Rule-Based Algorithm to Identify Recurrent Non-Hodgkin Lymphoma in Electronic Health Data

Area of Science:

• Oncology
• Health Informatics
• Epidemiology

Background:

• Standardized capture of recurrent cancers in US tumor registries is lacking, hindering research on recurrence risk factors.
• Electronic health data offers a potential source for identifying cancer recurrence, but requires robust algorithms.

Purpose of the Study:

• To develop and validate rule-based algorithms for identifying recurrent diffuse large B-cell lymphoma (DLBCL) and follicular lymphoma (FL) using electronic health data.
• To improve the accuracy of identifying cancer recurrence for population-based epidemiological studies.

Main Methods:

• Developed a rule-based algorithm using pharmacy and procedure codes to define recurrent DLBCL and FL based on treatment patterns.
• Validated a baseline algorithm using claims data from Fallon Health and electronic health records from Henry Ford Health.
• Revised the algorithm to reduce false-positive rates and calculated sensitivity, specificity, and predictive values through chart review.

Main Results:

• The revised R2D-non-Hodgkin lymphoma (NHL) algorithm identified 60 recurrent cases with a 10% false-positive rate at Henry Ford Health.
• The R2D-NHL algorithm demonstrated a sensitivity of 74% and specificity of 90% in identifying recurrent NHL.
• Positive and negative predictive values were 83% for the R2D-NHL algorithm, with slight variations between DLBCL and FL subtypes.

Conclusions:

• A validated rule-based algorithm (R2D-NHL) can effectively identify recurrent DLBCL and FL in electronic health data.
• This algorithm facilitates population-based research on cancer recurrence by standardizing case identification.
• The developed algorithm addresses a critical gap in cancer registry data for studying NHL recurrence.