Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1745
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-17T22:02:22.877268Z
status: published
lens: g_space
sector: health
headline: One-day-ahead prediction of severe hyponatremia in patients with cancer using medication and diagnostic trajectories without electrolyte indicators
dek: Intro Hyponatremia is a common complication in patients with cancer and is associated with poor prognosis; however, early detection remains challenging. Existing models often rely on laboratory parameters directly related to sodium levels, limiting early risk identification before electrolyte testing. This study developed a machine learning model to predict severe hyponatremia 1 day before its occurrence using longitudinal prescription and diagnostic data without relying on serum electrolyte variables. Methods T…
gain_title: A gradient-boosting model using only 30-day prescription and diagnosis trajectories predicted severe hyponatremia one day ahead with an AUC of 0.853, close to a lab-inclusive model's 0.862.
problem_title: (none)
trace_subject: (none)
gain_reading: A gradient-boosting model using only 30-day prescription and diagnosis trajectories predicted severe hyponatremia one day ahead with an AUC of 0.853, close to a lab-inclusive model's 0.862.
gain_evidence: XGBoost achieved an area under the curve (AUC) of 0.853 and an F1-score of 0.411, comparable to those in Case 2 (AUC 0.862 and F1-score 0.426) | Severe hyponatremia can be predicted 1 day in advance with clinically meaningful predictive performance
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers built machine learning models on records from 33,160 adult cancer patients treated between 2010 and 2021 at a single center, converted to the OMOP common data model. About 8.8 percent of patients developed severe hyponatremia, defined as serum sodium of 125 mmol/L or lower. Models were fed 30 days of time-series data and tested across three feature sets.

The finding matters because it suggests high-risk patients could be spotted before any electrolyte test, for example during treatment planning or when routine monitoring is not done, allowing more targeted testing and earlier clinician action. The authors say this still awaits external validation, and the model was developed at one center.
limitation: The study is retrospective and from a single center, and the authors state that its use for targeted testing and earlier intervention depends on external validation.
tag: Evidence-backed gain
key_points: The study covered 33,160 adult patients, of whom 2,916 (8.8 percent) developed severe hyponatremia. | XGBoost consistently outperformed the Transformer model across the feature sets. | The medication-focused set reached an AUC of 0.853 and F1-score of 0.411, versus 0.862 and 0.426 with expanded laboratory variables. | SHAP and attention analyses highlighted cisplatin, lung cancer, male sex, dexamethasone, and trimethoprim/sulfamethoxazole as key predictors.
rundown: Existing prediction models usually depend on laboratory values tied to sodium levels, which limits risk identification before electrolyte testing takes place. This study aimed to predict severe hyponatremia a day ahead using only prescription and diagnostic history, without serum electrolyte variables.

Three feature configurations were compared: a baseline set, an information-rich set with expanded laboratory variables, and a medication-focused set. The XGBoost and Transformer architectures were each evaluated, with XGBoost performing better throughout.

Interpretability analyses pointed to specific drugs and patient characteristics driving predictions. The authors conclude the approach could support more targeted electrolyte testing and earlier intervention, pending external validation.
sources:
- peer_reviewed | International Journal of Medical Informatics | https://doi.org/10.1016/j.ijmedinf.2026.106743 | 2026-10-07
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

a8622427d6585980da833130e56bb27704d699a49818c4ca05b0da630c0e789e

previous

0000000000000000000000000000000000000000000000000000000000000000