Mission-critical healthcare database platforms support clinical workflows, enterprise reporting, integration services, and operational decision-making. Performance degradation in these systems can appear as slow dashboards, query timeouts, failed maintenance jobs, capacity alerts, unstable execution plans, or increased operational risk. This preprint proposes a KPI-driven AI-assisted framework for database reliability and performance optimization in heterogeneous healthcare data platforms. The framework collects operational signals such as P95 query latency, blocking, index fragmentation, log usage, TempDB version store, Query Store regressions, stale statistics, tablespace pressure, invalid objects, and backup-readiness indicators. These signals are normalized into an auditable health model, classified into actionable severity levels, analyzed using AI-assisted reasoning, governed by DBA safety gates, and validated using before/after measurements. A preliminary evaluation was performed using anonymized weekly DBA operational reports from five maintenance reporting periods spanning SQL Server fleet operations and Oracle production tuning cases. The SQL Server evidence covers 142 instance-week observations. Across these reports, critical health states were reduced from 80 to 1; healthy states increased from 43 to 107; approximately 2,349 high-fragmentation indexes were addressed; and cumulative P95 query latency reduction reached approximately 3,521,603 ms. Oracle case studies demonstrate controlled plan stabilization through SQL Patch, SQL Profile, SQL Plan Baseline, SQL Plan Management, and invisible indexes. No patient-level data, protected health information, organization names, IP addresses, usernames, database names, schema names, or internal operational identifiers are disclosed. Raw operational reports are confidential; only anonymized aggregate metrics are used.
Mohamed Ezzat Shady (Sun,) studied this question.