Empiric therapy is practical and must be begun promptly; the specific
regimen chosen must be based upon local conditions and epidemiology. I
t must be recalled that subgroups of patients are not necessarily equi
valent to the majority, i.e., there are low-risk patients for whom amb
ulatory and/or oral therapy is appropriate and, conversely, there are
high-risk patients who have a potential for a high mortality and who,
while perhaps few in number, are of critical importance. Further, many
of these patients are very complex., and this leads, to a high level
of physician concern and insecurity. This physician concern, in turn,
leads to a tendency to modify regimens, given that the physician all t
oo often is dealing with inadequate diagnostic information owing to th
e patient situation. The physician's choice of modification is highly
dependent upon knowledge of the regimen the patient is already receivi
ng. There is a need for clear definition of endpoints, and these must
be established before the study is initiated. All too many published s
tudies are too small to evaluate the endpoint that has been defined, a
nd many others, although sufficient in size, have all of the problems
inherent in studies conducted at multiple sites by multiple individual
s with differing degrees of commitment or enthusiasm toward the study
at hand. A few implications for study design and evaluation seem evide
nt: it is critical to define endpoints and execute the study according
ly. This means determining the size of the population needed and deter
mining the presence or absence of risk groups. Patients to be excluded
e.g., those in whom infection is doubted must be selected on the basi
s of objective data by an observer blinded to both the outcome and the
treatment. Similarly, the classification of response should preferabl
y be done by an observer not influenced by knowledge of the therapy be
ing given. Finally, and similarly, the decision to modify therapy (esp
ecially if modification is equivalent to defining failure with the reg
imen) should not be influenced by knowledge of the therapy being admin
istered.