One Scoring System Does Not Fit All Healthcare Settings.

To the Editor—We read with interest the article by Peinado-Acevedo and coworkers [1] describing the validation of the Predicting Risk of Endocarditis Using a Clinical Tool (PREDICT) and VIRSTA scores in a large Colombian cohort of patients with Staphylococcus aureus bacteremia (SAB). PREDICT and VIRSTA are scoring systems intended to guide the use of echocardiography to detect infective endocarditis (IE) in patients with SAB [2, 3].

External validation of scoring systems is essential to evaluating their clinical usefulness. Therefore, the study by Peinado-Acevedo et al [1] is of interest. Importantly, it was performed in a healthcare system different from those used in the PREDICT and VIRSTA studies [2, 3]. The cohort studied was very different in composition from those used to generate the PREDICT and VIRSTA scores [2, 3]. Most importantly, SAB in this cohort was mainly a nosocomial complication of the use of central venous access, and only 16% of patients had community-acquired infection.

The main finding reported by Peinado-Acevedo et al [1] was that the VIRSTA score has high sensitivity (97%), while the PREDICT score has relatively low sensitivity (52%). The conclusion, therefore, was that transesophageal echocardiography could safely be omitted in patients with negative VIRSTA results, but not in those with negative PREDICT results. This conclusion is reasonable given the conditions in the Colombian cohort.

Different factors likely contribute to the low sensitivity of PREDICT in this study. First, PREDICT uses community acquisition as one variable in the score, and the low proportion of such patients partly explains the low sensitivity [2]. Second, a very large proportion of patients with IE were receiving hemodialysis, and among these patients PREDICT had an even lower sensitivity. Thus, PREDICT might be particularly unsuited to detecting IE in this subgroup. The sensitivity of PREDICT was higher among patients not receiving hemodialysis (65%).

Peinado-Acevedo and coworkers [1] stated that there is no external validation of PREDICT and VIRSTA, but Abu Saleh et al [4] and Kahn et al [5] have performed external validations of PREDICT. Their studies were from the United States and Sweden and demonstrated sensitivities of 100% and 81%–95%, respectively. VIRSTA was also validated in the Swedish cohort, showing high sensitivity (85%–100%) but moderate specificity (44%–55%) [5].

Time to blood culture positivity (TTP) is a feature readily available with automated blood culturing systems. A low TTP, indicative of a high bacterial concentration in blood, is a feature of intravascular infections [5-7]. Kahn et al [5] demonstrated that TTP could be included in a scoring system called POSITIVE, which had a high sensitivity and specificity for detecting IE in a cohort of patients with SAB, separate from the generation cohort [5]. It would be very interesting to evaluate the performance of POSITIVE in a different cohort of patient with SAB, such as that presented by Peinado-Acevedo and coworkers.

The results reported by Peinado-Acevedo et al [1] clearly demonstrate that scoring systems cannot be universally applied and that the performance of a given system needs to be validated before implementation. The performance of PREDICT is likely better in clinical settings other than that described by these authors.