Antibiotic trials in intra-abdominal infections. A critical evaluation of study design and outcome reporting.

The introduction of several new antibiotics, including cephalosporins and ureido-penicillins, has been a stimulus for clinical trials with these agents for intra-abdominal infection. Despite marked differences in antibacterial spectra, substantial differences in treatment results have not been documented. We reviewed published trials of antibiotic therapy for intra-abdominal infection to determine factors in study design that might impair identification of clinically important differences between regimens. Sixteen articles were identified that provided sufficient numbers of cases and data for analysis. Eight were prospective comparative trials, the remainder "single-armed" studies. The mortality rate was 3.5%, and the overall success rate was 84% for aminoglycoside plus clindamycin (range 52%-96%), 89% (range 83%-93%) for aminoglycoside plus metronidazole, and 93% (range 61%-95%) for cephalosporin-based regimens. Several defects in study design were identified. (1) Exclusionary criteria employed generally prevented enrollment of seriously ill patients or infections associated with high failure rates: Patients were excluded if even mild renal impairment was present or if antibiotic therapy had been recently administered, thereby excluding patients with postoperative or recurrent infections. Several studies allowed entry of contaminated but not infected patients. (2) Criteria used for reporting infectious diagnosis, premorbid health status, severity of infection, and outcome were nonuniform, and few studies provided such information. (3) Despite the small number of treatment failures, data reported did not allow determination of the basis for failure. For example, only four studies provided information on the operations performed upon treatment failures. Whether treatment failures were due to inadequate antibiotic therapy could therefore not be determined. Enrollment of a variety of low mortality infections precluded demonstration of any differences in regimens. Use of stratified randomization, stratifying for site of infection and severity of infection, and inclusion of greater numbers of patients would increase the likelihood of identifying differences between regimens. Such study design would likely require a multicenter trial to enroll sufficient numbers of cases for statistical analysis.