Sequential drug decision problems in long-term medical conditions: a case Study of Primary Hypertension Eunju Kim ba, ma, msc

4) Other local search and problem-specific heuristic methods

86 studies, which were classified as other local search, employed a form of neighbourhood search within their own heuristic framework or together with other optimisation methods. Rather than examining all possibilities to the problem, local searches focused on searching a neighbourhood of particular solutions by introducing a mechanism to define a neighbourhood structure and to evaluate the costs or benefits of moving from the current solution to others within the same neighbourhood.

290 studies were classified as the problem-specific heuristic category because the heuristic methods cannot be classified as any of the categories discussed earlier or the author(s) did not clearly state the specific information about the type of heuristic methods. Most problem-specific heuristics implied a specific heuristic approach to a particular problem based on the given prior information in that area, whereas meta-heuristics is a set of heuristic concepts that can be used to a wide set of different problems.

5) Sequential decision problems in healthcare

Among the 512 identified studies, 19 studies addressed healthcare-related sequential decision problems. There were six studies associated with identifying an optimal treatment strategy in clinical practice^{[59, 60, 130, 132, 187, 188]}, two studies associated with dynamic treatment regimen in clinical trials^{[189, 190]}, two studies associated with screening strategies^{[191, 192]}, two studies associated with public health policies to control outbreaks of infectious disease^{[58, 193]}, four studies associated with the optimal diagnosis by medical imaging or diagnostic device^[194-197] and two studies were associated with hospital management^{[198, 199]}. The dynamic decision process with patient’s stochastic response to the treatment decision was depicted in partially observable MDP^[59], dynamic limited-memory influence diagrams^[60] and probabilistic Boolean Networks^[187]. For the optimisation procedure, most of them used DP where patient level data was available.