An empirical comparison of Markov cohort modeling and discrete event simulation in a capacity-constrained health care setting

Abstract

The MM and DES-no-DQ ICER estimates differed due to the MM having implicit delays built into its structure as a result of having fixed cycle lengths, which are not a feature of DES. The non-DQ models assume that queues are at a steady state. Conversely, queues in the DES-DQ develop flexibly with supply and demand for resources, in this case, leading to different estimates of resource use and CE. The choice of MM or DES (with or without DQ) would not alter the reimbursement of OPSC as it was highly cost-effective compared to UC in all analyses. However, the modeling method may influence decisions where ICERs are closer to the CE acceptability threshold, or where capacity constraints and DQ are important features of the system. In these cases, DES-DQ would be the preferred modeling technique to avoid incorrect resource allocation decisions.

Discrete Event Simulation-Based Resource Modelling in Health Technology Assessment

“…RM performed at an aggregate level). Out of the 10 articles, 5 studies applied RM at an organisational level [19][20][21][22][23]] and 5 at a national level [24,[26][27][28][29].…”

“…The outcomes that were observed in these studies included total flow time [19,20,[27][28][29], waiting time [19][20][21][22][23][24][26][27][28][29], waiting probability [28], patient throughput [19-21, 23, 27, 28], resource utilisation [19-21, 23, 26, 28], queue length [21,23,24,28,29], probability of abandonment [28], and service level [28]. In one study observed [22], only one RM outcome was mentioned (waiting time); the remainder of these studies reported multiple outcomes.…”

“…7 of the 10 studies [19-21, 23, 24, 26, 27] used fixed constraint, by having identical staffing mixes between shifts and/or constant availability of resources. Meanwhile, the remaining 3 studies [22,28,29] assessed time variant constraints, by using different staffing mixes between shifts [28] and/or limiting service throughput to different number of patients accepted in different days of the week (e.g. using appointment scheduling [22]).…”

“…The majority of the reviewed studies [19][20][21][26][27][28][29] used simple capacity constraints. That is, these studies neglected the realistic details such as the effects of queuing or prioritising severity of illness.…”

“…The most common was assumption (n=9), followed by medical records (n=2) and patient logs (n=1). Only three models [22,23,29] combined multiple sources of data constraints, while the remaining focused on a single source. It should also be noted that out of the 5 national level models, only 1 [24] used national constraint data (e.g.…”