Navigating Adverse Event and Medical History Clinical Data Anonymization  

In the ever-evolving landscape of clinical trials, managing data responsibly while adhering to regulatory and company defined disclosure requirements is crucial. This post will explore a best-in-class   methodology to assess and anonymize  Adverse Event (AE) and Medical History (MH) data.  

Understanding Adverse Event Assessment

The Objective

At the heart of Adverse Event assessment is the goal to accurately evaluate and report AEs while minimizing the risk of patient re-identification. Achieving this requires a systematic process that incorporates both quantitative and qualitative methods.

A Hybrid Approach

At RLS, we employ a hybrid approach to navigate the complexities of AE assessment. This method integrates quantitative models—which provide metrics, automation, and efficiency—with qualitative insights that derive from domain knowledge and the specific context of the trial and the participant narrative itself. The quantitative component primarily focuses on risk-based anonymization assessments applicable to participant quasi-identifiers, which help us define equivalence classes for the data.

By evaluating similar trials a broader participant population may be applied resulting  in equivalence classes smaller than traditional classes. This allows us to confidently retain more data while protecting participant privacy. Other methods include setting a risk of re-identification threshold—say, 9% depending on how the data will be shared. Using assessment capabilities such as reference populations and risk of re-identification thresholds combined with a qualitative review of the participant narratives found in clinical documents, informed decisions regarding retaining the AE data can be concluded.

The Assessment Process

The RLS assessment process unfolds in three key steps:

1. Automated AE Processing: The first step involves filtering AEs through the RLS internal Rare Sensitive and Observable database. This flags AEs that require further scrutiny, allowing us to focus on those that are truly rare and sensitive - which are most important to evaluate more closely.
2. Quantitative Assessment: In the second step, we apply the quantitative assessment results to retain AEs occurring 11 or more times, or occurring in classes or groups suggested by the k-sample. The k-sample value is the result of the analysis of quasi-identifiers which creates smaller groupings that adhere to the threshold.  
3. Contextual Review: For AEs that appear fewer than four times, we conduct a detailed contextual review. This step assesses each of the remaining AEs against the specific characteristics of the trial such as trial indication, safety drug profile or re-identifiability if it is observable, knowable or replicable, ensuring that we make informed decisions about anonymization or retention based on clinical relevance.

Through this systematic process, we are able to  maximize data utility while the necessary safeguards and risk of re-identification are fully understood before submitting the data  for regulatory and third party review.

Maximizing Data Utility

Balancing Re-Identification Risk and Usefulness of the Resulting Data

The overarching goal of our approach is to maximize the data utility of AEs.After all, AEs are a critical component for understanding the trial results and how it may impact secondary research efforts. Even if an event is rare or sensitive, it may still be crucial to the context of the study. For instance, if Schizophrenia appears among two participants as an AE in a trial focused on mental health, we prefer to retain this information due to its relevance, whereas other less pertinent AEs might be generalized to higher level terms. 

We often visualize the filtering process as a funnel. Starting with the broad pool of all AEs that have occurred in a trial, we progressively narrow down our focus through rigorous assessment through quantitative and qualitative means, ultimately retaining only those AEs that provide valuable insights without compromising patient privacy outside the desired re-identification risk threshold.

Examples of Contextual Review

To illustrate the significance of contextual review, consider three hypothetical AEs that occur less than four times within a study:

1. Schizophrenia: If associated with a study on psychiatric disorders, this AE might be retained due to its relevance.
2. Immunodeficiency Syndrome: If linked to one participant and viewed as potentially stigmatizing, it may be generalized to a higher level term to protect the individual’s identity.
3. Burns: This could be transformed to a system organ class due to its sensitivity and potentially something that is permanently observable, therefore posing a higher risk of re-identification.

These examples emphasize that our recommended approach is not a one-size-fits-all solution; context plays a  significant role in finalizing the determination to retain, generalize or possibly redact.

Anonymization of Medical History Data

While managing AEs is critical, anonymizing Medical History data presents its own set of challenges. It’s essential to evaluate the information based on specific criteria to determine whether it should be retained, redacted, generalized, or suppressed.

Key Considerations

We identified four crucial considerations when anonymizing medical history data:

1. Entities: Who does the Medical History reference pertain to? Is it participant-level, aggregate-level, or non-participant level (e.g., family history)? The answers to these questions will likely determine the approach taken to redact or retain per entity.
2. Rule Level: What specific decisions and anonymization strategies apply? When should  the standard/global redaction rules vs project specific rules? There needs to be a clear understanding and alignment that project specific rules will take precedence over standard rules. For example, if Nephrolithiasis is a term generally protected but it is present in a Urinary Disorder narrative, hence it will be retained because it is relevant to the context of the patient narrative section. Therefore It’s critical to understand the hierarchy of said rules, especially in cases of conflicting information while considering the trial characteristics.
3. Conditional Evaluations: Are there specific circumstances affecting how we treat certain Medical History terms? This ensures the approach is flexible and tailored to individual cases.
4. Standards: Establishing a database of common terms helps us maintain consistency. This ensures that widely recognized terms that pose minimal risk of re-identification are retained.

Practical Applications

In practice, the approach to Medical History anonymization can vary significantly based on trial characteristics. For example:

• Participant-Level Information: Generally, we recommend suppressing or redacting most Medical History unless it relates to inclusion criteria or safety profiles of the study.
• Aggregate Information: In larger studies, we generally recommend all aggregate Medical History data is retained, but in smaller trials, careful assessments are necessary to avoid re-identification risks.

Contextual Factors

The context of each Medical History term is critical in determining the most appropriate anonymization technique to apply. For example, if Diabetes is mentioned in a context that links it directly to an Adverse Event, it may be retained. However, if it appears without context, it might be redacted to protect against risk of re-identification.

Conclusion

In summary, navigating the complexities of anonymizing and disclosing Adverse Event and Medical History data requires a thoughtful and structured methodology. By applying a hybrid method that integrates quantitative and qualitative assessments, data utility can be  maximized while safeguarding patient privacy while adhering to a defined risk threshold. This careful balance is essential in the context of clinical trials where ethical,regulatory and company standards and policies must be met.

As we continue to refine our processes, we aim to contribute to a culture of transparency,  trust and participant privacy in clinical research, ensuring that vital data is preserved and utilized responsibly for the benefit of all stakeholders involved.