Outbox pattern: reliable event publishing from database transactions

Ensuring reliable event publishing from within database transactions is a common challenge for maintaining data consistency across services. We need to atomically update our database and publish a corresponding event; if one fails, the other shouldn't succeed. The outbox pattern seems like the most robust solution for this. The idea is to write the event to an 'outbox' table within the same database transaction as the business logic update. A separate process then polls this outbox table, publishes the events to a message broker (like Kafka), and marks them as published. This guarantees atomicity and at-least-once delivery. We're comparing this approach with using Change Data Capture (CDC) via Debezium, which captures database changes directly from the transaction log and publishes them as events. Debezium seems to reduce the application-level complexity, but adds a dependency on a separate CDC service. What are the key trade-offs between the outbox pattern and CDC for reliable event publishing, especially in terms of complexity, performance, and operational overhead?