Predictive modeling for adverse outcomes in patients with heart failure

Heart failure (HF) readmission and mortality rates remain high among patients with HF despite treatment advancements. Robust risk prediction models enable better monitoring, informed decision-making, targeted interventions, and improved outcome. The previous models have limitations including the use of non-contemporary cohorts for model development, lack of robust validations, and short follow-up times. Furthermore, the naïve model selection procedures employed also render these models susceptible to biases. Consequently, there is uncertainty regarding the utility of these models for predicting current long-term outcomes at the bedside in clinical settings.In this thesis, using a contemporary dataset from the VancOuver CoastAL Acute Heart Failure (VOCAL-AHF) registry with an extended follow-up time between April 1st 2015 to March 31st 2019 and an integrated approach to model selection (combining backward selection, Least Absolute Shrinkage and Selection Operator (LASSO), and expert opinion) we sought to overcome previous model limitations. Our primary outcome was the composite incidence of all-cause mortality or first HF recurrent hospital readmission. The cohort was comprised of 1,842 patients >18 years discharged alive following unplanned hospitalization with a primary diagnosis of HF. The data included baseline characteristics and comorbidities, examination findings, laboratory results, ECG and echocardiography results, procedures, and discharge medications. Multiple imputation (n=5) was used to address missing data. Hazard ratios were estimated using a Cox proportional hazard model. To account for uncertainty and improve generalizability, bootstrap Bayesian model averaging was used to derive the final risk model.Median follow-up time was 529 days (range 2-1459). 790 (43%) patients experienced the outcome, with 8.6% having the outcome within 30 days. Occurrences of events were observed to be more frequent in older individuals (76 vs 72 years, p<0.001) with underlying disease such as diabetes (42% vs 35%, p<0.003), atrial fibrillation/flutter (AF/AFL, 61% vs 50%, p< 0.001), and a previous HF diagnosis or hospitalization (62% vs 45%, p<0.001). There were no significant differences in the proportion of females between those with outcomes (44%) and those without outcomes (42%). The final risk model included 13 variables, of which seven were identified as the most important factors (with a posterior probability of >80%). Older age, prior HF diagnosis or HF-hospitalization, lower discharge hemoglobin level, absence of discharge prescription of angiotensin converting enzyme inhibitor (ACEI) or angiotensin receptor blocker (ARB) or angiotensin receptor neprilysin inhibitor (ARNI), increased dose of loop-diuretic at discharge, decreased systolic blood pressure at admission, and smoking significantly increased the risk of death or HF readmission risk over the follow-up time. Other variables with lower posterior probabilities were ranked as follows: AF/AFL rhythm at admission, urea at discharge, left ventricular ejection fraction, diabetes, triple guideline-directed medical therapy, and serum potassium at discharge. The C-statistic for the model was 0.65.In this thesis, a clinically-oriented model for predicting HF hospitalization or death was developed. Further validation, both internally and externally, is needed to ensure the robustness and generalizability of the model for clinical use

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.