Concept Architecture
Concept
Theoretically, Finite Mixture Model is a statistical model that represents a population as a weighted combination of two or more distinct probability distributions. It is founded on mixture distribution theory and latent class modelling, where observations are assumed to arise from unobserved subpopulations with different parameter values. In health economics, finite mixture models are used to model population heterogeneity, identify latent patient groups and improve the estimation of costs, outcomes and healthcare utilisation.
Mathematically, a finite mixture model expresses the overall probability distribution as a convex combination of component distributions, each weighted by its mixing proportion. Model parameters include the component-specific parameters and the mixing probabilities, which sum to one. Parameters are typically estimated using maximum likelihood estimation implemented through the Expectation-Maximisation (EM) algorithm.
In practice, finite mixture models are fitted by specifying the number of latent components, selecting appropriate probability distributions and estimating mixing proportions and component parameters. They are applied in health economics to model skewed cost distributions, treatment response heterogeneity, disease progression and latent patient subgroups.
Purpose
Used to model heterogeneous populations, identify latent subgroups, estimate mixture distributions and improve statistical modelling of complex health economic data.
Mathematical Formulae
Primary Formula
f(x) = ????? �? f?(x | ??)
where:
- f(x) = overall probability density
- K = number of mixture components
- �? = mixing proportion for component k
- f?(x | ??) = probability density of component k
- ?? = parameters of component k
subject to:
????? �? = 1
Supporting Formulae
Posterior probability of component membership:
P(k | x) = [�? f?(x | ??)] � [????? �? f?(x | ??)]
Log-likelihood:
? = ?? ln[????? �? f?(x? | ??)]
Related Mathematical Methods
- Expectation-Maximisation Algorithm
- Maximum Likelihood Estimation
- Latent Class Analysis
- Gaussian Mixture Models
- Bayesian Mixture Models
- Model-Based Clustering
- Information Criteria (AIC, BIC)
Example
A health economist analyses annual healthcare costs for 5,000 patients with diabetes. A two-component finite mixture model identifies one subgroup with relatively low routine management costs and another with substantially higher complication-related costs. The estimated mixing proportions are 0.72 and 0.28, indicating that approximately 28% of patients belong to the high-cost latent subgroup, supporting targeted resource allocation analyses.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| SUMPRODUCT | =SUMPRODUCT(B2:B4,C2:C4) | Calculate weighted mixture probabilities |
| LN | =LN(A2) | Compute log-likelihood contributions |
| EXP | =EXP(A2) | Calculate probability density components |
| Solver | Optimise log-likelihood | Estimate mixing proportions and component parameters using maximum likelihood |
VBA (Optional)
Automate finite mixture model estimation by iteratively implementing the Expectation-Maximisation algorithm and exporting parameter estimates and subgroup classifications.
Sources
- McLachlan G, Peel D. Finite Mixture Models.
- McLachlan GJ, Krishnan T. The EM Algorithm and Extensions.
- Titterington DM, Smith AFM, Makov UE. Statistical Analysis of Finite Mixture Distributions.
- Briggs AH, Claxton K, Sculpher MJ. Decision Modelling for Health Economic Evaluation.
- McLachlan GJ. Mixture Models in Statistics.
Related Concepts (2)
Library
Publications
1
Bayesian Methods in Health Economics — Gianluca Baio, 1st Edition ed., 2012 (Chapman & Hall / CRC Press)
An overview of Bayesian statistical methods for the analysis of health economic data, covering economic evaluation concepts, statistical cost-effectiveness analysis, Bayesian computation and MCMC, and applied health economic evaluation.
BookView source →
Frequently Asked Questions (6)
What is a finite mixture model?
A statistical model representing a population as a combination of a finite number of distinct subgroups, each following its own probability distribution.
Source: McLachlan & Peel 2000
What does a finite mixture model assume a population is made of?
A finite mixture model assumes that a population is made up of a limited number of distinct subgroups, each following its own probability distribution, blended together in the observed data. It untangles this blend by estimating both the proportion in each subgroup and the distribution within it, without the subgroups being labelled in advance. This suits data that look like several overlapping distributions combined, such as costs that split into low and high spenders. Treating the data as a blend of hidden subgroups is its premise. Hastie and colleagues (2009) describe such models.
Source: Hastie et al. 2009
How does a finite mixture model work?
A finite mixture model works by assuming the data come from a specified number of components, each with its own distribution and a weight giving its proportion in the population, and estimating the components' parameters and weights, often by the expectation-maximisation algorithm, which iteratively assigns observations to components and updates the estimates. So a finite mixture model works by fitting a weighted combination of component distributions to the data, inferring the characteristics of each subgroup and the probability that each observation belongs to each, which allows the unobserved subpopulations and their properties to be estimated from data that mix them together.
Source: McLachlan & Peel 2000
When is a finite mixture model used?
A finite mixture model is used when a population is thought to consist of distinct but unobserved subgroups that differ in their distributions, such as different disease subtypes or response patterns, and the aim is to identify these subgroups and their characteristics. It is also used for flexible modelling of complex distributions. So a finite mixture model is used to uncover and describe latent subpopulations within heterogeneous data, which is valuable when a single distribution does not fit well and the heterogeneity is believed to reflect meaningful subgroups, allowing those subgroups to be identified and their proportions and features estimated.
Source: McLachlan & Peel 2000
How is the number of components chosen in a finite mixture model?
The number of components in a finite mixture model is chosen by fitting models with different numbers of components and comparing them using measures such as information criteria that balance fit against complexity, along with interpretability and substantive knowledge. There is no single definitive rule. So the number of components is selected by comparing candidate models and weighing statistical fit against parsimony and meaning, since too few components underfit the heterogeneity and too many overfit or yield uninterpretable subgroups, which is why the choice is made carefully, using formal criteria together with judgement about whether the resulting subgroups are plausible and useful.
Source: McLachlan & Peel 2000
What are the limitations of finite mixture models?
The limitations of finite mixture models include difficulty in choosing the number of components; the risk of identifying subgroups that are statistical artefacts rather than real subpopulations; sensitivity to distributional assumptions for the components; and computational challenges such as convergence to poor solutions. So finite mixture models are applied and interpreted with caution, since the subgroups they identify require validation and may not correspond to genuine subpopulations, and the results depend on modelling choices and assumptions, which is why the plausibility and replicability of the components are assessed rather than assumed from the model fit alone.
Source: McLachlan & Peel 2000
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 16 Dec 2025
Content version: 1.0.0
Canonical Identity
- Term code
- HE-ES-SA-066
Stable URI · Machine-readable · Resolvable · CC BY 4.0