TOWARDS PREFERENCE-CONGRUENT CARE AT THE END OF LIFE: A MARKOV MODEL TO INFORM ADVANCE DIRECTIVE DECISIONS FOR PATIENT WITH COPD

Purpose: To facilitate preference-congruent care at the end of life using a Markov model to inform advance directive decision making in patients with COPD.

Method: We constructed a 5-state Markov model comparing QALYs for patients with severe COPD who endorse a “Full Code” (i.e., may use invasive mechanical ventilation) advance directive versus a “Do Not Intubate” (DNI, i.e., no invasive mechanical ventilation) advance directive. All patients were followed until death. Probabilities, utilities, and life expectancies were obtained from published estimates or expert opinion, and data were pooled where appropriate using the random effects method of Der Simonian and Laird. The health outcomes measured were: complications of mechanical ventilation; discharge location, including permanent institutionalization; survival; and QALYs. We considered different patient preferences regarding permanent institutionalization and complications of invasive mechanical ventilation using the Torrance time tradeoff method.

Result: For patients only willing to trade off minimal amounts of life expectancy to avoid complications of invasive mechanical ventilation or permanent institutionalization Full Code was the preferred advance directive (1.1 QALY, Full Code; vs. 1.0 QALY, DNI), resulting in a maximum increase in survival of 4% and an increase in cumulative reward of 0.1 QALYs. Being willing to trade off greater amounts of life expectancy to avoid complications of invasive mechanical ventilation or permanent institutionalization resulted in DNI being the preferred advance directive. For example, when willing to trade off 6 months of life expectancy DNI became the preferred strategy (0.98 QALY, DNI; vs. 0.80 QALY, Full Code). Changing preferences impacted Full Code (1.1 QALYs to 0.8 QALYs, a difference of 0.3 QALYs) more than DNI (1 QALY to 0.98, a difference of 0.02 QALYs). Our model was robust in sensitivity analysis for the non-preference variables with wide variability in outcomes found for only two variables: the probability of having a complication from mechanical ventilation and the probability of failing noninvasive mechanical ventilation.

Conclusion: For patients with severe COPD, endorsing a Full Code advance directive improved survival by 4% however this improvement in survival was offset by decreasing QALYs when patients had strong preferences against permanent institutionalization, or if setting-specific outcomes were poor. Future work will incorporate our model into a decision aid that elicits patient preferences and can be used at the point of care.