This was a retrospective cohort study of patients admitted with heart failure (HF) from January 2021 to December 2024. The study was conducted at a Laquintinie Hospital Douala, a second level hospital in Cameroon serving a socioeconomically diverse population with limited access to advanced cardiac care. Data were extracted from medical records and patient charts in the cardiac unit.

Eligible participants were patients with a clinical diagnosis of HF admitted during the study period. Patients with incomplete data on outcome or not hospitalised for an acute episode of heart failure as primary diagnoses were excluded. Participants were followed from admission until death or discharge.

The sample size was calculated to detect predictors of in-hospital mortality, assuming a 34% mortality rate from regional heart failure studies in sub-Saharan Africa [8] . To ensure model stability and prevent overfitting, we aimed for an events-per-variable (EPV) ratio of 10 based on recommendations for logistic regression, as outlined in statistical literature [12] . Targeting 10 predictors, a minimum of 100 mortality events was required, necessitating at least 340 patients (100/0.34).

The primary outcome was in-hospital mortality, defined as death during hospitalization. HF diagnosis was confirmed using ESC 2021 criteria, incorporating clinical symptoms, and/or echocardiography or biomarker data where available, to minimize misclassification [13] . Covariates assessed included age, sex, comorbidities (hypertension, kidney disease, anemia, via laboratory tests), and use of recommended medications including renin-angiotensin-aldosterone system inhibitors, beta-blockers, mineralocorticoid receptor antagonists, or sodium-glucose co-transporter-2 inhibitors.

Data were sourced from hospital medical records, echocardiography reports, and laboratory results. Trained personnel used standardized protocols to extract data, with 10% of records cross-checked for accuracy. Data were collected using a structured tool that included: 1) demographic information (such as age, sex, marital status, and place of residence); 2) clinical parameters (including ejection fraction categorized as HFrEF <40%, HFmrEF 40% - 49%, and HFpEF ≥50%, systolic blood pressure, and oxygen saturation); 3) laboratory results (such as haemoglobin levels and blood urea nitrogen [BUN]); 4) prescribed medications (including RAAS inhibitors) before admission, during admission and at discharge; and 5) outcomes (mortality and length of hospital stay). Patients with missing outcome data were excluded from analysis. Potential confounding by socioeconomic factors was addressed by adjusting for relevant covariates such as marital status in the analysis.

Missing key severity markers such as NYHA class, natriuretic peptides, and ejection fraction (EF) for 45% of patients introduces potential biases. NYHA class was not routinely documented due to inconsistent practices, risking misclassification bias and underestimating mortality risk (bias toward null). Natriuretic peptides were unavailable due to limited diagnostic resources, causing confounding bias by omitting a key prognostic factor, potentially over- or underestimating covariate effects. Missing EF data for 45% of patients, due to scarce echocardiography access, introduces selection bias, as tested patients may differ (e.g., wealthier or sicker). This may underestimate mortality risk in severe cases (e.g., undiagnosed HFrEF), biasing results toward the null. Mitigation included MICE imputation, covariate adjustment (e.g., age, comorbidities), and ROSE for class imbalance. Residual confounding persists, necessitating future studies with robust diagnostic data.

The study developed a logistic regression model to predict in-hospital mortality using a robust methodology. Normality of variables was assessed with the Shapiro-Wilk test, and descriptive statistics (medians, interquartile ranges, percentages) summarized baseline characteristics and mortality rates due to non-normal distributions. Univariable logistic regression identified factors associated with mortality (p < 0.250 and clinical relevance) for inclusion in the multivariable model. The dataset was pre-processed by renaming columns for clarity and ensuring appropriate data types. Missing data were handled using Multiple Imputation by Chained Equations (MICE), generating five datasets, with the first used for analysis; imputation assumptions were based on observed variable distributions, and sensitivity analyses confirmed robustness. To address moderate class imbalance (17% mortality), the Random Over-Sampling Examples (ROSE) technique was applied to balance discharge and death outcomes, as it improved model sensitivity for detecting mortality events in preliminary analyses. Predictor selection involved univariate screening, multicollinearity assessment (removing variables with Pearson’s |r| > 0.7 or variance inflation factors [VIF] > 5; all included variables had VIF < 5), and stepwise regression (forward/backward selection via AIC). Clinically relevant interaction terms were cautiously included. Adjusted odds-ratio estimates and standard errors in multivariable model were pooled across all imputed datasets, Model performance was assessed via AUC, sensitivity, specificity, and Hosmer-Lemeshow test. Missing data imputation and sensitivity analyses were conducted to ensure robustness. Model performance was evaluated using AUC, Gini coefficient, accuracy, sensitivity, specificity, and the Hosmer-Lemeshow test (p > 0.05 indicating good calibration). Calibration was visualized using calibration plots, and model robustness was tested via bootstrap validation (1000 iterations). Analyses were conducted using SPSS v26 and R v2025.05.0 + 496.

The study was approved by the Regional Ethics Committee for the Littoral under reference N 2024/CE/CRERSH-LITTORAL and informed consent was waived for retrospective data.

This retrospective cohort study included 757 patients hospitalized for heart failure (HF) at Laquintinie Hospital in Douala, Cameroon, from 2021 to 2024 as shown in Figure 1 . Eligible patients were identified from hospital records with a confirmed HF diagnosis, admitted between January 2021 and December 2024. No patients were excluded due to missing outcome data (in-hospital mortality). The cohort was distributed across admission years: 18.0% (n = 136) in 2021, 26.9% (n = 204) in 2022, 24.3% (n = 184) in 2023, and 30.8% (n = 233) in 2024.

The median age was 63 years (IQR: 51 - 74), with 51.3% female (n = 388) and 48.7% male (n = 369). Most patients resided in Douala (87.5%, n = 643), and 47.6% (n = 360) were married, while 21.5% (n = 163) were single, 23.0% (n = 174) widowed, and 6.6% (n = 50) unreported as shown in Table 1 . Professionally, 32.8% (n = 248) were unemployed or retired. Prevalent comorbidities included hypertension (48.6%, n = 368), chronic heart failure (34.2%, n = 259), diabetes mellitus (16.9%, n = 128), and history of lung pathology (8.6%, n = 65) as shown in Table 2 . Routine cardiovascular medications included ACE inhibitors (12.9%, n = 98), beta-blockers (11.9%, n = 90), and calcium channel blockers (16.5%, n = 125). At admission, median systolic blood pressure was 137 mmHg (IQR: 114 - 162), with 4.6% (n = 35) having hypotension (SBP <90 mmHg) as shown in Table 3 . Diagnostic procedures included echocardiography (54.4%, n = 412) and ECG (46.5%, n = 352). Common HF aetiologies were hypertensive heart disease (28.5%, n = 216), dilated cardiomyopathy (12.9%, n = 98), and Cor pulmonale (9.1%, n = 69) as shown in Table 4 . Median ejection fraction was 39% (IQR: 26 - 60) and HF subtypes were distributed as follows: HFpEF (36.9%, n = 152), HFmrEF (10.2%, n = 42), and HFrEF (52.9%, n = 218).

In-hospital mortality was 17.0% (n = 129), with rates of 11.8% (2021), 15.2% (2022), 19.0% (2023), and 20.2% (2024). Among the 628 patients discharged alive, 74.0% (n = 560) had authorized discharges, 6.9% (n = 52) were discharged against medical advice, and 2.1% (n = 16) were transferred. The median hospitalization duration was 8 days (IQR: 5 - 12) as shown in Table 5 .

Univariate logistic regression identified predictors significantly associated with in-hospital mortality (p < 0.05). Admission in 2023-2024 (vs. 2021-2022), cerebrovascular accident, chronic liver disease/hepatitis, hypotension (SBP <90 mmHg), elevated blood urea nitrogen (>60 mg/dL), Cor pulmonale, and high-output heart failure increased mortality risk. Conversely, leucocytosis was observed in 22.0% (n = 93/424) of patients (20.4% alive, n = 74/363; 30.6% deceased, n = 19/62), and anemia (Hb <12 mg/dL) in 60.0% (n = 273/455) (58.5% alive, n = 227/388; 68.7% deceased, n = 46/67). Hypertensive heart disease, and hypertensive crisis were linked to lower mortality. Dobutamine use increased mortality, while beta-blockers and RAAS inhibitors reduced it.

The logistic regression models were evaluated for predicting in-hospital mortality, with the stepwise model selected as the best due to its superior calibration (Hosmer-Lemeshow p = 0.811) and balanced performance metrics as shown in Figure 2 and Figure 3 . The stepwise model achieved an AUC of 0.769, indicating good discriminatory ability, and a Gini coefficient of 0.538. Its accuracy was 0.680, with a sensitivity of 0.783 and specificity of 0.659, reflecting effective identification of both mortality and survival cases. The model outperformed the initial (AUC = 0.767, accuracy = 0.674), refined (AUC = 0.767, accuracy = 0.674), and optimized (AUC = 0.738, accuracy = 0.679) models, which either had lower calibration (optimized: p = 0.068) or similar but less optimal metrics. The stepwise model’s most potent predictors identified as statistically significant (p < 0.05), were: Blood urea Nitrogen (OR 1.01; 95% CI 1.01 - 1.02, p < 0.001); Systolic pressure on admission mmHg (OR 0.99; 95% CI 0.98 - 1.00, p = 0.009); Dobutamine use (OR 2.85; 95% CI 1.30 - 6.25, p = 0.009); Hypertensive heart disease (OR 0.40; 95% CI 0.21 - 0.77, p = 0.006)); Age in years (OR 1.01; 95% CI 1.00 - 1.02, p = 0.069), Year of admission 2023-2024 (OR 1.79; 95% CI 1.15 - 2.80, p = 0.010) were visualized in Figure 4 . Other predictors, including age, calcium channel blockers, RAAS inhibitors/ARNi, beta blockers, and ischemic/dilated cardiomyopathy, were included but not significant.

COR^a = Crude Odds Ratio.

ACEI = Angiotensin Converting Enzyme Inhibitors, ARAII = Angiotensin II receptor antagonists, COPD = Chronic Obstructive Pulmonary Disease, CV = Cardiovascular, CVA = Cerebrovascular accident, MRA = Mineralocorticoid Receptor Antagonist, SGLT2 = Sodium-glucose cotransporter 2, TB = Tuberculosis. ^bHistory of lung pathology (Asthma, Pulmonary Embolism, COPD, TB).

COR^a = Crude Odds Ratio; ^bRef = Normal potassium level.

COR^a = Crude Odds Ratio.

RAAS = Renin-Angiotensin Aldosterone System; SGLT2 = Sodium-glucose cotransporter 2; COR^a = Crude Odds Ratio.

The 17.0% mortality rate and low use of SGLT2 inhibitors highlight the need for health system reforms, including subsidized diagnostics, clinician training on 2022 AHA/ACC/HFSA guidelines, and task-shifting to improve guideline uptake in Cameroon. More studies should be conducted to evaluate the impact of guideline adherence on mortality.