Paper - Selection Bias in Clinical Trials of Lyme Disease

Cameron DJ, Mt. Kisco, New York, USA, board member, International Lyme and Associated Diseases Society (ILADS)

Objective: To assess the prevalence of biases from selection of patients in a recently completed randomized placebo-controlled clinical trial.

Background: A recently completed clinical trial by Klempner et al (N Engl J Med 2001;345(2):85-92.) found no difference in outcome between 3 months of antibiotics (1 month ceftriaxone and 2 months doxycycline) and placebo. The authors described a population of Lyme disease patients ill for 4.7 years who had been treated a mean of 3.2 times before enrollment in their clinical trial.

Method: To evaluate the potential for bias in selection, patients in the clinical trials were compared with those treated in a surveillance database. Only surveillance database patients evaluated from 1997 to 2001 confirmed by at least 5 bands on an IgG Western blot were considered. To help control for baseline differences in the groups, patients in the clinical trial were matched with the 36 patients in the surveillance database who had been treated with antibiotics at least 3 times. Outcomes were compared between the two groups.

Result: The surveillance database patients were less likely to fail treatment than the clinical trials patients (33% vs 73%). Selection of patients from the surveillance database without treatment delay and with successful initial treatment would have led to a failure rate as low as 14%. Selection of surveillance database patients with either a treatment delay or failed initial treatment would have led to a failure rate of 47%. Selection of patients with both treatment delays and failed initial treatment would have led to failure rates of as high as 66%. Although the matching was between two differing designs, these data provide some evidence that Lyme disease patients who enroll in clinical trials do have different characteristics than patients who do not.

Conclusion: Using an ongoing surveillance database as a control, these results confirm the potential for bias associated with differential selection of subjects. Delays in treatment and failed previous treatment apparently were not considered when enrolling patients in the clinical trial. Further clinical trials are needed to evaluate the outcome of prolonging intravenous antibiotics longer than 4 weeks, raising the dose of doxycycline from 200 mg to 400 mg, and evaluation for co-infections for Lyme disease patients with delayed treatment or failing initial treatment.