To The Editor:
We read with interest "The Value of Intraoperative Gram Stain in Revision Total Knee Arthroplasty" (2009;91:2124-9), by Morgan et al., but are concerned that the article suggested that the practice of routinely performing a Gram stain at the time of revision total knee arthroplasty may safely be abandoned. The authors failed to recruit a consecutive sample of patients who underwent tests for white blood-cell count, erythrocyte sedimentation rate, C-reactive protein, and aspiration, and this introduces biases and jeopardizes the conclusion. From their data, we calculated the commonly used summary indices for diagnostic tests: accuracy, Youden index1, and diagnostic odds ratio. The accuracies for Gram stain, white blood-cell count, erythrocyte sedimentation rate, C-reactive protein, and aspiration are 80.35%, 80.28%, 78.82%, 77.33%, and 66.74%, respectively. The Youden indices for Gram stain, white blood-cell count, erythrocyte sedimentation rate, C-reactive protein, and aspiration are 0.2698, 0.2731, 0.2731, 0.2677, and 0.2487, respectively. The diagnostic odds ratios for Gram stain, white blood-cell count, erythrocyte sedimentation rate, C-reactive protein, and aspiration are 250.45, 247.88, 247.88, 178.12, and infinity, respectively. The commonly used summary indices for Gram stain are not the most inferior ones and are superior to C-reactive protein. The positive likelihood ratio for Gram stain, 182.83, is the highest among the tests. The negative likelihood ratio for Gram stain, 0.73, is the same as those of white blood-cell count, erythrocyte sedimentation rate, and C-reactive protein. The negative and positive predictive values for Gram stain (78.90% and 98.53%, respectively) are higher than those of C-reactive protein (75.39% and 98.31%, respectively). The negative and positive predictive values are easily influenced by the disease prevalence and become unstable. The sensitivity of Gram stain (27.13%) is better than those of C-reactive protein (26.98%) and aspiration (24.87%). The specificity of Gram stain (99.85%) is also better than C-reactive protein (99.79%). It is still too early to abandon the Gram stain, especially while awaiting the results of bacterial culture, and the results of the Gram stain will help with selecting the proper empiric antibiotics. The results shown in Table III reflect the spectrum effect2 and do not argue against the Gram stain.
Last, there are several proofreading errors. First, in the fifth column of Table I, the total number of patients with aspiration does not add up and should be 427. Second, the authors misnamed the Mann-Whitney U test to a Mann-Whitney t test with Gaussian approximation. Third, the authors misused the term chi-square analysis to describe a decision matrix.