Advances in the study and application of digital technology in the clinical practice of atopic dermatitis

The complexity and challenges of AD

Atopic dermatitis (AD) is a chronic relapsing, heterogeneous, inflammatory dermatosis with complex pathophysiological mechanisms involving skin barrier dysfunction (e.g. mutations in the filaggrin gene), abnormal Th2-type immune responses and microbiome dysregulation (e.g. the process of colonisation by Staphylococcus aureus, along with the interactions between genetic and environmental factors, has been identified as a contributing factor.^1,2 The clinical manifestations of AD are characterised by a range of features, including intense itching, eczema-like lesions, and dry and painful skin. These manifestations may be accompanied by various morphological changes, such as erythema, exudation and mossy lesions. ³ Furthermore, the heterogeneity of AD can be attributed to the diversity of clinical phenotypes and molecular endotypes, such as the delineation of early versus late-onset and exogenous versus endogenous AD. These differences highlight the urgent need for individualised and precise treatment. ³

The multidimensional impact of AD on quality of life

The impact of AD on patients’ quality of life (QoL) is multidimensional and far-reaching. At the physical level, intense pruritus can result in the onset of sleep disorders (e.g. the condition is characterised by a number of symptoms, including difficulty in falling asleep and nocturnal awakenings). In addition, patients frequently report skin pain and recurrent infections, which can have a significant impact on daily life.^4,5 On the psychosocial level, patients often suffer from the appearance damage caused by the skin lesions and the recurrence of the symptoms. This can lead to anxiety, depression (correlated with the score of the Patient Health Questionnaire-9) and even social avoidance. Limited employment and family tensions may also occur.^6,7 It has been demonstrated that children are more prone to reduced attention span and fluctuating academic performance and that family members (especially family caregivers) have significantly lower QoL. ⁸ From an economic perspective, patients with moderate-to-severe AD face significant financial strain, largely attributable to the frequency of medical visits, the high cost of medication, and the consequent loss of productivity. ⁹ At present, instruments such as the Dermatology Life Quality Index (DLQI), the Patient-Oriented Eczema Measure (POEM) and the SCORAD are extensively utilised for the quantification of these indicators. ¹⁰

Limitations of the traditional AD treatment model

The prevailing AD diagnosis and treatment model is currently facing several challenges. At the diagnostic level, due to the absence of specific biomarkers, clinicians rely on the Hanifin-Rajka criteria, which can lead to confusion with other eczematous diseases and a high misdiagnosis rate for atypical cases. ¹ With regard to the treatment of the condition, conventional therapeutic interventions primarily encompass the utilisation of topical glucocorticosteroids, calcium-modulated phosphatase inhibitors, and the systemic administration of immunosuppressants, amongst other modalities. However, the long-term utilisation of glucocorticosteroids or immunosuppressants may result in skin atrophy, an increased risk of infection, and a lack of responsiveness to available therapeutic interventions. ¹⁰ In addition to traditional treatments, biologics are also preferred drugs for the treatment of AD. Currently, biologics used to treat AD primarily target specific immune pathways, offering new options for moderate-to-severe patients who have not responded to traditional treatments. Dupilumab is the first biologic agent approved for the treatment of AD, demonstrating significant efficacy and good safety by inhibiting the Interleukin-4/Interleukin-13 (IL-4/IL-13) pathway.^11,12 In addition to dupilumab, novel biologics such as tralokinumab (anti-IL-13 monoclonal antibody) are also being used in clinical practice. ¹³ These drugs have shown outstanding efficacy in improving clinical symptoms and reducing relapses, particularly in paediatric patients. ¹⁴ However, biological agent therapy may be associated with adverse reactions such as local infections and pustular psoriasis, ¹⁵ and some patients may experience immune shift phenomena. ¹⁶ Although biological agents have made breakthrough progress in AD treatment, their long-term safety and efficacy differences among patients with different phenotypes still require further research. ¹⁷ In the context of disease monitoring, conventional evaluation instruments (e.g. EASI and SCORAD) depend on physician interpretation, which can hinder the capture of in-real-time alterations in patient condition and subjective symptom improvement (e.g. fluctuations in itch intensity).

Digital technology: A new path to breaking through bottlenecks

In recent years, advancements in digital technology have engendered new pathways to circumvent the aforementioned bottlenecks. In the realm of contemporary applications, remote monitoring and patient-reported outcomes (PROs) have emerged as pivotal instruments in the collection of real-time data on patients’ itch intensity and sleep quality. These data are captured through mobile applications and wearable devices, which, when integrated with tools such as the DLQI, have been shown to enhance patient participation in diagnostic and therapeutic processes.^18,19 The advent of artificial intelligence (AI)-assisted diagnostic systems has ushered in a new era of automation, with the capacity to expedite the scoring of EASI (Eczema Area and Severity Index), thereby mitigating subjectivity bias through the analysis of skin lesion images (e.g. as demonstrated in references^20,21 the utilisation of machine learning algorithms facilitates the identification of erythema and exudation). Digital phenotyping technologies have the capacity to integrate genomic and transcriptional features, thereby reducing subjective bias. EASI scoring is automated to reduce subjective bias.^20,21 Digital phenotyping technology can integrate genomic, transcriptomic, and clinical data, and identify intramolecular phenotypes through AI algorithms to guide precision-targeted therapy. ²²

The development of future trends may encompass the creation of predictive models for disease progression (e.g. an early warning system for the “itch-scratch cycle”) to facilitate early intervention, ²³ the utilisation of virtual reality (VR) to distract from itchiness and enhance psychological well-being, ²⁴ and the establishment of blockchain-powered real-world data (RWD) platforms to expedite clinical research, ²⁵ amongst other potential avenues. Notwithstanding the challenges posed by data privacy (e.g. General Data Protection Regulation (GDPR) compliance) and technology standardisation, Digital Therapeutics (DTx) is anticipated to function in conjunction with biologics to establish a closed-loop monitoring-intervention-feedback system. This will ultimately result in a paradigm shift in the treatment of AD and facilitate the rational improvement of patient prognoses. ²⁶

Skin ultrasound: Quantitative assessment of skin lesion characteristics

The diagnosis of AD is contingent on the observation and assessment of skin lesions by physicians. In 1980, Hanifin and Rajka published a list of 23 clinical signs and symptoms of AD which is still utilised as a benchmark for clinical research. ²⁷ Skin ultrasound has been demonstrated to provide a means of quantitatively analysing the characteristics of skin lesions in AD through high-resolution imaging. Research has demonstrated that epidermal thickness is considerably augmented in regions of skin lesions in patients with AD, and there is a positive correlation with inflammatory cell infiltration, such as mast cells and T cells. ³ A study demonstrated that the extent of epidermal thickness reduction after treatment with the combination of Olea Europaea Leaf Extract+Spirodela Polyrhiza Extract (OLE + SPE) was synchronised with a decrease in CD23/B220(+) B-cell infiltration in a 1-Chloro-2,4-Dinitrobenzene (DNCB)-induced animal model of AD. ²⁸ With regard to the phenomenon of dermal vasodilatation, skin ultrasound has been demonstrated to detect vasodilatation phenomena associated with the inflammatory response and the severity of pruritus. However, the precise mechanism by which this occurs remains to be fully elucidated within the context of haemodynamics. Furthermore, research has indicated that epidermal thickness difference (δET) may serve as a potential biomarker of disease severity, with a reduced δET (i.e. a diminished difference between lesional and non-lesional areas) indicating a more severe condition. ²⁹

AI: Improving diagnostic objectivity and efficiency

The core applications of AI in the diagnosis of AD encompass severity grading and differential diagnosis. In one study, a convolutional neural network (CNN)-based model analysed 9192 images of AD lesions and compared them with annotated Investigator's General Assessment (IGA) scores from five dermatologists, and the results were in high agreement with expert assessment data. ³⁰ CNN represent a type of Deep Learning. Deep learning is an advanced form of AI that facilitates the training of computers to analyse data in a manner analogous to the human brain. Deep learning has been demonstrated to exhibit numerous advantages over conventional machine learning algorithms, including KA-nearest neighbour, support vector algorithms and regression approaches. It has been demonstrated that deep learning models have the capacity to process data of a greater degree of complexity than that which can be handled by traditional machine learning methods. ³¹ The AI can integrate multimodal data (e.g. lesion morphology and distribution) and optimise it for classification purposes, rendering it particularly suitable for patients with AD who exhibit a high degree of heterogeneity. In distinguishing between other forms of dermatitis, AI has been shown to enhance diagnostic precision through the utilisation of feature difference analysis. In the context of contact dermatitis (CD), for instance, AI is capable of differentiating based on the clarity of lesion boundaries and the history of local exposure. As stated in the relevant literature, occupational allergens and typical lesion sites (e.g. eyelids and hands) are to be considered. ³² In the case of neurodermatitis (ND), differentiation is achieved through the identification of mossy lesions and itch-scratch cyclic behaviours (e.g. night-time scratching frequency monitoring). ³³

Digital tools: Multi-parameter integration and optimisation assessment

Existing digital tools have been shown to enhance objectivity in AD assessment through multi-parameter integration. In terms of the quantification of symptoms, wearable sensors primarily monitor nocturnal scratching behaviour through the use of acoustic signals, thereby reducing bias in subjective reporting. ¹⁹ Concurrently, a team has proposed methodologies for the objective, continuous assessment of nocturnal scratching and sleep, utilising accelerometer data captured by wearable devices. ³⁴ Deep learning image segmentation techniques automatically calculate the percentage of lesion area. In combination with digitised measurements of transcutaneous water loss (TEWL), these techniques enable the quantification of the extent of skin barrier damage. ³⁵ In the context of severity grading, conventional scoring systems (e.g. SCORAD and EASI) predominantly depend on subjective evaluation by physicians and are often laborious and time consuming. Conversely, digital tools have the capacity to enhance the assessment process by constructing composite indicators (e.g. δET, TEWL and vasodilatation index). Research has demonstrated a robust correlation between δET and the SCORAD score, thus suggesting its utilisation as an objective complementary index. Furthermore, the multi-parameter integration model has been shown to enhance the efficiency and consistency of severity grading. ²⁹

Limitations and future directions

Nevertheless, prevailing AI diagnostic systems continue to exhibit significant limitations in the cross-domain identification of atypical skin lesion samples. To enhance the generalisation ability of the model, future research should be devoted to further optimising feature extraction through a transfer learning framework or improving its clinical applicability by constructing a large-scale and diverse database of dermatopathological images.

Optimise local treatment plans

Conventional localised treatment protocols are predicated on intermittent clinical assessments and patient self-reporting, a method which is susceptible to overlooking subtle fluctuations in condition. Conversely, digital technologies have been demonstrated to enhance treatment response speed and accuracy through continuous data collection and automated analysis. ³⁶ It has been observed that remote dermatology platforms have the capacity to reduce the efficacy assessment cycle from weeks to days, thereby reducing the risk of exacerbation due to untimely assessment. ³⁷ The utilisation of AI-driven remote assessment tools has the potential to assist physicians in the selection of more suitable topical medications, such as glucocorticosteroids or calcium-modulated phosphatase inhibitors, by analysing skin lesion image features, including erythema area and the degree of erythroderma. This approach aims to mitigate treatment bias arising from empirical differences. ³⁶ A study demonstrated that CNN models can predict patient response to specific drugs based on the morphological characteristics of skin lesions. This capability enables optimisation of the initial treatment regimen. ³⁰ Furthermore, the integration of wearable sensors (e.g. devices for monitoring humidity and temperature) has the potential to provide real-time feedback on alterations in the skin microenvironment. This, in turn, could offer an objective basis for the selection of drug dosage forms (e.g. creams or ointments), thereby further enhancing the adaptability of topical treatments. ³⁸ Concurrently, digital technologies facilitate real-time adjustment of medication dosage through continuous monitoring of medication administration behaviour (e.g. electronic adherence recording devices) and dynamic changes in skin conditions. It has been hypothesised that customised digital interventions (e.g. smart medication reminders combined with patient feedback) have the potential to significantly improve treatment adherence, thereby reducing fluctuations in efficacy due to over- or under-dosing. ³⁹ Furthermore, AI models can predict drug efficacy based on individualised data (e.g. skin lesion evolution trends, biomarker levels), thus assisting clinicians in developing precise medication regimens. ⁴⁰

Digital phototherapy equipment: Precision and safety

Although traditional phototherapy (e.g. narrow-spectrum ultraviolet B (UVB) therapy) has been proven to be effective for AD, the assessment of its efficacy is more dependent on the subjective judgement of clinicians. The integration of digital technology facilitates the acquisition of skin barrier function parameters (e.g. transcutaneous water loss) and inflammatory markers in real time. These parameters can then be analysed in conjunction with real-time efficacy assessment algorithms, enabling dynamic adjustments to phototherapy dosing, frequency and duration. This, in turn, enhances the accuracy and safety of the treatment. ⁴¹ Despite the absence of direct elucidation within extant literature concerning the precise mechanism of action of digital phototherapy devices (e.g. UV therapy combined with digital technology), relevant studies have demonstrated their potential value in disease monitoring and efficacy assessment. In one study, an AI-based CNN model was found to enable objective assessment of AD severity by analysing skin lesion images, providing data support for optimising phototherapy regimens.^30,41 However, further research is required to fully elucidate the direct mechanism of action of digital phototherapy devices and to compare them with traditional phototherapy.

Telemedicine: Expanding access to specialised services

Telemedicine platforms (e.g. AI-based-assisted diagnosis systems) overcome geographical limitations and facilitate access to specialised services for patients with AD without the need for physical presence.^37,42 Clinicians can utilise real-time uploaded images of skin lesions and symptom data (e.g. itch scores, sleep quality) to swiftly evaluate the condition and make timely adjustments to the treatment plan, such as upgrading the intensity of phototherapy or switching to a biologic agent (e.g. dupilumab). ³⁶ For patients diagnosed with moderate-to-severe AD, the utilisation of remote monitoring platforms to assess the efficacy and adverse effects of biologics has been shown to markedly reduce the frequency of hospital visits, enhancing patient comfort and compliance. ⁴³ Interactive digital tools (e.g. AI-powered educational apps) have the potential to facilitate patient education regarding disease mechanisms and self-management strategies through the provision of personalised content (e.g. acute care guidelines, allergen avoidance strategies). ⁴⁴ Furthermore, VR technology has been shown to simulate daily care scenarios, thereby enhancing patients’ proficiency in essential tasks such as moisturising and drug application. ⁴⁵

Comprehensive improvement of treatment compliance and optimisation of resource allocation

The utilisation of remote follow-up systems, underpinned by wearable devices and electronic logs, has been demonstrated to facilitate the synchronised tracking of patient data, including symptoms, medication records and quality of life. This approach has been shown to enhance treatment adherence by more than 30%. ⁴³ A plethora of studies have confirmed that regular feedback from digital platforms (e.g. early warnings of worsening symptoms, and medication reminders) can effectively reduce treatment interruptions due to forgetfulness or misunderstanding.^39,43 Furthermore, telemedicine has been demonstrated to engender a substantial enhancement in overall satisfaction by reducing time-related and financial expenditures for patients. ⁴⁶ Telemedicine can triage patients with mild-to-moderate AD to primary care facilities based on real-time monitoring data. This, in turn, enables specialised resources to be focused on complex cases. For instance, an AI-mediated triage system can automatically assess treatment priorities based on the severity of the patient's condition and optimise the efficiency of resource allocation.^38,42 This model has been shown to reduce pressure on specialist care services, while also promoting the implementation of a hierarchical diagnosis and treatment system.

Smart skin care device: Multidimensional monitoring and intervention

Smart skincare devices (e.g. smart moisturisers, smart dressings) have the potential to enhance the efficacy of care for patients with AD through multi-dimensional monitoring and intelligent intervention. Real-time monitoring technology based on smart dressings (e.g. electronic skin patches) can dynamically track skin temperature, humidity, pH and exudate composition (e.g. inflammatory factor levels), thus providing an objective basis for assessing the skin barrier repair process. ⁴⁷ It has been indicated that 3D bioprinted smart dressings using hyaluronic acid not only allow for personalised drug release but also enable optimised repair protocols through sensor feedback data. ⁴⁸ Furthermore, portable biosensors (e.g. sweat sensing systems) have the potential to facilitate continuous monitoring of the recovery process of skin barrier function in patients in AD by quantifying the efficacy of moisturising products. ⁴⁹ The integration of smart moisturising devices with humidity sensing technology facilitates dynamic adjustment of the intensity of local skin moisturisation, thereby reducing dryness-induced itching and inflammation, and thus optimising symptom management. ⁵⁰ The utilisation of temperature-sensitive smart materials (e.g. bifunctional films) has been demonstrated to be an effective measure in the regulation of the skin microenvironment, thereby alleviating symptoms such as burning and itching in patients with AD. ⁵¹ The data captured by these devices can also provide clinicians with objective indicators such as the degree of inflammation and skin integrity. This, in turn, can assist clinicians in developing individualised and precise treatment plans, such as adjusting drug dosages or the frequency of dressing changes based on real-time monitoring results. ⁴⁸

Digital nursing management platform: Precision and efficiency

The digital care management platform promotes AD management towards precision by integrating data from multiple sources. The platform has the capacity to combine patient history, genetic data, environmental exposure information (e.g. temperature, humidity) and real-time monitoring data to identify AD intra-molecular phenotypes (e.g. Th2 inflammation-dominant or barrier-deficient) and subsequently generate personalised care plans.^52,53 The Health Circuit platform, for instance, has been developed to incorporate adaptive case management technology that dynamically adjusts interventions (e.g. type of moisturiser, frequency of phototherapy) in conjunction with individualised data to achieve continuous optimisation of care programmes. ⁵⁴ In terms of the enhancement of the quality of care, digital platforms have been demonstrated to reduce subjective bias through standardised AD severity assessment (e.g. SCORAD score). ⁴¹ A prospective study in Taiwan has designed a multimedia mixed reality(MR) game. ⁵⁵ The present study is the inaugural mixed reality interactive shared decision-making (MR-SDM) game to target children with AD, and it combines clinical decision-making processes with gamification. This combination renders the game more acceptable to children and less costly than VR. The game has been found to effectively promote the participation of paediatric patients in SDM, enhancing their understanding and acceptance of treatment. Conversely, automated data analytics and remote collaboration features have been shown to enhance healthcare responsiveness and reduce the reliance on emergency care or hospitalisation (studies have demonstrated that digital interventions result in a decline in AD hospitalisation rates). ⁵⁶ Furthermore, the platform integrates electronic health records (EHRs) with PROs to support long-term efficacy tracking (e.g. itch frequency, sleep quality) and uses machine learning to predict recurrence risk, providing a basis for dynamically adjusting management strategies.^42,57

Patient-side digital tools: Empowering self-management

The utilisation of patient-side digital tools has been demonstrated to enhance AD control outcomes by facilitating self-management. Mobile applications (e.g. AD Control Tool) assist patients in documenting itch intensity, lesion extent and medication usage, in conjunction with visual trend graphs to facilitate the identification of triggers such as stress and environmental allergens. ⁵⁸ It has been demonstrated that certain applications incorporate image recognition technology intending to automatically analyse the severity of skin lesions. This is achieved by the application capturing an image of the lesion, thereby facilitating the monitoring process. ⁴¹ The health education and behavioural intervention module has been shown to improve patient self-care behaviours through the AD knowledge base (e.g. moisturising tips, medication instructions) and interactive courses (e.g. video teaching). ⁴⁵ Clinical studies have demonstrated that the integration of a smartphone application with educational training programmes results in a substantial enhancement in adherence, with a maximum improvement of 30% observed in AD severity scores. ⁴⁵ The social features of digital management platforms (e.g. online forums) and smart assistants (e.g. Chatbots) further extend the disease support network. The former facilitates experience sharing and psychological support, thereby alleviating disease-related anxiety. ⁵⁸ The latter reinforces adherence to long-term management through medication reminders and real-time Q&A. ⁵⁹

Big data analysis reveals heterogeneity in pathological mechanisms

Big data analysis has systematically revealed the heterogeneous pathological mechanisms of AD by integrating multidimensional data, such as clinical phenotype, gene expression, immune microenvironment and skin microbiome. It has been demonstrated that studies employing cluster analysis have successfully identified various clinical phenotypes (e.g. early/late hairstyles, with/without allergic diseases) and molecular typing (e.g. Th2/Th17/Th22 dominant phenotypes) of AD, thereby providing a scientific basis for the development of precise treatment strategies.^53,60 Furthermore, CNN technology has been employed to automatically grade over 9000 AD skin images, in conjunction with clinician-assessed IGAs and EASIs. This facilitates objective quantification of disease severity and significantly enhances the efficiency and accuracy of clinical decision making.^30,36 Concurrently, epidemiological studies utilising Internet-crowdsourced data in multiple European countries have further elucidated the unmet treatment needs of patients with AD and their socioeconomic burden, thereby substantiating data support for the optimisation of public health policy. ⁶¹

Computer-aided drug design accelerates targeted therapy research and development

The advent of computer-aided drug design technology has had a profound impact on the efficiency of AD-targeted drug development, through the implementation of a multi-disciplinary, cross-cutting strategy. In the target identification and validation stage, some researchers combined network pharmacology and molecular dynamics simulation techniques to screen the core targets of key AD signalling pathways (e.g. IL-4/IL-13, Janus Kinase-Signal Transducer and Activator of Tran Scription (JAK-STAT), Phosphodiesterase 4 (PDE4)) and predicted the binding patterns of small molecules and targets through deep learning models. This approach successfully shortened the cycle of traditional drug development.^62–64 In the drug optimisation stage, molecular docking and free energy calculation techniques were utilised to enhance the affinity and selectivity of drug candidates. In one study, Computer-Aided Drug Design (CADD) was indicated to ultimately achieve clinical efficacy by validating the binding conformation of drugs targeting IL-4Rα (e.g. dupilumab) to the receptor. ⁶⁴ Furthermore, computer-assisted synthetic planning (CASP) technology has been demonstrated to generate candidate molecules that are amenable to synthesis in conjunction with constrained laboratory resource conditions, thereby efficaciously reducing the cost of drug development. ⁶⁵ New digital biomarkers (e.g. quantitative monitoring of night-time scratching behaviour) provide objective evidence for AD efficacy analysis. Current AD treatment faces challenges such as clinical phenotype heterogeneity and treatment response variability. ⁶⁶ Digital biomarkers have the potential to compensate for the subjective limitations of traditional visual scoring by continuously collecting behavioural data.^41,67 This would assist in achieving precision medicine. To illustrate this point, machine learning models have been employed in the context of AD diagnosis and treatment efficacy assessment, while the establishment of a digital health framework has yielded novel tools for disease monitoring. ⁶⁸ Nevertheless, such technologies must still meet scientific standards, digital ethics and ethical issues related to their application in paediatric patients. ⁶⁹ The clinical translation of these findings necessitates multi-dimensional validation in conjunction with molecular characteristics, including immunological biomarkers. In the future, data analysis models integrating digital biomarkers and molecular typing may provide a more comprehensive evidence chain for the regulatory approval of novel therapies.^70,52

Promoting innovation in personalised medical devices and smart drug delivery systems

Digital technologies are also driving innovation in personalised therapeutic devices and smart drug delivery systems. In the domain of device development, the integration of computer-aided design (CAD) and three-dimensional printing has emerged as a pivotal approach in the fabrication of personalised transdermal patches or microneedles. These patches have been engineered for the targeted delivery of glucocorticosteroids or JAK inhibitors, thereby markedly diminishing the occurrence of adverse systemic effects.^35,71 In the field of drug delivery, it has been demonstrated that peptide-based intradermal delivery technology (IDDT) effectively enhances drug retention time in inflamed skin by optimising the particle size and surface charge of drug-carrying nanoparticles.^72,73 Furthermore, the integration of AI-driven portable optical sensors with machine learning algorithms has the potential to facilitate the real-time monitoring of changes in skin barrier function (e.g. the rate of trans-epidermal water loss) and inflammatory markers (e.g. IL-31 levels). This real-time data can then inform the dynamic adjustment of treatment regimens, as evidenced by studies.^41,74 The combined application of these technologies has been demonstrated to enhance the accuracy of AD treatment and to facilitate new avenues for patient self-management.