Trusted Data: The Foundation for AI in Healthcare

AI adoption has accelerated across the healthcare industry in recent years. Whether it is care coordination, patient and member engagement, clinical decision making, or utilization management, an increasing number of providers and payers are making substantial investments in AI to achieve better outcomes and efficiency.

However, there is a critical question these organizations must ask themselves: Can the data fueling these technologies be trusted?

The Risks of Bad Data

When an AI initiative fails, it is typically not because of the underlying technology, but rather due to data issues. If the data powering AI is incomplete, low-quality, lacking in context, then the quality of the output will suffer. The old saying ‘garbage in, garbage out’ is very relevant here.

In healthcare, privacy and regulatory compliance is another major challenge. Without patient and member consent being built into the data, in addition to delivering poor outcomes, AI can amplify the risk of privacy breaches and non-compliance in light of constantly evolving privacy regulations such as HIPAA, GDPR, CCPA, etc.

Unlike less regulated industries, bad data feeding AI initiatives in healthcare could mean serious financial liability and reputational damage for organizations.

Examples of Risks in Practice

• Incomplete Data: A hospital implements an AI model to predict which patients are likely to miss appointments. The model is trained purely on clinical encounter data and doesn’t take into account social determinants of health.

This could lead to an incorrect calculation of appointment adherence risk for certain demographics, and as a result, missed opportunities for intervention.

• Lack of Context: A hospital is currently using an AI model to support physicians in clinical decision making.

If the data powering the AI model lacks context such as patients’ life circumstances, the insights gleaned from clinical data alone might be impractical or even harmful.

• Missing Consent: A cancer treatment center launches an AI-powered chatbot to interact with patients and help them get to helpful resources faster. However, patients were never asked for consent whether they wanted to be engaged in this way.

One patient, seeing the chatbot branded with an unfamiliar vendor name, finds it intrusive and wonders whether their health information is being shared with vendors inappropriately. Rather than helping patients, the chatbot ends up causing trust issues.

This example illustrates how data that is incomplete, lacking context, or doesn’t have consent factored in can lead to poor health outcomes, increased privacy risks, and damaged patient trust.

Why Trusted Data Must Be the Foundation for AI in Healthcare

Trusted data is data that has the following characteristics:

• Unified across touch points: Patient and member data must be unified across all touch points (portals, websites, call centers, EHRs, etc.) into a single view of the patient. Without this unification, AI models can only work off of fragments of the patient’s or member’s journey, leading to incomplete insights.
  
• High-quality and contextual: Patient and member data powering AI must be accurate, complete, and reliable. While unification helps eliminate data gaps and silos, the data powering AI models must be relevant and rich in context. To generate actionable insights, AI models need the full story behind the data.
  
• Consented: Patient and member consent and permissions must be granular and explicit. Consent must be integrated directly into the data so that there is no ambiguity around whether the data can be used in a particular way. This ensures compliance with HIPAA, GDPR, CCPA, and other regulations while preserving patient and member trust.
  
• Real-time: AI models perform best when they have access to up-to-date patient and member data. For example, a chatbot recommending resources to a patient can provide truly helpful guidance only if it sees the patient’s most recent interactions and data, rather than relying solely on historical data.

These four pillars form the foundation for successful AI initiatives in healthcare to deliver meaningful outcomes while preserving regulatory compliance and patient and member trust.

The Way Forward

Healthcare organizations looking to create a trusted data framework for AI can benefit from the following actionable steps:

• Integrated Consent Management: Implement a consent management tool that allows the capture of granular patient and member consent, preferences, and opt-outs. Captured consent must be explicit and tightly integrated with your data collection infrastructure.
  
• Secure Real-time Data Collection and Unification: Collect consented patient and member data securely across touchpoints such as portals, CRMs, Data Warehouses, EHRs, call centers, etc. Unified, up-to-date patient and member profiles enable AI models to have a full picture of the journey, leading to more accurate insights.
  
• Data Quality Standards: Establish and enforce data quality standards to ensure that the data you are sending downstream is accurate and complete. Technologies like CDPs can help automate enforcement, but defining the standards is a governance responsibility.
  
• Data Anonymization and Obfuscation: Maintain the ability to anonymize and obfuscate Protected Health Information (PHI). Removing identifiers while retaining full-fidelity data ensures flexibility across AI use cases.
  
• Real-time Data Activation: Invest in real-time data pipelines so AI models always have access to the most current data. This ensures that AI-driven engagement, such as chatbots, can deliver meaningful benefits.

AI has tremendous potential to transform healthcare, but without trusted data, initiatives often fail, and may expose organizations to regulatory non-compliance risks while undermining patient trust.

By prioritizing a trusted data framework at the center of your ecosystem, you can unify data across all touch points, integrate consent, anonymize data where necessary, enforce data quality standards, and make high-quality data available to AI models in real-time.

Healthcare leaders who invest in trusted data today can create a strong foundation for AI success, generate actionable insights, and improve health outcomes at scale.