Introduction

Digital phenotyping is an innovative health monitoring method that utilizes smart devices, sensors, and mobile apps to continuously and real-time collect data on an individual’s behavior, psychological, and physiological states, thereby providing a comprehensive and dynamic portrayal of an individual’s health1,2,3,4. This concept was first introduced by researchers at Harvard University in 2015, aiming to meticulously capture behavioral characteristics in daily life through digital means, thus offering new avenues for research and management of mental health and physiological diseases3.

The breadth and potential of digital phenotyping lie in its ability to cover all aspects of everyday life. By collecting a variety of information such as smartphone usage patterns, social media activity, exercise data, heart rate, and sleep quality4, digital phenotyping can help healthcare professionals identify potential health risks and provide early warnings about changes in an individual’s health5. Additionally, this data can be used for personalized medicine, helping to develop more precise treatment plans.

With the widespread adoption of smart devices and mobile technology, digital phenotyping has gradually become a hot topic in medical research. It not only provides new tools for assessing mental health, such as the early detection of depression and anxiety, but also shows great potential in chronic disease management and rehabilitation monitoring. At the same time, the implementation of digital phenotyping has sparked discussions about data privacy and ethical issues. Balancing technological advancements with the protection of individual privacy has become a pressing challenge that needs to be addressed.

Classification of digital phenotyping

Digital phenotyping can be classified based on various dimensions such as data sources, collection methods, and analysis objectives (Fig. 1a). Below are the main classification approaches:

Fig. 1: Classification of digital phenotyping.
figure 1

a Represents the classification of digital phenotypes based on data sources. b Represents the classification of digital phenotypes based on data collection methods. c Represents the classification of digital phenotypes based on analysis goals. d Represents the classification of digital phenotypes based on application scenarios.

Full size image

Classification based on data sources

Behavioral phenotyping

This type of phenotyping uses monitoring and analysis of an individual’s daily behavior data to describe their health status. The data typically comes from smartphones, wearable devices, etc., and includes information such as step count, physical activity, sleep patterns, phone usage time, and social media activity (Fig. 1a)6,7. These data can reflect a person’s daily activity level, sleep and wake patterns, and social behaviors6,7.

Physiological phenotyping

This category involves monitoring an individual’s physiological parameters to assess their health. Sources of data include wearable medical devices, smartwatches, heart rate monitors, etc (Fig. 1a)8,9,10,11,12,13. Common physiological data include heart rate, blood pressure, blood glucose levels, body temperature, and respiratory rate. These data help evaluate cardiovascular health, metabolic status, and other physiological functions8,9,10,11,12,13.

Psychological phenotyping

This type monitors data related to an individual’s psychological state to assess their health, often involving emotions, stress levels, cognitive functions, etc. Data can be obtained through surveys, psychological assessment tools, and sentiment analysis (such as through social media or voice data) (Fig. 1a)14,15,16. Psychological phenotyping plays a crucial role in identifying and managing mental health issues like depression and anxiety14,15,16.

Environmental phenotyping

This type of phenotyping tracks an individual’s geographical location, travel patterns, and common destinations using mobile GPS. It monitors environmental factors such as air quality, noise levels, and temperature, typically using smartphone sensors or specialized devices (Fig. 1a)17,18.

Social phenotyping

This type analyzes call logs, text messages, and social media activities to assess social contacts, interaction frequency, and patterns (Fig. 1a). It evaluates the complexity of language, choice of vocabulary, and conversational patterns through text or voice analysis to understand an individual’s social and psychological state1,19.

Medical phenotyping

This category includes medication adherence medical phenotyping, and symptom tracking medical phenotyping. Medication adherence medical phenotyping monitors and records an individual’s timely medication intake using smart pillboxes and mobile reminders (Fig. 1a)20,21. Symptom tracking medical phenotyping employs mobile apps or online platforms to record and track changes in a patient’s symptoms, such as pain, fatigue, dizziness, etc20,21.

Classification based on methods of data collection

Active data collection

This type of data collection requires active participation or input from the user1,2,22,23,24. For example, users may log their diet and exercise routines through health apps or regularly complete psychological assessment questionnaires. This method generally yields more detailed and specific data but may be limited by the user’s level of participation and accuracy (Fig. 1b)1,2,23,24.

Passive data collection

This method involves data being automatically collected from a user’s devices or environment without any active input from the user22. For instance, a smartwatch might automatically monitor heart rate and step count, while a smartphone tracks usage time and geographical location. This approach allows for continuous, seamless data collection, reducing the burden on users, but it may lack specific contextual information (Fig. 1b)1,2.

Classification based on application scenarios

Public health phenotyping

This type of phenotyping is used in large-scale public health research to help analyze population health trends and formulate public health policies (Fig. 1c)25,26. By aggregating and analyzing the digital phenotyping data of a large number of individuals, researchers can identify health risk factors in specific regions or populations (Fig. 1c)25,26.

Personalized health phenotyping

This type of phenotyping is used for individual health management, providing personalized health advice and interventions (Fig. 1c)27,28,29,30. It customizes health plans based on an individual’s behavioral and physiological data, improving quality of life and health outcomes.

Through these classifications, digital phenotyping can provide targeted support for different health management needs, playing a key role in precision medicine, public health, and personalized healthcare.

Classification based on analysis objectives

Diagnostic phenotyping

This type of phenotyping is primarily used to support the diagnosis of diseases. By analyzing abnormal changes in specific behaviors or physiological patterns, digital phenotyping can help identify early signs of a disease (Fig. 1d)29,31,32. For example, significant changes in sleep patterns may be an indicator of depression31,32.

Predictive phenotyping

Predictive phenotyping aims to forecast future health events or risks of diseases (Fig. 1d)30,33,34. For instance, by monitoring physical activity and heart rate data over time, it is possible to predict an individual’s risk of developing cardiovascular diseases33,34.

Preventive phenotyping

This type of phenotyping is used to prevent the onset of diseases. By identifying potential health risk factors early, individuals can take preventative measures to halt the progression of diseases (Fig. 1d)28,35,36. For example, continuous monitoring of fluctuations in blood glucose levels can help prevent diabetes35,36.

Monitoring phenotyping

Used for ongoing monitoring of diagnosed health conditions or disease progression (Fig. 1d)37,38,39. For example, patients with diabetes can use digital phenotyping to continuously monitor their blood sugar levels, aiding in disease management37,38,39.

Clinical benefits of digital phenotyping

In the early 21st century, American biologist Professor Leroy Hood proposed the theory of P4 Medicine, emphasizing that in the era of precision medicine, the four principles of Predictive, Preventive, Personalized, and Participatory are core tenets40. Digital phenotyping is a concrete form of P4 Medicine, emerging as a result of advancements in digital sensor technology, wearable devices, and other digital technologies. It underscores the importance of digital monitoring, dynamic surveillance, disease heterogeneity, disease subphenotypes, and personalized treatment41.

Predictive

Predictive capability is a fundamental attribute of the P4 medicine framework, emphasizing that the comprehensive analysis of large-scale individual data (such as genomics, environment, behavior, lifestyle, and electronic health records) can achieve early disease risk prediction30,33,34. With the development of digital phenotyping, especially continuous and dynamic tracking of phenotypic data, it is not only possible to identify early signs and risks of diseases but also to provide effective health management strategies to slow down disease progression30,34,41. For example, Lakhtakia et al.42 used smartphone-based digital phenotyping as an important monitoring technology to explore the impact of smartphone usage patterns on the early identification of schizophrenia. The results indicated that smartphone-based digital phenotyping technology was associated with clinical outcome assessments42. Subsequently, Onnela et al.3, further confirmed that smartphone-based digital phenotyping, such as changes in smartphone usage patterns3,34,42, sleep quality39, and social behavior43, might signal the onset of mental health issues like depression or anxiety. By dynamically monitoring and detecting these abnormal behaviors early, it not only provides data support for clinical diagnoses but also offers valuable reference for early intervention in disease management (Fig. 2a).

Fig. 2: Clinical benefits of digital phenotyping.
figure 2

a represents the potential clinical benefits of digital phenotyping. b represents the potential clinical benefits of digital twins under the framework of large language models (LLMs).

Full size image

Preventive

Similar to predictive characteristics, prevention is also an important aspect of P4 medicine, but it particularly emphasizes the core role of dietary changes, lifestyle modifications, and other factors in health management and disease prevention28,35,36. Prevention is not only about early intervention for future disease occurrence, but also an active approach to health management, aiming to reduce disease risk by improving an individual’s lifestyle. Digital phenotyping, with its extensive and continuous data collection capabilities, undoubtedly provides a solid data foundation for disease prevention in medical and health management. For example, Kim et al.44 explored the impact of exercise on weight, HbA1c (Glycated Hemoglobin A1c), fasting blood glucose, and other indicators in diabetic patients using a randomized controlled trial44. The results indicated that digital phenotyping technologies based on smartphone personal health record apps could help improve diabetes-related indices44. Furthermore, dynamic monitoring of chronic diseases such as obesity45,46, diabetes44,47,48,49, and cardiovascular diseases50,51,52 has shown that digital phenotyping can improve disease-related indices through lifestyle interventions such as exercise and diet (Fig. 2a).

Personalized

Personalized treatment and care are key features of P4 medicine, emphasizing that patient treatment and health management in the era of precision medicine should comprehensively consider the genetic and personalized characteristics of patients or care recipients. With the widespread adoption and application of digital phenotyping, researchers studying patients with the same disease phenotype have discovered significant heterogeneity in subphenotypic characteristics, medication adherence, and drug dosage28,29,30. For example, Inomata et al. utilized smartphones and the DryEyeRhythm application as research tools to evaluate and analyze the potential of digital phenotyping in assessing the phenotypic diversity and patient heterogeneity of dry eye disease41. The results indicated that the smartphone-based DryEyeRhythm application demonstrated strong patient stratification capabilities and significant potential for application in personalized treatment41. Subsequently, Kim et al.‘s study further demonstrated that a personalized exercise recommendation program based on a mobile application had positive effects on blood glucose control and weight management in individuals with type 2 diabetes (T2D)44.The extensive use of digital phenotyping not only enables healthcare providers to tailor medication dosages based on each individual’s unique lifestyle and health data but also allows for the customization of health-supportive treatment recommendations and plans by collecting and analyzing personal behavioral patterns, emotional changes, exercise habits, and other data28,29,30 (Fig. 2a).

Participatory

A key advantage of digital phenotyping is its ability to increase patient engagement in their own health management20,21,43. Through mobile applications or other digital tools, patients can access real-time information about their health status and receive personalized health advice, enabling them to actively participate in health decisions1,2,22,23,24. This highly participatory model not only raises patient awareness but also promotes better treatment adherence, improving clinical outcomes (Fig. 2a).

In summary, in the era of P4 medicine, digital phenotyping not only enables dynamic, real-time, and comprehensive monitoring of individual health data to achieve early prediction of diseases or disease risks9,11,30,34,38,53 but also helps mitigate disease risks through preventive, personalized, and participatory approaches8,54. Moreover, digital phenotyping plays a vital role in dynamic monitoring and management in fields such as exercise-assisted therapy45,50,55,56,57,58 and music-assisted therapy59,60, significantly enhancing the positive impact of lifestyle changes and increased physical activity on health. Therefore, digital phenotyping not only has extensive clinical application prospects but also holds the potential to become a driving force for transformative change in healthcare (Fig. 2a)1,5,22,24,61,62,63. In these cases or implementation examples, biobanks, especially large-scale biobanks like the UK Biobank64,65 and All of Us66, have demonstrated the immense value of digital phenotyping in scientific research in the era of P4 medicine, as well as its role in showcasing the clinical benefits and exemplary impacts brought by lifestyle changes and increased physical activity (Box 1).

Challenges and barriers to widespread adoption

Although digital phenotyping has broad application prospects, its promotion still faces multiple obstacles, primarily in terms of technology, privacy, ethics, and data integration (Fig. 3a).

Fig. 3: Challenges & barriers and future directions of digital phenotyping.
figure 3

a represents the challenges and barriers of digital phenotyping. b represents the future directions of digital phenotyping.

Full size image

Data privacy and security

Digital phenotyping involves the collection and analysis of large amounts of sensitive personal data, such as physiological indicators, behavioral patterns, and psychological states. If this data is misused or leaked, it could pose serious threats to personal privacy and security (Fig. 3a)2,67,68. Ensuring privacy protection during the collection and use of such data, and establishing transparent privacy policies and data protection measures, is the primary challenge to the widespread adoption of digital phenotyping2,68,69.

Ethical issues

The promotion of digital phenotyping also faces ethical dilemmas, particularly concerning data ownership, usage rights, and informed consent (Fig. 3a)2,67,68. Do patients fully understand and consent to how their data will be used when collected? Will the data be used for commercial purposes? How can transparency and fairness in data usage be ensured? These are pressing ethical issues that need to be addressed. Additionally, digital phenotyping may lead to data discrimination or inequality, such as unequal access to health data or medical services across different populations2,68,70.

Technological and infrastructure limitations

Digital phenotyping relies on the widespread use of smart devices, sensors, and mobile applications4,5. However, in technologically underdeveloped or resource-limited areas, the necessary hardware and network infrastructure may be lacking. Furthermore, device accuracy, continuity of data collection, and ease of use can affect the reliability and adoption of digital phenotyping. If the accuracy of devices is insufficient or the user experience is poor, it may lead to biased data or user dissatisfaction, which could hinder its promotion (Fig. 3a)4.

Data integration and standardization issues

The data generated by digital phenotyping comes from a variety of sources, including smartphones, wearable devices, and health applications. This data often exists in different formats and lacks uniform standards, making integration and analysis more challenging4. Healthcare systems need to establish standardized frameworks to ensure seamless integration of data from different sources, providing reliable support for medical decision-making (Fig. 3a)4.

Clinical validation and regulatory issues

Although digital phenotyping shows great potential, its effectiveness and safety still require further clinical validation67,68. A lack of sufficient research evidence may hinder its clinical adoption. Additionally, the regulatory framework is not yet fully developed, and different countries and regions manage health data differently (Fig. 3a)67,68. Digital phenotyping applications must reach consensus on compliance and legality to avoid legal disputes.

User acceptance and health literacy

Not everyone is willing or able to easily adapt to using digital devices to monitor their health. Older adults or users unfamiliar with technology may face difficulties using these tools. Moreover, some users may feel uncomfortable with continuous monitoring of their health data, fearing privacy breaches or misinterpretation of the data. These psychological and behavioral barriers could affect the widespread adoption of digital phenotyping4,21,67.

In conclusion, while digital phenotyping holds great potential in the medical field, achieving widespread adoption will require overcoming multiple obstacles, including privacy and security concerns, technical standards, ethical dilemmas, and improving user acceptance and data processing capabilities.

Future directions

As an emerging tool for health monitoring and management, digital phenotyping has several key future development directions focused on technological advancements, data integration, application expansion, and ethical regulations (Fig. 3b). Below are some key areas of development:

Technological innovation and intelligence

In the future, digital phenotyping will benefit from more advanced technologies, including artificial intelligence (AI)94, machine learning (ML)2, 5 G communication, the Internet of Things (IoT), and edge computing4. These technologies can further improve the accuracy and real-time capabilities of data collection:

AI and ML

AI and ML algorithms are being widely applied to the analysis and interpretation of complex digital phenotypic data94. Through deep learning models, systems can more accurately identify health patterns, predict health risks, and even provide personalized health recommendations2,19,71. In recent years, with the advancement of ML methods, particularly deep learning and meta-learning, deep learning-based digital twin technology has been extensively applied to the analysis and modeling of digital phenotypic data72,73,74. This technology plays a pivotal role in personalized medicine, disease progression monitoring, disease tipping point identification, chronic disease management, multi-center data sharing, and reducing medical risks and costs (Fig. 4a)72,73,74.

Fig. 4: Clinical benefits of the digital phenotyping analysis framework based on digital twin models.
figure 4

a presents the framework for digital phenotyping analysis based on digital twin models and its clinical benefits. b showcases the unique advantages of digital twins under large language models (LLMs).

Full size image

For example, Alam et al. explored digital twin models based on dementia, Alzheimer’s disease, and amyotrophic lateral sclerosis (ALS), revealing that multiple neurodegenerative diseases, such as Alzheimer’s disease and ALS, undergo a gradual transition phase75. These diseases show distinct tipping point characteristics in motor digital biomarkers, speech digital biomarkers, salivation digital biomarkers, and swallowing digital biomarkers75.

In recent years, digital twin models based on LLMs have been widely used in disease digital modeling. LLM-based digital twins not only demonstrate significant advantages in integrating semi-structured and unstructured digital phenotypic data in medical contexts, but they also excel in efficiently integrating multimodal data (Fig. 4b)75,76. Furthermore, they show greater potential in areas such as self-supervised learning and few-shot learning, efficient simulation of complex systems, multi-turn dialog and multi-scenario interactions, as well as privacy protection (Fig. 4b)76.

Sensor technology

The accurate monitoring of digital phenotypes relies on diverse sensor technologies and devices77,78. Currently, health and environmental monitoring sensor technologies, such as blood oxygen sensors, electrocardiogram (ECG/EKG) sensors, electrodermal activity (EDA/GSR) sensors, accelerometers, and ambient light sensors, are widely applied in digital phenotype monitoring77,78. However, these sensor technologies generally face challenges such as limited miniaturization, insufficient sensitivity, and poor compatibility. Therefore, exploring sensor technologies with higher compatibility, greater sensitivity, and advanced miniaturization—such as non-invasive blood glucose monitoring sensors4,79,80 and respiratory component monitoring and analysis sensors81,82—will enable digital phenotypes to encompass a broader range of health indicators and demonstrate greater application potential and value.

Edge computing

As an efficient computing architecture in the era of intelligent medicine, edge computing offers a secure, real-time, and fast-response solution for the efficient processing of data in the digital phenotype era24,77,83. Exploring digital phenotypes based on edge computing will provide novel approaches and methodologies for computation within digital architectures.

Cross-platform data integration and interoperability

Digital phenotyping involves diverse data sources, requiring seamless integration and interoperability across different devices and platforms (Fig. 3b):

Data standardization

Establishing unified data formats and exchange standards will improve compatibility between different devices and applications4,22. This includes not only physiological data but also behavioral and psychological data for comprehensive health monitoring.

Multi-source data integration

Future developments will focus on integrating digital phenotyping data with traditional biomedical data (such as genomic data and clinical records)4,22. By merging data from multiple sources, researchers and clinicians can gain a more comprehensive view of health, leading to more accurate health assessments and decisions.

Personalized health management and precision medicine

As digital phenotyping technology matures, it will be more deeply applied in personalized health management and precision medicine (Fig. 3b):

Personalized health advice

Through in-depth analysis of individual behavioral, psychological, and physiological data, systems can provide more personalized health advice, including recommendations on diet, exercise, sleep, and more27,30,48. This personalized management will significantly improve health outcomes.

Precision medicine

Digital phenotyping data can help doctors better understand how a patient’s lifestyle and environmental factors affect disease, leading to more precise treatment plans. For example, continuous monitoring of mental health data can help tailor treatments for depression28,32.

Public health and epidemiological applications

In the field of public health, digital phenotyping can provide new tools for epidemiological research and public health policy formulation (Fig. 3b):

Disease surveillance and early warning

By collecting and analyzing large-scale digital phenotyping data, public health agencies can more rapidly identify and respond to disease outbreaks and even predict potential epidemics22,61,63.

Health trend analysis

Digital phenotyping data can be used to analyze population health trends, helping to create more effective public health interventions22,63. For instance, emotional data from social media can help monitor changes in societal mental health22,63.

Data privacy and ethical issues

As digital phenotyping becomes more widespread, data privacy and ethical issues will become important topics (Fig. 3b):

Data privacy protection

Future development requires stricter data privacy protection measures, including data encryption, anonymization, and distributed storage technologies. Users should have full control over their data and decide how and to what extent it is used69.

Establishment of ethical frameworks

As digital phenotyping applications expand, establishing transparent and responsible ethical frameworks is essential. This includes informed consent, data use transparency, fairness, and other areas to ensure that technology does not lead to social inequities or discrimination69.

Global collaboration and policy development

The development of digital phenotyping requires global collaboration and coordination (Fig. 3b):

International cooperation

Countries need to cooperate internationally on technical standards, data sharing, and privacy protection to promote the global application and research of digital phenotyping42,54,63,84.

Policy and regulatory frameworks

Governments will need to formulate relevant policies and regulatory frameworks to govern the collection, use, and sharing of digital phenotyping data, ensuring the healthy development of the technology42,54,63.

In summary, the future development of digital phenotyping encompasses multiple areas, including technological innovation, data integration, personalized applications, public health impact, ethical and privacy protection, and global cooperation (Fig. 3b). As these fields continue to progress, digital phenotyping is expected to become a critical tool in the healthcare sector, driving the realization of precision medicine and personalized health management.

Conclusion

Digital phenotyping has a profound impact on clinical medicine, offering new possibilities for personalized treatment plans based on continuous individual data streams, which in turn improve therapeutic outcomes and shift healthcare from reactive care to proactive, preventive care, thus reducing medical costs. Furthermore, digital phenotyping plays a transformative role in public health, enabling researchers to precisely monitor population health trends, identify risk factors, and provide strong support for public health policy-making.

However, to fully realize these benefits, ongoing research and development are crucial. Challenges such as data privacy, security, integration, and standardization must be addressed to ensure the safe and effective application of digital phenotyping in clinical settings. Additionally, the complexity of integrating digital phenotyping data with traditional biomedical data requires innovative approaches in data science and bioinformatics. International collaboration is essential for harmonizing ethical and regulatory standards, ensuring that this technology is applied equitably across different regions and populations.

In conclusion, while digital phenotyping has already made a significant impact on healthcare, continued research and development are necessary to unlock its full potential. By addressing current limitations and enhancing global collaboration, digital phenotyping is poised to become a cornerstone of modern healthcare, driving advancements in precision medicine, personalized healthcare, and public health.