A drug-based model to predict hyponatremia in outpatients of a geriatric clinic

Chronic hyponatremia in older people is associated with adverse outcomes including gait disturbances, falls, osteoporosis, fractures, cognitive impairment, and cardiovascular disease. Diagnosis in outpatient settings is challenging due to the non-specific nature of its symptoms. While hyponatremia is well-studied in hospitalized patients, little research has focused on outpatient settings. This study aimed to develop a drug-based model to non-invasively predict hyponatremia in older adults attending a geriatric outpatient clinic.

Cross-sectional data from 2181 outpatients aged ≥ 55 were analysed using logistic regression. Polypharmacy, 27 specific drug groups, sex, age, and BMI were considered as potential risk factors. Predictors were selected using stepwise backward logistic regression for the complex model and LASSO regression for the simple model. Internal validation was performed through bootstrapping, and model performance was evaluated by constructing a receiver operating characteristic (ROC) curve.

The prevalence of hyponatremia was 10.5%, with higher occurrence in women. The complex model identified predictors including sex, age, BMI, polypharmacy, and 11 drug groups, achieving an area under the curve (AUC) of 0.75, [95% CI 0.72-0.79], indicating a reasonably good ability to distinguish between hypo- and normonatremia. The simple model, including only polypharmacy, had limited predictive performance (AUC = 0.64 [95% CI 0.60-0.68]).

The complex, drug-based model predicts hyponatremia risk in outpatients of a geriatric clinic. Timely recognition may prevent inappropriate treatments for undiagnosed cases and associated harms. The model merits further development for clinical use.