Caroline B Hing and Yasmin Antoniou of AI for Good explore the future of AI in healthcare and clinical practice.
Automation during the industrial revolution led to a profound change in working practices across the 18th and 19th centuries. Currently, we are in the midst of a fourth industrial revolution, with globalisation and automation affecting every aspect of our working lives and leisure activities. The COVID-19 pandemic has provided a further driver for change, altering how we work and interact with remote working and reduction of human interaction at the centre of global initiatives to try and reduce the spread of the virus.
Healthcare systems generate large quantities of complex datasets pertaining to patients. Artificial Intelligence (AI) can offer solutions to medical problems by attempting to replicate decisions that would otherwise require human intelligence. Specific algorithms can be created in order to make associations between data and predict future outcomes. The COVID-19 pandemic has accelerated change within healthcare systems and driven interest in automated algorithms capable of assisting hospitals in diagnostics, decision making and repetitive clerical tasks thus reducing the potential ‘footfall’ of staff on site. Automation and intelligent algorithms that learn and improve with further iterative cycles require data and the ethics behind large personal data sets, the challenges of anonymisation remain in their infancy.
AI applications within healthcare can be broadly categorised into diagnosis, research, management and system analysis. Ultimately in the time of COVID, workforce adaptations have led healthcare providers to use their workforce wisely, reducing the staff requirements on site and moving towards remote working.
Electronic health records and data analysis
Healthcare records in most countries have moved from paper-based records and notes to digital media. The electronic health record (EHR) allows the capture of large quantities of data across patient groups. Future patient care aims to develop tailored treatments for patients in a cost-effective manner.
The use of large datasets requires an element of data curation. Data needs to be retrieved from multiple disparate data sources. Data needs to then be ‘cleaned’ to remove anomalies and ‘harmonised’ to ensure similar data sets are compared across patient records. An element of these processes needs human oversight to ensure the correct data is fed into the algorithms.
Automation and standardisation
Administrative applications are resource-heavy and repetitive. These applications include workflow management such as uploading referral letters from primary care, setting up referral assessment services and booking patients to the correct service provider in secondary care. Robotic process automation (RPA) consists of computer programmes that ‘obey’ rules for these manual, resource-heavy tasks.
Additionally, process management systems (PMS), while embedded within commercial businesses, are still in their infancy within a healthcare application. Patients are individuals and tailoring care to them necessitates being able to react to changing physiological parameters within the confines of the organisation. In order to standardise care, clinical pathways have been developed to manage patient care from referral, to ordering diagnostic tests and eventual treatment pathway. While standardisation across patient groups allows for automation of some of these processes (templates for referral, standard test orders), allowing a more tailored approach which is patient-specific is the ultimate goal; thus generating a conflict to resolve between standardisation and a tailored patient approach specific to their individual needs.
Flexibility within a process management system requires technological skills to allow tasks to be postponed or reorganised. However, healthcare professionals lack the technical skillset to implement this, requiring a user-friendly interface. Further work in user interface and user experience (UI/UX) is therefore required to ensure a system that allows flexibility without losing the advantages of rapid automation and processing of large numbers of patients within the pathway management systems.
Automating healthcare administration
Increasingly call centre staff for websites have been replaced by chatbots that use natural language processing (NLP) to provide callers with information and manage queries. A chatbot is a type of AI programme that can conduct an intelligent conversation via text or auditory methods. It is predicted that by 2025 the global chatbot market will be worth $1.23bn. For hospitals dealing with upwards of 10,000 patient appointments per week, the use of chatbots to handle patient queries regarding appointment queries is still in its infancy when compared to more established sectors such as banking or commerce.
AI in screening and diagnosis
The current COVID-19 pandemic has highlighted the need for rapid screening and testing of patients to improve treatment pathways and also reduce the risk of cross infection. Clinical testing requires taking biological samples from patients which can be resource heavy and incorporates a time lag before results are available from real-time polymerase chain reaction testing (RT-PCR). The use of AI within this environment accessing electronic health records (EHR) of routinely ordered tests and vital signs can produce an effective tool to screen patients in emergency departments and hospital admission units.
Predictive analytics utilise AI algorithms to analyse healthcare data from EHRs to predict future outcomes thus aiming to improve outcomes and the patient experience as well as reducing costs. Data collected from EHRs can be supplemented with data from wearable technology and medical devices. Risk prediction models utilising AI would improve with successive data collection cycles aiming to supplement decision making by clinicians. Applications include management of chronic diseases such as chronic renal failure, diabetes and cardiovascular disease. Specific patient populations can vary across geographical healthcare providers and the ability of a predictive model to ‘learn’ from its local population provides an advantage over established static modelling. Scalability across healthcare providers can therefore be challenging due to differences in socio-economic factors and populations based on geographical location. The ethical implications in terms of health insurance and risk stratification are in their infancy; and issues around data governance and data sharing may have a significant impact that is yet to be fully regulated.
Looking beyond COVID-19
Diagnostic applications of AI technological advances have exploded over the past ten years with multiple applications. Imaging studies such as breast mammograms or histological analysis rely on skilled scientists or clinicians performing repetitive tasks to manually identify abnormalities. An inaccurate diagnosis can have serious consequences for patient care. AI programmes can be trained to perform these tasks and have shown an accuracy in correctly diagnosing abnormalities which in some studies has been shown to be as accurate as a trained clinician. Further future applications in diagnostic imaging include the field of radiomics which extracts nuanced features peculiar to imaging modalities such as wavelength, texture and shape. This additional information can provide further data for diagnosis and prognostic indicators specific to patients.
Towards an AI-powered future
With the potential application of AI within the healthcare setting, the question remains how will this impact the workforce? The fourth industrial revolution has 50% of companies to predict that by 2022, automation will decrease their numbers of full-time staff and that by 2030 robots will replace 800 million workers across the world (McKinsey Global Institute reports). Automation of clerical processes and care pathways could potentially impact on the non-clinical workforce within a healthcare setting. Specialties such as radiology where imaging reports can be automated and produced by AI algorithms may soon be the reality. The ethics of data sharing and the implications for patients and their insurers is a further area of controversy. We enter a brave new world.
Caroline B Hing
AI for Good