Background: Ventilator-associated pneumonia (VAP) is a major cause of morbidity and mortality in intensive care units (ICUs), particularly in low- and middle-income countries (LMICs). Evidence-based practice (EBP) bundles have shown effectiveness in reducing VAP; however, the implementation in Bangladesh remains limited. This study aimed to evaluate the effectiveness of EBP to reduce the incidence rate of VAP among adult ICU patients in Bangladesh. Methods: A quasi-experimental study with a historical control group was conducted among 347 eligible ICU patients from October 2024 to April 2025. The intervention included nurse training on VAP bundle practices with advanced equipment support. Data on VAP incidence as a primary endpoint and VAP-related patients’ outcomes were analyzed. Results: The clinically suspected VAP incidence was 30.1 and 51.1 per 1000 ventilator-days, and the prevalence decreased significantly in the intervention group compared to the control group (26.9% vs. 46.1%; p < 0.001), respectively. Logistic regression indicated VAP bundle implementation was associated with reduced VAP (Exp(B) = 0.417, 95% CI: 0.262–0.666), while ventilation ≥96 h was a significant risk factor (Exp(B) = 2.6, 95% CI: 1.385–4.881). Early-onset VAP was reduced (25.0% vs. 10.2%), though late-onset predominated in the intervention group (75.0% vs. 89.8%). Conclusion: Implementation of an EBP-based VAP bundle by trained nurses significantly reduced VAP incidence. However, increased overall ICU mortality highlights the need for broader critical care improvements, including advanced comorbidity management and comprehensive ICU services. This study underscores the feasibility and effectiveness of VAP bundle implementation in the ICU of an LMIC.

Patient satisfaction—particularly with doctor–patient communication—is a key indicator of hospital service quality and a driver of operational improvement. Traditional approaches to improving communication, such as staff training or system reforms, are often resource-intensive and difficult to implement broadly. Existing doctor recommendation systems typically either rely heavily on patient health information, overlooking patient preferences, or are designed for online consultations, limiting their applicability to routine offline care. To address these limitations, we propose a satisfaction-driven doctor recommendation strategy that utilizes only real-world Hospital Information System (HIS) data, without relying on patients’ subjective preferences or simplistic rating scales. Drawing on 49,372 outpatient visits from a tertiary hospital in China, the strategy constructs behavioral features for both patients and doctors, applies unsupervised clustering, and uses a Random Forest classifier to predict satisfaction levels. The model achieves AUCs of 0.85 (negative), 0.82 (neutral), and 0.81 (positive), with validation and test accuracies of 77.05% and 78.22%, respectively. Its effectiveness is supported by significant improvements in satisfaction: 94 cases shifted from negative to positive, and 225 from neutral to positive, after applying the recommendation (chi-square and symmetry tests, both, p < .001). The strategy enables real-time, operationally neutral recommendations without increasing clinical workload and is easily adaptable across institutions using standardized HIS data. Its process-oriented feature engineering enhances interpretability, generalizability, and consistency with real patient experiences. Future work will explore the inclusion of environmental variables and multi-center validation.