Abstract
{ "background": "District hospitals are critical nodes in sub-Saharan African health systems, yet robust evidence on the impact of systemic interventions on clinical outcomes remains scarce. Evaluations often lack rigorous counterfactual designs, limiting causal inference.", "purpose and objectives": "This study aimed to quantify the causal effect of a bundled health systems intervention—comprising staff training, supply chain reinforcement, and clinical protocol standardisation—on inpatient mortality and length of stay in Ethiopian district hospitals.", "methodology": "We employed a quasi-experimental difference-in-differences design, analysing longitudinal patient administration data from 24 intervention and 18 control hospitals. The primary model was specified as $Y{it} = \\beta0 + \\beta1 (Interventioni \\times Postt) + \\beta2 X{it} + \\alphai + \\gammat + \\epsilon{it}$, where $Y_{it}$ is the outcome for hospital $i$ in period $t$. Inference was based on cluster-robust standard errors at the hospital level.", "findings": "The intervention was associated with a statistically significant reduction in risk-adjusted inpatient mortality of 3.2 percentage points (95% CI: -5.1, -1.3). No significant effect was observed for average length of stay. The mortality reduction was most pronounced for medical, as opposed to surgical, admissions.", "conclusion": "The bundled systems intervention demonstrated a clinically meaningful, causal effect on reducing inpatient mortality in a resource-constrained hospital setting.", "recommendations": "Policy should prioritise integrated systems strengthening over single-component reforms. Future scale-up should tailor support to medical departments and ensure sustained investment in supply chains and supervision.", "key words": "health systems strengthening, quasi-experimental design, causal inference, inpatient mortality, sub-Saharan Africa", "contribution statement": "This paper provides novel empirical evidence on the mortality impact of a complex health system intervention in a low-resource setting, employing a robust causal inference framework rarely applied