Abstract
Digital medicine is already well established in respiratory medicine through remote monitoring digital devices which are used in the day-to-day care of patients with asthma, COPD and sleep disorders. Image recognition software, deployed in thoracic radiology for many applications including lung cancer screening, is another application of digital medicine. Used as clinical decision support, this software will soon become part of day-to-day practice once concerns regarding generalisability have been addressed. Embodied in the electronic health record, digital medicine also plays a substantial role in the day-to-day clinical practice of respiratory medicine. Given the considerable work the electronic health record demands from clinicians, the next tangible impact of digital medicine may be artificial intelligence that aids administration, makes record keeping easier and facilitates better digital communication with patients. Future promises of digital medicine are based on their potential to analyse and characterise the large amounts of digital clinical data that are collected in routine care. Offering the potential to predict outcomes and personalise therapy, there is much to be excited by in this new epoch of innovation. However, these digital tools are by no means a silver bullet. It remains uncertain whether, let alone when, the promises of better models of personalisation and prediction will translate into clinically meaningful and cost-effective products for clinicians.
Shareable abstract
Digital medicine used in remote monitoring technologies, the electronic health record and image analysis in radiology have changed routine care. Future applications include digital healthcare assistants and AI that will improve predictive analysis. https://bit.ly/4deVc38
Background
“AI will never replace physicians – but physicians who use AI will replace those who don't”
Jesse Ehrenfeld, American Medical Association President, July 2023
Digital medicine is an umbrella term used to describe innovations in healthcare enabled by a variety of digital technologies. There are three broad categories of digital medicine, namely connected health, eHealth and precision medicine. Connected health involves the collection and analysis of continuously recorded remotely monitored physiological data and health behaviours [1]. These data provide granular information on an individual's health status [2–6]. eHealth refers to the use of information technology to support healthcare delivery; examples of its use include digital pharmacy services and platforms such as the electronic patient health record [7]. Precision medicine involves advanced statistical analysis of genetic, clinical, behavioural and physiological data, which is used to gain unique person-specific diagnostic, treatment and prognostic insights [8].
The opportunities for these new tools in the field of respiratory medicine are immense. For example, personalised treatments for lung cancer are increasingly being recommended for individuals based on data obtained from advanced statistical analysis of digitally formatted molecular and genomic data [9]. Even in common clinical conditions such as COPD, patients have many co-existing health conditions as well as individual health behaviours [3]. This means that to understand an individual's symptoms, such as breathlessness, remote monitoring tools are required. These can help to provide insights into the many potential causes, such as levels of exercise, medication adherence and spirometry [3, 10].
Outside of the potential application of digital medicine to improve direct clinical care, digital technologies offer huge opportunities for the delivery of healthcare. Electronic health records (EHRs) are an example of this aspect of digital medicine. The development of artificial intelligence (AI) tools such as large language models (LLMs) and chatbot assistants designed specifically to perform routine tasks are opportunities for digital technologies to improve the clinician's workflow [11].
Despite their potential, there are many important limitations to these technologies. Central to these are privacy concerns and the risk of systemic bias. Sharing health data with agents outside of the care team, as well as data theft from security breaches, are real areas for privacy concern [12, 13]. Remote monitoring digital tools provide important insights into social and human behaviour; however, they can also be considered intrusive and thus also raise privacy concerns [14]. The potential for social divide, with those with poor access to the internet being left behind, and the inherent racial bias in certain AI models, are major concerns with regard to perpetuating systemic bias as the use of digital health tools expands [15, 16]. Reliability, consistency and translatability are also key concerns to be addressed before any AI tool can be widely distributed. AI utilises complex statistical functions to apply advanced algorithms to health data in an attempt to mimic human clinical reasoning. Different clinical settings can influence AI outputs, leading to inconsistencies and challenges in reliably reproducing or translating them across other healthcare settings [17].
Healthcare delivery has become fragmented as patient care is shared and delivered by a combination of increasingly sub-specialised physicians, primary care physicians, emergency departments and out-of-hours urgent care services [18]. This fragmentation emphasises the need for shared, objectively collected health data. The recent experience of telemedicine, which at first seemed so convenient but has not been persistently adopted because it failed to address the need for in-person interactions with clinicians, illustrates the fundamental limitation of digital medicine. Central to the adoption of these innovations will be how they impact the interactions between clinicians and patients. In this article, we outline digital developments in respiratory medicine to show the opportunities as well as the threats of this new field of medicine and discuss how these developments may be incorporated into clinical practice.
Methods
The focus of this review is on how innovations in these areas may alter how healthcare is delivered by respiratory clinicians under the three broad themes of connected health, digital information technologies and precision medicine. A PubMed search of respiratory medicine airways, sleep, radiology, machine learning, AI and digital technology was performed to support this review. A glossary of some of the more frequently used methods referred to in this article is listed in table 1.
Telemedicine clinics, virtual wards, digital therapeutics and remote monitoring as adjuncts to current clinical care
As one of the best recognised examples of digital medicine, telemedicine serves as an alternative to traditional in-person clinic visits [19]. There are many cases in respiratory medicine where telemedicine has developed into a sustained and practical way of delivering healthcare. For example, telemedicine is well suited to manage patients with sleep and ventilation disorders. Using remotely monitored data from ventilation devices and connecting virtually with patients most in need of care rather than routine in-person clinic visits makes for a more efficient delivery of care [20]. Similarly, pulmonary rehabilitation can be successfully delivered online, leading to a wider participation by patients who might otherwise not be able to travel to in-person classes [19, 21]. These examples address some challenges associated with traditional in-person clinics, notably time constraints and patient convenience. The approach also leads to enhanced patient retention and reduced carbon emissions by minimising travel requirements [22]. Patients have high rates of reported satisfaction with telemedicine delivered care, demonstrating that they are not only practical but well-received [23]. However, clinician enthusiasm for telemedicine has significantly declined since the coronavirus 2019 (COVID-19) pandemic, when its use flourished [24]. Some of this waning enthusiasm may reflect the difficulty in financial reimbursement, as well as a return to the “old ways”, wherein clinicians feel that they deliver better care in person providing the “human touch”. Clinician concerns include the lack of same-day diagnostics, language barriers, cultural differences and technological issues, all of which can hinder communication when done remotely [25]. As an adjunct to standard clinical care, telemedicine is a viable alternative for rural populations, as well as some racial and ethnic minorities, alongside other historically underserved communities. For these groups it will not replace current care models, rather it will just serve as an adjunct to support care.
Emerging during the COVID-19 pandemic, virtual wards were developed to monitor patients at home and thus avoid unnecessary hospitalisation [26]. One example during the COVID-19 pandemic was their use in remotely monitoring patients who may have required escalation or hospital admission for “silent hypoxia” in the absence of associated breathlessness [26, 27]. Another is home sleep apnoea testing, which, when used appropriately in selected populations, increases convenience and accessibility while reducing waiting times for appointments, diagnosis and treatment initiation [20]. Home wards have the potential to worsen socioeconomic division, as they require a patient to have a suitable infrastructure to remain at home. This requirement means that socially or economically disadvantaged groups may not be able to utilise these services, meaning that this is not a model that can be adopted globally in its current form, despite the potential for cost saving and patient convenience [26].
Digital therapeutics (DTx) deliver medical interventions directly to patients using evidence-based, clinically evaluated software aimed at treating and preventing a broad spectrum of diseases and disorders [28]. DTx are increasingly being used in respiratory medicine for conditions such as smoking cessation [29]. Virtual cognitive behavioural therapy platforms, accessible online or via app-based programmes, are now recommended by National Institute for Health and Care Excellence guidelines for the treatment of insomnia [30]. Digital therapeutics for dysfunctional breathing, a common yet debilitating condition, are also in development [31]. A major threat of these digital therapeutics is from social media apps and influencers which might spread misinformation and unproven therapies through unregulated applications [32, 33]. However, on balance, regulated digital therapeutics offer tailored treatments for benign chronic conditions and are a welcome suitable replacement for clinic activities and reduce the burden on services.
Wearable devices, such as smart watches, which collect biometric data including oxygen saturation, heart rate variability and sleep patterns are used in telemedicine, as an adjunct or replacement for some laboratory tests [34]. Examples in respiratory medicine include bespoke devices used to screen for sleep apnoea [35, 36] and home spirometry, with data stored on person-specific platforms for monitoring patients with idiopathic pulmonary fibrosis, cystic fibrosis and following lung transplant [37–39]. Digital inhalers, used to monitor adherence and technique, when paired with lung function, have been shown to reduce the need for biologic add-on therapy in asthma and subsequently reduce the cost to the healthcare system (with the financial effect of different rates of add-on therapy estimated to reduce costs by €3000 per patient per year or a lifetime saving of over €60 000) [40–42]. Telemedicine, virtual wards, digital therapeutics and remote-monitoring technologies have all developed through the innovation cycle from initial enthusiasm, through a phase of disappointment and are emerging with better identified adjunct roles supporting aspects of healthcare.
Table 2 provides further examples of the use of digital health in optimisation of routine care in respiratory medicine.
The EHR – a cause of division
The EHR is an underappreciated but essential tool of digital medicine. Thought to be essential for patient safety [43, 44], the potential benefits of EHRs are also clear from a research perspective. The landmark COVID-19 RECOVERY trial demonstrated the value of the EHR. Baseline demographics, patient randomisation and online follow-up were all collected from the EHRs and this was pivotal to the success of the trial that randomised over 6000 patients to the study in a 3-month period [45]. On the other hand, EHRs impose significant burdens and frustrations among clinicians and have been linked with clinician burnout [46, 47]. Responding to this frustration, AI-powered algorithms that streamline data entry and centralise documentation of patient care are being developed. Examples include natural language processing models, which process and analyse free text and have been shown to increase data accuracy and reduce human error in medical record keeping [48]. GPT-4 and other bespoke LLMs are increasingly being used to create structured “pseudo-personalised” documentation, such as letters to patients explaining results and summarising data for other healthcare professionals [49, 50]. Future uses include opportunities to incorporate data from remote monitoring devices into LLM-enabled chatbots, which would deliver autonomous monitored care. In short, novel AI technologies have the potential to move EHRs from foe to the embodiment of digital medicine. Major areas of concern still persist regarding data privacy and security; when breaches occur, they can be devastating, with both loss of clinical data and trust in the providers [44, 51]. Despite these concerns, there are clear opportunities for AI to change how the EHR serves clinical care.
AI in respiratory medicine
One of the key features of AI that makes it suitable for clinical applications is its ability to recognise patterns. A glossary of some of the more frequently used AI methods referred to in this article are listed in table 1. Some cases are illustrated to show the potential of these technologies, the steps required before broad adoption and the need to retain the patient–clinician relationship.
In chest radiology, AI, in particular using deep-learning techniques, can identify a variety of conditions, such as interstitial lung disease, pneumothorax, cystic lesions and pulmonary nodules, in particular those most likely to be malignant [52–55]. Utilising AI to detect these conditions has the potential to improve diagnostic accuracy through reducing human error as well as supporting nonspecialist centres to deliver care at the same standard as advanced ones. Hampering their widespread clinical deployment are some important practical issues. For example, in the case of the radiologic assessment of pneumothorax, the presence of a chest drain effects the accuracy of the AI model [54]. Recognition of such artefacts has led researchers to realise that there needs to be a robust system of testing with a “human in the loop” [56]. In other words, before AI-enabled technology is made commercially available, robust testing, validation and certification are needed and, once used in practice, repeated re-audits are required.
Sleep medicine is ideally placed to benefit from these technologies and deliver more personalised diagnosis and treatment. Data-driven machine-learning algorithms have been shown to be effective through precision-based patient stratification to identify where patient testing should be performed [20, 57]. Further development of these technologies and analysis of data from smartwatches and other wearable devices could lead to the automation of the diagnostic and therapeutic assessment of patients with sleep apnoea. It is not difficult to imagine an app deployed on a phone's operating system that, when linked to a smart watch, detects that the wearer has sleep apnoea, recommends a continuous positive airway pressure provider, monitors the effectiveness of therapy and provides practical information on mask fitting and adherence. The entire management could thus be provided with no clinician input. The regulatory and indemnity risks to manufacturers of these remote monitoring devices and software are large and potentially unavoidable barriers to the complete replacement of clinicians. However, the ethical concern of not being able to attend to the long waiting lists of people with sleepiness is an important issue to consider when evaluating the threat to clinician livelihood of fully automated systems.
AI models that predict sepsis and acute renal failure that outperform rule-based models have already been developed [58, 59]. The first-generation models had low specificity (many false positives), which affected their accuracy and practical usability [60]. One reason for their low specificity was that the input data used to train the models was derived from retrospective EHR data [61]. Missing data or imprecise recording of the timing of events, such as when blood tests were taken or the precise times that clinical notes refer to, significantly impacted the model's performance [61]. Such limitations could be overcome by prospective studies where data is comprehensively collected. In time, we foresee that AI-informed clinical decision aids will replace generic guidelines and rules of thumb for clinical decision-making.
While the above-mentioned models may replace guidelines and assist clinicians in detecting potential risks, it will be a good deal more challenging for the models to develop to the point where they can predict when an event will occur [62]. It will also be a challenge for them to suggest a treatment that would be better than that suggested by experts, particularly where determining causality is required [63]. In addition to what the models do, the next issue will be whether a model applies to the individual being treated. Models developed in one group of patients may not transfer to another. Transparency of a model in terms of analytics and algorithms is important for patient safety and to earn the trust of the treating clinician.
The methodologies used in chest radiology, sleep medicine, sepsis and other fields, with examples, are demonstrated in tables 3 and 4.
Clinician and patient involvement in the development and delivery of novel digital medicine applications
While the above examples illustrate the near-term potential of digital medicine to make healthcare safer and more efficient, they focus on specific applications of particular technologies. However, the real impact of these innovations will be appreciated when all of the aspects of digital medicine are incorporated into a complete solution for a particular healthcare need. For example, a truly digital model of lung cancer screening might involve AI that identifies patients at risk of lung cancer from an EHR, with generative AI being used to send out personalised invitations to participate, with smart chatbots being used to schedule appointments and address patient concerns, while an AI-trained radiology system analyses the scan and arrange follow-up tests. How would such an intervention be received? Concerns regarding the robustness of AI algorithms in finding patients, scheduling and interpreting complex respiratory data is a legal risk. The erosion of clinical autonomy, as well as litigation fears, would interfere with the patient-centred approach of traditional doctor–patient consultations [64]. Clinical leadership in scenarios such as this will be a critical decider of how healthcare delivery is shaped by digital technologies.
The integration of the novel tools of digital health requires clinicians to see the need, clinical utility and potential downsides of these innovations. It is they who will need to adapt to new workflows and practices, but they cannot do so alone. Collaborating with different disciplines such as technology developers, engineers and information technology, with patient involvement, to design user-friendly devices will increase the likelihood of successful implementation. Regulation should focus on clinical effectiveness not simply that the system works.
Privacy and protection of patient data remain of paramount importance, from both the patient and clinician perspective. Patient data remains subject to cyber-attacks [51], with a subsequent risk of data leaks to insurance companies, and subsequent clear cost and confidentiality implications. Privacy violations and potential data breaches of sensitive information may create a barrier to universal uptake from a patient point of view. From a patient perspective, access and affordability are also important. Patients from underserved communities may encounter problems including a lack of devices, connectivity issues and digital literacy. Access to a readily available internet service remains a major barrier. For example, it is estimated that access to mobile internet in sub-Saharan Africa remains at 40%, while availability of uncensored information is also frequently lacking globally [65]. This has the potential to exacerbate existing disparities in healthcare. Sufficient support also needs to be widely available to those who experience technical problems or require assistance with device setup or troubleshooting. Digital devices, however convenient, are not a one size fits all and this must be considered if medical technology is to be rolled out universally across the world. Even in the acute hospital setting, concerns have been raised about medical devices used in diagnostics relying on racially based algorithms that have been linked to under-recognition and under-reporting of conditions in certain populations [66].
What next for patients, researchers and clinicians?
Telemedicine, digital therapeutics, as well as remote monitoring have already changed many aspects of everyday care for many patients with respiratory conditions. Near-term developments will involve developing trusted platforms for patient education and engagement. The need for trusted platforms is not just because of the opportunity but because misinformation is widely spread and shared online [65, 67]. From a clinician perspective, leadership in devising how these technologies are designed, tested and implemented into healthcare is pivotal.
Conclusions
Connected health in the form of telemedicine and remote digital monitoring are already established in many domains of respiratory medicine, including asthma, pulmonary rehabilitation and sleep medicine. Machine learning tools, including imaging recognition software and their use in respiratory radiology, will almost certainly become part of day-to-day practice in the coming years, once concerns around generalisability have been addressed. AI in many forms may improve administration, record keeping and other administrative tasks. It remains uncertain whether the promises of better models of personalisation and prediction will translate into clinically meaningful and cost-effective products for clinicians. As AI evolves and evidence emerges from real-world experience, clinicians will have an obligation to work with other healthcare providers and regulatory agencies to establish clinical guidelines, quality metrics and standards of care for the use of digital health in clinical settings. Collaborating with different disciplines such as technology developers, engineers and information technology along with patient involvement to design user-friendly devices will increase the likelihood of successful implementation. Healthcare providers and training bodies must provide staff and trainees with continuous training to enhance proficiency in utilising digital health technologies. Within the field of respiratory medicine, we will soon need to establish which areas are suited to virtual care models from health outcomes and the perspective of patient satisfaction. Thereafter, the patient–clinician relationship may shift dramatically and only time will tell how well this is tolerated on both sides. We cannot truly adapt these techniques until we are sure they are equitable for all and until ethical concerns regarding data protection, patient consent and socioeconomic, gender and race inequalities are overcome.
Footnotes
Provenance: Commissioned article, peer reviewed.
Previous articles in this series: No. 1: Greene CM, Abdulkadir M. Global respiratory health priorities at the beginning of the 21st century. Eur Respir Rev 2024; 33: 230205. No. 2: Bush A, Byrnes CA, Chan KC, et al. Social determinants of respiratory health from birth: still of concern in the 21st century? Eur Respir Rev 2024; 33: 230222. No. 3: Domingo KN, Gabaldon KL, Hussari MN, et al. Impact of climate change on paediatric respiratory health: pollutants and aeroallergens. Eur Respir Rev 2024; 33: 230249. No. 4: Vasse GF, Melgert BN. Microplastic and plastic pollution: impact on respiratory disease and health. Eur Respir Rev 2024; 33: 230226.
Number 5 in the Series “Environment and lung health in a rapidly changing world” Edited by Sara de Matteis, Catherine M. Greene, Zorana Jovanovic Andersen and Renata L. Riha
Conflict of interest: R.W. Costello reports grants from GSK; consulting fees from PMD solutions; lecture honoraria from Aerogen, BioNTech, GSK and TEVA; and travel support from AstraZeneca. R.W. Costello also reports involvement on the following patents: patent on acoustics to assess inhaler use; patent on assessing adherence; and a patent on detecting and predicting exacerbations. R.W. Costello acted as Education Council Chair for ERS. All other authors have nothing to disclose.
- Received December 5, 2023.
- Accepted July 31, 2024.
- Copyright ©The authors 2024
This version is distributed under the terms of the Creative Commons Attribution Non-Commercial Licence 4.0. For commercial reproduction rights and permissions contact permissions{at}ersnet.org