Digital Health and AI 2024

Telehealth or real-time remote communication between patients and clinicians is not new and has been providing healthcare support to populations with limited access in many countries.13 It is already replacing many face-to-face clinical consultation and monitoring visits to free up pressurized clinician time and refocus it where and when it makes most sense. Additional benefits include a reduced risk to healthcare professionals during infectious disease outbreaks, as was seen during the COVID-19 pandemic. AI-assisted telehealth has the potential to support and direct health advice based on risk information. Algorithms can generate a “stepped care” approach to organizing the delivery of care and treatment. Here the most effective, least intrusive and least resource-intensive treatments are delivered to patients first, with “self-correcting” mechanisms in place so that those people who do not benefit from initial treatments can be “stepped up” to access more intensive treatments as they need them.14 These developments are particularly helpful in workplace settings where early interventions can be promoted to provide some immediate medical support and prevent issues from getting worse before a face-to-face appointment is made. AI-enabled clinical equipment is being developed to allow low-cost, novice-performed “point of care” data assessments that drive critical decisions with similar accuracy to that of hospital expert assessments; recent studies have found that ultrasound assessments of gestational age using AI-enhanced ultrasonography offer great promise for establishing gestational age for improved decision-making and overall pregnancy care, especially in LMICs.15,16 Use of AI and machine learning has also been proposed as a means of expanding access to vision screening for diabetic retinopathy (DR), particularly in remote areas where healthcare access is limited. Automated detection and grading of DR and predictions of progression from retinal photographs all have the potential to increase case detection and triage referrals at a time when sight can be preserved. Such applications have potential benefits for workers, organizations and the public, especially those in jobs with specific visual acuity requirements, such as vocational drivers. Similar approaches to the delivery of a wide range of health interventions can be considered as part of an overall employee health benefits strategy. At the same time, the challenges of implementing such systems without access to reliable and fast internet connections and meeting demands for those requiring a clinical intervention should not be underestimated, especially in low-income countries. A significant proportion of the costs associated with health and healthcare relates to labour-intensive administrative activity. Automating hospital logistics is necessary to improve resource allocation and meet ever-increasing healthcare demands and operations. Significant savings can also be made when activities such as appointment scheduling and billing can be automated.17 The same is true in occupational settings, where AI can be used to optimize the identification of exposed populations for health surveillance, to schedule appointments in electronic health record systems and to remind employers of their accountabilities, when they need to control exposure and before any irreversible harm is caused to workers. Of course, people of working age are also the beneficiaries of wider AI-driven improvements in population health,18 including personalized healthcare, accelerated drug discoveries and precision medicine, all of which can reduce the time they may need to be away from their jobs. At most, a very small proportion of AI-developed health tools have been fully evaluated to assess their impacts on health outcomes and workforce productivity. Integrating healthcare access opportunities as employee benefits in workplace settings is an area in which a close working relationship between chief health officers, academics and reward managers can be beneficial. Areas for further focus and extra vigilance Just as AI has the potential to improve worker health and organizational productivity, areas of concern are being identified every day in which caution is needed to ensure that the benefits are not outweighed by the harms to health and reputation. Workforce dissatisfaction and anxiety related to job loss/disruption Workplace relationships are fundamental to the long-term success of all safe, effective and productive organizations. The World Economic Forum’s Future of Jobs report estimates that new technologies will be at the heart of almost 69 million new jobs in the next five years.19 However, with up to 83 million new jobs being put at risk at the same time (corresponding to 2% of employment of the time of the report), current jobholders will face significant change and uncertainty, which is known to be associated with anxiety. Use of AI technologies and “people analytics” are becoming integral components of business decision-making. If widely used in operational settings to increase the efficiency of work processes, they can, however, lead to greater intensification of work. AI-driven task assignment, monitoring and scheduling of activities and breaks can reduce worker involvement in decisions that affect them. Such increases in workload and reductions in worker autonomy are well known to increase work stress and risk of burn-out.20,21 At the same time, there is no doubt that computers are better at rapidly processing multiple, complex datasets than humans and have the potential to reduce digital overload when sympathetically deployed. There is an important role for health and well-being advisers to inform organizations of the potentially unexpected consequences of AI-driven technical advice. They can also be instrumental in shaping more successful implementation strategies that include support for the mental health and well- being of employees as they navigate these changes.22 Bias in datasets driving AI-based decision-making and distribution of benefits AI has the potential either to improve or to exacerbate existing workplace safety and health inequities. AI algorithms are dependent on large datasets, and yet, in many of the existing health databases driving these algorithms, women, ethnic minorities and other underserved populations are underrepresented in the core data. In employment settings, the datasets may also be relatively small, magnifying such inequalities. This makes the conclusions drawn and advice given less relevant to those populations and in some cases harmful.