You are absolutely correct, the Business Model rules because it is the model that contains all of the business requirements.
The Data Model models the data about which we describe the Business Model, from which the database is derived.
The Application Model, however does not describe the business nor the data, merely the business requirements as to how the data is processed. However, how do you model the process until you have all of the inputs, the business requirents and data structures?
To often, the Data Model (if there is one at all) is constructed or derived only after the application process has been constructed. It is this that I contend that is backwards.
As for your clinical trials, clearly you have provided insufficient information from which to derive any meaningful discussion. However, as far as what you optimize, optimization is last. You model the business; you model the data, for integrity because without it, the storage is meaningless, even if optimal, I assume, for speed.
The whole point of normalization is for the reduction of data anomolies, from which the data maintains its consistency. Without this, the rest is fruitless. A side effect of normalization is performance optimization. Now, this is argued against quite often due to the JOINs that must take place. However, if you consider all of the performance issues, not just the JOINs, but include the time necessary to maintain the integrity of the system, you will see that the normalized database is more efficient.
I really do not believe either one of us will agree with the other on this point, as you have brought up your clinical trial data before. However, you have yet to provide the structures of this data, either as you have designed it, or as we have suggested.