Predicting High Excess Risk of Hyponatremia Among Thiazide Users

Importance: Hyponatremia is a potential serious adverse drug reaction to treatment with thiazide diuretics. Whether individuals with a high risk of developing thiazide-induced hyponatremia can be predicted before treatment initiation is unknown. Objective: To identify patients with a high risk of thiazide-induced hyponatremia following treatment initiation. Design: Population-based cohort study, using national register data on demography, comorbidities, comedication, and blood test results to predict individual-level risk of thiazide-induced hyponatremia following treatment initiation based on models trained with the causal forest method. Models were validated in a separate cohort and compared by concordance-for-benefit (C-for-benefit) and calibration plots. Setting: Denmark, 1 January 2014 to 31 December 2020. Participants: Patients were ≥40 years old and new users of thiazide or non-thiazide antihypertensive drugs (renin-angiotensin-system inhibitors or calcium channel blockers) and split into a development (n=185,699; from 2014-2018) and a validation cohort (n=75,030; 2019-2020). Exposure: Thiazide or non-thiazide antihypertensive treatment. Main outcome and measure: Predicted excess risk of moderate-to-severe hyponatremia defined as plasma sodium levels <130 mmol/L, measured within the first 120 days of treatment. Results: Individual-level excess risk could be parsimoniously described by a four-covariate model that included information on age and plasma sodium, hemoglobin, and C-reactive protein levels at baseline with good calibration and C-for-benefit (0.66; 95% CI, 0.66-0.67) in the validation cohort. The average 120-day excess risk of hyponatremia among thiazide-treated patients was 1.8% (95% CI, 1.3% to 2.2%), with individual-level heterogeneity ranging from -1.6% to 15.9%. The 10% of thiazide-treated with the highest excess risk of hyponatremia had an average excess risk of 7.4% (95% CI, 4.4% to 10.5%). If this high-risk group were instead assigned a non-thiazide antihypertensive, the excess risk within the remaining thiazide-treated population would be reduced by 0.7% (95% CI, 0.1% to 1.2%), corresponding to a 38% relative reduction. Conclusion and relevance: The population-level burden of thiazide-induced hyponatremia can potentially be markedly reduced by assigning a small group of patients at highest risk, who would otherwise be assigned a thiazide, to an alternative antihypertensive drug. This high-risk group can be identified using a small set of simple baseline information.