Concept Architecture
Concept
Theoretically, a Mixture Model is a statistical model that represents a population as a combination of two or more latent subpopulations, each described by its own probability distribution. It is founded on finite mixture theory, which assumes that observed heterogeneity arises because individuals belong to different unobserved groups. In health economics, mixture models are used to capture population heterogeneity, identify latent patient subgroups and improve the estimation of costs, health outcomes and survival.
Mathematically, a mixture model expresses the overall probability distribution as a weighted sum of component distributions, where the mixture weights represent the probability of membership in each latent subgroup. The component distributions may belong to the same or different distributional families, and both the mixing proportions and component parameters are estimated simultaneously. Parameter estimation is commonly performed using maximum likelihood estimation, frequently implemented through the Expectation-Maximisation (EM) algorithm.
In practice, mixture models are fitted to individual-level data to identify latent classes that cannot be directly observed. Model selection is based on likelihood-based criteria, information criteria and assessment of clinical plausibility. In health economics, mixture models are applied in latent class analysis, finite mixture regression, survival modelling, health preference modelling and cost-effectiveness analyses involving heterogeneous patient populations.
Purpose
Used to model heterogeneous populations by representing observed data as a combination of latent subpopulations, improving estimation, prediction and decision modelling in health economic analyses.
Mathematical Formulae
Primary Formula
f(x) = ????? �?f?(x)
where:
�? � 0
????? �? = 1
�? = mixing proportion for component k
f?(x) = probability density of component k
Supporting Formulae
L = ?? ????? �?f?(x?)
P(z? = k | x?) = (�?f?(x?)) � ????? �?f?(x?)
where:
L = likelihood function
P(z? = k | x?) = posterior probability that observation i belongs to component k
Related Mathematical Methods
- Finite Mixture Model
- Expectation-Maximisation (EM) Algorithm
- Maximum Likelihood Estimation
- Latent Class Analysis
- Mixture Cure Model
- Gaussian Mixture Model
- Model-Based Clustering
Example
Healthcare costs are analysed for 1,000 patients. A two-component mixture model identifies one subgroup with relatively low healthcare utilisation (70% of patients) and another with substantially higher resource use (30% of patients). The estimated mixing proportions are �? = 0.70 and �? = 0.30. The model provides a substantially better fit than a single-distribution model, reflecting the presence of distinct latent patient populations.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| SUMPRODUCT | =SUMPRODUCT(B2:C2,$F$2:$G$2) | Calculate weighted mixture probabilities. |
| EXP | =EXP(H2) | Evaluate component density functions during likelihood calculations. |
| LN | =LN(I2) | Calculate log-likelihood values. |
| Solver | Maximum likelihood optimisation | Estimate mixing proportions and component parameters. |
VBA (Optional)
Automate Expectation-Maximisation iterations to estimate mixture model parameters and compare alternative numbers of latent components.
Sources
- McLachlan GJ, Peel D. Finite Mixture Models. Wiley.
- Titterington DM, Smith AFM, Makov UE. Statistical Analysis of Finite Mixture Distributions.
- McLachlan G, Krishnan T. The EM Algorithm and Extensions.
- Everitt BS, Hand DJ. Finite Mixture Distributions.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
- NICE. Health Technology Evaluation Manual.
Related Concepts (2)
Library
Publications
1
NICE DSU Technical Support Document 15: Cost-effectiveness modelling using patient-level simulation — Davis, Stevenson, Tappenden & Wailoo, TSD 15 ed., 2014 (NICE Decision Support Unit (University of Sheffield))
Guidance on individual patient-level (microsimulation) cost-effectiveness modelling — when to use it in preference to cohort models, how to structure it, and how to handle the associated computational and uncertainty challenges.
Frequently Asked Questions (6)
What is a mixture model?
A statistical model representing a population as a combination of two or more distinct subgroups, each following its own underlying distribution.
Source: McLachlan & Peel 2000
Why use a mixture model when a population is not uniform?
When a population contains distinct subgroups that behave differently, a single distribution fitted to all of them describes none well, blurring the groups into an average that fits poorly. A mixture model instead supposes the data arise from two or more component distributions combined in unknown proportions, and estimates both the components and their weights. This lets it represent, for instance, a group at high risk alongside one at low risk without knowing in advance who belongs to which. It recovers hidden structure in the population. McLachlan and Peel (2000) describe mixture models.
Source: McLachlan & Peel 2000
How does a mixture model work?
A mixture model works by representing the overall distribution of the data as a weighted combination of component distributions, one for each subgroup, with weights giving the proportion of the population in each. The model estimates the component distributions and their weights, often without knowing which subgroup each observation belongs to, using methods such as the expectation-maximisation algorithm. This allows the model to identify the underlying subgroups and their distributions from data that mix them, capturing heterogeneity as a blend of components.
Source: McLachlan & Peel 2000
When are mixture models used?
Mixture models are used when a population is thought to comprise distinct subgroups with different distributions that are not directly identified, such as different disease subtypes, responder groups, or latent classes, and when the overall data show features, like multiple modes, suggesting a blend. They are also used to represent flexible distributions and to classify observations into components. In survival analysis, cure models are a form of mixture model, separating cured and uncured subgroups, so mixture models underlie various analyses of heterogeneous populations.
Source: McLachlan & Peel 2000
How do mixture models relate to cure models?
Mixture models relate to cure models in that a mixture cure model is a specific type of mixture model, representing the population as a mixture of a cured subgroup, with no residual disease risk, and an uncured subgroup following a survival distribution. The general mixture model framework, of combining component distributions with weights, underlies the cure model's structure. So cure models apply the mixture-model idea to survival data with a cured fraction, and the broader mixture-model theory supports their estimation and interpretation.
Source: Boag 1949
What are the challenges of mixture models?
Mixture models can be difficult to estimate, since the subgroup membership is usually unobserved, and the estimation may converge to poor solutions or be sensitive to starting values, with the number of components sometimes uncertain and hard to determine. Identifying the components and confirming they are real rather than artefacts requires care, and adding components risks overfitting. These challenges mean mixture models are fitted and interpreted carefully, with attention to whether the identified subgroups are meaningful and the number of components justified.
Source: McLachlan & Peel 2000
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 22 Oct 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/mixture-model
- Term code
- HE-EM-SM-053
Stable URI · Machine-readable · Resolvable · CC BY 4.0