Reply to “Diagnostic Value of a PCR-Based Technique for Prosthetic Joint Infection”

REPLY

We greatly appreciate the interest of Li and Yu (1) in our article (2) that evaluated the use of PCR assays for diagnosis of prosthetic joint infection (PJI). In their letter, Li and Yu raised some thoughtful comments regarding the selection of the included studies and performed an updated meta-analysis. They indicated that two similar studies (3, 4) by Kobayashi et al. were performed during the same period and suggested including the one with the larger sample size (4). However, the small one (which includes 23 cases) (3) has more-detailed descriptions and a higher quality score than the large one (which includes 36 cases) (4). Therefore, after careful discussion and consideration, we included the study (3) with the higher quality score. Regarding the study conducted by Moojen et al. (5), most of the patients were enrolled for PCR evaluation for diagnosis of PJI, and we finally included this study after comprehensive consideration. Further sensitivity analysis showed that the exclusion of the study conducted by Moojen et al. did not influence the result (the pooled sensitivity and specificity were 0.84 [95% confidence interval {CI}, 0.74 to 0.90] and 0.91 [95% CI, 0.80 to 0.96], respectively).

In the updated meta-analysis by Li and Yu, the pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under the curve (AUC) were 0.79 (95% CI, 0.76 to 0.82), 0.86 (95% CI, 0.84 to 0.87), 8.22 (95% CI, 4.66 to 15.16), 0.25 (95% CI, 0.18 to 0.34), 49.03 (95%CI, 30.72 to 78.25), and 0.93 (95% CI, 0.92 to 0.94), respectively (Table 1). Also, they indicated that the diagnostic value of the PCR-based technique might be overestimated in our study (2). However, the discrepancy between our meta-analysis and that of Li and Yu may be mainly attributable to the different statistical methods used. The meta-analyses of Li and Yu used Meta-DiSc (version 1.4), based on a traditional univariate model, which had difficulty in providing the 95% CI for AUC, and our meta-analysis used the Midas module for Stata version 11 based on a bivariate mixed-effects regression model (6–9). Further, we reanalyzed our data using another two statistical methods, including a univariate model with Meta-DiSc (version 1.4), and both the bivariate model and the hierarchical summary receiver operating characteristic (HSROC) model with the Metandi module of Stata (version 11). The results showed that using different statistical methods may produce different outcomes (the pooled sensitivities were 0.81 [95% CI, 0.77 to 0.85], 0.86 [95% CI, 0.77 to 0.92], and 0.86 [95% CI, 0.77 to 0.92] and the pooled specificities were 0.86 [95% CI, 0.84 to 0.87], 0.91 [95% CI, 0.81 to 0.96], and 0.91 [95% CI, 0.81 to 0.96] for Meta-DiSc, Midas, and Metandi, respectively) (Table 1). The univariate model may lead to a lower statistic than that for the bivariate model as well as the HSROC model. Further updated meta-analysis is required to assess the use of PCR assays for diagnosis of PJI using multiple statistical methods.

TABLE 1.

Statistical results of our study using three different methods and the study of Li and Yu (1) using Meta-DiSc^a

Estimate	Value for study and method(s) (95% CI)
	Our study			Li and Yu, Meta-DiSc
	Stata, Midas	Stata, Metandi	Meta-DiSc
Sensitivity	0.86 (0.77–0.92)	0.86 (0.77–0.92)	0.81 (0.77–0.85)	0.79 (0.76–0.82)
Specificity	0.91 (0.81–0.96)	0.91 (0.81–0.96)	0.86 (0.84–0.88)	0.86 (0.84–0.87)
PLR	9.1 (4.6–18.2)	9.1 (4.6–18.2)	6.9 (4.01–11.69)	8.22 (4.66–15.16)
NLR	0.16 (0.10–0.25)	0.16 (0.10–0.25)	0.24 (0.17–0.32)	0.25 (0.18–0.34)
DOR	59 (29–118)	59 (29–118)	48.71 (27.38–86.68)	49.03 (30.72–78.25)
AUC	0.94 (0.91–0.95)	0.94 (0.91–0.95)	0.92	0.93 (0.92–0.94)

^aAbbreviations: PLR, positive likelihood ratio; NLR, negative likelihood ratio; DOR, diagnostic odds ratio; AUC, area under the curve; CI, confidence interval.