Latent class analysis-based prognostic classification in laryngeal carcinoma: beyond TNM staging.

Abstract

Introduction: The aim of this study is to identify latent clinical-biological profiles using latent class analysis (LCA) and to compare their prognostic capacity with that of the TNM system.

Materials and methods: A retrospective study of 489 patients with squamous cell carcinoma of the larynx (2000-2020). LCA was performed using eight indicator variables: age, sex, toxic consumption, sub-location, T category, N category, Karnofsky index, and host index. TNM staging was grouped into three categories (stages I-II, III, and IV) as no survival differences were observed between stages I and II (p = 0.056). Prognostic capacity was compared using the C-index, Royston-Sauerbrei D-statistic, and integrated Brier score (IBS).

Results: The LCA identified three classes: Class 1 (n = 257, 52.6%), early glottic carcinoma, N0; Class 2 (n = 143, 29.2%), advanced supraglottic carcinoma with good functional status; and Class 3 (n = 89, 18.2%), defined by functional impairment and an unfavorable inflammatory-nutritional profile. Five-year disease-specific survival was 93.9%, 74.4%, and 40.8%, respectively (p < 0.001). The LCA achieved a higher Royston-Sauerbrei D-statistic (6.73 vs. 5.45; p < 0.001) and a lower prediction error (IBS: 0.086 vs. 0.092) than the simplified TNM, with no significant difference in Harrell's concordance index (0.758 vs. 0.732; p = 0.39). In the cross-validation analysis, the simplified TNM lost its discriminatory power within Class 3 (C-index = 0.560; p = 0.130).

Conclusions: LCA identifies three prognostic profiles in laryngeal carcinoma with greater risk-stratification capacity and lower prediction error than the simplified TNM, highlighting a high-risk subgroup determined by host status rather than tumor extent.