Hypokalaemia in hypertensive patients : data-driven epidemiological insights and machine learning-based risk analysis

Abstract

Hypertension, a leading risk factor for cardiovascular disease, is commonly managed with diuretic therapy. Diuretic-induced hypokalaemia, a well-recognised electrolyte imbalance, carries the risk of precipitate ventricular arrhythmias or even sudden death, potentially undermining the efficacy of diuretics in reducing cardiovascular morbidity and mortality. However, the burden of hypokalaemia is challenging to quantify due to its subtle or non-specific symptoms and the diverse characteristics of the patient population affected by this condition. A comprehensive understanding of its prevalence, risk factors, and management is, therefore, imperative. In exploring the intricate relationship between hypokalaemia and hypertension, this thesis addresses a critical research gap— quantifying the risk and identifying predictors of hypokalaemia among hypertensive patients, especially treated with diuretics. Through a series of data-driven investigations, this body of work synthesizes findings from national and territorial datasets and machine learning models to inform clinical strategies and patient management.

The initial investigation into hydrochlorothiazide use in the treatment of hypertension highlights a notable hypokalaemia prevalence of 12.6%, with demographic factors and treatment duration identified as significant risk contributors. A subsequent study on indapamide shows a low incidence of severe hypokalaemia, with less than 1% necessitating hospitalisation, predominantly occurring early in the treatment course. These findings underscore the importance of careful drug selection and monitoring practices in clinical protocols.

Diving deeper into predictive analytics, this thesis employed machine learning algorithms to the broader hypertensive cohort. The Random Forest algorithm, in particular, was found to outperform four other common algorithms in identifying key predictors of hypokalaemia, including but not limited to demographic factors, medication use, and comorbid conditions. For patients with concurrent cardiovascular disease, additional risk factors such as specific treatment regimens were identified. A derived nomogram from this model provides clinicians with a practical tool to effectively identify hypertensive individuals at risk of hypokalaemia. This predictive capability suggests a pathway for artificial intelligence-augmented risk stratification and personalised medicine approaches.

Moreover, the application of consensus clustering, an unsupervised machine learning method, reveals three distinct cluster subgroups within the hypokalaemic hypertensive population, each with unique clinical characteristics and differential mortality risks. Cluster 1 comprises younger patients with fewer comorbidities; Cluster 2 includes older patients with prevalent cardiovascular conditions and exhibits the highest all-cause mortality risk; Cluster 3 consists of the oldest patients, with a higher prevalence of arthritis and cancer, and the highest cardiovascular mortality risks. This nuanced classification not only delineates the heterogeneity of the condition but also signals the need for tailored management strategies to mitigate the identified elevated risks of all-cause and cardiovascular mortality.

Collectively, these studies aim to highlight the multifaceted nature of hypokalaemia in the context of hypertension. This thesis not only fills a crucial knowledge gap by providing robust epidemiological evidence and advanced analytical risk assessments but also advocates for an evolution in the management of hypertensive patients. Through strategic monitoring, individualised therapy, and the integration of innovative machine learning tools, practitioners are better equipped to navigate the complexities of hypokalaemia, ultimately improving patient outcomes.