Table 2.

Consensus statements on ethics of mental health applications of digital phenotyping.

Statementa			Agreement levelb
			Necessity		Feasibility
Evidence of validity for the intended use
	Algorithms incorporated into a digital phenotyping tool, especially at a large scale, have to be thoroughly evaluated in terms of performance and accuracy, including false positives and false negatives.	Strong		Strong
	Implement processes for review of digital phenotyping tools’ effectiveness after implementation, including review of updates, and monitoring and reporting of adverse events caused by an algorithm’s findings.	Strong		Moderate
	Digital phenotyping tools that are intended for use in health care should use relevant standards for data systems to support the goal of interoperability with existing health data systems.	Strong		Moderate
	Digital phenotyping tools for mental health applications should respond to real-world needs and concerns of the intended users, such as clinicians, patients or consumers, in order to enhance user engagement and provide value.	Strong		Strong
Transparency
	Explanations of the processes, risks, limitations, and results that are relevant to different stakeholders should be provided to them in an appropriate format and reading level.	Strong		Strong
	Processes involved in the collection, storage, and dissemination of raw data, as well as data processing and the architecture of the algorithms, should be explainable.	Strong		Moderate
Accountability
	Development and use of digital phenotyping tools (eg, plans for data collection or validation) should be reviewed for potential ethical issues by an independent interdisciplinary group with relevant expertise, starting early in the development process.	Strong		Moderate
	Provision of appropriate educational and training materials for IRBsc handling review of digital phenotyping projects is also necessary.	Strong		Moderate
Consent
	Consent should be required from individuals when their personal data are collected for digital phenotyping tools.	Strong		Moderate
	Consent for collection of digital phenotyping data should include information at a sixth-grade level regarding the types of data collected, the inferences that can be drawn from the data, the reports made from the data, who the data and reports would be shared with, the potential risks and benefits to the user, and the limitations that apply to the findings.	Strong		Moderate
	Include relevant stakeholders in efforts to formulate and disseminate relevant information for disclosure (eg, data storage, utilizing appropriate languages and formats for relevant stakeholders, such as health care providers, government institutions, advocacy organizations, patients, consumers, or the public).	Strong		Strong
Data security and privacy
	Data and findings that are identifying should not be collected, used or shared with third parties without the informed consent of that individual.	Strong		Strong
	Sharing of data to advance scientific research and the validity of the tools remains an important goal.	Strong		Moderate
	If data will be shared with third-party researchers, clear information, written at sixth-grade reading level, must be given to the individual user about third-party researcher and how they plan to store, use and/or share the data.	Strong		Strong
	The individual user also must have an option to opt out of sharing their data with third parties.	Strong		Moderate
	Raw data that is nonidentifying, and nonidentifying summary statistics, may be shared without consent.	Strong		Strong
	There should be periodic review to re-evaluate whether identifying information can be drawn from the raw data, particularly when combined with other available data.	Moderate		Moderate
	Raw data should always be encrypted when stored or transmitted; potential identifiers in data (eg, phone numbers and IP addresses) should be replaced with surrogates (eg, hashed or encrypted).	Moderate		Moderate
	Standards and approaches to minimize risk of reidentification of individuals, such as differential privacy measures, should be implemented.	Strong		Moderate
	The security standards for data storage, sharing, and use of the individual’s data, as well as the process for monitoring compliance with these standards, should be clearly defined and communicated to users of digital phenotyping tools.	Strong		Moderate
	Security reviews and audits of data practices should also be implemented.	Strong		Moderate
Fairness
	Encourage collaborative research and partnerships to develop ways to identify and minimize bias or discrimination in the development of digital phenotyping tools and to identify and minimize any potential bias that may occur because of how the tools may be used in different communities or local contexts.	Moderate		Moderate
	Conduct research into and implement methods to mitigate bias in different levels of algorithm development, including in the training data, in the algorithmic process or focus, in the transfer of digital phenotyping tools to different contexts, and in the interpretation of digital phenotyping findings.	Strong		Strong
	Identify the specific ways that mental health and clinical care may impact the potential for bias in these areas. Periodic review and re-evaluation of the methods for addressing and mitigating bias at the different levels of algorithmic development may be needed.	Strong		Strong

^aThe statements represent the ethical issues in digital phenotyping for mental health applications resulting from the interviews and the first survey.

^bThe agreement rating listed represents the level of consensus for statements that were determined through the second survey.

^cIRB: institutional review board.