Research

The problem

Embolization has largely replaced surgery for stable blunt splenic injury, but WSES guidelines still route every hemodynamically unstable patient to the operating room. The evidence behind that instruction compared embolization against watchful waiting rather than against the operation, and almost never in unstable patients, who cannot be randomized in an emergency.

What she did

Registered the protocol on PROSPERO before screening began, fixing the analysis plan in advance: random-effects REML with Hartung-Knapp correction, a prespecified subgroup at systolic pressure of 90 mmHg or below, ROBINS-I, GRADE. Led six investigators across four countries and wrote a deduplication rule so overlapping national registries did not enter the pooled estimate twice.

Added trial sequential analysis to distinguish a real effect from accumulated random error, and corrected a misreading of the required information size carried in the team's earlier documents.

Result

Mortality was 5.2 percent with embolization against 14.5 percent with surgery, and the cumulative evidence crossed the required information size, so the question does not need another study. The benefit held in the unstable subgroup, with heterogeneity falling from 74 percent to zero.

What was not claimed

Length of stay favored embolization but had not reached the required information size and was published as provisional. The thromboembolism benefit disappeared when a single study was removed and was reported as fragile. Certainty was not upgraded despite the large effect, because confounding by indication could not be excluded.

The problem

Evidence synthesis at scale fails on expert access, not on methodology. The organization needed dependable clinical input across specialties it did not employ, available continuously rather than assembled project by project, and had no structure for obtaining it.

What she did

Designed the network and recruited into it, launching in October 2025 with two specialists per specialty so no manuscript depended on one reviewer's availability. Built a nine-person ambassador layer reporting to her, and ran the reporting cycle that made engagement measurable rather than anecdotal.

As participation from outside Brazil grew, led the move from Portuguese to English as the operating language and rebuilt the model around it, shifting to shared infrastructure with an opportunities board, specialist spotlight content and sprint integration.

Result

The 17 specialty groups supported roughly 200 manuscripts. The project closed in July 2026 with its scope delivered, having operated in two languages from a standing start.

The decision behind it

A wind-down had been prepared as the alternative. She recommended restructuring instead, and designed the ambassador incentive structure that carried the model through its final cycle.