Name in original language
RADAR, sistema de alerta temprana de anomalias en el gasto y frecuencia de servicios de salud
Initiative overview
Context and problem
ADRES manages large-scale public health financing flows. Oversight teams often face fragmented datasets, manual checks and retrospective reviews, which limit early detection of unusual behaviours and slow the feedback loop to improve data quality and payment controls.
What the initiative does
RADAR-Desviaciones turns routine administrative health data into actionable early-warning signals. It combines (1) data-quality gates and deterministic business rules (e.g. structural validity, internal consistency and known red-flag patterns) with (2) anomaly detection methods based on statistical baselines and, where appropriate, machine-learning scoring. Outputs are presented as ranked worklists and dashboards that explain why a case was flagged.
How it is used
RADAR is a decision-support system: it does not automate sanctions or final decisions. Alerts are reviewed by human analysts/auditors who confirm findings and provide feedback. The system keeps records of inputs, transformations and rule/model versions, enabling auditability and continuous improvement.
How it scales
The programme is designed to start with a controlled pilot (selected cohorts, regions or payment lines) and expand as data quality stabilises and operating procedures mature. Governance covers access control, privacy and security safeguards, documentation and monitoring (performance, drift and operational impact).
Results, outcomes and impacts
Status: pilot or roll-out to be confirmed (2025–2026). Early deliverables include:
- (a) an agreed deviation catalogue and risk indicators developed with domain experts;
- (b) automated data-quality checks and traceable pipelines;
- and (c) dashboards that rank cases by risk and show key drivers.
Impact is measured through: reduction in time to triage, audit confirmation rate (hit rate), coverage (records screened per period) and cycle time from alert to action.
