Strategic planning to reduce the burden of stroke among veterans: using simulation modeling to inform decision making

Abstract

Background and purpose: Reducing the burden of stroke is a priority for the Veterans Affairs Health System, reflected by the creation of the Veterans Affairs Stroke Quality Enhancement Research Initiative. To inform the initiative's strategic planning, we estimated the relative population-level impact and efficiency of distinct approaches to improving stroke care in the US Veteran population to inform policy and practice.

Methods: A System Dynamics stroke model of the Veteran population was constructed to evaluate the relative impact of 15 intervention scenarios including both broad and targeted primary and secondary prevention and acute care/rehabilitation on cumulative (20 years) outcomes including quality-adjusted life years (QALYs) gained, strokes prevented, stroke fatalities prevented, and the number-needed-to-treat per QALY gained.

Results: At the population level, a broad hypertension control effort yielded the largest increase in QALYs (35,517), followed by targeted prevention addressing hypertension and anticoagulation among Veterans with prior cardiovascular disease (27,856) and hypertension control among diabetics (23,100). Adjusting QALYs gained by the number of Veterans needed to treat, thrombolytic therapy with tissue-type plasminogen activator was most efficient, needing 3.1 Veterans to be treated per QALY gained. This was followed by rehabilitation (3.9) and targeted prevention addressing hypertension and anticoagulation among those with prior cardiovascular disease (5.1). Probabilistic sensitivity analysis showed that the ranking of interventions was robust to uncertainty in input parameter values.

Conclusions: Prevention strategies tend to have larger population impacts, though interventions targeting specific high-risk groups tend to be more efficient in terms of number-needed-to-treat per QALY gained.

Keywords: Veterans; comparative effectiveness research; computer simulation; health planning; stroke.

Figure 1

Depicted in the diagram are the stocks (solid rectangles) and flows (arrows), which capture the states and changes in health status of the Veteran enrollee population over time. The dashed rectangles show the descriptive segmentation of the Veteran population based on history of TIA or stroke. The flows in the model manipulate the transitions between stocks which shift individuals between states over time and ultimately affect modeled outcome variables. “VA users without prior TIA or stroke” are not tracked as a stock, but rather a flow into indicated stocks.

Figure 2

15 stroke intervention scenarios are defined, with each including a target subpopulation, current and projected level of care and estimated effectiveness of the intervention. Footnote:*Calibrated within the model; †Baseline (the comparator, “current level of care”); ‡ Workgroup consensus. SBP: Systolic Blood Pressure (measured in mmHg); DM: Diabetes Mellitus; HTN: Hypertension; AF: Atrial Fibrillation; TIA: Transient Ischemic Attack; ED: Emergency Department; tPA: Tissue plasminogen activator; TTR: Time in Therapeutic Range; RRR: Relative Risk Reduction; CEA: Carotid Endarterectomy; mRS: Modified Rankin Scale

Figure 3

The median number-needed-to-treat (NNT) per quality-adjusted life year (QALY) gained across 10,000 replications of the model is indicated by the line inside each box. The box spans the first to third quartiles, and the whiskers include the maximum and minimum values of NNT per QALY gained, excluding outliers. Outliers are depicted by solid circles. The vertical axis is on a logarithmic scale. Footnote:*P-value < 0.001 when Mann–Whitney U test is applied to examine whether NNT per QALY gained for each intervention is significantly different from each other intervention. NNT: numbers-needed-to-treat; QALY: quality-adjusted life year