When we say a drug is intended to lower blood pressure, we abbreviate. The desired outcome is that the patient will not die, will be able to tolerate the drug, will not require switching to or adding another drug, and will have lower blood pressure than otherwise. The relevant data are then how many patients died, how many dropped out for toxicity, how many switched, and what the blood pressures were for the rest. The usual, unjustified, disastrous practice has been to lump death, toxicity, and switching as “missingness” and handle them analogously to missing data in surveys where a well-defined value of a variable exists but has not been ascertained. In general, a treatment may affect any of these aspects of outcome: it may kill or save patients, it may cause dropout for toxicity, it may keep patients from needing another drug, or it may change the blood pressure of patients in whom it does none of the other things. The effect of the drug is a combination of effects on all these aspects of outcome. The problem is to define a summary measure that can be estimated or a hypothesis that can be tested, under plausible assumptions, and that meaningfully indicates the benefit of treatment. Some bad and some reasonably good ways of doing this are discussed.