Clinical Phenotypes of Chronic Hypoxemia Among Hospitalized Adults in Kenya: A Cluster Analysis

Abstract Rationale: Chronic hypoxemia is an important contributor to morbidity and mortality among individuals with chronic lung disease. However, etiologies and outcomes of chronic hypoxemia in high-HIV and high-tuberculosis burden settings such as Kenya are not well understood. Thus, we sought to identify clinical phenotypes of chronically hypoxemic lung disease among individuals admitted to a tertiary care hospital in Kenya. Methods: A k-prototype cluster analysis was performed using data from a case-control study on chronic hypoxemia among adults admitted to Moi Teaching and Referral Hospital in Eldoret, Kenya between 2019 and 2022. Chronic hypoxemia was defined as an SpO2 ≤ 88% at either 1-month post-discharge follow-up or, for patients who died prior to follow-up, a documented SpO2 ≤88% during a previous hospital discharge or outpatient visit within the last 6 months. Of a possible 78 variables, 21 were chosen based on data completeness, clinical relevance, and for chest x-ray data, inter-reader reliability. Silhouette index was assessed to determine the optimal number of clusters among participants with chronic hypoxemia. Kruskal-Wallis and chi-square tests were used to compare differences in characteristics and mortality outcomes between clusters and the control group of participants without hypoxemia. Results: The dataset included 348 participants, of whom 108 (31%) were chronically hypoxemic. Four distinct clusters were identified among chronically hypoxemic participants (Table 1). Cluster 1 (n=25) is characterized by males with high rates of tobacco use and pleural effusion, thickening or pneumothorax on CXR. Cluster 2 (n=4) is characterized by younger males with high rates of kidney disease (75%). Cluster 3 (n=37) is characterized by older females with high rates of tobacco use and hyperinflation on CXR, and Cluster 4 (n=42) is characterized by younger females without a tobacco history, high rates of HIV co-infection and enlarged cardiac silhouettes on CXR (50%). All 4 clusters had moderately to severely elevated right-heart pressures by echo. Cluster 2 participants had the highest inpatient mortality (75%), while Cluster 4 participants had the highest rates of 1-month post discharge mortality (19%). Conclusions: There are distinct clinical phenotypes among individuals with chronic hypoxemia admitted to a national referral hospital in Kenya that can be identified using routinely available data. These phenotypes inform mortality and mental health outcomes and can be utilized in implementation efforts of medical oxygen and long-term oxygen therapy. Future research is needed to validate these phenotypes in another similar cohort.