# Natural language processing to identify lupus nephritis phenotype in electronic health records

**Authors:** Yu Deng, Jennifer A. Pacheco, Anh Chung, Chengsheng Mao, Joshua C. Smith, Juan Zhao, Wei-Qi Wei, April Barnado, Chunhua Weng, Cong Liu, Adam Cordon, Jingzhi Yu, Yacob Tedla, Abel Kho, Rosalind Ramsey-Goldman, Theresa Walunas, Yuan Luo  
**arXiv:** [2112.10821](https://arxiv.org/abs/2112.10821)  
**Date:** 2021-12-20  
**Summary page:** https://www.summarizepaper.com/en/paper/2112.10821v1/

## Key points

- Systemic lupus erythematosus (SLE) is a rare autoimmune disorder characterized by unpredictable flares and remission with diverse manifestations.
- Lupus nephritis, a major manifestation of SLE, can cause organ damage and mortality.
- Accurate identification of lupus nephritis is crucial for large cohort observational studies and clinical trials.
- Procedure codes and structured data in electronic health records (EHRs) can help recognize lupus nephritis, but critical information like histologic reports and medical history narratives require sophisticated text processing.
- Researchers developed algorithms to identify lupus nephritis using EHR data, with and without natural language processing (NLP).
- Four algorithms were created: a rule-based algorithm using only structured data as the baseline, and three algorithms utilizing different NLP models.
- The best performing NLP model showed significant improvement in F measure compared to the baseline algorithm in both datasets used for validation.
- NLP-based algorithms have the potential to accurately identify lupus nephritis in EHRs.
- This has important implications for recruitment, study design, analysis in large cohort observational studies and clinical trials focused on SLE.

## Summary

Systemic lupus erythematosus (SLE) is a rare autoimmune disorder characterized by unpredictable flares and remission with diverse manifestations. Lupus nephritis, a major manifestation of SLE, can cause organ damage and mortality, making its accurate identification crucial for large cohort observational studies and clinical trials. While procedure codes and structured data like laboratory tests can help recognize lupus nephritis in electronic health records (EHRs), critical information such as histologic reports from kidney biopsies and medical history narratives require sophisticated text processing. In this study, the researchers developed algorithms to identify lupus nephritis using EHR data, with and without natural language processing (NLP). They created four algorithms: a rule-based algorithm using only structured data as the baseline, and three algorithms utilizing different NLP models. The NLP models were based on regularized logistic regression and incorporated various features like positive mention of concept unique identifiers (CUIs) and the number of appearances of CUIs. To validate their algorithms, the researchers conducted external validation on a dataset from Vanderbilt University Medical Center (VUMC). The best performing NLP model, which incorporated features from both structured data, regular expression concepts, and mapped CUIs, showed significant improvement in F measure compared to the baseline algorithm in both the NMEDW dataset (0.41 vs 0.79) and VUMC dataset (0.62 vs 0.96). The findings highlight the potential of NLP-based algorithms in accurately identifying lupus nephritis in EHRs. This has important implications for recruitment, study design, analysis in large cohort observational studies and clinical trials focused on SLE. By leveraging sophisticated text processing techniques to mine information from pathology reports and medical history narratives, researchers can enhance their understanding of lupus nephritis phenotypes for improved patient characterization.