Assessing the Concordance of MRSA Carriage Screening With MRSA Infections Free

This retrospective cohort study included patients admitted to the PICU, NICU, or general pediatrics floors at our 175-bed urban children’s hospital in the Midwestern United States during a 20-year period (January 1, 2002–December 1, 2022). Patients were included if they had a culture-confirmed MRSA infection from any site during a given admission and if they had electronic medical record (EMR) documentation of MRSA surveillance screening performed in the 365 days preceding that confirmed infection. If a patient had multiple screens preceding infection, only the most recent screen was used. Patients may have been admitted multiple times in the antecedent 365-day period, with the understanding that multiple admissions may engender MRSA carriage.

Because this study sought to determine the utility of MRSA surveillance in guiding the need for empirical anti-MRSA treatment, only patients without any history of MRSA infection, at any point in their life, were included. This was based on the assumption that many physicians, if beginning empirical antibiotic therapy, would likely include anti-MRSA therapy should a patient have had a previously documented invasive MRSA infection. As such, patients were excluded if the patient’s EMR had a past documented (nonsurveillance) culture with MRSA, or documentation indicating past MRSA infection at any hospital (including outside hospital systems, as viewable on the Care Everywhere Network).¹⁶ 

EMR review determined the source of the screen (eg, nares versus axillae/groin), the method of screening (eg, polymerase chain reaction [PCR] versus culture), age (days) of screen after positive culture, and infection site (eg, bloodstream).¹⁷  If a patient had multiple sites with positive MRSA cultures at the time of infection, all affected sites were counted. For patients younger than age 6 months (≤182 days), additional chart review identified if their infection occurred while in the NICU or elsewhere.^18,19  For bacteremic patients, chart review identified if the patient had a central venous catheter (including umbilical arterial/venous catheters) present at the time of infection or if their infection had been documented as a central line–associated bloodstream infection (CLABSI). In both of these scenarios, the patient was considered to have had a CLABSI. This analysis of CLABSIs was performed to delineate whether central lines predisposed bacteremic patients toward higher concordance, with the notion that central lines might serve as conduits for MRSA into the bloodstream. For patients with multiple MRSA surveillance screens, intermittent positivity was recorded.

Study approval was obtained from our center’s institutional review board. All data were deidentified via systematic removal of all patient identifiers and assignment of random numbers to represent individuals. Therefore, the need for informed consent was waived. All procedures followed were in accordance with institutional and national ethical standards.

For all children with documented MRSA infection, descriptive analyses with unweighted frequencies and weighted proportions were compared according to type of MRSA infection, MRSA screening site(s) (eg, nares versus axillae/groin), screening method (PCR versus culture), patient age, and MRSA screen age. For each characteristic, raw sensitivity was calculated (number of positive MRSA screens/number of patients with documented MRSA infection), and statistical significance was assessed via the χ² statistic. The prevalence of many types of infections (eg, biliary, pleural) was too low to use normal approximation to the binomial, therefore Wilson score intervals were used for 95% confidence intervals (CIs). Additionally, patients were categorized as having tested as always positive, always negative, or intermittently positive on historical MRSA screens. The association between MRSA infections and MRSA screening characteristics was described via multivariable logistic regression. To identify generalizable patterns, variables were selected that would be expected to be found in any data set of MRSA screens: days elapsed since an MRSA screen was performed, the method of MRSA screening, and the number of screening sites (eg, nasal = 1, nasal/axilla/groin = 3). For patients with bacteremia, the secondary analysis of CLABSIs was performed to delineate whether central lines were predisposing factors to higher concordance in bacteremic patients.

Binomial regression described the association between days elapsed since MRSA screen and subsequently documented MRSA infection; a polynomial line of best fit was added. For the special population of children younger than age 6 months (≤182 days), a sensitivity analysis was performed to assess the effect of replacing the covariate of having been admitted to any hospital unit at the time of MRSA infection, with a covariate of exclusively NICU admission. This analysis yielded similar results (Supplemental Table 4).

To calculate the theoretical utility of rescreening patients for MRSA carriage, number needed to treat, in this case number needed to test, was calculated via an assumption about what might happen to patients who did not receive such a rescreening. This theoretical control arm of patients would receive no rescreening compared with the theoretical treatment arm who would receive rescreening, on a given number of days.

Analyses were conducted using R (version 4.2.2 [2022-10-31], R Foundation for Statistical Computing, Vienna, Austria).

From 2002 through 2022, there were 246 patients with no history of MRSA infection who developed a culture-proven MRSA infection subsequent to a screening. Among these patients, 39.0% had a previous concordant MRSA screen (Table 1). The median age of patients at infection was 1.3 years (interquartile range, 0.3–5.0). The majority of patients, 206 (83.7%), were aged 9 years or younger at the time of MRSA infection. The median time between MRSA screening and MRSA infection was 7.0 days (interquartile range, 0–40.3). Nearly two-thirds of screens, 151 (61.4%), were obtained in the 2 weeks before infection.

Thirteen total infection types were recorded. The majority, 102 (41.5%), were SSTIs. ET/respiratory comprised 81 (32.9%) infections and 42 bacteremia (17.1%) infections. Sensitivity for bacteremia (eg, positive MRSA screens/patients with bacteremia) was 50.0% (n = 42), for ET/respiratory 44.4% (n = 81), and for SSTI 29.4% (n = 102). For children younger than age 6 months, sensitivity was 35.9% (n = 78); when selecting only NICU patients, sensitivity decreased to 24.5% (n = 53). MRSA screens performed on the day of birth were 6.7% (n = 45) sensitive.

Multivariable analysis showed a significant association between days elapsed since MRSA screening and infection (adjusted odds ratio [aOR], 0.98; 95% CI, 0.97–0.99) (Table 2). For bacteremic patients specifically, this association was not significant (aOR, 0.84; 95% CI, 0.68–1.0), but it remained significant for patients with SSTI (aOR, 0.98; 95% CI, 0.97–0.99) and ET/respiratory infections (aOR, 0.97; 95% CI, 0.96–0.99). Screening method (PCR versus culture) and number of screening sites (eg, nasal/axilla/groin = 3) were not significantly associated with concordance.

Binomial regression analyzing the relationship between MRSA infection and days elapsed since MRSA screening modeled the probability of concordance over time (Table 3, Fig 1). This association was found to be significant for “all MRSA infections” (odds ratio [OR], 0.98; 95% CI, 0.97–0.99), as well as bacteremia (OR, 0.83; 95% CI, 0.70–0.99), SSTI (OR, 0.98; 95% CI, 0.97–0.99), and ET/respiratory infections (OR, 0.98; 95% CI, 0.96–0.99). The probability of concordance fell below 50.0% for “all MRSA infections” on day 4 after screening, on day 6 for bacteremia, and on day 12 for ET/respiratory infections. The probability of concordance was never higher than 50.0% for SSTI (Fig 1). For children younger than age 6 months, the probability of concordance fell to below 50.0% on day 11 after screening and on day 9 for NICU-specific patients (Supplemental Fig 3).

A number needed to treat analysis showed that, in a theoretical scenario in which patients are rescreened for MRSA carriage daily, 1.7 patients would need to be screened to predict 1 MRSA infection (Fig 2). If screened every 7 to 13 days, 2.1 patients would be needed. If screened every 14 to 27 days, 5.3 patients would be needed. This equates to a 56.9% to 58.8% absolute risk reduction for patients rescreened daily to weekly and a 10.3% to 11.7% absolute risk reduction for patients rescreened monthly or less.

Past scholarship on the concordant or discordant nature of MRSA screening in children has focused on screening in select, typically low-risk groups of children. Sands and colleagues found the sensitivity of MRSA nares screens to be 42.9%, Bradford and colleagues found it to be 16.2%, and our research found overall sensitivity among all types of infections in line with these numbers, at 39.0%. Yet, when considering that all 246 children in our study had known MRSA infections, this equates to 150 children whose MRSA screenings did not concord. This study sought to characterize the common features found in discordance and to use statistical analysis to predict trends of concordance and discordance.

Interestingly, among the types of infections with the largest sample sizes (SSTI, ET/respiratory, and bacteremia) MRSA screening for bacteremia showed the highest sensitivity. Intuitively, one might expect such a result if positing that a strong skin floral burden of MRSA would engender bacteremia because of line infection. Yet, when isolating for CLABSI only, sensitivity paradoxically fell to 40.9%. For children with documented ET/respiratory infections, sensitivity was 44.4%, which is closer to the overall sensitivity found in our study.

Multivariable analysis demonstrated that the only covariate of significance, among all infection types, was the number of days elapsed since the MRSA screen. This seems intuitive: a MRSA screen will be less indicative of a patient’s flora as time elapses. Yet, it is compelling to see the rapidity at which binomial regression modeled probabilities of concordance to drop below 50.0%. Essentially, after 1 week, MRSA screening was less likely than a coin flip to concord for patients with bacteremia or ET/respiratory infections. It is also interesting that the other covariates that did not prove significant: differences in screening method (PCR versus culture) did not, nor did the number of body sites that were swabbed. Given that previous literature has led to the inference that PCR-based swabs may remain positive for a longer period than culture-based swabs,²⁰  we performed a post hoc analysis on the positivity of MRSA swabs over time by screening method. This analysis did not find a higher likelihood for PCR swabs to return positive over a longer period than cultures (Supplemental Table 5). This agreed with our multivariable analysis, which found that screening type was not significantly associated with concordance (Table 2).

For children younger than age 6 months, regression analysis and prediction modeling showed similar results: the only significant covariate predicting concordance was days elapsed since MRSA screen. Moreover, after 7 days, a MRSA screening in the NICU showed only ∼50.0% concordance with infection. With consideration that many NICU children are screened for MRSA on the day of their birth and never again, rescreening may partially mitigate the effects of missed MRSA carriage resulting from long-term NICU hospitalizations.^21,22 

Finally, within the context of past studies which found high negative predictive values for negative MRSA screens in children, this study reflects a different concern. Our study did not include the totality of children screened for MRSA within the 20-year period, but only those children with confirmed MRSA infections. As such, we could not calculate NPV, PPV, and specificity within our study, and we would not suggest that our results should alter clinicians’ a priori assumptions about the likelihood of MRSA infection among a broader population of children being tested for MRSA. Considering that from among tens of thousands of patients, this study found only 246 patients meeting criteria over a 20-year span, the NPV is likely still to be low in this broader pediatric population.

However, previous studies excluded children at high risk, whereas our study included children within NICUs and other children at high risk such as those with prolonged hospital stays. Moreover, our study’s MRSA prevalence was, by design, 100%. The resultantly calculated concordance within this cohort that included higher risk children was far lower than NPVs found in past studies. One might consider if those previous study results were driven by low prevalence, and the exclusion of higher risk children, to inform high NPVs. Therefore, we submit that this study should be balanced against previous studies with lower pretest probabilities of MRSA infection. We do not advocate for the indiscriminate use of anti-MRSA agents when treating the broader pediatric population. But we would suggest that, among children at high risk with high pretest probabilities for MRSA infection, the initiation or removal of empirical anti-MRSA treatment should not be based solely on MRSA carriage screens. Moreover, MRSA carriage screens should not replace actual cultures of infected body sites.

This study has several limitations. First, this study did not calculate NPV, PPV, nor specificity because it did not include all patients screened for MRSA, but only those with documented infections. Second, our research is limited by biases inherent in retrospective reviews, such as the possibility of confounding from risks not measured nor considered. Additionally, patients with hospitalizations not viewable on the Care Everywhere Network may have been inappropriately included or excluded, and misclassification may have occurred. Third, our hospital’s catchment area serves a primarily urban population, and findings may not be generalizable to a general US population. Fourth, testing practices for MRSA carriage are innately facility and clinician dependent, and conceivably different at other hospital systems. As well, this study spans a 20-year period, and the detection abilities for both cultures and PCR samples have improved over time. This may have led to an underestimation of MRSA carriage and infections. Fifth, this study did not exclude patients who may have received anti-MRSA antibiotics before nasal carriage screening, Though we presume that most patients admitted from an emergency department to an ICU would have received 0 to 1 doses of anti-MRSA antibiotics, and only in the immediate hours before screening, it is possible that these antibiotics could engender negative surveillance swabs.¹⁹  Finally, our multivariable analysis sought to select covariates that would be expected to be found in any MRSA screening data set, but residual or additional biases may exist.

Using data from more than 20 years, among 246 pediatric hospitalizations, our study demonstrates an overall concordance of only 39.0% between MRSA positive screens and all types of invasive MRSA infection. Among the most frequent types of infection, bacteremia demonstrated 50.0% concordance, ET/respiratory infections 44.4% concordance, and SSTI 29.4% concordance. The number of days elapsed since obtaining a MRSA screen proved to be the only significant covariate in predicting concordance to infection. When that covariate modeled the probability of MRSA screening concordance with infection, 7 days represented the point at which the probability of concordance with infection would fall to or drop below 50.0% for most types of infections. This was notably also true for the special population of children in the NICU. Strikingly, the sensitivity of MRSA screens performed on the day of a child’s birth was only 6.7%. Given that only patients with known MRSA infections were included in our study, the concordance of screens was far lower than NPV found in past studies, and previous study results may have been impacted by low prevalence and the exclusion of children at high risk to inform high NPV. For all of these reasons, we suggest that negative MRSA carriage screens should not invalidate reasonable suspicion for MRSA infection in patients with high pretest probabilities for infection.

The authors thank Drs Allison Bartlett, Daniel Johnson, Julia Rosebush, Chris Lehmann, and Jack Flores for their guidance and feedback in this project.