This Multi-center prospective cohort study was conducted in Southern Ethiopia teaching referral hospital ICUs from October 2020 to November 2022. Three teaching and referral hospital ICUs were selected randomly from ten teaching hospitals in Southern Ethiopia, namely Hawassa university referral hospital (HURH), Dilla University referral hospital (DURH), and Wolaita Sodo referral hospital (WURH). The ICUs are providing a similar level of care with almost similar staff profiles, monitoring modalities, ICU infrastructure, medical supplies, and admission patterns.
Study inclusion and exclusion criteria
All patients over the age of 12 who received mechanical ventilation at DURH, Wolaita Sodo University Hospital, and Hawassa University Specialized Hospital ICUs and stayed there for longer than 24 h of admission were included; whereas, patients receiving non-invasive oxygen supplementation, patients who were discharged, transferred to, or left against medical advice before 24 h of mechanical ventilation, patients who were admitted to ICU on re-admission, and patients without attendants were excluded.
The cumulative incidence of 28-day mortality, duration of a mechanical ventilator, incidence of complications, pattern of disease, and length of stay in ICU were dependent variables, while socio-demographic characteristics: Age, sex, height, weight, BMI; Admission characteristics: date of ICU admission, date of discharge, vital sign, diagnosis, admission category, time of admission, laboratory index, disease severity score; Mechanical ventilation parameters: date of initiation, initial mode, date of tracheostomy, and date of weaning were independent variables.
The data were collected prospectively from three hospital ICUs patients in whom mechanical ventilation was initiated. After 2 h of mechanical ventilation, trained ICU nurses followed eligible participants for 28 days. The data were collected with a validated questionnaire and tools adopted from previous studies [11, 21,22,23]. A total of 630 patients were recruited consecutively from each ICU from October 2020 to November 2022 as per proportion allocation to size (Fig. 1).
The data collection included Socio-demographic characteristics: Age, Sex, Height, Weight, BMI); Admission characteristics: date of ICU admission, date of discharge, vital signs, diagnosis, admission category, time of admission, laboratory index, disease severity score; Complications: nosocomial infection, ventilator-associated pneumonia, sepsis, ARDS, aspiration, unplanned extubation, endotracheal tube blockage, tracheostomy loose/stenosis/fistula, cardiac arrest, acute kidney injury, bedsore); comorbidities: hypertension, diabetes mellitus, cardiovascular disease, Asthma/COPD, liver failure, renal failure, others, and clinical outcomes including mortality, duration of a mechanical ventilator, and complications were measured for 28 days. Furthermore, the disease severity scores including sequential organ assessment (SOFA) score, modified SOFA score, Acute Physiology, and Chronic Health Evaluation (APACHE) II score, and modified APACHE II score were recorded at admission and thereafter till discharge every seven days to compare the predicting ability and 28-day mortality of these tools.
The data were checked, coded, and entered into Epi-info version 7.0 and imported to SPSS version 22 for analysis. However, STATA version 14 was used to produce some graphs, multilevel analysis and pool the incidence of clinical outcomes. Descriptive statistics were summarized with tables and figures. Categorical variables were reported in frequency and percentage. The numerical data were reported in mean ± SD for symmetric and median (Interquartile range) for asymmetric numeric data. The outlier of the data was checked with standardized residuals while Shapiro–Wilk tests were employed for the normality test. The multicollinearity among independent variables was checked by the Variance inflation factor (VIF), and a VIF of greater than 10 was considered multicollinearity. The association of demographic characteristics, admission category, indications for MV, and complications of MV with mortality were analyzed using multilevel binary logistic regression as there was clustering as depicted with an intraclass correlation coefficient (ICC = 8.5%). All variables showing significance on multilevel bivariate analysis at a p-value less than 0.25 will be taken to multilevel multivariate analysis. In multivariate analysis, a p-value of less than 0.05 was considered for the statistical association. This study was conducted in compliance with Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for observational studies .
The primary endpoints were the incidence of 28-day mortality and duration of mechanical ventilation, while the incidence of complications during MV was the secondary clinical outcome.
Definition incidence of clinical outcomes refers to the occurrence of new clinical events including mortality, morbidity, and prolonged mechanical ventilation in 28 days of ICU stay. Morbidities are defined as any complication developed during Mechanical ventilation which includes nosocomial infection, ventilator-associated pneumonia, sepsis, ARDS, aspiration, unplanned extubation, endotracheal tube blockage, tracheostomy loose/stenosis/fistula, cardiac arrest, acute kidney injury, and bedsore. Prolonged mechanical ventilation refers to the duration of mechanical ventilation for over 21 days [25, 26].
Model building and comparison
The hierarchical nature of the hospital ICU care in which patients are nested within clusters, a multivariable multilevel logistic regression analysis was employed to account for this clustering effect. Thence, four models containing variables of interest were fitted for this study as follows: Model I (Null model) was fitted without explanatory variables to test random variability in the intercept and to estimate the intraclass correlation coefficient (ICC), Model II assessed the effects of individual-level predictors, Model III assessed the effects of hospital-level predictors, and Model IV (Full model) examined the effects of both individual and hospital-level characteristics simultaneously. The Akaike’s Information Criterion (AIC) and Bayesian Information Criterion (BIC) were used to select the model, and the model with low AIC and BIC values was considered the best-fitted model. Based on AIC and BIC, the full (model with individual and hospital-related variables) model had the smallest AIC and BIC value. Therefore, the full model best fits the data.
This study was reviewed and approved by Dilla University, College of Health Science and Medicine Research Ethics and Review Board (RERB), and a Unique Identification Number (UIN- duirb/001/21-08) was received. Besides, the protocol was registered retrospectively in ClinicalTrials.gov (NCT05303831) on March 30, 2022. The study was conducted in compliance with the Helsinki declaration for human and animal studies . Informed consent was received from all participants’ legal attendants or guardians, and all patient identifiers were kept anonymous at all times. Besides, a formal letter was written to each University hospital ICU director to get permission.