Using system dynamics modelling to estimate the costs of relaxing health system constraints: a case study of tuberculosis prevention and control interventions in South Africa

Health system constraints are increasingly recognized as an important addition to model-based analyses of disease control interventions, as they affect achievable impact and scale. Enabling activities implemented alongside interventions to relax constraints and reach the intended coverage may incur additional costs, which should be considered in priority setting decisions. We explore the use of group model building, a participatory system dynamics modelling technique, for eliciting information from key stakeholders on the constraints that apply to tuberculosis infection prevention and control processes within primary healthcare clinics in South Africa. This information was used to design feasible interventions, including the necessary enablers to relax existing constraints. Intervention and enabler costs were then calculated at two clinics in KwaZulu-Natal using input prices and quantities from the published literature and local suppliers. Among the proposed interventions, the most inexpensive was retrofitting buildings to improve ventilation (US$1644 per year), followed by maximizing the use of community sites for medication collection among stable patients on antiretroviral therapy (ART; US$3753) and introducing appointments systems to reduce crowding (US$9302). Enablers identified included enhanced staff training, supervision and patient engagement activities to support behaviour change and local ownership. Several of the enablers identified by the stakeholders, such as obtaining building permissions or improving information flow between levels of the health systems, were not amenable to costing. Despite this limitation, an approach to costing rooted in system dynamics modelling can be successfully applied in economic evaluations to more accurately estimate the ‘real world’ opportunity cost of intervention options. Further empirical research applying this approach to different intervention types (e.g. new preventive technologies or diagnostics) may identify interventions that are not cost-effective in specific contexts based on the size of the required investment in enablers.