Latent Transition Analysis

Concept

---

Theoretically, Latent Transition Analysis (LTA) is a longitudinal extension of Latent Class Analysis that models changes in unobserved categorical class membership over time. It assumes that individuals belong to one of several latent classes at each measurement occasion and estimates the probabilities of transitioning between these classes across successive time points. LTA exists to characterise dynamic changes in latent population structure while accounting for measurement error and unobserved heterogeneity.

Mathematically, LTA is formulated as a finite mixture model incorporating latent class probabilities at each time point and transition probabilities between latent classes. The model combines measurement models describing class-specific response probabilities with a Markov transition model describing movement between latent classes over time. Parameters are estimated using maximum likelihood, typically via the Expectation-Maximisation (EM) algorithm, and model selection commonly relies on likelihood-based criteria such as the Bayesian Information Criterion (BIC).

In practice, LTA is fitted to repeated categorical measurements collected from the same individuals over multiple time points. In health economics, it is used to evaluate changes in patient health states, healthcare utilisation profiles, treatment adherence patterns, behavioural risk groups and disease severity categories. Estimated transition probabilities provide valuable information for forecasting patient trajectories, identifying intervention effects and informing decision models that incorporate changes in latent patient characteristics over time.

Purpose

---

Used to model transitions between unobserved population subgroups over time, estimate transition probabilities, evaluate longitudinal heterogeneity and characterise changes in patient behaviour or health status within health economic research.

---

Mathematical Formulae

Primary Formula

P(Y?, ?, Y?) = ??? ? ??? �?? ? ????? P(C? = k? | C??? = k???) ? ????? P(Y? | C?)

where:

• �?? = initial latent class probability
• C? = latent class at time t
• P(C? = k? | C??? = k???) = transition probability
• P(Y? | C?) = conditional response probability

Supporting Formulae

Transition probability:

P(C? = j | C??? = i)

Posterior class probability:

P(C? = k | Y)

Bayesian Information Criterion:

BIC = ?2ln(L) + p ln(n)

where:

• L = maximum likelihood
• p = number of estimated parameters
• n = sample size

Related Mathematical Methods

• Latent Class Analysis
• Finite Mixture Models
• Hidden Markov Models
• Maximum Likelihood Estimation
• Expectation-Maximisation Algorithm
• Growth Mixture Modelling
• Latent Profile Analysis

---

Example

A health economist follows 1,200 patients with diabetes over three annual assessments using six indicators of disease management.

Latent Transition Analysis identifies three latent classes:

• Well-controlled disease
• Moderately controlled disease
• Poorly controlled disease

Estimated annual transition probabilities include:

• Well-controlled ? Well-controlled = 0.88
• Well-controlled ? Moderately controlled = 0.10
• Well-controlled ? Poorly controlled = 0.02

These transition probabilities are incorporated into a long-term economic model to estimate future healthcare utilisation and intervention costs based on changing patient risk profiles.

---

Excel Implementation

Function	Example Formula	Health Economics Application
SUMPRODUCT	=SUMPRODUCT(B2:D2,E2:G2)	Calculate weighted latent class probabilities.
MMULT	=MMULT(B2:D4,F2:H4)	Multiply transition probability matrices across time periods.
LN	=LN(B2)	Calculate log-likelihood contributions during model estimation.
EXP	=EXP(A2)	Convert log probabilities into probabilities.
Solver	Maximise the overall log-likelihood by varying transition and class probability parameters.	Estimate latent transition model parameters using maximum likelihood.

---

VBA (Optional)

A VBA routine can automate calculation of transition probability matrices, summarise class membership over successive time points and prepare longitudinal datasets for latent transition modelling software.

---

Sources

• Collins LM, Lanza ST. Latent Class and Latent Transition Analysis. Wiley.
• Collins LM, Wugalter SE. Latent Class Models for Stage-Sequential Dynamic Latent Variables. Multivariate Behavioral Research. 1992.
• Muth�n B, Muth�n LK. Mplus User's Guide.
• Vermunt JK, Magidson J. Latent GOLD User's Guide.
• Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford University Press.
• ISPOR Good Practice Reports.