Predicting Antimicrobial Resistance Prevalence and Incidence from Indicators of Antimicrobial Use: What Is the Most Accurate Indicator for Surveillance in Intensive Care Units?

A Pragmatic Machine Learning Model To Predict Carbapenem Resistance

Infection caused by carbapenem-resistant (CR) organisms is a rising problem in the United States. While the risk factors for antibiotic resistance are well known, there remains a large need for the early identification of antibiotic-resistant infections. Using machine learning (ML), we sought to develop a prediction model for carbapenem resistance. All patients >18 years of age admitted to a tertiary-care academic medical center between 1 January 2012 and 10 October 2017 with ≥1 bacterial culture were eligible for inclusion. All demographic, medication, vital sign, procedure, laboratory, and culture/sensitivity data were extracted from the electronic health record. Organisms were considered CR if a single isolate was reported as intermediate or resistant. Patients with CR and non-CR organisms were temporally matched to maintain the positive/negative case ratio. Extreme gradient boosting was used for model development. In total, 68,472 patients met inclusion criteria, with 1,088 patients identified as having CR organisms. Sixty-seven features were used for predictive modeling. The most important features were number of prior antibiotic days, recent central venous catheter placement, and inpatient surgery. After model training, the area under the receiver operating characteristic curve was 0.846. The sensitivity of the model was 30%, with a positive predictive value (PPV) of 30% and a negative predictive value of 99%. Using readily available clinical data, we were able to create a ML model capable of predicting CR infections at the time of culture collection with a high PPV.

Risk factors for development of aminoglycoside resistance among gram-negative rods

PURPOSE

Development of scoring systems to predict the risk of aminoglycoside resistance and to guide therapy is described.

METHODS

Infections due to aminoglycoside-resistant gram-negative rods (AR-GNRs) are increasingly common and associated with adverse outcomes; selection of effective initial antibiotic therapy is necessary to reduce adverse consequences and shorten length of stay. To determine risk factors for AR-GNR recovery from culture, cases of GNR infection among patients admitted to 2 institutions in a major academic hospital system during the period 2011-2016 were retrospectively analyzed. Gentamicin and tobramycin resistance (GTR-GNR) and amikacin resistance (AmR-GNR) patterns were analyzed separately. A total of 26,154 GNR isolates from 12,516 patients were analyzed, 6,699 of which were GTR, and 2,467 of which were AmR.

RESULTS

In multivariate analysis, risk factors for GTR-GNR were presence of weight loss, admission from another medical or long-term care facility, a hemoglobin level of <11 g/dL, receipt of any carbapenem in the prior 30 days, and receipt of any fluoroquinolone in the prior 30 days (C statistic, 0.63). Risk factors for AmR-GNR were diagnosis of cystic fibrosis, male gender, admission from another medical or long-term care facility, ventilation at any point prior to culture during the index hospitalization, receipt of any carbapenem in the prior 30 days, and receipt of any anti-MRSA agent in the prior 30 days (C statistic, 0.74). Multinomial and ordinal models demonstrated that the risk factors for the 2 resistance patterns differed significantly.

CONCLUSION

A scoring system derived from the developed risk prediction models can be applied by providers to guide empirical antimicrobial therapy for treatment of GNR infections.

Impact of Chlorhexidine Bathing on Antimicrobial Utilization in Surgical Intensive Care Unit

BACKGROUND

This secondary analysis compared antimicrobial utilization among surgical intensive care unit patients randomized to every other day chlorhexidine bathing (chlorhexidine) versus daily soap and water bathing (soap-and-water) using data from the CHlorhexidine Gluconate BATHing trial.

MATERIALS AND METHODS

Antimicrobial utilization was quantified using defined daily dose (DDD)/100 patient-days and agent-days/100 patient-days for systemic antimicrobials. Antivirals (except oseltamivir), antiparasitics, and prophylaxis agents were excluded. The 2018 anatomic therapeutic chemical/DDD index was used to calculate DDD. Agent-days were calculated as the sum of calendar days where antimicrobials were administered. Patient-days were defined as time patients were at risk for health care-acquired infections plus up to 14 d. Primary analyses were conducted using linear regression adjusted for baseline Acute Physiology and Chronic Health Evaluation II scores.

RESULTS

Of 325 CHlorhexidine Gluconate BATHing trial patients, 312 (157 in soap-and-water and 155 in chlorhexidine) were included. The median (interquartile range) of total antimicrobial DDD/100 patient-days was 135.4 (75.2-231.8) for soap-and-water and 129.9 (49.2-215.3) for chlorhexidine. The median (interquartile range) of total antimicrobial agent-days/100 patient-days was 155.6 (83.3-243.2) for soap-and-water and 146.7 (66.7-217.4) for chlorhexidine. After adjusting for Acute Physiology and Chronic Health Evaluation II scores, chlorhexidine bathing was associated with a nonsignificant reduction in total antimicrobial DDD/100 patient-days (-3.9; 95% confidence interval, -33.9 to 26.1; P = 0.80) and total antimicrobial agent-days/100 patient-days (-10.3; 95% confidence interval, -34.7 to 14.1; P = 0.41).

CONCLUSIONS

Compared with daily soap and water bathing, every other day chlorhexidine bathing did not significantly reduce total antimicrobial utilization in surgical intensive care unit patients.

Risk Factors for Development of Carbapenem Resistance Among Gram-Negative Rods

Background

Infections due to carbapenem-resistant Gram-negative rods (CR-GNR) are increasing in frequency and result in high morbidity and mortality. Appropriate initial antibiotic therapy is necessary to reduce adverse consequences and shorten length of stay.

Methods

To determine risk factors for recovery on culture of CR-GNR, cases were retrospectively analyzed at a major academic hospital system from 2011 to 2016. Ertapenem resistance (ER-GNR) and antipseudomonal (nonertapenem) carbapenem resistance (ACR-GNR) patterns were analyzed separately. A total of 30951 GNR isolates from 12370 patients were analyzed, 563 of which were ER and 1307 of which were ACR.

Results

In multivariate analysis, risk factors for ER-GNR were renal disease, admission from another health care facility, ventilation at any point before culture during the index hospitalization, receipt of any carbapenem in the prior 30 days, and receipt of any anti-methicillin-resistant Staphylococcus aureus (anti-MRSA) agent in the prior 30 days (c-statistic, 0.74). Risk factors for ACR-GNR were male sex, admission from another health care facility, ventilation at any point before culture during the index hospitalization, receipt of any carbapenem in the prior 30 days, and receipt of any anti-MRSA agent in the prior 30 days (c-statistic, 0.76).

Conclusions

A straightforward scoring system derived from these models can be applied by providers to guide empiric antimicrobial therapy; it outperformed use of a standard hospital antibiogram in predicting infections with ER-GNR and ACR-GNR.

Risk Factors for Colistin Resistance among Gram-Negative Rods and Klebsiella pneumoniae Isolates

Infections due to colistin-resistant (Col^r) Gram-negative rods (GNRs) and colistin-resistant Klebsiella pneumoniae isolates in particular result in high associated mortality and poor treatment options. To determine the risk factors for recovery on culture of Col^r GNRs and Col^rK. pneumoniae, analyses were chosen to aid decisions at two separate time points: the first when only Gram stain results are available without any bacterial species information (corresponding to the Col^r GNR model) and the second when organism identification is performed but prior to reporting of antimicrobial susceptibility testing results (corresponding to the Col^rK. pneumoniae model). Cases were retrospectively analyzed at a major academic hospital system from 2011 to 2016. After excluding bacteria that were intrinsically resistant to colistin, a total of 28,512 GNR isolates (4,557 K. pneumoniae isolates) were analyzed, 128 of which were Col^r (i.e., MIC > 2 μg/ml), including 68 of which that were Col^rK. pneumoniae In multivariate analysis, risk factors for Col^r GNRs were neurologic disease, residence in a skilled nursing facility prior to admission, receipt of carbapenems in the last 90 days, prior infection with a carbapenem-resistant organism, and receipt of ventilatory support (c-statistic = 0.81). Risk factors for Col^rK. pneumoniae specifically were neurologic disease, residence in a skilled nursing facility prior to admission, receipt of carbapenems in the last 90 days, receipt of an anti-methicillin-resistant Staphylococcus aureus antimicrobial in the last 90 days, and prior infection with a carbapenem-resistant organism (c-statistic = 0.89). A scoring system derived from these models can be applied by providers to guide empirical antimicrobial therapy in patients with infections with suspected Col^r GNR and Col^rK. pneumoniae isolates.

A Simulation Study to Assess Indicators of Antimicrobial Use as Predictors of Resistance: Does It Matter Which Indicator Is Used?

OBJECTIVE

Indicators of antimicrobial use have been described previously, but few studies have compared their accuracy in prediction of antimicrobial resistance in hospital settings. This study aimed to identify conditions under which significant differences would be observed in the predictive accuracy of indicators in the context of surveillance of intensive care units (ICUs).

METHODS

Ten resistance / antimicrobial use combinations were studied. We used simulation to determine if Québec's network of 81 ICUs or the National Healthcare Safety Network (NHSN) of 2952 ICUs are large enough to allow the detection of predetermined differences between the most accurate and 1) the second most accurate indicator, and 2) the least accurate indicator, in more than 80% of simulations. For each indicator, we simulated absolute errors in prediction for each ICU and each 4-week period, for surveillance lasting up to 5 years. Absolute errors were generated following a binomial distribution, using mean absolute errors (MAEs) observed in 9 ICUs as the average proportion; simulated MAEs were compared using t-tests. This was repeated 1000 times per scenario.

RESULTS

When comparing the two most accurate indicators, 80% power was reached less often with the Québec network versus the NHSN (0/20 versus 2/20 scenarios, with 5 years of surveillance data), a finding reinforced when comparing the most and least accurate indicators (3/20 versus 20/20 scenarios). When simulating 1 year of data, scenarios reaching an 80% power dropped to 0/20, comparing the two most accurate indicators with the larger network, and to 1/20, comparing the most and least accurate indicators with the smaller network.

CONCLUSION

Most of the time (72%), identifying an indicator of antimicrobial use predicting antimicrobial resistance with a better accuracy was not possible. The choice of an indicator for an eventual surveillance system should rely on criteria other that predictive accuracy.