1 Medical Doctor) initiated and completed the training. Trainees subsequently performed a total of 429 cardiac POCUS examinations in their clinical practices over a 9 week period. Stratified by trainee, the median percent of images that were of sufficient quality to be interpretable was 76.7% (range 18.0-94.2%). For five of eight trainees, 75% or more of images were interpretable. For detection of pre-specified key pathologies, kappa statistics for agreement between trainee and cardiologist ranged from - 0.03 (no agreement) for detection of pericardial effusion to 0.42 (moderate agreement) for detection of tricuspid valve regurgitation. Mean kappa values across the key pathologies varied by trainee from 0 (no agreement) to 0.32 (fair agreement).

CONCLUSIONS

The 5-day training program was sufficient to train most clinicians to obtain basic cardiac images but not to accurately interpret them. Proficiency in image interpretation may require a more intensive training program.

Feasibility of Training Clinical Officers in Point-of-Care Ultrasound for Pediatric Respiratory Diseases in Aweil, South Sudan

Lower respiratory tract infections (LRTIs) are the leading cause of deaths in children < 5 years old worldwide, particularly affecting low-resource settings such as Aweil, South Sudan. In these settings, diagnosis can be difficult because of either lack of access to radiography or clinical algorithms that overtreat children with antibiotics who only have viral LRTIs. Point-of-care ultrasound (POCUS) has been applied to LRTIs, but not by nonphysician clinicians, and with limited data from low-resource settings. Our goal was to examine the feasibility of training the mid-level provider cadre clinical officers (COs) in a Médecins Sans Frontières project in South Sudan to perform a POCUS algorithm to differentiate among causes of LRTI. Six COs underwent POCUS training, and each subsequently performed 60 lung POCUS studies on hospitalized pediatric patients < 5 years old with criteria for pneumonia. Two blinded experts, with a tiebreaker expert adjudicating discordant results, served as a reference standard to calculate test performance characteristics, assessed image quality and CO interpretation. The COs performed 360 studies. Reviewers rated 99.1% of the images acceptable and 86.0% CO interpretations appropriate. The inter-rater agreement (κ) between COs and experts for lung consolidation with air bronchograms was 0.73 (0.63–0.82) and for viral LRTI/bronchiolitis was 0.81 (0.74–0.87). It is feasible to train COs in South Sudan to use a POCUS algorithm to diagnose pneumonia and other pulmonary diseases in children < 5 years old.

Cardiac ultrasound in resource-limited settings (CURLS): towards a wider use of basic echo applications in Africa

Background

Point-of-care ultrasound is increasingly being used as a diagnostic tool in resource-limited settings. The majority of existing ultrasound protocols have been developed and implemented in high-resource settings. In sub-Saharan Africa (SSA), patients with heart failure of various etiologies commonly present late in the disease process, with a similar syndrome of dyspnea, edema and cardiomegaly on chest X-ray. The causes of heart failure in SSA differ from those in high-resource settings. Point-of-care ultrasound has the potential to identify the underlying etiology of heart failure, and lead to targeted therapy. Based on a literature review and weighted score of disease prevalence, diagnostic impact and difficulty in performing the ultrasound, we propose a context-specific cardiac ultrasound protocol to help differentiate patients presenting with heart failure in SSA.

Results

Pericardial effusion, dilated cardiomyopathy, cor pulmonale, mitral valve disease, and left ventricular hypertrophy were identified as target conditions for a focused ultrasound protocol in patients with cardiac failure and cardiomegaly in SSA. By utilizing a simplified 5-question approach with all images obtained from the subxiphoid view, the protocol is suitable for use by health care professionals with limited ultrasound experience.

Conclusions

The “Cardiac ultrasound for resource-limited settings (CURLS)” protocol is a context-specific algorithm designed to aid the clinician in diagnosing the five most clinically relevant etiologies of heart failure and cardiomegaly in SSA. The protocol has the potential to influence treatment decisions in patients who present with clinical signs of heart failure in resource-limited settings outside of the traditional referral institutions.

Effect of anaemia on the diagnosis of rheumatic heart disease using World Heart Federation criteria

BACKGROUND

There is overlap between pathological mitral regurgitation seen in borderline rheumatic heart disease using World Heart Federation echocardiography criteria and physiologic regurgitation found in normal children. One possible contributing factor is higher rates of anaemia in endemic countries.

OBJECTIVE

To investigate the contribution of anaemia as a potential confounder in the diagnosis of rheumatic heart disease detected in echocardiographic screening.

METHOD/DESIGN

A novel Server 2012 data warehouse tool was used to incorporate haematology and echocardiography databases. The study included a convenience sample of patients from 5 to 18 years old without structural or functional heart disease that had a haemoglobin value within 1 month prior to an echocardiogram. Echocardiogram images were reviewed to determine presence or absence of World Heart Federation criteria for rheumatic heart disease. The rate of rheumatic heart disease among anaemic and non-anaemic children according to gender- and age-based norms groups was compared.

RESULTS

Of the 935 patients who met the study inclusion criteria, 406 were classified as anaemic. There was no difference in the rate of echocardiograms meeting criteria for borderline rheumatic heart disease in anaemic (2.0%, 95% CI 0.6-3.3%) and non-anaemic children (1.3%, 95% CI 0.3-2.3%). However, there was a statistically significant increase in rates of mitral regurgitation of unclear significance among anaemic versus non-anaemic patients (8.6 versus 3.6%; p = 0.0012).

CONCLUSION

Anaemia does not increase the likelihood of meeting echocardiographic criteria for borderline rheumatic heart disease. Future studies should evaluate for the correlation between anaemia and mitral regurgitation in endemic settings.

Task Sharing in the Diagnosis, Prevention, and Management of Rheumatic Heart Disease: A Systematic Review

BACKGROUND

Globally, rheumatic heart disease (RHD) is a major contributor to the burden of cardiovascular disease. Major gaps in RHD prevention and treatment have been documented at all levels of health systems in low- and middle-income countries. Task sharing is an approach that could prove effective in remediating bottlenecks in RHD-related care.

OBJECTIVES

This study conducted a systematic review to assess the state of the evidence for the use of task sharing in the diagnosis, prevention, and management of RHD.

METHODS

Guided by a previously published protocol, we searched various databases using a systematic search strategy including MeSH and free-text terms for (1) group A streptococcus, acute rheumatic fever, and RHD and (2) strategies of task sharing in limited-resource settings. Two investigators independently screened the search outputs, selected the studies, extracted the data, and assessed the risk of bias, resolving discrepancies by discussion and consensus.

RESULTS

The publications search yielded 212 records, of which 18 articles were deemed as potentially eligible for inclusion. None of the studies, however, met with the inclusion criteria.

CONCLUSIONS

There is a lack of evidence for the use of task-sharing approaches in scaling up RHD prevention and treatment services in limited-resource settings. Considering the persistent burden of group A streptococcus, acute rheumatic fever, and RHD in low- and middle-income countries, this work highlights the urgent need to develop and test models of RHD-related care utilizing an evidence-based approach to task sharing. [Task Sharing in the Diagnosis, Prevention, and Management of Rheumatic Heart Disease: A Systematic Review; CRD42017072989].

Integration of echocardiographic screening by non-physicians with remote reading in primary care

INTRODUCTION

Heart disease (HD) accounts for high morbidity and mortality in Brazil. Underserved populations often suffer long delays in diagnosis. We aimed to evaluate the feasibility of integrating screening echocardiography (echo) with remote interpretation in the established primary care system (PC) in Brazil and to assess HD prevalence.

METHODS

Over 11 months, 20 healthcare workers (four physicians, four nurses, and 12 technicians) at 16 PC centres were trained on simplified handheld echo protocols. Three screening (SC) groups, including all consented patients aged 17-20, 35-40 and 60-65 years, and patients referred (RF) for clinical indications underwent focused echo. Studies were remotely interpreted through telemedicine. Significant HD was defined as moderate/severe valve disease, ventricular dysfunction/hypertrophy, pericardial effusion, wall-motion abnormalities and congenital heart disease.

RESULTS

Total 1004 patients underwent echo; 299 (29.8%) in the SC group. Median age was 51±18 years, 63.9% females; 42.7% had cardiovascular symptoms. Significant HD was found in 354 (35.3%) patients (23.4% in SC vs 40.3% in RF group, p<0.001). Prevalence was higher in patients in the SC group aged >60 years (29.2%), compared with 35-40 (14.9%) and under 20 (16.5%), p=0.012. Comparing SC to RF groups, moderate/severe left ventricular dysfunction was observed in 4.1% vs 8.1%, p=0.03, mitral regurgitation in 8.9% vs 20.3%, p<0.001 and aortic stenosis in 5.4% vs 4.3%, p=0.51.

CONCLUSIONS

Integration focused echo into PC is feasible in Brazil as a strategy to deliver cardiovascular care to low-resourced areas through task shifting. The burden of HD observed suggests this tool may improve early diagnosis and referral.

Prevention and control of rheumatic heart disease: Overcoming core challenges in resource-poor environments

Rheumatic heart disease (RHD) has long receded as a significant threat to public health in high-income countries. In low-resource settings, however, the specter of RHD remains unabated, as exemplified by recent data from the Global Burden of Diseases Study. There are many complex reasons for this ongoing global disparity, including inadequate data on disease burden, challenges in effective advocacy, ongoing poverty and inequality, and weak health systems, most of which predominantly affect developing nations. In this review, we discuss how each of these acts as a core challenge in RHD prevention and control. We then examine key lessons learnt from successful control programs in the past and highlight resources that have been developed to help create strong national RHD control programs.