Optimal dynamic empirical therapy in a health care facility: an artificial intelligence approach

Abstract

We propose a solution to the problem of finding an empirical therapy policy in a health care facility that minimizes the cumulative infected patient-days over a given time horizon. We assume that the parameters of the model are known and that when the policy is implemented, all patients receive the same treatment at a given time. We model the emergence and spread of antimicrobial resistance at the population level with the stochastic version of a compartmental model. The model features two drugs and the possibility of double resistance. Our solution method is a variant of the Monte-Carlo tree search algorithm. In our example, this method allows to reduce the cumulative infected patient-days over two years by 22% compared to the best standard therapy.

Surveillance based dynamic empirical therapy in a health care facility: an artificial intelligence approach

“…In Figure 2, we compare the performance of OPTIDYN over two years with that of the following policies: NONE, that consists in administering no treatment; COMBO, that consists in administering treatment 12; CYC-30, that consists in alternating 30 days of treatment 1 with 30 days of treatment 2; MONO-1 that consists in treatment 1 in monotherapy; and MONO-2, that consists in treatment 2 in monotherapy. Finally, we compare OPTIDYN with OPTI, a policy that consists in choosing a treatment regimen for two years with drug switches allowed every 30 days, based on no other information than the population dynamics parameters, as explained in [6]. 3 Without treatment (NONE), we obtain an average of 18,215.4 (95% CI: 18,126.2 -18,304.6) cumulative infected patient-days over two years.…”

“…We summarize the considered events in Table 1 and the parameters in Table 2. A detailed description and discussion of the model can be found in [6].…”

“…We solve this dynamic optimization problem with a variant of the Monte-Carlo tree search algorithm, a solution method borrowed from the field of artificial intelligence. See [6] for an introductory application of this method to empirical therapy optimization, and [7,8] for an application to chemotherapy regimen optimization. As for the underlying epidemiological model, our optimization method relies on the stochastic version of a compartmental model derived from those presented in [17].…”

Informed and uninformed empirical therapy policies

“…An empirical therapy policy specifies which antimicrobial is to be administered "by default" before the exact cause of an infection is known. 1 Empirical therapy policies aimed at reducing the evolution and spread of antimicrobial resistance include alternating drugs (drug cycling or metronomic therapies, either following a fixed rotation schedule or adaptively, see [14]), and assigning drugs randomly to patients (drug mixing).…”

“…in [21]). See also [15] for a dynamic optimization approach relying on periodic screening of the population, and a comparison with the expected performance of an optimization strategy [14] that does not use this information. Another source of uncertainty that is more difficult to control is the way policies are actually implemented in the field; see [4] for a stochastic differential equation approach to this problem.…”