Change in Methicillin-Resistant Staphylococcus aureus Testing in the Intensive Care Unit as an Antimicrobial Stewardship Initiative

Abstract

Background: Methicillin-resistant Staphylococcus aureus (MRSA)-associated infections are a cause of morbidity and mortality in the intensive care unit (ICU). Vancomycin is a treatment option but is not without risks. Methods: A MRSA testing change-the switch from culture to polymerase chain reaction-was implemented at 2 adult (tertiary and community) ICUs located in a Midwestern US health system. Data from 2016 to 2020 were included in the study, and the median change in time to test results was examined. Results: During the study period, 71% of 19,975 patients seen at the 2 ICUs received MRSA testing. In the preintervention period, 91% and 99% of patients at the tertiary and community hospitals received testing via culture, respectively. Culture testing was used 1% and ∼0% of the time at the tertiary and community hospitals, respectively, in the postintervention period. A counterfactual estimate showed 36 (95% credible interval [CrI], 35, 37) and 32 (95% CrI, 31, 33) fewer hours until results were available at the tertiary and community hospitals, respectively. Conclusion: After the testing change, MRSA results were available in less time. Obtaining results sooner can assist with antimicrobial stewardship through the potential delay in initiating therapies such as vancomycin and/or quicker de-escalation of such therapies.

Keywords: Anti-infective agents; Staphylococcus aureus; diagnosis; diagnostic techniques and procedures; intensive care units; methicillin-resistant Staphylococcus aureus; vancomycin.

Figure.

Plotted time in hours until receipt of methicillin-resistant Staphylococcus aureus test results for patients admitted to 2 adult intensive care units by study period day. The top panel (A) presents tertiary hospital data, and the bottom panel (B) presents community hospital data. The solid black vertical line represents when the test change occurred; the solid white lines are the estimated time to receipt of test results per period; the dashed black lines represent the counterfactual estimate at the midpoint of the postintervention period with a diamond located at the estimate. In the bottom panel after the vertical dashed line is a slight possible reduction in time to receipt of test results at the community hospital when it changed laboratories. Sixteen patients (0.1%) were dropped from the analyses because the time to receiving MRSA test results was >4 days. Symbols: x=culture test; +=polymerase chain reaction test.

Figure A1.

The top panel presents the priors used in the tertiary hospital model, and the bottom panel presents the priors used in the community hospital model. Priors for slope values for the 2-line segments were the same and mostly overlap in the figures. (For readers of the print publication, a color version of this figure is available at https://www.doi.org/10.31486/toj.22.0103.)

Figure A2.

Tertiary hospital posterior estimates. The x-axes represent the Bayesian quantile regression model coefficients related to receipt of methicillin-resistant Staphylococcus aureus results. Preintervention is the period before the switch from culture testing to polymerase chain reaction testing. Slopes represent change in time (hours) to receipt of results per day increase during the designated period (ie, preintervention or postintervention).

Figure A3.

Community hospital posterior estimates. The x-axes represent the Bayesian quantile regression model coefficients related to receipt of methicillin-resistant Staphylococcus aureus results. Preintervention is the period before the switch from culture testing to polymerase chain reaction testing. Slopes represent change in time (hours) to receipt of results per day increase during the designated period (ie, preintervention or postintervention).

Figure A4.

Two alternative model fits. The green line represents the quantile regression from the paper. The purple line is a robust regression (ie, Huber), while the orange line is a deep artificial neural network (ie, 3 hidden layers with rectified linear units activations, initial null bias and Glorot normal weights, batch normalization, and 10% dropout). The neural network shows what a more adaptive nonlinear prediction can look like given the presence of increased noncompliance in default testing method near the end of the study period (ie, dashed vertical line). ICU, intensive care unit; MRSA, methicillin-resistant Staphylococcus aureus. (For readers of the print publication, a color version of this figure is available at https://www.doi.org/10.31486/toj.22.0103.)