Study design
This is a retrospective study in women admitted to the HDU of Princess Christian Maternity Hospital (PCMH), Freetown, Sierra Leone, from 2nd October 2017 to 2nd October 2018. The study protocol was approved by the Sierra Leone Ethics and Scientific Review Committee on the 18th December 2018. The need for informed consent was waived. The study was registered on ClinicalTrials.gov (study identifier NCT04121234).
Study setting
With 129 beds, the PCMH is the largest maternity referral hospital in Sierra Leone with a reference population of 1.5 million inhabitants. PCMH is a primarily obstetric institution with approximately 9,000 admissions and 6,500 deliveries per year (9,10). One–third of the parturients develop major obstetric emergencies, including peripartum haemorrhage, sepsis and pre–eclampsia (9,10). Operating theatre facilities are essential, anaesthetic service is available with a majority of spinal anaesthesia procedures.
The HDU is a 4-bed medium care unit, with a basic model of intermediate intensive care defined on few essential criteria such as a high nurse/patient ratio, close monitoring of vital signs, personalized intravenous fluid and vasopressor therapy management, a rational use of oxygen and a very basic point of care laboratory (11). Electricity and clean water are continuously available, while oxygen is generated through bedside oxygen concentrators with a maximal output of 10 l/min and maximal purity of 96%. No mechanical ventilators or dialysis apparatus are available in the unit or hospital. A basic neonatal ICU is available as a separate entity.
Patients
Patients were eligible if (1) admitted to the HDU, (2) being pregnant or within 42 days after termination of pregnancy (12), (3) between 2nd October 2017 and 2nd October 2018. No exclusion criteria were used.
Study endpoints
The primary endpoint was the association between several potentially modifiable and non–modifiable factors, as standardly collected in the patient chart of each patient, and mortality during HDU stay. Secondary endpoints were demographic characteristics, reasons of admission, obstetric early warning score (OEWS) at admission, treatments received, vital parameters on presentation, admission time (night/day or weekend/working days), length of stay in HDU, discharge destination and hospital mortality.
Data collection
A set of predefined variables was assessed at hospital admission, HDU admission and at discharge from HDU. The primary data source was the HDU patient chart, with data crosschecked with the hospital patient charts and the HDU admission book for quality control purpose. Data on hospital deliveries, admissions and mortality were extracted from the hospital register and the maternal mortality hospital database. Data were retrospectively collected by a study physician (CM) and included: patient demographics; admission date and source; main reason for admission in hospital (classified as by the WHO handbook on Monitoring emergency obstetric care (12) as: complication of abortion; ante-partum haemorrhage (APH), ectopic pregnancy, obstructed labour, postpartum haemorrhage (PPH), pre-Eclampsia(PE)/eclampsia, puerperal Sepsis, uterine rupture (UR), others); main reason for admission to the HDU were classified at the source as: haemodynamic instability, sepsis, haemorrhage, acute renal failure, neurological impairment, respiratory distress, severe malaria, coagulopathy or other diagnoses. These were standardized diagnoses based on the clinical assessment of the attending physician and thus not based on strict research case definitions.
Vital signs and treatments collected at admission included body temperature, heart rate, respiratory rate, neurological status according to the AVPU scale, systolic and diastolic blood pressure, transcutaneous peripheral saturation (SpO2). The ratio between SpO2 and fraction of inspired oxygen (SpO2/FiO2) was computed to better assess oxygenation in patients receiving oxygen. A modified obstetric modified early warning score (OEWS) was also computed from vital parameters (13).
Specific treatments received at any point during HDU stay were extracted from the patient file and included: oxygen supplementation, use of vasopressors, whole blood transfusions, antibiotic therapy, use of magnesium sulphate for eclamptic seizures prevention and use of hydralazine for antihypertensive purposes. Point of care laboratory parameters such as capillary lactates levels and haemoglobin (Hb) levels were collected when available. Date and time of hospital admission was collected whenever available. Length of stay in HDU (LOS) and patient outcomes (classified as death in HDU, discharge to ward, or transfer to other facility) were reported at discharge.
Definitions
The modified obstetric early warning score (OEWS) uses core physiological parameters such as systolic (SBP) and diastolic blood pressure (DBP), heart rate (HR), respiratory rate (RR) and temperature to identify deteriorating obstetric patients demanding extra attention (13). The OEWS used in this study, and in the hospital at the time of the data collection, ranges from 0 to 10 points, with a green colour code granted with a total OEWS of 1–2, a yellow colour code for total OEWS of 3 to 5, and a red colour code attributed to any patient with a OEWS above 5, or any danger sign (any among a SBP>160 mm Hg or <80 mm Hg, a DBP of > 110 mm Hg, a HR<60 or >120, RR >31 or <10 and a body temperature >38.5 °C or < 35°C). Patients were stratified according to severity on HDU admission in two groups i.e. with patients having a red OEWS in one group and patients having a green or yellow OEWS in the less severe group.
The AVPU is a simplification of the Glasgow Coma Scale useful to rapidly grade a patient’s gross level of mental status based on four criteria (Alert, Verbally responsive, responsive to Painful stimulus, Unresponsive). The ratio between SpO2 and fraction of inspired oxygen (FiO2) was computed by deriving the FiO2 from the oxygen therapy in litres by the following formula FiO2 = 0.21 + O2 (l/min) x 0.03 (14). Death during HDU stay was defined as death occurred from referral to HDU to discharge from HDU. Hospital mortality was defined as death occurring during the whole hospital stay. Time from admission to hospital to referral to HDU was computed from the relevant variables whenever these where available.
Statistical analysis
No formal sample size was calculated a priori, and the study included all women admitted during the first year of activity of the HDU. Continuous data were expressed as median and interquartile range (IQR), and categorical data as number and percentage. Mortality rate, AVPU classification and OEWS score were compared among reasons for referral to HDU using Chi Square test. The proportion of patients admitted with a red code per HDU bed per week was calculated by dividing the number of patients with red code per HDU bed over the study period. Patients with a red code at the admission were compared with the ones presenting with a yellow or green code.
The association between mortality and each clinically relevant variable was explored using logistic regression models (unadjusted analysis). Independent risk factors for mortality were investigated using a logistic regression model including a set of candidate predictors at admission (adjusted analysis). The limited number of deaths restricted the number of candidate predictors that could be included in the first stage of model selection. Owing to collinearity with OEWS, some variables (temperature, heart rate, respiratory rate, SpO2/FiO2 ratio, systolic and diastolic blood pressures) were not included in the model (15). Other variables (age, Hb, admission during night and weekend admission) were not included in the model according to unadjusted analysis of mortality. The final model included source of admission, OEWS, SpO2, AVPU, use of oxygen and use of vasopressors at admission. Model selection was performed by Akaike’s information criterion reduction. Effect sizes were presented as odds ratios (OR) with 95% confidence intervals (CI).
All analyses were 2-sided and a p-value less than 0.05 was considered statistically significant. Statistical analysis was performed using R 3.5 (R Foundation for Statistical Computing, Vienna, Austria) (16).