Choosing a caching strategy: read-through, write-through, or write-back?

I'm working on a service that needs to significantly reduce database load and latency for frequently accessed data. We're currently debating the best caching strategy. I'm leaning towards a read-through cache for simplicity, but I'm curious about the trade-offs, especially regarding data consistency and performance implications for write-heavy workloads with write-through vs. write-back. What are your experiences or recommendations for choosing between these, or even implementing a hybrid approach?