In his presentation at AIMed Cardiology took place on Monday (17 June), Dr. Partho Sengupta, Chief, Division of Cardiology, Director of Cardiovascular Imaging, and Professor of Medicine at West Virginia University Heart & Vascular Institute, comprehensively shared how machine learning (ML) can be deployed to overcome some of the challenges his specialty faces today.
Dr. Sengupta said in some of the hospitals where he is presently based, cardiovascular specialists typically receive 30-40 patients on a daily basis. Along with that, they will also read about 70-80 echocardiogram. The demand often put a strain on physicians’ workload and result in burnout or fatigue-induced errors over
Moreover, ML may help in patient phenotyping. The amount of information gathered from patients could be enormous but integrating all the data distilling out a relevant pattern to improve existing care can be tedious. In the case of aortic stenosis, physicians understand it as a common valve-related heart condition. With network tomography, where a multi-parametric integration of data is being mapped out, physicians can now have an insight into the severity of the patients, generate similarity analyses that propagate precision medicine and eventually, leading to a new taxonomy of how complex diseases should be identified.
To be more specific, we have chosen two case studies which were presented at AIMed Cardiology, to further illustrate the power of ML in solving real-life challenges in the field.
Case study 1: Machine Learning from fetal waveforms to predictive adverse perinatal outcomes
Dr. Devyani Chowdhury, Director of Cardiology Care for Children presented her cross-institutional collaborative project on the use of ML to predict neonatal mortality in middle-lower income regions. Karachi, Pakistan was chosen to be the venue where the ML algorithm will be developed because it has one of the highest neonatal deaths in the World. Thus far, close to 700 expectant ladies were recruited, with an estimated 47.6% between 22-26 weeks of gestation and the rest between 30-34 weeks of gestation.
Basic maternal information such as blood pressure were taken from the ladies. Fetal echocardiogram, maternal echocardiogram, and Doppler ultrasound were performed. Rather than traditional Doppler analyses which involved peak velocity, mean velocity, calculating slopes, the research team had digitized the waveforms of the Doppler using automation, so that they can analyze it from every aspect. Six sites were identified based on fetus physiology to decide which area makes the most sense in generating the mortality rate of the unborn. K-Fold validation and other measures were also taken into consideration so that the model will not overfit the hypothesis and there is no bias against the minorities.
The ML model was able to achieve 85% accuracy for
Dr. Chowdhury expressed, the project’s goal was to identify pregnancies that are at risks and ML had demonstrated its capability to perform that identification so that some form of interventions could be offered. There will also be a better allocation of resources and develop prevention strategies for a
Case study 2: Cardio-metabolic disease prevention at population level
Dr. Rajesh Dash, Assistant Professor of Cardiovascular Medicine and Preventive Cardiologist from Stanford University, talked about the company, HealthPals, which he established together with data scientist Sushant Shankar about four years ago. Primarily, the initiative employs data from electronic health records, combined with artificial intelligence and medical guidelines, to conjure a smart platform – CLINT (short for clinical intelligence), to undercover gaps within the healthcare system that prevent individuals from receiving adequate care.
Dr. Dash believes, at the moment, there is a mismatch between the amount of medical information made available and the amount of engagement that physicians or healthcare providers actually have, in order to incorporate them into their practice. Besides, back in his own clinic, Dr. Dash noticed when patients arrived for the very first time, their lab test results often suggested that they should be on some form of medications to control their blood pressure, diabetes or high cholesterol for basic prevention. However, most of them were not and the gap is rather concerning.
Thus, the dashboard of CLINT consolidates information on the risks of cardiovascular disease, who is having anticoagulant or other types of preventive medications and who is not, as well as comorbidities and gaps in those comorbidities, from the population level all the way down to the clinic or zip-code level, so that preventive specialists can optimize them to devise appropriate measures.
By looking at the longitudinal data and tapping onto insights generated by ML, specialists can underline clusters of non-respondents to a certain type of drugs and start to delve more deeply into the reasons behind the response. Eventually, specialists are able to determine, based on the feature of a particular patient, whether he or she will respond well to the prescribed drugs. Dr. Dash is confident that the CLINT model is applicable across the World, especially in regions where there a health system to incentivize their citizens to pay less for a better outcome, as in the case of Australia, India, Singapore,
The partnership between Dr. Dash and data scientist Sushant Shankar is also featured in the latest issue of AIMed magazine. You may wish to re-visit the AIMed Cardiology seminar and panel discussions here. Do follow us on Twitter, Instagram, Facebook and Youtube for more event updates.
A science writer with data background and an interest in the current affair, culture,