Immune checkpoint inhibitor-associated severe hyponatremia: evidence mapping of SIADH-like phenotypes and endocrine immune-related adverse events

OBJECTIVE: Severe or symptomatic hyponatremia during immune checkpoint inhibitor (ICI) therapy may reflect syndrome of inappropriate antidiuresis (SIADH), endocrine immune-related adverse events (irAEs), cancer-related factors, or mixed mechanisms. We mapped the evidence and reclassified diagnostically extractable cases to distinguish SIADH-like phenotypes from endocrine irAEs. METHODS: PubMed, Embase, Web of Science Core Collection, Scopus, and the Cochrane Library were searched through 24 May 2026. Reports were assigned to A1 case-level, A2 population-level, or A3 diagnostic-framework layers. A1 summaries used patient/case denominators after a report-versus-patient audit. Diagnostic categories were reviewer-derived. Reporting completeness was described using six prespecified domains. We performed source-label and full-length-only sensitivity analyses. RESULTS: All 194 reports sought for retrieval were assessed, and 146 were included: 127 A1 reports describing 144 patients/cases, 16 A2 reports, and 3 A3 reports. Strict classification identified 127/144 (88.2%) confirmed/probable adrenal-axis irAEs, 1/144 (0.7%) confirmed/probable SIADH, 3/144 (2.1%) SIADH-like phenotypes with incomplete endocrine exclusion, 2/144 (1.4%) thyroid-related cases, and 11/144 (7.6%) mixed, confounded, or non-endocrine mechanisms. Ten patients/cases carried an explicit source-level SIADH label; strict review classified six as adrenal-axis irAEs, one as SIADH, and three as SIADH-like with incomplete endocrine exclusion. After exclusion of 81 abstract/database-only A1 reports, 46 full-length reports described 59 patients: 52 adrenal-axis irAEs, 1 SIADH, and 6 mixed/confounded or non-endocrine cases. CONCLUSIONS: Published classifiable A1 cases were dominated by adrenal-axis irAEs. Strictly supported SIADH occurred but was rare in this selected evidence base, and the direction of findings persisted in the full-length-only analysis. These proportions are not incidence estimates. The proposed algorithm supports diagnostic sequencing and transparent reporting; it is not a standalone practice guideline.