Correctly applied, artificial intelligence, (AI), has major potential to drastically improve equity, access and quality of healthcare in South Africa, Dr Tapiwa Chiwewe, Research Manager for IBM Research in South Africa.
Speaking at the Hospital Association of South Africa (HASA) conference being held in Cape Town, he said IBM was already helping South Africa to reduce cancer reporting delays (which currently run to four years), by applying deep machine learning algorithms to classify text-based pathology reports, mainly in breast cancer, thus updating and rendering more current the cancer registry so essential for health records and planning population-wide interventions.
The company was also helping improve SA’s health systems for care coordination and implementing clinical workflows by sharing data between clinics and pharmacies and referral hospitals, the underlying blockchain technology also helping control access to patient records. By analysing different patient cohorts with similar characteristics or symptoms such as hypertension or diabetes, machine learning could automatically cluster groups of patients together, increasing treatment efficiency.
“Superhuman” low error rate
Describing artificial intelligence as ‘augmented intelligence,” helping clinicians and its human operators, Dr Chiwewe said that AI was now so advanced that its error rate was 3%, two percent lower than human beings. This was dependent on accurate data being inputted for processing. With the modern data deluge accounting for 60% of determinants in health volume, variety, velocity and veracity, generating 11 000 terabytes per person per lifetime, and genomics data accounting for six terabytes per person, it was obvious that healthcare staffers and researchers could not cope without AI.
Giving an example of AI-value in population health, Dr Chiwewe said a Minister of Health concerned about regular disease outbreaks could digitise paper records to begin analysing data to correlate historical records from different geographic locations, seasonal variations in rainfall, windspeed, home roof designs in order to predict and plan for future outbreaks. Front line healthcare workers in rural areas could quickly diagnose a child with fever and a rash, initially suspected of suffering from Dengue Fever, by inputting a picture of the rash, and quickly diagnosing it as a spider bite or malaria, referring onwards for appropriate specialist treatment. He said the more data there were, the better AI performed. It was particularly effective in research and development, he said citing the $2 billion it cost a develop a new pharmaceutical drug, where 10% of drugs in development never made it to market.
“We estimate there’s about $2 trillian in waste in the healthcare industry globally, via things like unnecessary tests and variability of care or failed clinical trials. When you add government fraud, waste and abuse and the cost of treating chronic disease, we estimate AI can create cost savings of around $360 billion annually,” he added.
Some of the challenges to data innovation included unrepresentative datasets, data availability and quality. Black and minority ethnic groups were more likely to opt out of volunteering their health data for research and thus became more “invisible” to scientists. Data protection rules were also complex, leading to data holders becoming risk-averse and often restricting access to it. However, advocacy and the technology itself could help lower these barriers. A major deterrent to data scientists was that data could be ‘messy and complex,” driving academics into higher paid jobs in the technology sector. The utility of wearable apps included early detection of depression or suicide risk, monitoring healthcare worker movements during a working day to unblock treatment bottlenecks, and early warnings for Alzheimer’s Disease, he added. It’s all about collecting as rich a dataset as you can,” he added.