Natural language processing to identify documented pain preceding radiotherapy for bone metastases

JNCI Cancer Spectr. 2026 Feb 4:pkag010. doi: 10.1093/jncics/pkag010. Online ahead of print.

ABSTRACT

BACKGROUND: Radiotherapy (RT) plays a crucial role in managing cancer-related symptoms. This study characterized symptom documentation, especially pain, preceding bone metastasis (BM) diagnosis and initiation of RT for BM, using natural language processing (NLP) approaches.

METHODS: A de-identified cohort of patients who received RT for BM at a single tertiary-care institution (2013-2023) was created. Clinical data, notes, and metadata were computationally extracted. A previously validated NLP pipeline based on the clinical Text Analysis and Knowledge Extraction System was used to extract CTCAE-encoded symptoms from all notes in the 30 days preceding BM diagnosis and each course of RT for BM. Logistic regression analyses examined the association between clinical and demographic variables and pain documentation.

RESULTS: 1,061 patients (median [IQR] age, 64 [54-72] years; 582 [54.9%] men) received 1,718 courses of RT for BM. The most common documented symptoms before BM diagnosis and first RT for BM, respectively, were BM-related pain (52.5% vs 91.6%, p < .001), nausea (20.8% vs 48.9%, p < .001), and constipation (12.8% vs 34.2%, p < .001). Prior to BM diagnosis, multiracial/other race (OR = 0.61 [95% CI 0.38-0.99], p = .045) was associated with decreased pain documentation compared to white race. Prior to RT for BM, women (OR = 1.48 [95% CI 1.02-2.15], p = .039) had increased pain documentation compared to men.

CONCLUSIONS: Women and multiracial/other race patients experienced a relative increase in pain documentation from BM diagnosis to RT for BM. This may reflect differential decision-making for which patients are offered RT for BM sooner in the symptom trajectory. Interventions are needed to increase equitable distribution of RT.

PMID:41639010 | DOI:10.1093/jncics/pkag010