Why digital innovation may not reduce healthcare’s environmental footprint

Environmental effects of digital innovation in healthcare

Digital innovations have the potential to decrease the environmental harm from health systems in several ways (box 1). First, digital innovations are expected to help reduce the greenhouse gas emissions associated with existing healthcare facilities by improving their efficiency. In the UK the NHS has predicted carbon savings through the use of realtime monitoring, including artificial intelligence, to better control buildings (eg, lights, heating, and cooling) and to forecast resource allocation more effectively.6 Use of digital technologies to predict electricity and water consumption across various healthcare facilities has allowed hospital managers to identify variation in usage and deal with the causes.7

Second, digital innovations are expected to provide a substitute for services or processes that use materials and energy. For example, digital imaging infrastructures have cut chemical use for film. Telemedicine reduces transport related emissions compared with equivalent in-person appointments.8 Remote monitoring of outpatients can also reduce transport associated carbon emissions.910 It may also reduce energy and material use through the earlier detection of clinically important symptoms, thereby reducing the need for intensive healthcare intervention.1112

Third, the increasing move to telemedicine and remote monitoring is expected to mean that fewer people will attend hospitals, which will mean smaller healthcare facilities with lower carbon emissions.13 Pilot studies already suggest the effectiveness of remote realtime monitoring of patients’ blood pressure14 and blood glucose levels.15 Preliminary evidence also suggests digitally delivered therapy can help support people struggling with their mental health.1617 Use of AI to improve the effectiveness and accessibility of healthcare delivery is well under way in high income countries and of growing interest in low and middle income countries.18

At the same time, some research is showing evidence of increased consumption patterns associated with telemedicine. Quality Watch—a joint research programme for the UK Health Foundation and UK Nuffield Trust—found that more follow-up appointments were required and there was a lower rate of discharge from an online service compared with face-to-face appointments. Moreover, online appointments led to increased rates of new prescriptions and referrals.19 Further complexities include ensuring patient privacy during virtual appointments and internet access problems, both of which may mean in-person follow-up appointments are also needed. Such problems may affect disadvantaged socioeconomic groups disproportionately, leading to widening inequities.

Growth in digital infrastructure

Although generally positioned as reducing environmental impact, digital technologies produce emissions from the energy used to collect, store, process, and analyse data. Greenhouse emissions associated with digital technologies are difficult to quantify and will depend on (among other factors) when and where data are stored and processed. Estimates suggest that the digital sector accounts for 1.8-2.8% of all global emissions.20 The carbon footprint of storing and processing data depends on numerous factors (eg, location, type of storage, and server used) but is thought to be around 10 kg (range 4-28 kg) carbon dioxide equivalents (CO₂e) a year for 1 TB of data storage.21 Emissions will rise with increased storage, but the extent depends on the use of renewables, types of data storage, and hardware and software efficiencies.

Over the next few years, healthcare is primed to grow faster than other sectors in the digital sphere because of advances in healthcare analytics. Some industry predictions state that data requirements will exceed 10 billion TB (10 zettabytes) by 2025.22 Advances in digital pathology are a useful example of such growth, with clinically applicable deep learning support already available in cancer pathology. A German study estimated that deep learning led analysis of worldwide pathology cases would create about 16 megatonnes of CO₂e a year (requiring up to 86 590 km² (0.22%) of world forest to sequester the emissions).23

A recent commentary on the use of AI for imaging and informatics raised sustainability concerns about the rapidly increasing computational intensity of AI models and argued for full lifecycle assessment of their emissions.24 Developments in genomics provide another example. Several NHS embedded research studies offer whole genome sequencing—for example, the newborn whole genome sequencing study. Although only a tiny proportion of the genome (<0.01%) will be analysed to screen for genetic conditions and informing the newborn’s healthcare,25 the rest of the genomic data will be stored, creating much higher storage costs than current targeted genetic testing. The fact that such ventures straddle both research and healthcare settings—the drive to store whole genomes is research led, yet the healthcare benefits are only available if parents take part in the research—means the footprint for healthcare is difficult to measure, but the whole genome sequence of half a million participants in UK Biobank requires about 27.5 petabytes (27 500 TB) of storage.26

The manufacture, use, and disposal of digital technologies also create environmental harms. These include the environmental cost of mining and depleting natural resources, the use of water to provide systems to cool digital servers, and the disposal of electronic hardware waste.272829 Such processes often rely on extractivist and exploitative practices that particularly affect people in low and middle income countries,30 For example, electronic waste is often disposed of in low and middle income countries, where informal recycling provides an income. However, this informal work is part of broader neocolonial processes that can widen inequities and often involve practices such as burning and acid bathing that may harm the health of those working and living in the vicinities.31

Over the past five years, improvements in digital capabilities have allowed for increasing efficiency so that energy and resource consumption has not increased in line with societies’ growing appetite to gather and process ever more data. However, efficiency gains are unlikely to keep up with the drive to create and gather ever more data.

Rebound effect

Behaviour has been shown to change in response to perceived cost and energy savings, and this can lead to energy savings being less than expected (the rebound effect32) or even increasing (backfire effect). A highly efficient refrigerator may still consume more energy than a highly inefficient one because—thinking it is efficient—we might open the door more, clean it less, or purchase additional appliances with the savings. Equally, in healthcare, if we use any savings made to collect and process more data, overall savings reduce. Health and social care systems in England have recently moved data to a cloud based server to improve storage and transmission,33 but this may have the effect of allowing the collection and processing of more data—for example, gathering whole genome sequences routinely rather than using clinically guided genetic testing.

Remote monitoring may also lead to increased digital consumption because more powerful and resource intensive algorithms are used on patient data. This is particularly the case for digital phenotyping, where data from smart devices create a holistic digital picture of behaviours, often using AI.3435 Google has reported that AI represents 10-15% of its power use.36

Moving forward

Some of the concerns about the energy consumption of digital technology might be lessened by moves towards renewable energy. Health systems and facilities in industrialised countries are already making increased use of renewable energy,37 and there are many examples of new health systems in low and middle income countries being developed alongside renewable infrastructures to ensure energy sufficiency.38 Furthermore, many large technology companies providing cloud services to the healthcare sector have already declared net zero targets.39

However, the sector’s support of renewable energy does not mean that growth of this sector has no environmental consequences. For instance, data centres may outrun renewable power consumption or leave a shortfall for other sectors.20 Furthermore, many healthcare systems in low and middle income countries are still powered by fossil fuels, especially in places where electricity outages and the need for back-up generators are common.40 Overall, this is not a reason to refrain from using digital technology, especially if it clearly benefits patients. What is important, though, is that this patient benefit is assessed rather than assumed, and that digital technologies are implemented in ways that mitigate their associated environmental harms as much as possible.

To do this, efficiencies gained using digital technologies must not be seen as a fix for the environmental harms of healthcare systems but as one way to contribute to making healthcare more environmentally sustainable. Monitoring and evaluating the positive and negative environmental and health effects of digitalisation will be important. (Of course, any evaluation will itself require data and have an environmental impact.) Such assessments are complex because digital pathways are often missing from databases that collate information about emissions associated with healthcare activities,41 but they need to be included. Assessments should also evolve to include measurement of rebound effects through more deliberative qualitative and quantitative reflection.42 The SusQI framework, developed by the UK Centre for Sustainable Healthcare, is one attempt to do this. Although it is not specific for digitalisation, it provides a useful starting point.43

Incorporation of assessment frameworks into broader international and local digital health governance structures such as the World Health Organization’s digital health strategy44 or the UK National Institute for Health and Care Excellence evidence standards,45 will allow environmental issues to be assessed alongside considerations such as safety, equitability, and effectiveness. Environmental effects also need to be embedded into procurement strategies for digital software, hardware, and services.46 Countries will differ in how best they can consider environmental implications given competing (health) priorities, as well as other potential constraints (for instance resources and data sovereignty issues). These differences may be greatest between high income countries and lower income countries.

Healthcare systems face the daunting challenge of rapidly decarbonising and becoming environmentally sustainable as part of the global response to climate change. Although many hope the ongoing digitalisation of healthcare will make healthcare systems sustainable, getting to that goal is nuanced and complex. More comprehensive ways to measure and understand the environmental costs of digitalisation of healthcare systems are needed together with informed debate about assumptions that more data and advanced data processing are always a good thing for healthcare.