Bigger Isn’t Always Better: Data Considerations for Latent Class Analysis

Increasingly large datasets of clinical information present a significant opportunity to stratify patients for a more personalised approach to care. However, these data can be sparse and noisy, requiring processing of this data to be carefully considered. Here, we apply different minimum prevalence thresholds to diagnosis codes in a cohort of Multiple Sclerosis patients. We use this data to stratify patients using latent class analysis and examine the effect of different prevalence thresholds on the classes generated. We also examine computation efficiency using several disease-specific datasets.