Informed consent in the face of digital healthcare in dermatology

INTRODUCTION

Informed consent is no longer just a piece of paper that provides documentary evidence to serve the purpose of clinical protocol and standard operating procedures. With machine learning and deep learning-based digital systems being increasingly used, there is an increasing need to modify existing descriptions of informed consent. Visual diagnostics are at the core of specialties such as dermatology and other subjects that are inherently clinical imaging/lesion-driven (radiology, pathology, and ophthalmology). This translates to real-time and inadvertent sharing of images across different platforms, pushing patient consent and privacy to the backseat.^[1] Despite efforts to incorporate artificial intelligence (AI) into medical care, moving toward trustworthy digital systems remains gray. These technologies can introduce additional layers of complexity to the consent process as AI’s involvement is dynamic and not straightforward.

The informed consent process lacks uniformity, standardization, and access.^[2] Legal systems remain lax on the requirement of informed consent in deidentified data.^[3]

TRADITIONAL INFORMED CONSENT VERSUS AI IN THE LOOP

Conventional informed consent has always been about facilitating communication between the provider/physician party and the patient party. Table 1 highlights the components/checklists of informed consent. A mutual understanding and agreement lay the foundation of the informed consent process. This allows due process of initiating the required examination, procedure, drug administration, photography, and decision-making in critical case scenarios. With the receiver party sufficiently comprehending the conveyed information, in the absence of coercive force, a healthcare professional gets the required authorization to act.

The said exercise is still far from ideal.^[4] In the face of AI and healthcare information, transparency is a key to fostering trust among all stakeholders, particularly when data storage and use take center stage.^[5] Data regulations lack clarity, are subject to an open access collection, and are under the influence of algorithms that drive and orchestrate clinical tools, treatment regimens, or “second opinion” care.

HOW AI COMES INTO PLAY IN INFORMED CONSENT?

• AI-assisted patient-friendly language models can generate simplified, patient/case-specific explanations and visualize risk profiles.

• Preliminary counseling, frequently asked questions – their answers, chatbot-based systems with emergency helplines of the dermatologist/researcher can be utilized with the consent-taking process.

• Advancements in consent by audio/video recording or photography can be done with AI assistance, which has higher legal value.

• Data sharing or future use of samples/information for research and compliance with regulatory bodies.

• Ease of assessment and periodic review by the ethics committee.

• Partial/segmental and continued consent that may be required for procedures to be done in different sessions can be obtained with the help of AI technology.

• Special consideration cases (e.g., minors, vulnerable populations, illiterate participants, and physically challenged) – AI can assist in drafting and obtaining consent based on the individual needs and circumstances.

BIG DATA AND HEALTHCARE

When it comes to Big Data (referring to large and ever-expanding data sets) use, we argue that informed consent is not taken as seriously yet.^[6] Physicians, let alone patients, cannot perceive the extent of risks at the time of consenting. Although expected threats are real, AI is too new, at the current juncture, to foresee or experience tangible digital threats.

Addressing how this asymmetry is handled, along with the callous dealings by big technology companies, is a matter of concern.^[7]

Allaying patient’s discomfort at the prospect of their health/disease information being synced with AI tools will require careful monitoring and scrutiny of tools in practice, with healthcare providers and set-ups bearing responsibility for potential mishaps.

DIGITAL LITERACY AND CONSENT

Use of AI in medical settings is more high-stakes than acknowledged, and a smooth user interface that can be understood and navigated by an average user may be counted as digital literacy in informed consent. There is already a pre-existing gap in information comprehension owing to the lengthiness of the document, the use of technical jargon, and an overall mental state of disease vulnerability. Adding enormously detailed terms and conditions forms for patients to proceed with a “click” for consenting defeats the purpose of patient comprehension. The components of AI-assisted informed consent are shown in Figure 1.

Close

Figure 1:

Components of artificial intelligence-based informed consent.

Export to PPT

Simplifying word content by condensing vital information and intents, by a tech aggregator/stakeholder that plays a hand in data control, is essential.^[8] Stunkel et al. highlight the need to shorten consent forms for better understanding.^[9] Using the ability of large language models to carry out the task of simplifying and paraphrasing can truly facilitate and bring real meaning to the informed consent process.^[10]

ALGORITHMIC PROCESSES

The doctor–patient relationship is undergoing a paradigm shift with the integration of AI in many subsets, making it liable to come under the litigious scanner. Regulatory frameworks are scant and evolving, lacking standardization.^[11] The liability on the AI system developer is unclear, posing increasing risks to the user end.^[12]

BLACK BOX IN HEALTHCARE AI SYSTEMS

Opacity around software and AI tools with regard to their intricate and complete networks is often called a black box in AI. When developers do not disclose or adequately explain how algorithmic systems generate results, it maintains the black box phenomenon, particularly concerning evaluations and approvals by regulatory authorities such as the Food and Drug Administration (FDA).^[13] Uncertainty ups the risk of apprehension, at the cost of needed deployment, and can further jeopardize patient outcomes. Many questions remain in the face of non-interpretability.

While prima facie consenting to something may be simple enough such as patient (verbally) consenting to use of electronic health record for administrative workflows, it could, at some point in the future lead to third party access, that collects data to construct patterns and predictions of individuals, to generate and influence demographic data, use statistics to analyze and display targeted advertisements or content. Data can rear many heads.

Based on the user requirements, AI developers should implement interfaces to enable end users to use the AI model effectively, annotate the input data in a standardized manner, and verify the AI inputs and results. Given the high-stakes nature of medical AI, human oversight is essential, and its need is increasingly felt. Human-in-the-loop mechanisms should be designed and implemented to perform specific quality checks (e.g., to flag biases, hallucinations, data misuse, malware errors). The risk of cybercrime, data theft, and misuse of data in the advent of inadvertent leaks or malware needs vital care.

OPEN CONVERSATIONS AROUND AI: TIME TO REFRESH

Encouraging bilateral communication around the use of digital systems, emphasizing human oversight, as opposed to the widely imagined fully “AI-doctor” or “AI-hospital,” will enable validation around this evolving dynamic and preserve a certain level of transparency and fairness. Timely and relevant adoption of services is needed. AI-assisted informed consent for surgery was found to have a positive effect on early comprehension, without any negative effect on satisfaction or anxiety.^[14] In pediatric care, it has been purported to have superior outcomes.^[15]

Thus, consent forms in dermatology may incorporate the following aspects, keeping in mind the evolving needs of digital healthcare [Table 2].^[16-18]

CONCLUSION

With the increasing use of AI in dermatology, there is an inevitable confluence and merger of many facets of tech in healthcare. It is vital that dermatologists learn to adapt to the hybrid landscape of traditional and AI-based interfaces. This shift requires transparent communication that emphasizes both the benefits and limitations of AI. In dermatology, where image-based datasets form the cornerstone of AI-based use, maintaining the patient’s trust while encouraging open communication for mutual help will be crucial to the ethical application of AI. Applying the basic principles of beneficence, non-malevolence, and a patient-centric approach would guide better decision-making in navigating the choppy waters of AI in medicine.