Study design
This was a prospective observational cohort study conducted between December 2018 and April 2019.
Study participants and settings
The study was a hospital based at the Mulago National Referral Hospital (NRH) (https://health.go.ug/content/mulago-national-referral-hospital), which is also the teaching hospital for the School of Medicine, College of Health Sciences-Makerere University (https://www.mak.ac.ug/). The hospital is located in the Kampala Capital City of Uganda. Eligible adult patients who presented to accident and emergency (A&E) department with CT confirmed extra-axial hematoma were recruited consecutively. This tertiary hospital has a formal bed capacity of 1500 and bed occupancy rate of between 180 and 200%. However, due to the renovations and upgrade of the hospital at the time of data collection, the number of beds at casualty and neurosurgical wards were 35 and 42, respectively. The 42 beds of the neurosurgical ward are inclusive of 8 high dependence unit (HDU) beds. The neurosurgical patients who require intensive care have access to the general hospital intensive care unit (ICU). The ICU has 8 bed capacities with an average of 25 patients per month. If there is no free bed, the patients have to wait at the A&E, neurosurgical ward or HDU. At the time of this study, the hospital had access to two functional computed tomographic (CT) scanners: one in a private hospital about 1.6 km from the casualty and the second one in a public hospital that is 15 km away from the casualty. Both scanners require an out-of-pocket payment of about 70 USD, since majority of patients attending MNRH have no medical insurance.
Study procedure
The participants were recruited after review by a multidisciplinary team following interpretation of a brain CT image to characterize the nature of TBI. In general, patients at Mulago National Referral Hospital are received and triaged by the medical team on duty at the accident and emergence department. The first contact clinician is usually a general doctor who then consults a general surgery or neurosurgery resident, general surgeon, trauma surgeon, neurosurgeon, radiologist, and or maxillofacial and plastic surgeons when there is a need. The team routinely carries out several ward rounds in a day at the accident and emergence to determine if there is a need to amend the initial treatment decisions. The recruitment process and flow of participants is summarized in Fig. 1
Inclusion criteria
We included adults with clinical-radiological diagnosis of ASDH and or AEDH based on head and brain CT imaging, following trauma, who presented at Mulago NRH during the data collection period. Whereas ASDH and AEDH are seemingly two different pathologies, they have several aspects in common that were of interest to this study. For example, both hematomas are extra-axial and surgical evacuation is thus the treatment of choice [26]. Also, once surgery is indicated, the duration from decision to actual surgical intervention is the most important determinant of outcome for both hematomas [13, 22]. In this regard, the authors desired to recruit a reasonable sample size of patients in the limited approved duration for which surgical decompression is the major determinant of outcome. However, to avoid loss of any information, we report the overall outcomes alongside the individual outcomes for each hematoma.
The hospital did not have intracranial pressure monitoring facilities by the time of the study. However, patients with suspected cerebral edema based on brain CT scan findings were included as long as they had ASDH or AEDH. The treatment offered for cerebral edema involved hypertonic saline or mannitol with subsequent clinical radiological assessment. Also, participants with cranial and skull base fractures who concurrently sustained ASDH or AEDH were included in the study. This is because in practice, it is not uncommon for more than one category of TBI to occur in the same patient [12], more so with ASDH. Thus, patients with cranial and basilar fracture comorbidities were included in order to report an outcome that is a true reflection of the practical clinical picture.
Exclusion criteria
Patients with acute on chronic subdural hematoma were excluded because this entity has a variable pathogenesis. Although such hematoma may be related to trauma, theoretically the presence of aberrant friable blood vessels and a localized bleeding disorder that trigger hemorrhage even with trivial trauma cannot be ignored. Acute on chronic subdural hematoma may occur in the absence of trauma [27]. In some cases, the trauma may be as trivial as a slap [16]. This aetiology does not match that of ASDH or AEDH, where the severity of injury corresponds to the gravity of trauma involved [8]. On the other hand, acute on chronic subdural hematoma may take several weeks before a diagnosis is made. In such cases, the date of onset becomes difficult to define which is not in line with our study design, assessing the outcomes 30 days from onset of trauma. Also, patients with cerebral contusions, cerebral edema, cranial, or basilar fractures but without ASDH or AEDH were beyond the scope of this study.
Sample size estimation
For objective one, aimed at determining the proportion of participants with favorable outcome, Kish Leslie formula was used.
$$ \mathrm{ns}={\mathrm{Z}\upalpha}^2\ast \frac{\mathrm{P}\ \left(1-\mathrm{P}\right)}{\updelta^2} $$
where ns is the sample size, p is the expected proportion of patients with functional recovery outcome assumed to be 0.5, (1-P) is the proportion of patients with non-functional recovery outcome assumed to be 0.5, and Zα is the standard normal value corresponding to set level of confidence = 1.96 whereas δ = degree of accuracy = 0.05,
[(1.96)2 × (0.5 × 0.5)]/(0.05 × 0.05). Thus, ns = 384.
Adjusting for finite population
$$ \mathrm{Sample}\kern0.17em \mathrm{size}\ (N)=\frac{\mathrm{ns}}{1+\frac{\mathrm{ns}-1}{n}} $$
where N is the adjusted population size, ns is the estimated sample size, and n is the population under study = 125 (based on the hospital data registry).
$$ {\displaystyle \begin{array}{c}N=\frac{384}{1+\frac{383}{125}}\\ {}N=94\end{array}} $$
The second study objective aimed at determining factors associated with favorable outcome. Thus, the formula for sample size of two proportions in cohort studies was deemed suitable because it takes into account of binary outcome variables.
$$ N=\frac{{\left[{Z}_{\alpha /2}\sqrt{p\left(1-p\right)\left(\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${q}_1$}\right.+\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${q}_2$}\right.\right)}+{Z}_{\beta}\sqrt{p_1\left(1-{p}_1\right)\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${q}_1$}\right.+{p}_2\left(1-{p}_2\right)\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${q}_2$}\right.}\right]}^2}{{\left({p}_1-{p}_2\right)}^2} $$
N = minimum required sample size, Zα/2 = 1.96, the standard normal value corresponding to 95% confidence level, Zβ = 0.84, the standard normal value corresponding to 80% power, q1 = 0.847, proportion of patients who recovered with favorable outcome with a fall as the cause of the ASDH [3], p1 = 0.50, proportion of patients who recovered with unfavorable outcome with a fall as the cause of the ASDH [3], q2 = 0.153, proportion of patients who recovered with favorable outcome with road traffic crush (RTC) as the cause of ASDH [3], p2 = 0.769, proportion of patients who recovered with unfavorable outcome with RTC as the cause of ASDH [3]; p = p1q1 + p2q2 = 0.541.
Therefore, the minimum sample size required to answer objective 2 was 192 patients. We adjusted our sample size to take into account a 10% loss to follow-up as below, where n = requires sample size, n1 = 192.
$$ \mathrm{ns}={n}_1\ast \frac{1}{\left(1-10\%\right)}=192\ast \frac{1}{\left(1-0.1\right)}=213 $$
Adjusting the sample size for a finite population of 125 patients that were expected to be admitted during 5-month study period,
$$ (N)=\frac{\mathrm{ns}}{1+\frac{\mathrm{ns}-1}{n}} $$
where ns = 213, n = 125; thus, N = 79 participants. The larger sample size of 94 was considered.
Sampling procedure
Patient who met the inclusion criteria were consecutively recruited until the estimated minimum sample size was attained. This was intended to attain a sample size large enough for validity of the study.
Study variables
Our data tool captured independent variables including socio-demographic (age, sex, occupation, commute distance from nearest health facility) and clinical (Glasgow Coma Score (GCS) at admission-assessed at casualty after the initial resuscitation, post-operative care area—whether the patient was admitted to the general ICU or neurosurgical ward after surgery). In addition, we captured dependent variables such as disposition at 30 days after admission—either discharged, died, or still admitted; Glasgow Outcome Scale (GOS) at 30 days after injury—assessed in outpatient clinic or on phone interview with the patients and/or care takers and recorded as 1 = death—confirmed brain dead; 2 = persistent vegetative state—severe damage with prolonged state of unresponsiveness and a lack of higher mental functions; 3 = severe disability—severe injury with permanent need for help with activities of daily living; 4 = moderate disability—no need for assistance in everyday life, employment is possible but may require special equipment; and 5 = good recovery—light damage with minor neurological and psychological deficits. We dichotomized the GOS at two levels as (a) favorable outcome (4 or 5) and (b) unfavorable outcome (1, 2, or 3). The scope of this observational study was limited to these clinical variables that are routinely assessed during the care of patients with TBI.
Data analysis
Data was entered into EpiData software (Christiansen TB & Lauritsen JM (version 2.0.8.56) EpiData-Comprehensive Data Management and Basic Statistical Analysis System. Odense Denmark, EpiData Association, 2010) and exported to Stata software version 14 (StataCorp. 2015. Stata Statistical Software: Release 14. College Station, TX: StataCorp LP) for cleaning and analysis.
We summarize the participants’ socio-demographic and clinical categorical baseline characteristics using frequencies and percentages in tables. The mean and standard deviation were used for continuous participant characteristics that were normally distributed; otherwise, the median and inter-quartile range were used.
We used chi-square test (X2) to compare the patients in care and those lost to follow-up, for categorical variables, whereas the student’s t test was used for continuous variables.
We used the modified Poisson regression (with robust standard errors) model to determine the socio-demographic and clinical factors that influence favorable outcomes. At bivariate analysis, factors with p value less than 0.2 were assumed to be important and considered for multivariate analysis. At multivariate level, interaction and confounding were assessed in the model before the reaching the final adjusted model. The variables with p < 0.05, in the final model, were considered to be of statistical significance at 95% confidence interval.