The study was conducted in the medical wards and intensive care units of two public teaching and referral hospitals, Tikur Anbessa Specialized Hospital (TASH) and Yekatit-12 Hospital Medical College, in Addis Ababa, Ethiopia. Currently caring for more than 400,000 patients annually, TASH is the largest referral teaching hospital in the nation. The hospital has 13 neurologists, residents, nurses, and other supportive staff. Yekatit 12 Hospital, with its medical college, is currently serving more than 500,000 patients per year and it has facilities for a range of services, including preventive and rehabilitative activities.

There was no acute stroke unit or protocol in place at either hospital during the study period. Both hospitals have an emergency room, medical wards, and adult intensive care units. The hospitals have echocardiography and carotid Doppler ultrasound. Some laboratory investigations (anticardiolipin antibody and lupus anticoagulant) and thrombectomy are lacking and patients were referred to private centers. The TASH has both computed tomography scan and magnetic resonance imaging but is usually not functional, though it used to work. Unless the patient has health insurance, all patients are expected to pay for all interventions or services they are receiving. Data were collected from December 1, 2020 to November 30, 2021.

A multicenter, incidence-based, prospective, cross-sectional cost-of-illness study was conducted among 99 adult acute stroke patients using a consecutive sampling technique in the two hospitals.

The study comprised adult patients (≥18 years old) whose diagnosis of stroke events (primary and recurrent) was verified by a computerized tomography (CT) scan and/or magnetic resonance imaging (MRI).

Patients were excluded from the study if they had at least one of the following conditions that developed during their hospital stay: cancer, fatal renal, hepatic, or patients with severe strokes that need mechanical ventilation, disabling and progressive neurological diseases like multiple sclerosis and Parkinson’s disease, and dementia.

A pretested, structured, interviewer-administered questionnaire adapted from previous cost-of-stroke studies was used to collect data from patients and their caregivers. Six experienced nurses conducted the interviews at the two hospitals. The questionnaire captured sociodemographic characteristics (age, sex, educational status, occupational status, and income level) and clinical characteristics, including co-morbidities and stroke type (ischemic or hemorrhagic; first-ever or recurrent). Stroke severity at admission was assessed using the National Institutes of Health Stroke Scale (NIHSS).

The questionnaire also included an economic module that captured both direct and indirect costs related to the index stroke. Cost data were recorded at three points: (1) during the index hospital admission, (2) at discharge, and (3) during all stroke-related health-care contacts occurring after discharge. We defined “subsequent visits” as any stroke-related contact with a health facility within 12 months of the index admission, including neurology or medical outpatient appointments, emergency department visits, and any stroke-related re-admissions at the study hospitals or other health facilities. Data on resource use during hospitalization and at subsequent visits were collected prospectively from patients and caregivers using the same structured questionnaire.

Cost information was collected prospectively from the time of the index hospitalization through 12 months of follow-up for each patient. We applied an incidence-based micro-costing approach from the perspective of patients and their families. Costs are reported in 2021 US dollars, using the average exchange rate during the study period (44.30 Ethiopian birr per 1 USD). The economic questionnaire was designed to capture only expenditures that patients or caregivers attributed to the stroke episode. For each resource item, we recorded the quantity of use at the individual level (e.g., number of hospital days, number and type of investigations, and quantities of medications) and multiplied these quantities by the corresponding unit costs. Total direct cost per patient was then calculated as the sum of all items.

Direct costs were defined as all medical and non-medical payments made by the patient or family to obtain care for the index stroke. In the study hospitals, unless the patient has health insurance, all services are paid out-of-pocket; therefore, the recorded direct costs predominantly represent out-of-pocket expenditures. We did not adjust costs for any subsequent insurance reimbursements. Direct costs were grouped into five categories:

For services delivered within the two study hospitals, unit costs for inpatient bed-days, investigations, and medications were obtained from hospital billing and finance offices and from hospital pharmacy price lists. For services obtained outside the study hospitals (private diagnostic centres, laboratories, imaging facilities, and community pharmacies), we used patient receipts when available. When receipts or exact amounts could not be produced, we collected current price lists from diagnostic centres, laboratories, and pharmacies in Addis Ababa and calculated the mean unit price for each investigation and medication. These mean prices were then applied as unit costs. If the patient or caregiver was unable to recall the precise amount paid for a given service, we assigned the corresponding average unit cost.

Direct cost estimates covered the entire 12-month period after the index stroke, not only the initial admission. Thus, in addition to costs accrued during the index hospitalization, we included post-discharge expenditures for stroke-related outpatient visits, emergency department attendances, additional investigations, medications, transportation for follow-up, and any stroke-related re-admissions within 1 year.

Indirect costs reflected productivity losses due to stroke for both patients and their main caregivers over the same 12-month period. We used the human capital approach: for employed patients and caregivers, each day of work lost because of hospitalization, outpatient visits, or caregiving responsibilities was valued at the self-reported average daily income from their main occupation. For unpaid caregivers, we used a replacement cost approach to value time spent providing informal care. Total indirect cost per patient was calculated as the sum of productivity losses for the patient and caregiver. When a patient or caregiver could not recall the precise income loss, we used the average value observed among other respondents. Supplementary Table 1 details investigation costs; Supplementary Table 2 presents individual-level cost components.

Data were collected using a structured, interviewer-administered paper questionnaire. The completed paper forms were entered into Microsoft Excel spreadsheets and then exported to SAS software, version 9.4 (TS1M9; SAS Institute Inc., Cary, NC, USA) for data management and analysis. All datasets and variables were checked and cleaned prior to analysis, including verification of ranges, internal consistency, and patterns of missingness.

Because the cost variables were right-skewed, continuous variables were summarized by median and inter-quartile range (IQR), and categorical variables as frequencies and percentages. We used the Mann–Whitney U test for comparing continuous or ordinal variables between two groups, and the Kruskal–Wallis test for more than two groups.

To identify factors associated with total acute stroke care cost per patient, we modeled total cost using a generalized linear model with a gamma distribution and log link, which is appropriate for strictly positive, right-skewed, and potentially heteroskedastic cost data. Candidate predictors were selected based on clinical relevance and bivariable analyses; variables with a p-value <0.10 in bivariable analyses were entered into the multivariable model, and statistical significance at p < 0.05 was used to inform the final model specification.

Model diagnostics based on Pearson and deviance residuals indicated adequate fit, with no apparent systematic patterns and most residuals falling within ±1.5 standardized units. Residual heteroskedasticity was explored using a Breusch–Pagan–type test applied to the deviance residuals. Multicollinearity among covariates was evaluated using variance inflation factors (VIFs). Overall model performance was summarized using deviance statistics and a Cox–Snell–type pseudo-R², which provides a descriptive measure of relative model fit rather than a direct analogue of the R² from linear regression.

Deterministic sensitivity analyses were conducted to assess the robustness of cost estimates. One-way analyses varied key cost drivers within empirically observed ranges, and scenario analyses evaluated low-, base-, and high-cost assumptions, including a conservative scenario accounting for the exclusion of mechanically ventilated patients. Costs were also examined under alternative currency conversion assumptions and expressed as a proportion of national GDP per capita.

Out of a total of 118 participants, 99 patients with complete data were included in the study, giving a response rate of 84%. The largest age group was 61–80 years, comprising 43% patients, followed by 31% patients aged 41–60 years. More than half of the patients were male (57.6%). Most patients resided in urban areas (77.8%), and the most common occupation category was private employment (42.4%), followed by unemployed (30.3%) and retired (16.2%). Among the variables examined, marital status (p = 0.0099) and area of residence (p = 0.0008) showed statistically significant differences in total cost per patient, whereas age group, sex, educational status, occupation, and religion were not significantly associated with total cost (Table 1).

The median length of hospital stay was 9 days (IQR 6–12). Only 14% patients arrived at the hospital within 4.5 h of symptom onset: 7% arrived within 3 h, 7% between 3 and 4.5 h, while the majority (85.9%) arrived after 4.5 h. Ischemic stroke was the predominant type, accounting for 67.7% patients, whereas 32.3% had hemorrhagic stroke. Previous stroke was reported in 9.1% patients, and 69.7% had at least one comorbid illness such as hypertension, diabetes mellitus, or dyslipidemia.

At admission, stroke severity based on NIHSS was: 26.3% minor (1–4), 46.5% moderate (5–15), 17.2% moderate-to-severe (16–20), and 10.1% severe (21–42). Non-parametric analysis of total cost per patient showed significant associations with onset-to-arrival time (p < 0.0001), stroke type (p = 0.002), presence of medical complications (p = 0.01), admission NIHSS category (p = 0.03), place of admission (general ward vs. ICU; p = 0.02), and presence of comorbid illness (p = 0.002). History of previous stroke was not significantly associated with total cost (p = 0.14) (Table 2).

In this study, the total direct costs of acute stroke care included the sum of costs for transportation, laboratory tests and imaging modalities, medications, inpatient services, and other healthcare facility visits for the current illness before reaching the study hospitals. The minimum and maximum direct costs are $55.7 and $844.2, respectively. The median direct cost of acute stroke care was $193.77 per patient; likewise, the median cost of investigations was $76.26 per patient during their course of care in the hospital. The median cost of inpatient service was $50.16 and the median cost of medication was $23.98 per patient during their course of care in the hospital. Neuroimaging and laboratory costs drove the majority of direct acute stroke care costs.

Ninety-seven (98%) patients underwent non-contrast brain CT scan imaging and two patients had MRIs from the beginning. The median cost of a non-contrast brain CT scan was $45.13, and 11 patients (11%) required additional brain MRI to settle their diagnosis. The median cost of a brain MRI was $90.25. The total median cost of brain imaging, including both CT scans and MRIs, was $45.13. The median laboratory cost was $31.7, which included a complete blood cell count, coagulation profile, liver function test, renal function test, serum electrolytes, erythrocyte sedimentation rate, c-reactive protein, viral markers, urine analysis, fasting blood sugar, random blood sugar, lipid profile, and cardiac biomarkers. The median cost of other investigations (ECG, echocardiography, chest X-ray, and carotid Doppler) was $2.71.

In this study, indirect costs were considered the monetary value of lost productivity due to the absence of patients from work and their caregivers during the acute stroke care period, so total indirect costs were the sum of indirect patient costs and indirect caregiver costs. The median indirect patient cost was $9.41 and that of the caregivers was $54.6. Total acute stroke care costs were the sum of direct and indirect stroke care costs. Accordingly, the median cost of acute stroke care in these hospitals was $286 per patient during their course of care in the hospital. Figure 1 demonstrates the components of the cost of acute stroke care. Based on stroke type, it was $361.37 for ischemic stroke and $263.67 for hemorrhagic stroke per patient during their course of care in the hospital (Table 3).

Higher between-group costs were independently associated with delayed hospital presentation, ICU admission, advanced age, comorbid conditions, and rural residence. Patients who arrived at the hospital more than 4.5 h after symptom onset had substantially higher expenditures than those presenting earlier, with mean total costs increased by about 78% (adjusted cost ratio [CR] 1.78, 95% CI: 1.35–2.35), highlighting delayed presentation as a major driver of resource use.

Admission to the ICU was also associated with higher expenditure (CR 1.56, 95% CI: 1.00–2.42; p = 0.049), consistent with the greater intensity and cost of intensive care. Older age contributed to increased costs: patients aged >60 years incurred approximately 27% higher mean costs than those aged ≤60 years (CR 1.27, 95% CI: 1.06–1.53). Similarly, having at least one comorbid condition was associated with a 27% higher mean cost compared with no comorbidity (CR 1.27, 95% CI: 1.03–1.56). Additionally, rural residence was associated with higher costs: rural residents had 40% higher mean expenses than urban patients (CR 1.40, 95% CI: 1.14–1.73), highlighting the extra economic burden on rural households (Table 4).

Deterministic sensitivity analyses showed that total cost estimates were robust across plausible variations in key cost drivers, with length of hospital stay and ICU admission remaining the main contributors to cost variability (Supplementary Table 3; Supplementary Figure 1).