Randomization and measurement
Prior to data collection, randomly assigned ID numbers were recorded on the bottom of 200 agar plates, which were then placed in a box. One member of our research team gave each clinician 2 plates. Participants imprinted their stethoscope head onto the first plate and handed it to another investigator, who recorded the prewash ID numbers. Participants then performed the handwashing and stethoscope rub and repeated the imprinting procedure with the second plate. This time, the investigator recorded the professional role of each participant (eg, resident, attending, nurse, faculty) as well as the post-wash ID numbers.
After 48 hours at 35°C incubation, the plates were arranged in numerical order. A member of the research team then counted the number and identified the type of bacterial colonies on each plate and recorded the findings on a data sheet by ID number.
Validation
In order to validate the bacterial counts, the supervisor of the hospital laboratory—who had 20 years’ experience in examining cultures and served as the gold standard—independently examined a random sample of plates. We agreed in advance that any count that deviated by more than 7 (approximately half the effect the study was powered to detect) from the gold standard would require another investigator to intervene. This proved unnecessary as no such deviation was found.
Coagulase studies were performed on all plates with bacterial isolates, and gram staining was performed on selected plates, along with identification of gram-negative stains, using the Microscan (Siemens, New York, NY). An “honest broker”—the only person authorized to match the plates with the stethoscopes’ ID numbers—then matched the prewash and post-wash data by stethoscope and type of health care provider. Another investigator analyzed the final data sheet for accuracy.
Power and sample size
A pilot study was performed to obtain estimates of the average and variance of the bacterial counts in a control group of stethoscopes and to determine whether the act of imprinting the stethoscope itself would significantly reduce the colony counts. The results established that there was no statistical change in either the summary statistics or the distribution of the bacterial counts over the course of multiple imprinting.
Estimates obtained from the pilot study indicated that 58 stethoscopes would be sufficient to yield 80% power (alpha=0.05, 2-tailed) for detecting an average difference of 15 colony counts between the prewash and post-wash samples. Seventy-eight stethoscopes would increase the power to 90%. We ultimately tested 92 stethoscopes.
Statistical analysis
Descriptive statistical measures were calculated to examine the bacterial counts. Linear regression analysis was used to compute the correlation in the validation data. This before-and-after design results in “paired data,” and both parametric and nonparametric statistical tests were used. We used a paired t-test to test the mean difference in bacterial counts between the pre- and post-wash samples, and a random effects model to estimate the individual components of variance. The difference in the median bacterial counts was tested using the signed rank test. We used various diagnostic measures to examine the assumptions of the statistical tests; and means, medians, 95% confidence intervals (CIs), and P-values (using P<.05 as statistical significance) to report the results. The Bonferroni multiple comparisons procedure was used to determine whether the bacterial counts were statistically different among subgroups of health care providers. All statistical analyses were performed using SAS (Cary, NC) software.
Results
A total of 184 culture plates showing before and after samples for 92 stethoscopes were analyzed. The provider breakdown of the sample consisted of nurses (39%), residents (30%), attending physicians (15%), faculty (13%), and medical students (3%). Thirty-five (approximately 1 in 6) of the 184 plates were randomly sampled for validation. There was a high degree of reliability between the investigator’s bacterial counts and the bacterial counts of the gold standard (r=+0.98, P<.001).
Bacterial counts. The distribution of the bacterial colony counts skewed right in both the prewash (0-198) and post-wash (0-48) samples. The FIGURE shows the skewed distributions in the actual bacterial counts for the 92 pairs of plates before and after hand and stethoscope washing. In the prewash sample, the mean bacterial count was 28.4 (95% CI, 20.2-36.6), vs a post-wash mean of 3.2 (95% CI, 1.8-4.6; P<.001). This resulted in an estimated difference in mean bacterial counts of 25.2 (95% CI, 17.2-33.3). The difference in the medians was also significant, with a prewash median of 11.5 and a post-wash median of 1.0 (P<.001). The difference between the pre- and post-wash periods remained significant even after using various transformations to normalize the data. Random effects modeling showed that very little (<5%) of the total variation was related to the type of health care provider.