Latent Class Analysis-Derived Classification and Cancer-Specific Death in Endometrial Carcinoma: A Population-Based Cohort Study.
Molecular classifications have advanced prognostic stratification in endometrial carcinoma (EC), but they do not explicitly account for cause-specific mortality (CSM) and host-clinical context. We aimed to derive a clinically accessible classification that links demographic/clinicopathologic features, CSM, and molecular landscapes and to test whether it refines molecular-based survival stratification.
We conducted latent class analysis (LCA) on 46,772 cases of primary EC from the SEER program to identify latent characteristic lineages. The resulting LCA scheme was mapped to The Cancer Genome Atlas (TCGA) data. Cumulative incidence functions and Fine-Gray competing risks regression were used to estimate CSM. The Fisher test was employed to evaluate its association with TCGA subtypes. Transcriptomic differential expression analysis and pathway enrichment analysis were performed where data were accessible. Likelihood ratio tests assessed treatment-LCA interactions.
Five clinically interpretable classes emerged. Class 4 had significantly increased subdistribution hazard for CSM (1.34, P = .031); classes 3/5 trended higher. Classes were associated with TCGA subtypes (P < .001). Specifically, class 3 was associated with copy number (CN)-high enriched sybtype and class 1 was associated with CN-low subtype. Incorporating LCA labels into molecular survival models increased C-index and robustness of prognostic discrimination. Transcriptomics revealed class specific signatures implicating metabolic reprogramming, DNA repair, and immune pathways. Treatment benefit varied by class: compared with surgery alone, class 3 demonstrated the most significant survival advantage with surgery-radiotherapy-chemotherapy. Class 4 exhibited sensitivity to radiotherapy, while class 5 benefited exclusively from the triple therapy. Conversely, class 1/2 did not benefit from combined treatment.
LCA of routine clinical variables identifies subgroups susceptible to CSM and complements TCGA molecular classification, improving survival stratification.
We conducted latent class analysis (LCA) on 46,772 cases of primary EC from the SEER program to identify latent characteristic lineages. The resulting LCA scheme was mapped to The Cancer Genome Atlas (TCGA) data. Cumulative incidence functions and Fine-Gray competing risks regression were used to estimate CSM. The Fisher test was employed to evaluate its association with TCGA subtypes. Transcriptomic differential expression analysis and pathway enrichment analysis were performed where data were accessible. Likelihood ratio tests assessed treatment-LCA interactions.
Five clinically interpretable classes emerged. Class 4 had significantly increased subdistribution hazard for CSM (1.34, P = .031); classes 3/5 trended higher. Classes were associated with TCGA subtypes (P < .001). Specifically, class 3 was associated with copy number (CN)-high enriched sybtype and class 1 was associated with CN-low subtype. Incorporating LCA labels into molecular survival models increased C-index and robustness of prognostic discrimination. Transcriptomics revealed class specific signatures implicating metabolic reprogramming, DNA repair, and immune pathways. Treatment benefit varied by class: compared with surgery alone, class 3 demonstrated the most significant survival advantage with surgery-radiotherapy-chemotherapy. Class 4 exhibited sensitivity to radiotherapy, while class 5 benefited exclusively from the triple therapy. Conversely, class 1/2 did not benefit from combined treatment.
LCA of routine clinical variables identifies subgroups susceptible to CSM and complements TCGA molecular classification, improving survival stratification.