Nine hundred ninety-eight adult patients admitted to intensive care units (ICUs) in the United States over a 24-hour period in 2011 were studied. Standard definitions from the CDC criteria to determine infection were used. Prolonged empiric antibiotic therapy (PEAT) was defined as the ratio of the total number of empiric antibiotics continued for at least 72 hours in the absence of adjudicated infection divided by the total number of antibiotics prescribed. A total of 660 unique antibiotics were prescribed to 364 patients as empiric therapy. Three hundred thirty-three of 660 (50%) of the empiric antibiotics were continued for more than 72 hours in the absence of adjudicated infection.

Vancomycin and piperacillin/tazobactam were the most commonly prescribed antibiotics. Suspected pneumonia accounted for 60% of the cases in which prolonged empiric antibiotics were used. ICUs using invasive techniques for the diagnosis of ventilator-associated pneumonia (VAP) had lower rates of PEAT (45%) than ICUs that did not use such techniques (60%). No other institutional factor was identified to be associated with PEAT.


An astounding percentage of patients admitted to ICUs (and general medical wards) receive empiric antibiotics. This study conducted in adult ICUs in the United States showed that 50% of all empiric antibiotics prescribed were continued more than 72 hours in the absence of adjudicated or documented infection.

The data presented in this study are concerning but probably not surprising to most of us who practice in tertiary hospitals and academic medical centers. My sense is that this problem has gotten worse in recent years. I suspect that some of the reasons why this has happened include: 1) concern that sepsis will be missed and adverse outcomes (including death) may occur; 2) the increasing use of automated “alerts,” in the emergency department and in the hospital, which trigger reflex protocol-driven orders such as lactate levels, initiation (or broadening) of antibiotics, and administration of IV fluid boluses; 3) one of the problems with these very sensitive “sepsis alerts” is that, while they are sensitive, they lack specificity in the diagnosis of sepsis; 4) perhaps a change in training in which fewer clinicians are confident in their ability to differentiate pneumonia from non-infectious causes of pulmonary infiltrates using a constellation of clinical and laboratory information; 5) the bias in most internal medicine training programs to “punish sins of omission” more harshly than “sins of commission,” at least when it comes to antibiotics; 6) the view that I have heard expressed by so many ICU fellows and attending physicians that, “the patient was so sick I felt uncomfortable stopping the broad-spectrum antibiotics.”

The bottom line is that we still need more sensitive and specific tests to diagnose sepsis.