Patient journey and care pathway analysis

Patient journey analysis is needed when the question is where a cohort of patients moves and where it is lost, not just what state it is in today. It applies when you already know how many patients are in each situation but not in what order they move between them, with what probability, or at which point on the path they abandon treatment, progress or drop out of follow-up. A cross-sectional snapshot describes the present; the trajectory reconstructs movement between states over time. The difference is the one between knowing that a percentage of patients abandon treatment and knowing at which specific transition they do so. Locating that point is what turns an aggregate figure into a decision about where to intervene.

You need an activity log with dated events per patient: an identifier, the milestones or states the patient passes through (diagnosis, visits, treatment, discharge, clinical events) and the date of each one. The client provides these records; the service defines the set of states and admissible transitions using your clinical judgement, reconstructs each patient's sequence and cleans the timing of the dates, on which the model depends entirely. The data do not need to arrive as an ordered sequence, but the events do need to be traceable over time and attributable to the same person or unit of follow-up. What determines the scope is temporal granularity: if the dates are imprecise or the states cannot be told apart in the data, the model stays at a coarser level.

Multistate models and Markov models belong to the family of state and transition models, which represent a trajectory as a process between discrete states. They apply when you want to quantify how likely a patient is to move from one state to another and how long they stay in each one. A Markov model assumes that the transition depends only on the present state, not on the path taken so far. Multistate models are a broader framework that allows Markov, semi-Markov or non-Markov formulations and the effect of covariates on each transition, as well as structures such as the illness-death model. The choice depends on the problem, for example on whether prior history matters, whether time is continuous or discrete, or whether there are absorbing states, and it is justified explicitly rather than assumed by default. The Markov assumption is a simplification that is tested, not a property taken for granted.

Sequence analysis and process mining reconstruct the trajectory a cohort actually followed from the event log, instead of starting from a theoretical path. They apply when you want to see actual practice, the real order of the steps, and compare it with the trajectory that a guideline or protocol expects, to locate where the two diverge. They make it possible to discover frequent paths, quantify how closely the cohort follows the expected path and detect the deviations worth analysing. The limit is an honest one: the method describes and compares observed trajectories; it does not by itself establish that a deviation causes a worse outcome, which requires a separate inference design. What it delivers is a map of how patients actually move through care, not a causal judgement on each deviation.

Patient journey analysis and HEOR can use the same technique, Markov models of health states, but they answer different questions depending on what the model is for. Patient journey analysis fits when the question is clinical or behavioural: where the cohort moves, at which transition it is lost and where to intervene. HEOR fits when the question is economic: whether an intervention offers value compared with its alternative, with the Markov model serving to calculate QALYs, DALYs and ICERs. The journey ends in a map of the trajectory and its critical transitions; HEOR ends in a cost-effectiveness ratio for a funding decision. The two can be combined, since the same state model can later feed an economic analysis, but one seeks to understand the patient's path and the other chooses between options by their value.

Patient journey analysis and patient flow optimisation can share the same kind of state model, but they look at different subjects. Patient journey analysis looks at the patient's clinical or behavioural trajectory: how their disease progresses, how their adherence evolves, at which point in their care they are lost. Patient flow optimisation looks at the operational process of the system: capacity, bottlenecks, bed management and waiting times, with techniques such as queueing theory and discrete-event simulation. The distinction lies in the objective: the journey seeks to understand the patient and their path; patient flow, to optimise hospital operations. When the same admission–ward–discharge path is studied in terms of clinical outcomes, adherence or loss to follow-up, it belongs to patient journey analysis; when it is studied to size resources, it belongs to patient flow optimisation. If the decision is about care, it is the patient journey; if it is about capacity, it is patient flow.

No: stratification, risk prediction and patient journey analysis answer three different questions about the same population. Patient stratification models give a static snapshot of which patient subgroups exist, using clustering and classification. Clinical risk and readmission prediction estimates an individual event: the probability that this particular patient will deteriorate or be readmitted. Patient journey analysis models the dynamics of the cohort across states: how it moves, in what order and with what probability it passes from one state to another over time. The difference is between a snapshot, an individual prognosis and the movement of the cohort. They complement each other naturally, since a stratification can define the starting states of a trajectory, but none of them replaces the others. If the question includes the words "when does it pass from one state to another?", it is patient journey analysis.

You receive a map of the cohort's trajectory: the set of states and transitions, the estimated transition probabilities and times, a visualisation of the flow, usually with Sankey diagrams and state occupation curves, and the identification of the critical transitions where attrition is concentrated. The service reconstructs and models the trajectory and indicates where an intervention could have the greatest impact, according to the observed dynamics and the model's assumptions; the clinical or management decision on which intervention to apply, and where, is made by the client, with their knowledge of the context. You do not receive a closed intervention recommendation, but the quantitative evidence on the cohort's dynamics that supports that decision. The deliverable is legible to a clinical or management committee, not just to an analysis team.