Table 2.

Summary of key steps and recommendations when setting up the data to perform Latent class analysis.

	Step	Description	Recommendation	Presentation
Study Design & Data Set-up	Indicator Selection	The indicators selected will dictate the nature of the clusters.	- Select indicators based on research question. - Exclude indicators that are composite of other indicators in the model. - Exclude outcome data as indicators.	- Present clear rationale for indicator selection.
	Data Processing	Transforming data to minimize extreme scales is more likely to yield informative classes.	- Categorical variables: Consider collapsing categories with less than 10% of the sample. - Non-parametric data should be transformed such that they are normally distributed and uniformly scaled.	- Clearly describe the procedures used for data transformation and collapsing of categories.
	Local Independence	Assumes that within class, observed indicators are independent.	- Test correlation of indicator variables in the complete dataset and within each class. - Consider removing one or more indicator if there is collinearity. - If a single pair is highly correlated consider relaxing the assumption.(31)	- Present the correlation coefficients of the most highly correlated indicators. - Clearly describe any variables that were excluded from the analysis.
	Sample Size	Power sample to: 1. Determine the true number of classes. 2. Detect pertinent differences between the classes.	- When N < 300 it is recommended to perform Monte Carlo simulation to determine adequacy of sample size.(38) - Standard power calculations should be performed to determine the sample size needed to detect significant inter-class differences.	- Present clear rationale for the sample size and any power calculations performed.
	Handling Missing Data	Approaches for missing data: 1. Full information maximum likelihood 2. Multiple Imputation	- Full information maximum likelihood and multiple imputation are recommended methods of dealing with missing data. - Lower levels of research biomarker assay detection (LLD) impute either LLD, LLD/2, or multiple imputation.(44)	- Present methods used for handling missing data. - Present differences in indicators and outcomes between missing and complete cases. - Sensitivity analysis with missing data / non-imputed data.