An institution-based cross-sectional study design was employed through a review of the medical records of patients who were admitted from 1 February 2020 to 1 March 2023. The data were extracted from 10 April to 28 May 2023.

The study was carried out at adult intensive care units of referral hospitals in Northwest Amhara, Ethiopia. In the northwest part of the Amhara region, there are five referral hospitals, including the University of Gondar Comprehensive Specialized Hospital (UOGCSH), Felege Hiwot Comprehensive Specialized Hospital (FHCSH), Debre Markos Comprehensive Specialized Hospital (DMCSH), Tibebe Gihon Comprehensive Specialized Hospital (TGCSH), and Debre Tabor Comprehensive Specialized Hospital (DTCSH). The catchment area for each referral hospital is thought to contain 5–7 million people (39). TGCSH is one of the teaching hospitals in Northwest Amhara, located in Bahirdar City, the capital of Amhara regional state, which is 565 km from the capital, Addis Ababa. There are two intensive care units (pediatric and adult). The adult ICU was equipped with 9 beds, 4 functioning mechanical ventilators, 7 patient monitors, and one bedside ultrasound. This unit is staffed with two anesthesiologists, internal medicine specialists and subspecialists, trained nurses, and medical and surgical residents (40). The pediatric ICU has two beds (41).

FHCSH is also the other referral hospital in Bahirdar City. The adult ICU is one of the 13 wards it has, where critically ill patients are admitted (41). Currently, it has 10 beds and 4 MVs. The UOGCSH is found in Gondar town, 700 km from Addis Ababa. UOGCSH started critical care service in 2011 with a four-bed ICU capacity, two motorized ventilators, one defibrillator, four non-invasive hemodynamic monitoring devices, and one ultrasound machine (42), and currently, it has four ICU departments divided based on specialty: medical ICU, surgical ICU, pediatrics ICU, and neonatal ICU. The adult medical and surgical ICUs of UOGCSH have 22 beds, 11 MVs, 22 monitors, 1 portable x-ray machine, 2 ultrasound machines, and 1 dialyzer machine. DTCSH is found in Debre Tabor town, the capital of the South Gondar zone. It is located approximately 665 km from the capital city of Ethiopia, Addis Ababa. It has three ICUs: 1 adult, 1 pediatric, and 1 neonatal. The adult ICU has 6 beds. DMCSH is located in East Gojam, which is located 300 km and 265 km from Addis Ababa and Bahir Dar, the capitals of Ethiopia and the Amhara regional state, respectively (43). The adult ICU has 4 beds, 3 functional mechanical ventilators, and 3 functional monitors.

All adult patients admitted to the ICU and received mechanical ventilation support at referral hospitals in Northwest Amhara were the source population. All adult patients admitted to the ICU and received mechanical ventilation support at referral hospitals in Northwest Amhara from 1 February 2020 to 1 March 2023 were the study population. All adult patients admitted to the ICU and who received mechanical ventilation support from 1 February 2020 to 1 March 2023 were included in the study. Those patients who were mechanically ventilated for less than 24 h were excluded from the study. Patient charts with variables not recorded like ICU admission and discharge date, MV initiation time, socio-demographics such as age and sex, GCS during admission, and unknown outcome variable (not recorded and the patient left against medical advice or referred to other hospitals) were declared as incomplete and excluded from the study.

For the first objective, a single population proportion formula was used to determine the sample size by taking a proportion, 57.1%, from a study in Saint Paul Hospital Millennium Medical College (SPHMMC) (23); d = margin of error, 5%, and Z α/2 = Z score of the 95% confidence level, 1.96.

For the second objective, the sample size was calculated as follows for significantly associated factors: sedation use, inotrope use, and duration of stay on MV (20), using Epi-info version 7.2.2.2, and the maximum sample size was 308. When comparing the sample sizes calculated for both objectives, the sample size obtained from the first objective (377) was found to be the highest. So, to get a maximum sample size, the sample size computed for the first objective was used.

A 15% contingency was considered for possible incomplete medical recording and possible lost charts, taking into account that a dead person’s inactive charts would be difficult to access, and the final sample size was 434.

All the five referral hospitals in the Northwest Amhara were included. After a proportional allocation of sample size made to the respective hospitals based on their 3-year data on MV use, a sampling frame was prepared using computer-generated random numbers by including all chart numbers of patients who were mechanically ventilated from February 2020 to March 2023. Finally, a simple random sampling technique was used to select the samples (Figure 1).

MODS: Failure of two or more organs at any time in the ICU, depending on ODINS criteria (44).

Required Hemodialysis: Presence of two or more findings from the indications of dialysis below:

Electrolyte imbalance: Uncontrolled Hyperkalemia (potassium >6.5 mmoL/L or rising).

Serum sodium level < 115 or > 165 mmoL/L.

Edema (fluid overload): Refractory fluid overload.

Uremia: excessive blood urea nitrogen (BUN) and creatinine levels or any uremic abnormalities such as uremic encephalopathy, uremic pericarditis, and the like.

Acidosis: severe metabolic acidosis with concomitant acute kidney injury, pH < 7.

Intoxication: life-threatening poisoning with a dialyzable drug, such as salicylates, lithium, isopropanol, methanol, or ethylene glycol (45).

Barotrauma: Radiographically confirmed pneumothorax, pneumomediastinum, or subcutaneous emphysema that could not be attributed to iatrogenic injury (46).

Comorbidity is the co-occurrence of two or more disorders or diseases at the same time (47). Comorbidity was declared if the patient has at least one chronic illness other than the acute indication for MV.

Incomplete Patient Chart: Patients’ charts were declared incomplete when they did not consist of complete baseline medical data, specifically for variables not recorded such as ICU admission date and discharge date, MV initiation time, socio-demographics such as age and sex, GCS during admission, and unknown outcome variables (not recorded and patient left against medical advice or referred to other hospitals).

Hypertension was defined as a blood pressure reading of systolic ≥130 and diastolic ≥80. At the same time, Normotensive was defined as a blood pressure reading of systolic 90–129 and diastolic 60–80 (48).

Hypotension was defined as a blood pressure reading of systolic <90 and diastolic <60 (49).

Tachypnea was defined as a respiratory rate > 30 breaths per minute (50).

Bradypnea was defined as a respiratory rate of <12 breaths per minute (51).

Sedation use was defined as having received an intravenous or intramuscular sedative (ketamine, benzodiazepines, dexmedetomidine, barbiturates, or propofol) for any period during the intensive care stay (52). This does not include the sedation used for procedures such as intubation.

Vasopressor use was defined as utilizing epinephrine, norepinephrine, vasopressin, dopamine, or phenylephrine (53) during ICU stay.

A data extraction checklist was developed from mechanical ventilation protocols and related literature (10, 16, 20–23). The data extraction tool comprised socio-demographic data, such as age, sex, and residence; clinical characteristics such as admission diagnosis, GCS at admission, serum albumin level, hemoglobin level, sepsis, indication for MV, vital signs, presence and types of comorbidity, presence and types of organ failure, and presence of MODS; and management-related characteristics, such as readmission, reintubation, sedation use, vasopressor use, required dialysis, initial ventilatory settings, and duration of stay in ICU and on MV. Five trained BSc nurses working in emergency wards of the respective hospitals collected the data using the Kobo toolbox, a mobile and tablet-based data collection platform. Four trained MSc nurses working in wards other than the adult ICU supervised the data collection process. The patients’ charts were found by taking the medical record number (MRN) from the log book at the ICU. Then, the charts were extracted from the card rooms of the corresponding hospitals. All randomly selected charts were roughly reviewed, and relevant data were extracted. For patients with readmission and reintubation, we used the last admission and intubation, respectively, to extract the data.

To control the quality of the data, data collectors and supervisors were trained separately for 2 days about the objectives of the study, confidentiality, and data collection techniques. The relevance of the variables in the tool was verified by consulting experts with a critical care specialty. Before the actual data collection, a preliminary chart review was conducted on 22 (5%) randomly selected charts at UOGCSH to check the accessibility of variables. Accordingly, variables that were repetitively not recorded in the patient recordings (inotrope use and I:E ratio) were excluded from the data extraction checklist. In addition, ventilation for less than 24 h was made as an exclusion criterion due to the absence of important variables for those ventilated for less than 24 h. During data collection, each filled checklist was cross-checked and revised daily by the investigator for completeness. Data cleaning was performed before analysis.

After completion of data collection, the data were exported from the Kobo Toolbox Server into Microsoft Excel 2019 for data cleaning and management, and then it was exported to Stata 17 for data management and analysis. The descriptive statistics were described using texts, frequency tables, percentages, and graphs, whereas mean with standard deviation and median with interquartile ranges were used for continuous variables after the data distribution was checked by histogram and skewness and kurtosis tests to characterize study participants. Mean imputation was used to manage missing values for the continuous variables, baseline serum albumin level, baseline hemoglobin level, and temperature at the initiation of MV. The category of interest (died) was coded as 1 and survived was coded as 0. All relevant variables were included. The chi-square assumption test was done for categorical independent variables. Multicollinearity was checked by the variance inflation factor (VIF) and the variables; the presence of complication and VAP were excluded due to VIF > 10. Bivariable analysis was conducted using the binary logistic regression model to determine the association between each independent variable and the outcome variable. Accordingly, variables with a p-value of ≤0.20 were considered for further analysis (multivariable analysis) to identify the net effect of each variable on the outcome variable. Finally, statistical significance was declared at p < 0.05, and the strength of associations was summarized using an adjusted odds ratio (AOR) with 95% confidence intervals (CI). The goodness of model fitness was checked using the Hosmer–Lemeshow goodness test.

A total of 404 charts of mechanically ventilated patients in the ICUs were included in the study, with a completeness rate of 93.1%. More than half (59.16%) of the patients were aged between 18 and 40 years, with a median age of 35.5 (IQR: 25–53). Two hundred twenty-five (55.69%) of the patients were male, and approximately 59.41% were rural residents (Table 1).

The main admission diagnosis for one-third of the study participants was respiratory, categorically. One hundred seventy-four (43.07%) of the study participants had a GCS of less than or equal to 8 or 7 with intubation (7T) during admission to the ICU. At the initiation of MV, more than half (62.13, 54.46, and 53.96%) of the patients were tachycardic, tachypneic, and hypoxic, respectively. Respiratory failure was the most common indication for MV (45.3%). Among the total study participants, the majority (81.44%) of them had at least one organ failure, and more than half (51.98%) had MODS. Only 87 (21.53%) patients had developed a complication, with VAP occupying the highest proportion (67.8%) (Table 2).

The median duration of ventilation for the study participants was 7 days (IQR: 3–12). Most of the patients (71.3%) had used sedatives, and nearly one-fourth (21.29%) of the patients were reintubated. The majority (87.87%) of the access to the airway was endotracheal tube and 46.53% of the patients were initiated by pressure control with assisted control mode. The median lengths of stay in the ICU and hospital for the study participants were 9 days (IQR: 5–15) and 12 days (IQR: 7–19), respectively. The mean initial Fi02 of the study participants was 83.28% ± 21.47 (Table 3).

In this study, the proportion of deaths among mechanically ventilated patients in the ICU was 254 (62.87%) with a 95% CI of (58.16–67.58) (Figure 2). The most common immediate cause of death registered was multiorgan failure, which accounts for 44.49% (Figure 3).

In the multivariable logistic regression analysis, age (41–70), sepsis, reintubation, and sedation use were found to be significant factors associated with the mortality of mechanically ventilated patients at ICUs at a p-value <0.05.

Keeping all other variables constant, the odds of mortality among the 41–70 years age group was 4.3 (AOR, 4.28, 95% CI: 1.89–9.62) times higher than those in the 18–40 years age group. While controlling for other variables, the odds of mortality among patients with sepsis was 2.4 (AOR, 2.43, 95% CI: 1.08–5.46) times greater than those without sepsis. Patients who were reintubated were 2.8 (AOR, 2.76, 95% CI: 1.06–7.21) times more likely to die than those non-reintubated patients, holding all other factors constant. However, using sedation decreased the odds of mortality by 59% (AOR, 0.41; 95% CI: 0.18–0.98) compared to patients not taking sedation, while other covariates remained the same (Table 4). The Hosmer–Lemeshow goodness of model fitness showed that the model is good-fitted at p = 0.4510 (Table 5). The mean VIF was found to be 2.28 (Table 6).

Since secondary data review was used, socio-economic, personal, nutritional status, and other socio-demographic characteristics were not explicitly included in this study. The effects of some important predictor parameters of mortality, such as APACHE, SOFA, and qSOFA, were not determined due to the inapplicability of the scores.