Unfair Discrimination vs Risk-Based Pricing: Where Digital Vitals Draw the Line

The expansion of digital health technologies in insurance underwriting has introduced a powerful new tension between two foundational concepts: permissible risk-based pricing and prohibited unfair discrimination. For chief medical officers and compliance leaders, the adoption of tools that measure digital vitals is not merely a technological decision but a complex regulatory challenge. These new data streams promise more accurate and efficient risk assessment, but they also create novel pathways for biases that can lead to unfairly discriminatory outcomes. Navigating this environment requires a clear understanding of where the regulatory line is drawn and how to build a compliance framework that can withstand scrutiny.

"at least 24 states have adopted or are in the process of implementing regulations based on the National Association of Insurance Commissioners (NAIC) Model Bulletin on the Use of Artificial Intelligence Systems, signaling a clear regulatory focus on fairness and accountability in algorithmic underwriting."

Drawing the line with digital vitals

The core business of insurance relies on risk-based pricing, the practice of charging premiums that correlate to the level of risk a policyholder represents. This is Permitted. Expected. However, all state insurance laws prohibit pricing models that result in "unfair discrimination". While historically this meant avoiding explicit use of protected characteristics like race or religion, the advent of artificial intelligence and machine learning has made the issue more complex. The central question for carriers today is whether their sophisticated new tools are creating proxies for these protected characteristics, leading to disparate impacts on certain populations.

This is the precise challenge of unfair discrimination risk-based pricing digital vitals. A digital vital sign measurement, such as heart rate derived from a video feed, appears objective. Yet, if the underlying algorithm is less accurate for individuals with darker skin tones due to technical factors like light absorption, a known issue in remote photoplethysmography (rPPG) research, its use in underwriting could lead to systemically biased outcomes. An inaccurate reading could place an individual in a higher-risk category not because of their actual health, but because of a technological limitation that correlates with a protected class. This creates a direct link between a seemingly neutral data point and a discriminatory result.

Principle	Permissible Risk-Based Pricing (Actuarially Sound)	Potentially Unfair Discrimination (Regulatory Scrutiny)
Data Input	Using direct, health-related measurements (e.g., blood pressure) that are validated for accuracy across all demographic groups.	Using raw data from a technology known to have performance variance across skin tones, genders, or age groups without correction.
Methodology	Applying actuarial analysis to data that has a demonstrated, causal link to mortality and morbidity risk.	Employing algorithmic models that generate correlations without a clear, explainable link to health outcomes, which may serve as proxies for protected classes.
Outcome	Individuals with similar risk profiles pay similar premiums, based on factors they can potentially control (e.g., lifestyle changes).	Individuals with similar risk profiles pay different premiums due to technological bias or demographic correlations embedded in the model.
Governance	Maintaining a robust model risk management program, including fairness testing, as outlined by the NAIC AI Model Bulletin.	Failing to document model testing, validate for bias, or establish human oversight for algorithmic outputs.

Industry applications and responsibilities

The burden of ensuring fairness falls across several departments within an insurance organization. The complexity of digital vital technologies requires a coordinated, cross-functional approach to compliance.

For underwriting departments

Underwriting teams must move beyond simply accepting the outputs of new technologies. They must be equipped to question the inputs and methodologies.

• Critically evaluate vendor claims about model accuracy.
• Demand transparency into the demographic makeup of the datasets used to train and test the underlying algorithms.
• Implement "human-in-the-loop" systems that flag anomalous or low-confidence readings for manual review.

For compliance and legal teams

Compliance officers are on the front lines of interpreting and implementing new regulatory guidance.

• Develop a comprehensive governance program for all AI and data-driven systems, consistent with the NAIC AI Model Bulletin.
• Conduct regular audits and fairness testing to proactively identify and mitigate potential biases.
• Document every step of the model validation process to create an evidence trail for regulators.

For reinsurance partners

Reinsurers are increasingly scrutinizing the technologies used by their carrier partners. They have a vested interest in ensuring that the underlying risk pools are priced accurately and fairly.

• Reinsurance treaties are evolving to include specific language about the use of AI and third-party data.
• Carriers should be prepared to provide evidence of their compliance and model risk management programs to their reinsurance partners.

Current research and evidence

The potential for bias in digital vital sign measurement is not theoretical. Scientific research into rPPG technologies has highlighted specific vulnerabilities. A 2022 narrative review published in a leading medical engineering journal on the representation biases in rPPG datasets found that most public datasets were heavily skewed toward lighter-skinned subjects.

Another key study, "Evaluation of biases in remote photoplethysmography methods," published through the National Institutes of Health (NIH) network, confirmed that algorithmic performance can vary across demographic groups. Researchers have consistently found that higher melanin content in darker skin can absorb more light, reducing the signal-to-noise ratio that rPPG relies on and potentially decreasing the accuracy of vital sign estimation. Without explicit mitigation strategies and demographically diverse testing, these technical biases can translate directly into underwriting inequities.

The future of digital vitals regulation

Regulators are moving from high-level principles to more prescriptive requirements. The NAIC's model bulletin is just the beginning. Future regulatory frameworks are expected to demand that carriers Attest to the fairness of their models. Provide detailed documentation of their testing and validation processes. The concept of "algorithmic accountability" will become a cornerstone of insurance compliance. Carriers that adopt digital vital technologies will need to demonstrate a proactive and sophisticated approach to model risk management, including ongoing monitoring for performance drift and emergent biases.

Frequently asked questions

What is the core difference between unfair discrimination and risk-based pricing? Risk-based pricing is the legally accepted practice of setting insurance premiums based on the quantifiable risk a policyholder presents. Unfair discrimination occurs when insurers charge different premiums to individuals with similar risk profiles or use legally protected characteristics, such as race or religion, as a basis for pricing, even if they are statistically correlated with risk.

How can digital vitals lead to proxy discrimination? Proxy discrimination, also known as disparate impact, can occur if a seemingly neutral data point is highly correlated with a protected characteristic. For example, if a digital vital sign algorithm is less accurate for individuals with darker skin, using its output for underwriting could systematically disadvantage a racial group, even if race is never explicitly considered. This makes the technology a "proxy" for a prohibited factor.

What does the NAIC Model Bulletin on AI require from insurers? The model bulletin requires insurers to establish a formal, written AI governance program. This program must include risk management controls, regular testing and validation of AI models to detect and mitigate bias, and clear lines of accountability for the outcomes produced by algorithmic systems. It emphasizes that insurers remain responsible for complying with all anti-discrimination laws, regardless of the technology they use.

How can carriers test digital vital technologies for bias? Carriers should conduct rigorous, targeted testing using datasets that are representative of their actual applicant pool. This involves comparing the technology's performance across different demographic subgroups, including race, age, and gender. The goal is to identify any statistical disparities in accuracy or outcomes and mitigate them before the technology is deployed in a live underwriting environment.

The challenge of balancing unfair discrimination risk-based pricing digital vitals requires a new generation of compliance infrastructure. As regulators intensify their focus on algorithmic fairness, carriers need tools built for this new reality. Circadify is focused on providing the compliance and data governance solutions that help carriers and reinsurers navigate these complex requirements confidently. To learn more about building a compliance-first approach to digital underwriting, explore our regulatory insights at circadify.com/industries/payers-insurance.