Optimizing Vaccine Clinic Operations in Low and Middle Income Countries

This dissertation focuses on two open questions in operating vaccination clinics in low and middle income countries. First, as a result of limited ``open vial life," clinicians face difficult tradeoffs between opening a multi-dose vial to satisfy a potentially small immediate demand versus retaining the vial to satisfy a potentially large future demand. Second, in low and middle income countries, governmental health organizations face tradeoffs between locating (additional) clinics or conducting outreach trips to vaccinate patients in remote locations.\ud \ud To answer the first question, we formulate Markov decision process models that determine when to conserve vials as a function of the time of day, the current vial inventory, and the remaining clinic-days until the next replenishment. The objective is to minimize ``open-vial waste" while administering as many vaccinations as possible. For the base model, we analytically establish that the optimal policy is of a threshold type; conduct extensive sensitivity analysis on model parameters; develop a practical heuristic policy; suggest operational approaches that do not overly inconvenience patients; define metrics for determining appropriate operating hours and sessions per inventory replenishment cycle; and study the impact of random vial yield. We then generalize the base model by considering a positive probability of return for patients who are not vaccinated on their first visit and incorporating non-stationary arrival rates. To study these enhancements, we perform extensive numerical experiments at the clinic level for a single replenishment cycle and then extrapolate to multiple clinics and the entire world on an annual basis. The results indicate potential savings on the order of millions of doses per year.\ud \ud To answer the second question, given a network of population centers we develop a mixed integer linear programming model that determines clinic locations and outreach activities. The model minimizes cost over a specified period of time subject to constraints on coverage, trip distance, trip size, trip frequency and patient travel. We address demand uncertainty; perform sensitivity analysis on key model parameters; compare the performance of the optimal solution to heuristic policies; and conclude that while counterintuitive it is often suboptimal to locate clinics in the largest population centers