In a remarkable achievement, CareFirst, the innovative healthcare web app developed by First Care Solutions, has secured 1st place in the Health Section at the 2024 Stuff Awards. Held annually by Stuff Magazine, these prestigious awards recognise excellence in digital innovation and technological advancement.
CareFirst has quickly emerged as a game-changer in South Africa’s healthcare landscape, offering users:
Instant medical consultations with qualified doctors, available 24/7 – 365
AI-powered Vital Scanning: Advanced technology to monitor key health metrics including blood pressure, respiratory rate, and heart rate.
Comprehensive Medical Services: Convenient access to doctors’ consultations, prescriptions, sick notes, and referrals.
‘’I’m really proud of what our team has created. CareFirst is a platform that can really transform the way people access healthcare,’’ says Dr Steve Holt, Chief Executive Officer – at First Care Solutions.
CareFirst represents a shift towards more accessible, efficient and personalised care, promising a future where quality care is available to everyone, regardless of their location or time constraint and this award is a testament to that.
This week the GIBS, (Gordon Institute of Business Science), held an on-campus Healthcare Industry Insights Conference aimed at healthcare professionals and others with an interest in this field to hear from experts providing insightful discussion and frank debate.
The sessions were each themed to different topics such as Innovation for Sustainable Access and Quality Care, Building a Skilled Workforce, navigating Public-Private Partnerships and Addressing Social Determinants.
The day ended with a focus on Digital Transformation and advances in medical device manufacturing, were discussed.
Dilip Naran, Vice President of Product Architecture at CompuGroup Medical South Africa, (an internationally leading MedTech provider), has over 25 years of dedicated service to the South African healthcare market, and was asked to share his thoughts on the next generation of digital health.
Naran has been actively involved in shaping both billing and clinical applications and has been a key player in the creation of cutting-edge cloud-based solutions that have revolutionised the way healthcare professionals operate in South Africa.
Improving workflow processes
The discussion focused on the AI and Electronic Health Records (EHRs), and how by harnessing the power of AI, healthcare providers can unlock unprecedented insights, enhance patient care and drive operational efficiencies.
The topical subject began by reminding the audience that AI has already improved the EHR data management. By extracting valuable insights from clinical notes, automation of repetitive tasks, analysing data to identify patterns and facilitating the seamless integration of multiple data sources. AI advances in HER and medical devices have reshaped the doctor / patient healthcare journey.
To continue this growth, AI powered tools must be implemented in EHRs to enable functionality that enhance the Dr/Patient journey. Some benefits of AI powered EHRs include:
Effective Clinical Decision Support
Intelligent Automation. This includes improvement in workflow by automating certain tasks
Smart Medication management . Ai can alert HCP to potential drug interactions and adverse effects
Predictive Analytics that are personalised based on patient history
Adoption in South Africa
Whilst some of the AI technologies are not yet available in South Africa, CGM’s recently launched Autoscriber solution which uses AI technologies such as Natural Language Processing NLP and a Large Language Model (LLM) has enabled South African HCPs to use this solution to create structured notes which includes diagnoses ICD10 and SNOMED coding. This assists the HCP in populating their HER without having to physically capture information.
At the moment the adoption rate of EHR in practices is around 30% in the private sector, with oncology leading the way.
With collaboration between government, private and public sector, existing technologies can forecast disease outbreaks, identify high-risk patients and optimise resource allocation.
Dilip Naran concluded by saying: “The use of AI technologies and processes can facilitate the meaningful use of data in EHRs and lead to better patient outcomes”
A new brain-computer interface (BCI) developed at UC Davis Health translates brain signals into speech with up to 97% accuracy – the most accurate system of its kind. The researchers implanted sensors in the brain of a man with severely impaired speech due to amyotrophic lateral sclerosis (ALS). The man was able to communicate his intended speech within minutes of activating the system.
ALS, also known as Lou Gehrig’s disease, affects the nerve cells that control movement throughout the body. The disease leads to a gradual loss of the ability to stand, walk and use one’s hands. It can also cause a person to lose control of the muscles used to speak, leading to a loss of understandable speech.
The new technology is being developed to restore communication for people who can’t speak due to paralysis or neurological conditions like ALS. It can interpret brain signals when the user tries to speak and turns them into text that is ‘spoken’ aloud by the computer.
“Our BCI technology helped a man with paralysis to communicate with friends, families and caregivers,” said UC Davis neurosurgeon David Brandman. “Our paper demonstrates the most accurate speech neuroprosthesis (device) ever reported.”
When someone tries to speak, the new BCI device transforms their brain activity into text on a computer screen. The computer can then read the text out loud.
To develop the system, the team enrolled Casey Harrell, a 45-year-old man with ALS, in the BrainGate clinical trial. At the time of his enrolment, Harrell had weakness in his arms and legs (tetraparesis). His speech was very hard to understand (dysarthria) and required others to help interpret for him.
In July 2023, Brandman implanted the investigational BCI device. He placed four microelectrode arrays into the left precentral gyrus, a brain region responsible for coordinating speech. The arrays are designed to record the brain activity from 256 cortical electrodes.
“We’re really detecting their attempt to move their muscles and talk,” explained neuroscientist Sergey Stavisky. Stavisky is an assistant professor in the Department of Neurological Surgery. He is the co-director of the UC Davis Neuroprosthetics Lab and co-principal investigator of the study. “We are recording from the part of the brain that’s trying to send these commands to the muscles. And we are basically listening into that, and we’re translating those patterns of brain activity into a phoneme – like a syllable or the unit of speech – and then the words they’re trying to say.”
Casey Harrell with his personal assistant Emma Alaimo and UC Davis neuroscientist Sergey Stavisky
Faster training, better results
Despite recent advances in BCI technology, efforts to enable communication have been slow and prone to errors. This is because the machine-learning programs that interpreted brain signals required a large amount of time and data to perform.
“Previous speech BCI systems had frequent word errors. This made it difficult for the user to be understood consistently and was a barrier to communication,” Brandman explained. “Our objective was to develop a system that empowered someone to be understood whenever they wanted to speak.”
Harrell used the system in both prompted and spontaneous conversational settings. In both cases, speech decoding happened in real time, with continuous system updates to keep it working accurately.
The decoded words were shown on a screen. Amazingly, they were read aloud in a voice that sounded like Harrell’s before he had ALS. The voice was composed using software trained with existing audio samples of his pre-ALS voice.
At the first speech data training session, the system took 30 minutes to achieve 99.6% word accuracy with a 50-word vocabulary.
“The first time we tried the system, he cried with joy as the words he was trying to say correctly appeared on-screen. We all did,” Stavisky said.
In the second session, the size of the potential vocabulary increased to 125 000 words. With just an additional 1.4 hours of training data, the BCI achieved a 90.2% word accuracy with this greatly expanded vocabulary. After continued data collection, the BCI has maintained 97.5% accuracy.
“At this point, we can decode what Casey is trying to say correctly about 97% of the time, which is better than many commercially available smartphone applications that try to interpret a person’s voice,” Brandman said. “This technology is transformative because it provides hope for people who want to speak but can’t. I hope that technology like this speech BCI will help future patients speak with their family and friends.”
The study reports on 84 data collection sessions over 32 weeks. In total, Harrell used the speech BCI in self-paced conversations for over 248 hours to communicate in person and over video chat.
“Not being able to communicate is so frustrating and demoralising. It is like you are trapped,” Harrell said. “Something like this technology will help people back into life and society.”
“It has been immensely rewarding to see Casey regain his ability to speak with his family and friends through this technology,” said the study’s lead author, Nicholas Card. Card is a postdoctoral scholar in the UC Davis Department of Neurological Surgery.
“Casey and our other BrainGate participants are truly extraordinary. They deserve tremendous credit for joining these early clinical trials. They do this not because they’re hoping to gain any personal benefit, but to help us develop a system that will restore communication and mobility for other people with paralysis,” said co-author and BrainGate trial sponsor-investigator Leigh Hochberg. Hochberg is a neurologist and neuroscientist at Massachusetts General Hospital, Brown University and the VA Providence Healthcare System.
Brandman is the site-responsible principal investigator of the BrainGate2 clinical trial. The trial is enrolling participants. To learn more about the study, visit braingate.org or contact braingate@ucdavis.edu.
Closeup of the pneumatic logic sensing device. (William Grover/UCR)
Medical engineers have developed a new, air-powered computer sets off alarms when certain medical devices fail. The invention is a more reliable and lower-cost way to help prevent blood clots and strokes – all without electronic sensors.
Described in a paper in the journal Device, the computer not only runs on air, but also uses air to issue warnings. It immediately blows a whistle when it detects a problem with the lifesaving compression machine it is designed to monitor.
Intermittent pneumatic compression (IPC) devices are pneumatic leg sleeves that periodically squeeze a patient’s legs to increase blood flow. This prevents clots that lead to blocked blood vessels, strokes, or death. Typically, these machines are powered and monitored by electronics.
“IPC devices can save lives, but all the electronics in them make them expensive. So, we wanted to develop a pneumatic device that gets rid of some of the electronics, to make these devices cheaper and safer,” said William Grover, associate professor of bioengineering at UC Riverside and corresponding paper author.
Pneumatics move compressed air from place to place. Emergency brakes on freight trains operate this way, as do bicycle pumps, tire pressure gauges, respirators, and IPC devices. It made sense to Grover and his colleagues to use one pneumatic logic device to control another and make it safer.
This type of device operates in a similar way to electronic circuits, by making parity bit calculations. “Let’s say I want to send a message in ones and zeroes, like 1-0-1, three bits,” Grover said. “Decades ago, people realized they could send these three bits with one additional piece of information to make sure the recipient got the right message.”
That extra piece of information is called a parity bit. The bit is a number – 1 if the message contains an odd number of ones, and 0 if the message contains an even number of ones. Should the number one appear at the end of a message with an even number of bits, then it is clear the message was flawed. Many electronic computers send messages this way.
An air-powered computer uses differences in air pressure flowing through 21 tiny valves to count the number of ones and zeroes. If no error in counting has occurred, then the whistle doesn’t blow.
If it does blow, that’s a sign the machine requires repairs. Grover and his students, in a video demonstrating the air computer, are shown damaging an IPC device with a knife, rendering it unusable. Seconds later, the whistle blows.
“This device is about the size of a box of matches. It replaces a handful of sensors as well as a computer,” Grover said. “So, we can reduce costs while still detecting problems in a device. And it could also be used in high humidity or high temperature environments that aren’t ideal for electronics.”
The IPC device monitoring is only one application for air computing. For his next project, Grover would like to design a device that could eliminate the need for a job that kills people every year: moving around grain at the top of tall silos.
Tall buildings full of corn or wheat, grain silos are a common sight in the Midwest. Often times, a human has to go inside with a shovel to break up the grains and even out the piles inside.
“A remarkable number of deaths occur because the grain shifts and the person gets trapped. A robot could do this job instead of a person. However, these silos are explosive, and a single electric spark could blow a silo apart, so an electronic robot may not be the best choice,” Grover said. “I want to make an air-powered robot that could work in this explosive environment, not generate any sparks, and take humans out of danger.”
Air-powered computing is an idea that has been around for at least a century. People used to make air-powered pianos that could play music from punched rolls of paper. After the rise of modern computing, engineers lost interest in pneumatic circuits.
“Once a new technology becomes dominant, we lose awareness of other solutions to problems,” Grover said. “One thing I like about this research is that it can show the world that there are situations today when 100-plus-year-old ideas can still be useful.”
By Margot Brews, Head Health Risk Management Strategy at Momentum Health Solutions
The healthcare sector has been on the cusp of substantial reform for quite some time. However, the introduction and application of artificial intelligence (AI) across various healthcare disciplines will surely stand out as one of the most revolutionary eras in the industry.
This rapidly evolving field has been lauded as the key to unlocking greater quality in healthcare services, introducing more efficient protocols and treatment pathways, as well as considerably increasing access to healthcare across all demographics.
When we place this in a local context, taking into consideration that steps to implement the NHI are already in motion, AI will be critical in helping implement specific elements, such as public health interventions. If we are to refer to the Covid-19 pandemic and its immense scale, AI was leveraged across various countries around the globe to predict the spread of the virus. In doing so, this allowed governments to implement protocols to curb its spread, as well as provide citizens with critical information in an effort to decrease its proliferation.
Looking at the medical schemes sector in South Africa, the industry aims to ultimately improve health outcomes for members and in doing so, encourage and maintain a better quality of life. AI has assisted Momentum Health Solutions in evaluating the delivery of healthcare in the future with clear goals that include increasing access to quality healthcare, and utilising the unmatched innovation that AI offers in assessing member profiles more comprehensively. This is to ensure that we are not only providing a service, but actually understanding in the broadest terms possible what type of care a member requires and partnering with them on that journey.
An example of this is closely analysing commonalities within a member’s treatment pathway. When we review clinical data such as doctors’ consultations, the discipline of the doctor and their particular field of expertise, along with the medication prescribed, we can more timeously start to see patterns developing. This indicates and therefore informs us that the member may have a more serious illness or chronic disease that requires clinical support on a more extensive scale, which we can then discuss with the member and facilitate.
To ensure we are providing tailored healthcare solutions that steer away from offering members generic benefits, we have partnered with Amazon Web Services (AWS), which provides the most comprehensive services, tools, and resources in artificial intelligence today. Through this partnership we have been able to provide members with unique services and individualised care that ultimately ensures their healthcare is a priority.
While AI is indeed key to creating a more efficient healthcare system, ethical considerations remain a concern for many when evaluating factors such as the protection and privacy of data and its ownership, as well as the accuracy of its outputs and conclusions. Having said this, risk mitigation protocols have been implemented to ensure that personal data is protected, and ethical standards are maintained at the highest level.
AI is certainly the most effective solution in the 21st century when investigating ways to solve the ongoing healthcare crisis, particularly in South Africa, where immense disparity exists between the public and private sectors. Leveraging AI, both from a medical scheme provider perspective and more broadly, will not only empower current and future workforces within the sector, but will also create greater opportunity for improved healthcare services that can be sustained. When implemented and utilised for the benefit of all, AI has the potential to be South Africa’s healthcare redeemer.
Dr Liza Street and Taryn Uhlmann, Co-founders of Ajuda
As technology continues to shrink the world, migration is becoming easier. According to Statista, nearly a million South Africans emigrated in 2020, and the semigration trend in recent years has seen hundreds of thousands of South Africans relocating between provinces. While belongings can go into a truck or shipping container, something even more important is slipping through the cracks: personal health records.
Two highly-motivated, entrepreneurial women are looking to change that with an innovative, user-friendly solution – a secure digital health vault called Ajuda, where anyone and everyone can store their health information and get quick access to it anytime, anywhere, using a secure login from their digital device.
Designed With You In Mind
Ajuda (which means ‘help’ in Portuguese) was founded and developed by both Dr Liza Street, a paediatrician, and Taryn Uhlmann, a technology and marketing business executive.
The two – both moms of three – met when Street began treating Uhlmann’s children. After COVID-19, with all the disruptions and innovations that arose during that time, they got chatting about how disempowering it was for moms not to have easy access to their children’s medical records and thus having to rely on memory when it came to their children’s developmental milestones, doctor consults and medication names. “We don’t realise how often we need to recall this information – for new schools, at doctors’ visits and even when relocating. This frustration, especially for busy parents, is where it all started,” says Uhlmann.
That conversation, three years ago, was the seed for what has blossomed into Ajuda. Uhlmann and Street recently welcomed a third member, Allan Sweidan, as an investor and advisor. Sweidan, a clinical psychologist, brings his experience from co-founding Akeso, Netcare’s mental healthcare clinics, and more recently the mental health app, October Health (Panda), to Ajuda.
Why a Central Health Information Vault?
While the idea may have been born out of a conversation about time-strapped moms managing their kids’ health, Uhlmann and Street soon realised that not having the means to securely and conveniently store one’s health information and have access to one’s medical history was a challenge faced by everyone.
Accurate medical records are necessary in emergency situations, anytime you’re having new medication prescribed, for insurance applications, school applications, visa applications, and in many other instances.
What’s more, not having access to personal health information makes it difficult for people to take control of their own health. Having accurate information on hand helps people keep track of which medications to take, in what dosages and when. It also helps healthcare providers make informed decisions around treatments and prescriptions, based on their patients’ health histories, and can lower the risk of adverse drug interactions. In an age of the ‘sandwich generation’, where adults today often take care of both their children and their parents, a centralised repository of their family’s health information is game-changing.
“We looked at all the challenges,” says Uhlmann. “Memory is a big challenge; migration and movement is a big challenge; and the fact that medical care is fragmented – you might have a scan at one hospital and a blood test at another, or maybe you take your child to a GP while you’re on holiday, and in the end, because you don’t have access to all those records in one place, it means no doctor has all the information required to ever look at you holistically. There are many separate medical apps for various healthcare organisations, but no centralised, consumer facing solution pulling it all together. ”
Everything In One Place
Ajuda addresses these concerns with a secure, easy-to use digital storage vault that users can access anywhere, anytime, free of charge.
A second time-saving feature of Ajuda is the ‘One Time Form.’ When signing up, a new user creates a profile for themselves and/or their children, which generates a ‘One Time Form’ and they’re set for life. Creating a profile is simple and interactive, and Ajuda users are guided through the process of completing and uploading their personal and health details, step by step, with user-friendly prompts and explanations.
“This is the same core information that you fill in every time you see a new doctor, have a blood test, or do any medical procedure, which becomes frustrating and time consuming,” says Street. “Now you just need to complete it once, unpressurised, in the comfort of your home, with the correct information at hand, and then take it each time you go to a healthcare provider. It’s a win-win for doctors and patients.”
If users don’t have all their information on hand, no problem – they can fill in the gaps later. Once they’ve completed their profile, they can then enjoy the peace of mind that their personal health information is safely, conveniently and accurately stored.
“For healthcare providers, it provides a comprehensive record of a patient’s health history, not only at their own practice but anywhere the patient has received treatment,” says Street.
Free and Independent
Ajuda is free to use, and you don’t need to be a member of a particular medical scheme or use a particular healthcare provider to access it. By making it free to use, Uhlmann and Street hope to empower everyone with the means to take control of their own health information.
For more information on Ajuda or to sign up, visit Ajuda.co.za
On Friday, users around the world began encountering a “blue screen of death”, signalling the start of a day of chaos. About 8.5 million Microsoft devices were affected by a bug, resulting in significant global disruption from airlines to finance and even small businesses. Healthcare infrastructure was also affected, which may have endangered an unknown number of lives because of missed appointments, inaccessible patient records, prescriptions and inventory data.
Worldwide, hospitals reported being unable to use their systems to access key information such as schedules, patient medical records and logistics. Reports emerged of cancelled procedures, and non-urgent patients being turned away.
“Many hospitals are cancelling elective procedures today. Patients should direct any questions to their providers because this is a practice-by-practice, hospital-by-hospital decision,” said the Massachusetts Department of Public Health in a statement.
In the UK, NHS England warned of delays and the British Medical Association advised of a backlog for normal GP service.
‘Like practising medicine in the dark ages’
All across Africa reported that many hospitals and clinics depend on Microsoft 365 and cloud services for crucial functions, Nehanda Radio reported. The outage highlights how critical infrastructure has become dependent on the stability of a handful of platforms.
“Our entire hospital was thrown into disarray. We couldn’t access patient files, schedule surgeries, or coordinate with suppliers,” said Dr Amina Salim, the chief medical officer at a major hospital in Abuja, Nigeria.
“It was like practising medicine in the dark ages. Our doctors and nurses were forced to resort to hand-written notes and countless phone calls just to provide basic care.”
“I went to refill my HIV medication and the pharmacist said their computers were down, so they couldn’t look up my prescription. I was worried I’d have to go without my treatment,” said Thembi Ndlovu, a patient in Johannesburg.
The problem was worsened in rural and underserved areas that are heavily reliant on the internet and cloud services for remote consultations, sharing of medical expertise and centralised databases.
“Our telemedicine program came to a screeching halt. We couldn’t video conference with specialists, access test results, or update patient records,” said Dr Khalid Elmahdi, the director of a rural health clinic in Morocco. “It was devastating for communities that have few other options for advanced care.”
The crashes were traced to an update from a security service provider, Crowdstrike – which ironically provides protection solutions against ransomware, a problem that has been plaguing healthcare.
While most services seem to be up and running after the weekend, experts say that it may take weeks for full recovery. Fixing the problem often requires physically accessing the system and installing a USB dongle with recovery software, which can be difficult in certain locations, such as remote clinic.
Report shows the alarming global rise of cyberattacks on the healthcare sector and the urgent need to prioritise cybersecurity
Photo by Nahel Abdul on Unsplash
KnowBe4 (www.KnowBe4.com), the provider of the world’s largest security awareness training and simulated phishing platform, released its International Healthcare Report. The report takes a closer look at the cybersecurity crisis currently experienced by the healthcare sector, in particular hospital groups, across the world.
Africa was the global region with the highest average number of weekly cyberattacks per organisation in 2023. One in every 19 organisations on the continent experienced an attempted attack every week. Although South Africa’s healthcare sector has managed to avoid a major attack since 2020, the alarming escalation of attacks in other sectors within the country suggests that it’s only a matter of time before the next attack strikes, making it a question of “when” rather than “if”.
Hospitals have become increasingly attractive targets for ransomware attacks due to their comprehensive patient databases, sensitive information, and their interconnectedness between systems and equipment. Moreover, poor security measures have made hospitals vulnerable to cyber threats. When attacked, cybercriminals can potentially take control of entire hospital systems, and gain access not only to patients’ health information but also their financial and insurance data.
Hospitals are severely impacted by cyberattacks (https://apo-opa.co/4csCXH4), which can lead to a reduction in patient care, loss of access to electronic systems, and a reliance on incomplete paper records. This can also result in the cancellation of surgeries, tests, appointments, and, in some cases, even loss of life.
Some shocking facts discussed in the report include:
In the first three quarters of 2023, the global healthcare sector experienced a staggering 1,613 cyberattacks per week, nearly four times the global average, and a significant increase from the same period the previous year.
The healthcare sector has seen a dramatic surge in cyberattack costs over the past three years, with the average cost of a breach reaching nearly $11 million, more than three times the global average. This makes healthcare the costliest sector for cyberattacks.
Ransomware attacks have been the most prevalent type of cyberattack on healthcare organisations, accounting for over 70% of successful attacks in the past two years.
The majority of cyberattacks (between 79% and 91%), across sectors, begin with phishing or social engineering tactics, which allow cybercriminals to gain access to accounts or servers.
According to KnowBe4’s 2024 Phishing by Industry Benchmarking Report (https://apo-opa.co/4csuiEB), healthcare and pharmaceutical organisations are among the most vulnerable to phishing attacks, with employees in large organisations in the sector having a 51.4% likelihood of falling victim to a phishing email. This means that cybercriminals have a better than 50/50 chance of successfully phishing an employee in the sector.
“The healthcare sector remains a prime target for cybercriminals looking to capitalise on the life-or-death situations hospitals face,” says Stu Sjouwerman, CEO of KnowBe4. “With patient data and critical systems held hostage, many hospitals feel like they are left with no choice but to pay exorbitant ransoms. This vicious cycle can be broken by prioritising comprehensive security awareness training to empower employees and cultivate a positive security culture as a strong defence against phishing and social engineering attacks.”
The report examines the state of cybersecurity in the healthcare sector in North America, Europe, the United Kingdom, Asia-Pacific, Africa, and Latin America. In addition it also highlights some of the most prolific global ransomware attacks that occurred between December 2023 and May 2024, the aftermath thereof and what healthcare organisations can do to protect themselves from cyberattacks.
To download a copy of KnowBe4’s International Healthcare Report, click here (https://apo-opa.co/3xIjjaY).
Artificial intelligence models often play a role in medical diagnoses, especially when it comes to analysing images such as X-rays. But these models have been found not perform as well across all demographic groups, usually faring worse on women and people of colour.
These models have also been shown to develop some surprising abilities. In 2022, MIT researchers reported that AI models can make accurate predictions about a patient’s race from their chest X-rays – something that the most skilled radiologists can’t do.
Now, in a new study appearing in Nature, the same research team has found that the models that are most accurate at making demographic predictions also show the biggest “fairness gaps”, ie having reduced accuracy diagnosing images of people of different races or genders. The findings suggest that these models may be using “demographic shortcuts” when making their diagnostic evaluations, which lead to incorrect results for women, Black people, and other groups, the researchers say.
“It’s well-established that high-capacity machine-learning models are good predictors of human demographics such as self-reported race or sex or age. This paper re-demonstrates that capacity, and then links that capacity to the lack of performance across different groups, which has never been done,” says senior author Marzyeh Ghassemi, an MIT associate professor of electrical engineering and computer science.
The researchers also found that they could retrain the models in a way that improves their fairness. However, their approached to “debiasing” worked best when the models were tested on the same types of patients they were trained on, such as patients from the same hospital. When these models were applied to patients from different hospitals, the fairness gaps reappeared.
“I think the main takeaways are, first, you should thoroughly evaluate any external models on your own data because any fairness guarantees that model developers provide on their training data may not transfer to your population. Second, whenever sufficient data is available, you should train models on your own data,” says Haoran Zhang, an MIT graduate student and one of the lead authors of the new paper.
Removing bias
As of May 2024, the FDA has approved 882 AI-enabled medical devices, with 671 of them designed to be used in radiology. Since 2022, when Ghassemi and her colleagues showed that these diagnostic models can accurately predict race, they and other researchers have shown that such models are also very good at predicting gender and age, even though the models are not trained on those tasks.
“Many popular machine learning models have superhuman demographic prediction capacity – radiologists cannot detect self-reported race from a chest X-ray,” Ghassemi says. “These are models that are good at predicting disease, but during training are learning to predict other things that may not be desirable.”
In this study, the researchers set out to explore why these models don’t work as well for certain groups. In particular, they wanted to see if the models were using demographic shortcuts to make predictions that ended up being less accurate for some groups. These shortcuts can arise in AI models when they use demographic attributes to determine whether a medical condition is present, instead of relying on other features of the images.
Using publicly available chest X-ray datasets from Beth Israel Deaconess Medical Center (BIDMC) in Boston, the researchers trained models to predict whether patients had one of three different medical conditions: fluid buildup in the lungs, collapsed lung, or enlargement of the heart. Then, they tested the models on X-rays that were held out from the training data.
Overall, the models performed well, but most of them displayed “fairness gaps” – that is, discrepancies between accuracy rates for men and women, and for white and Black patients.
The models were also able to predict the gender, race, and age of the X-ray subjects. Additionally, there was a significant correlation between each model’s accuracy in making demographic predictions and the size of its fairness gap. This suggests that the models may be using demographic categorisations as a shortcut to make their disease predictions.
The researchers then tried to reduce the fairness gaps using two types of strategies. For one set of models, they trained them to optimise “subgroup robustness,” meaning that the models are rewarded for having better performance on the subgroup for which they have the worst performance, and penalised if their error rate for one group is higher than the others.
In another set of models, the researchers forced them to remove any demographic information from the images, using “group adversarial” approaches. Both strategies worked fairly well, the researchers found.
“For in-distribution data, you can use existing state-of-the-art methods to reduce fairness gaps without making significant trade-offs in overall performance,” Ghassemi says. “Subgroup robustness methods force models to be sensitive to mispredicting a specific group, and group adversarial methods try to remove group information completely.”
Not always fairer
However, those approaches only worked when the models were tested on data from the same types of patients that they were trained on, eg from BIDMC.
When the researchers tested the models that had been “debiased” using the BIDMC data to analyse patients from five other hospital datasets, they found that the models’ overall accuracy remained high, but some of them exhibited large fairness gaps.
“If you debias the model in one set of patients, that fairness does not necessarily hold as you move to a new set of patients from a different hospital in a different location,” Zhang says.
This is worrisome because in many cases, hospitals use models that have been developed on data from other hospitals, especially in cases where an off-the-shelf model is purchased, the researchers say.
“We found that even state-of-the-art models which are optimally performant in data similar to their training sets are not optimal – that is, they do not make the best trade-off between overall and subgroup performance – in novel settings,” Ghassemi says. “Unfortunately, this is actually how a model is likely to be deployed. Most models are trained and validated with data from one hospital, or one source, and then deployed widely.”
The researchers found that the models that were debiased using group adversarial approaches showed slightly more fairness when tested on new patient groups than those debiased with subgroup robustness methods. They now plan to try to develop and test additional methods to see if they can create models that do a better job of making fair predictions on new datasets.
The findings suggest that hospitals that use these types of AI models should evaluate them on their own patient population before beginning to use them, to make sure they aren’t giving inaccurate results for certain groups.
By Ben Selier, Vice President: Secure Power, Anglophone Africa at Schneider Electric
The adage, knowledge is king couldn’t be more applicable when it comes to the collection and utilisation of data. And at the heart of this knowledge and resultant information lies the datacentre. Businesses and users count on datacentres, and more so in critical services such as healthcare.
Many hospitals today rely heavily on electronic health records (EHR), and this information resides and is backed up in on-premises datacentres or in the cloud. Datacentres are therefore a major contributor to effective and modernised healthcare.
There are several considerations when designing datacentres for healthcare. For one, hospitals operate within stringent legislation when it comes to the protection of patient information. The National Health Act (No. 61 of 2003), for example, stipulates that information must not be given to others unless the patient consents or the healthcare practitioner can justify the disclosure.
Datacentres form part of critical systems
To add an extra layer of complexity, in South Africa, datacentres should feature built-in continuous uptime and energy backup due to the country’s unstable power supply. Hospitals must therefore be designed to be autonomous from the grid, especially when they provide emergency and critical care.
Typically, datacentres are classified in tiers, with the Uptime Institute citing that a Tier-4 datacentre provides 99.995% availability, annual downtime of 0.4 hours, full redundancy, and power outage protection of 96 hours.
In healthcare and when one considers human lives, downtime is simply not an option. And whilst certain healthcare systems and its resultant availability are comparable to a typical Tier-3 or Tier-4 scenario, critical systems in hospitals carry a higher design consideration and must run 24/7 with immediate availability.
In healthcare, the critical infrastructure of a hospital enjoys priority. What this means is the datacentre is there to protect the IT system which in turn ensures the smooth running of these critical systems and equipment. There is therefore a delicate balance between the critical systems and infrastructure, and the datacentre, one can’t exist without the other.
Design considerations
To realise the above, hospitals must feature a strong mix of alternative energy resources such as backup generators, uninterrupted power supply (UPS) and renewables such as rooftop solar.
Additionally, like most organisations, storage volume and type and cloud systems will also vary from hospital to hospital. To this end, datacentre design for hospitals is anything but cookie cutter; teams need to work closely with the hospital whilst meeting industry standards for healthcare.
When designing healthcare facilities system infrastructure, the following should also be considered:
Software like Building Management Systems (BMS) are not just about building efficiency but also offer benefits such as monitoring and adjusting indoor conditions like temperature control, humidity, and air quality.
The BMS contributes to health and safety and critical operations in hospitals whilst also enabling patient comfort.
Maintenance – both building and systems maintenance transcend operational necessity and become a matter of life or death.
As mentioned, generators are essential when delivering continuous power which means enough fuel must be stored to run it. Here, hospitals must store fuel safely and in compliance with stringent regulations. In South Africa, proactively managing the refuelling timelines is also critical. The response times of refuelling these (fuel) bunkers can be severely hindered by issues such as traffic congestion as a result of outages and lights now working.
Selecting the right equipment for hospitals is therefore a delicate balance between technological advancement and safety. For instance, while lithium batteries offer many benefits, when used in hospitals, it is paramount that it is also stored in dry, cool and safe location.
Here, implementing an extinguishing system is a must to alleviate any potential damage from fire or explosions. That said, lithium batteries are generally considered safe to use but it’s important to be cognisant of its potential safety hazards.
Ultimately, hospitals carry the added weight of human lives which means the design of critical systems require meticulously planning and executed.
