LCA Class Membership & Outcome Probability Calculator

Live Demo

A Streamlit web app for computing Latent Class Analysis (LCA) class membership probabilities and outcome rates based on user-selected covariate values.

The live demo is pre-loaded with model weights from a study on data-driven health status classification of older adults with diabetes (docs/Diabetes_Latent_Class_Predictions.pdf). It classifies patients into three latent classes — Geriatric, Cardiac, and Healthy — based on 19 comorbidities (e.g., depression, congestive heart failure, dementia) and reports predicted outcome rates (microvascular/cardiovascular complications, infection, hypoglycemia, mortality) per 100 person-years.

Given a fitted LCA model (prior class probabilities, covariate–class probability estimates, and outcome rates), this tool lets users interactively select observed covariates and instantly see:

• The predicted latent class and its outcome rates
• A bar chart of class membership probabilities
• Covariate probability profiles for each latent class
• Outcome rates across all classes

Select comorbidities

Predicted class and outcome rates

Covariate profiles and outcome rate charts

How It Works

The app implements the standard LCA posterior probability formula:

$$P(j \mid x_{obs}) = \frac{\hat{\eta}_j \prod_{i=1}^{P} [\hat{\pi}_{ij}^{x_i} (1 - \hat{\pi}_{ij})^{1 - x_i}]^{r_i}}{\sum_{k=1}^{K} \hat{\eta}_k \prod_{i=1}^{P} [\hat{\pi}_{ik}^{x_i} (1 - \hat{\pi}_{ik})^{1 - x_i}]^{r_i}}$$

where covariates can be marked as Yes, No, or Unknown (unobserved covariates are excluded from the product).

Quick Start

1. Install dependencies:

   pip install -r requirements.txt

2. Configure your LCA model by editing config.yaml (see config-sample.yaml for the expected format):

   Study Name:
     covariate_label: "Conditions"
     prior_probabilities:
       class_1: 0.3
       class_2: 0.4
       class_3: 0.3
     covariate_class_probabilities:
       covariate_1: [0.8, 0.2, 0.1]
       covariate_2: [0.1, 0.7, 0.3]
     outcome_rates:
       outcome_1: [0.9, 0.05, 0.05]
     outcome_rate_unit: "per 100 person-years"

3. Run the app:

   streamlit run app.py

Configuration

All model parameters are defined in config.yaml:

Key	Description
covariate_label	Display label for covariates (e.g., "Conditions", "Symptoms")
prior_probabilities	Prior class membership probabilities (one per class)
covariate_class_probabilities	Probability of each covariate given class membership
outcome_rates	Outcome rates for each class
outcome_rate_unit	Unit label for outcome rates (e.g., "per 100 person-years")

License

Apache License 2.0 — see LICENSE for details.