Investigators queried the TriNetX US Collaborative Network to identify adult patients (≥18 years) with craniopharyngioma who underwent surgical resection between 2005 and 2026. The query used ICD-10 and CPT coding to define diagnoses and procedures. Patient-level data were deidentified and returned as aggregated cohorts by the platform; therefore, analyses leveraged population-level counts rather than raw individual charts. The unadjusted sample comprised 2,558 patients who underwent transsphenoidal resection (TSR) and 1,452 who underwent transcranial resection (TCR).
The study team performed both unmatched comparisons and propensity score matching (PSM). Matching variables included demographics, comorbidities, presenting clinical features, and prior radiation exposure. PSM produced 1,199 matched pairs for comparative analyses. Cox proportional hazards models were used to estimate hazard ratios (HRs) at prespecified time points—1, 5, 10, and 15 years after index surgery—allowing the investigators to evaluate longitudinal associations between surgical approach and a range of neurologic, endocrine, and postoperative outcomes.
Within the matched cohort, investigators reported a higher early risk of cerebrospinal fluid (CSF) leak after transsphenoidal resection (1-year HR: 1.65). Short-term mortality favored TSR, with a lower 1-year hazard of death in the TSR group (HR: 0.54). The matched analyses also showed reduced 1-year hazards of hydrocephalus and blindness following TSR (hydrocephalus HR: 0.35; blindness HR: 0.54). Rates of adjuvant radiation within the first year were lower among TSR patients (HR: 0.70).
These early differences suggest distinct trade-offs by approach: TSR carried higher risk of CSF-related wound complications but lower acute neurologic morbidity and lower early mortality in the matched population. The study text documents these specific hazard ratios; it does not supply granular operative technical details for each case within TriNetX's aggregated dataset.
At extended follow-up, findings diverged for endocrine endpoints. By 10 years after surgery, TSR was associated with a lower hazard of diabetes insipidus (DI) (HR: 0.69) and lower hazard of blindness (HR: 0.61). Simultaneously, TSR exhibited higher long-term rates of hypopituitarism (HR: 1.24) and a higher composite endocrine morbidity signal (HR: 1.23). The hazard for persistent CSF leak remained elevated in the TSR group at 10 years (HR: 1.32).
Investigators reported that the 5- and 15-year analyses were concordant with the 1- and 10-year results, indicating that these approach-specific patterns—fewer hydrocephalus and blindness events but more endocrine sequelae and CSF-leak complications after TSR—were durable over the study interval. The paper supplies hazard ratios and confidence intervals for these time points but does not present individual patient-level hormone testing or visual field data in the TriNetX export.
The matched cohort showed lower use of adjuvant radiation among TSR patients at 1 year. Lower early mortality was also observed following TSR in matched analyses. The authors reported that TCR patients had more frequent preoperative hydrocephalus and DI, while TSR patients had a higher prevalence of preoperative hypopituitarism in the unmatched dataset. Those baseline imbalances informed the matching strategy.
Beyond the reported hazard ratios, the TriNetX-derived dataset did not include granular indications for adjuvant radiation, radiation dose or fractionation, nor individual clinical decisions; those details were not available in the aggregated platform output and therefore are not reported in the study.
The investigators used a federated EHR network that returns aggregated, deidentified cohort data. As such, analyses depend on diagnostic and procedural coding (ICD-10 and CPT). The source explicitly states that detailed individual-level clinical variables, operative technical nuances, and chart-level confirmations are not provided in the TriNetX aggregate output. Coding inaccuracy and variation across contributing centers can influence observed rates.
The study authors described the dataset’s availability constraints: the underlying deidentified data are licensed through TriNetX and are not publicly accessible outside TriNetX data use agreements. Institutional review board approval and individual informed consent were not required, given the deidentified, aggregated nature of the data.
Remaining unanswered questions documented by the source include the extent to which operative technique, surgeon experience, tumor anatomy, and intraoperative decision-making—variables not captured in the TriNetX aggregate export—influence the observed long-term complication patterns. The paper reports hazard ratios and long-term associations but does not provide mechanistic or procedural detail to explain causal pathways. These limitations should inform interpretation and counseling based on the population-level estimates provided by the study.