Table 1.
Demographics of the final dataset.
Fig 1.
A: Probabilistic distribution of patients by latent class. Note that this represents the probability of a patient being classified in each subgroup, and is not equivalent to the distribution of cases in this study (Table 3). B: Outcome probabilities for each class. Each bar represents the probability of a parameter being positive in a particular class. Bars are colour coded based on the following groupings: HR/HER2 subtype (blue), distant organ metastases (orange), and lymph node metastases (green).
Table 2.
Note that BIC is minimized with 4 classes.
Table 3.
Latent class compositions.
Fig 2.
DAG plots for each latent class.
Arrow thickness represents the strength of association, which is inversely proportional to BIC. Dotted lines represent associations which are ≤5% of the strength of the strongest association. Node colours are grouped as follows: host features (blue), tumour biology (orange), disease burden (green), and overall survival (grey). Red boxes indicate interactions which were not validated by the SEM regression (p > 0.001). RMSEA = root mean square error of approximation; SRMR = standardized root mean square residual; “surv” = survival; “insur” = insurance.