Markov Cohort Model

A Markov model tracking a hypothetical cohort collectively as proportions across health states over cycles, rather than simulating each patient separately.

Source: Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Medical Decision Making. 1993;13(4):322-338. doi:10.1177/0272989X9301300409.

A Markov cohort model is a Markov model that tracks a hypothetical cohort collectively, as proportions of the cohort across the health states over cycles, rather than simulating each patient separately. At each cycle, the proportions are updated by the transition probabilities, and costs and effects accrue according to where the cohort is. The Markov cohort model is the standard and simplest form of Markov model in health economics, giving the expected experience of the population efficiently and transparently.

Source: Sonnenberg & Beck 1993

A Markov cohort model reports how a whole group is distributed across health states at each cycle and, from that, the totals accumulated over the run. Multiplying the proportion in each state by the cost and health value attached to it, and summing across states and cycles, gives the cohort's expected lifetime cost and quality-adjusted life years. The output is an average for the group rather than a result for any individual. This average is what feeds the cost-effectiveness comparison. Briggs and colleagues (2006) describe these outputs.

Source: Briggs et al. 2006

A Markov cohort model works by distributing the cohort across the health states, then at each cycle applying the transition probabilities to move proportions between states, updating the distribution, and accruing costs and health effects according to the proportions in each state. This is repeated over the cycles of the time horizon, and the accumulated costs and effects give the cohort's expected outcomes. The model follows the shares of the cohort in each state over time rather than individual patients.

Source: Sonnenberg & Beck 1993

A Markov cohort model is simple, transparent, and computationally efficient, since it tracks proportions using average transition probabilities rather than simulating many individuals, and it gives a deterministic result for given inputs, without the random variation of individual simulation. It is easy to build, check, and communicate, and it supports extensive sensitivity analysis quickly. For problems where outcomes depend on average effects and individual heterogeneity and history do not much matter, it provides valid results efficiently, which is why it is widely used.

Source: Sonnenberg & Beck 1993

A Markov cohort model cannot easily represent individual heterogeneity, patient history beyond the current state, or interactions between individuals, because it follows the cohort by average rates and the Markov assumption gives it no memory. Where future risks depend on past events, capturing this requires adding states, which can multiply their number, or moving to individual simulation. Where such individual-level features materially affect outcomes, a cohort model is inadequate, so its limitations stem from tracking proportions rather than individuals.

Source: Sonnenberg & Beck 1993

A Markov cohort model tracks the whole cohort as proportions across states using average transition probabilities, giving a deterministic result, whereas Markov microsimulation follows individual patients one at a time, each moving through states with their own randomly determined history. The cohort model is simpler and faster but cannot capture heterogeneity or history beyond the current state, while microsimulation can, at greater computational cost and with random variation. The choice depends on whether individual-level detail materially affects the results.

Source: Sonnenberg & Beck 1993