Selective and Cascade Reporting of Antimicrobial Susceptibility Testing Results and Its Impact on Antimicrobial Resistance Surveillance—National Healthcare Safety Network, April 2020 to March 2021

This study is the first to assess the extent of using selective and/or cascade antimicrobial susceptibility reporting for antimicrobial stewardship among U.S. hospitals and its impact on cumulative antibiograms in the context of electronic data exchange for national antimicrobial resistance surveillance. ABSTRACT Selective or cascade reporting (SR/CR) of antimicrobial susceptibility testing (AST) results is a strategy for antimicrobial stewardship. SR/CR is often achieved by suppressing AST results of secondary drugs in electronic laboratory reports. We assessed the extent of SR/CR and its impact on cumulative antibiograms (CAs) in a large cohort of U.S. hospitals submitting AST data to the CDC’s National Healthcare Safety Network (NHSN) through electronic data exchange. The NHSN calls for hospitals to extract AST data from their electronic systems. We analyzed the AST reported for Escherichia coli (blood and urine) and Staphylococcus aureus (blood and lower respiratory tract [LRT]) isolates from April 2020 to March 2021, used AST reporting patterns to assign SR/CR reporting status for hospitals, and compared their CAs. Sensitivity analyses were done to account for those potentially extracted complete data. At least 35% and 41% of the hospitals had AST data that were suppressed in more than 20% blood isolates for E. coli and S. aureus isolates, respectively. At least 63% (blood) and 50% (urine) routinely reported ciprofloxacin or levofloxacin for E. coli isolates; and 60% (blood) and 59% (LRT) routinely reported vancomycin for S. aureus isolates. The distribution of CAs for many agents differed between high SR/CR and low- or non-SR/CR hospitals. Hospitals struggled to obtain complete AST data through electronic data exchange because of data suppression. Use of SR/CR can bias CAs if incomplete data are used. Technical solutions are needed for extracting complete AST results for public health surveillance. IMPORTANCE This study is the first to assess the extent of using selective and/or cascade antimicrobial susceptibility reporting for antimicrobial stewardship among U.S. hospitals and its impact on cumulative antibiograms in the context of electronic data exchange for national antimicrobial resistance surveillance.

support antimicrobial stewardship. The use of SR/CR for stewardship is predicated on the precept that by omitting AST results for certain antimicrobials, those antimicrobials are less likely to be used by prescribers (4). Specific examples include: omitting results from agents that, despite in vitro susceptibility, are known not to be reliably effective at a particular site of infection (e.g., daptomycin in respiratory samples); or omitting results for drugs in patient populations for which there is contraindication (e.g., omitting fluoroquinolone results for pediatric patients). The omitted results for secondary agents are only reported on request or under specific circumstances. Cascade reporting (CR) is a type of SR in which results for secondary agents are only reported if an organism is resistant to primary agents (3) Guidelines for antimicrobial stewardship programs developed by the Infectious Diseases Society of America and the Society for Healthcare Epidemiology of America recommend SR and/or CR as an antimicrobial stewardship strategy (5). The Clinical and Laboratory Standards Institute's (CLSI's) M100 Performance Standards suggest grouping of antibiotics for AST reporting (4): group A agents are appropriate for routine testing and reporting of results; group B agents warrant primary testing, but may be reported selectively; group C agents are alternative or supplemental and may necessitate specific testing; group U includes agents that are used only or primarily for treating urinary tract infections (4).
Wide use of electronic health records (EHR) and laboratory information systems (LIS) has enabled more extensive use of SR and CR, largely because many prescribers use EHR to view AST results, and these platforms make it easier to selectively omit specific results from the view of prescribers based on the reporting/suppression algorithms (6). Because the suppressed results are stored in electronic databases, AR data that do not reach the EHR should be available for AR surveillance purposes, such as producing CAs (3). However, because AST results are produced from multiple sources and often managed in multiple information platforms, such as bacterial identification and antimicrobial susceptibility testing instruments, LIS, and EHR, suppressed results may not be widely available, or in some cases, are not even known by those preparing CAs. Using partially suppressed AST results to calculate CAs carries a risk of biased calculations.
Thus, it becomes important to understand the extent to which AST results suppression is performed by health care facilities and any inconsistencies in its implementation. Although the CLSI provides suggestions for how AST results can be reported for clinical purposes in its M100 Performance Standards (4), little is known about how clinical laboratory AST reporting aligns with the CLSI recommendations. In this study, we evaluated the extent of SR/CR, consistency or inconsistency of the practice relative to CLSI recommendations, and the impact of SR/CR on hospital CAs by analyzing the pattern of AST data that were extracted from hospital information systems and submitted to a public health antimicrobial resistance surveillance system.

RESULTS
A total of 750 hospitals were included in one or more of the following isolate-specimen type combinations: 675 (blood) and 734 (urine) submitted AST results for $90% E. coli isolates, 658 (blood) and 572 (lower respiratory tract) submitted AST results for $90% S. aureus isolates. Hospital characteristics are shown in Table S1.
(i) Blood cultures. Of hospitals reporting $30 isolates (n = 351), 2% to 19% of hospitals met the criteria for high SR/CR of various group A agents, and 8% to 36% met the criteria for high SR/CR of group B antimicrobial agents (Table 1). Overall, 62% hospitals showed high SR/CR for at least one group A or B agent. Compared to hospital CAs among routine reporting or low SR/CR hospitals, levels of susceptibility were generally lower for high SR/CR hospitals across CLSI group B agents. However, when comparing the median values between two groups, this difference was only significant for ampicillin-sulbactam/amoxicillin-clavulanate (median: 23% versus 57%, P , 0.0001) ( Table 2). In the case of group A agents, the impact of SR/CR on level of susceptibility was more mixed; however, hospitals with SR/CR for cefazolin had lower rates of susceptibility than low-or non-SR/CR hospitals (median: 61% versus 76%, P = 0.0004). A CA for certain organism-drug combinations could not be generated for many SR/CR hospitals due to apparent AST data suppression (i.e., ,30 AST results reported). (Table 2).
(ii) Urine cultures. Of hospitals reporting $30 isolates (n = 651), 1% to 8% of hospitals met the criteria for high SR/CR of various group A agents, 1% to 30% met the criteria for high SR/CR of group B agents (Table 1), and 59% hospitals showed high SR/CR for at least one group A, B, or U agent. Compared to CAs among low-or non-SR/CR hospitals, levels of susceptibility were uniformly lower in high SR/CR hospitals (Table 2).
Staphylococcus aureus. For blood and lower respiratory tract isolates, the top five antimicrobials that were reported routinely by most hospitals were oxacillin/cefoxitin (group A), vancomycin (group B), trimethoprim-sulfamethoxazole (group A), clindamycin (group A), and tetracyclines (group B). (Fig. 2).
(i) Blood cultures. Of hospitals reporting $30 isolates (n = 329), 2% to 23% of hospitals met the criteria for SR/CR of various group A agents, and 9% to 42% met the criteria for SR/ CR of group B antimicrobial agents (Table 3). Overall, 71% hospitals showed high SR/CR for one or more group A or B agents. Compared to the CAs among routine reporting and low SR/CR hospitals, levels of susceptibility of high SR/CR hospitals were higher for the group A agents azithromycin, clarithromycin, or erythromycin (median 57% versus 43%, P = 0.0009), and lower for the group A agent trimethoprim-sulfamethoxazole (median 88% versus 97%, P , 0.0001) ( Table 4). Fifty-seven percent of isolates from vancomycin routine reporting and low SR/CR hospitals were susceptible to oxacillin/cefoxitin (i.e., methicillin-susceptible S. aureus, MSSA). Among vancomycin high SR/CR hospitals, vancomycin was reported for 44% of isolates, of which, 89% were nonsusceptible to oxacillin/cefoxitin (i.e., methicillin-resistant S. aureus, MRSA).
(ii) Lower respiratory tract specimen cultures. Of hospitals reporting $30 isolates (n = 249), 2% to 25% of hospitals met the criteria for SR/CR of various group A and B antimicrobial agents (Table 3), and 57% hospitals showed high SR/CR for one or more group A or B agents. Fifty-three percent of isolates from vancomycin low-or non-SR/CR hospitals were   The cumulative antibiograms (%S) were prepared by the authors and do not include data from any hospital where fewer than 30 antimicrobial susceptibility results were submitted for the given organism-drug-specimen combinations. Sensitivity analyses. Of the hospitals reporting at least one agent for $90% isolates, 17% (blood) and 16% (urine) displayed no apparent data suppression across drugs that are commonly tested for E. coli isolates. Meanwhile, 22% (blood) and 25% (lower respiratory) of hospitals displayed no data suppression across drugs commonly tested for S. aureus isolates. Of hospitals reporting $30 isolates, 14% (blood) and 15% (urine) displayed no data suppression across drugs commonly tested for E. coli; 15% (blood) and 22% (lower respiratory tract) displayed no data suppression across commonly tested for S. aureus isolates. These indicate the percentages of hospitals that either did not perform any SR/CR for these drugs or submitted presuppression data to the NHSN.

DISCUSSION
This is the first evaluation of the impact of AST reporting practices on CAs that also provides insight into the clinical AST reporting practices of a large number of U.S. hospitals. The distribution of hospital CAs for many antimicrobials differed significantly between the hospitals with and those without patterns of SR/CR. However, our data could not tell us how these differences may have impacted antimicrobial prescribing in these hospitals. Approximately 62% (E. coli) and 71% (S. aureus) of reporting hospitals had AST data that were suppressed in more than 20% blood culture isolates for at least one of the CLSI group A or B drugs, likely as a result of SR or CR.
Enabled by the broad uptake of hospital information systems, the large amount of electronic AST data created for clinical use can now be re-used for national antimicrobial surveillance in the United States. This study is the first to assess the potential bias of electronic AST data exchange for this purpose. Obtaining all AST reports, whether or not they are delivered routinely to hospital prescribers, is a prerequisite for generating unbiased CAs (3). However, because AST data are often managed in multiple systems, such as testing instruments, LIS, and EHRs, pinpointing where and how SR/CR is applied can be challenging (7). This work requires collaboration with microbiologists and information technology  Numbers and percentages were calculated among those that reported 30 or more isolates for the given specimen type. b Not suggested to be tested or reported for respiratory specimens.
Selective AST Reporting in the U.S. Microbiology Spectrum The cumulative antibiograms (%S) were prepared by the authors and do not include data from any hospital where fewer than 30 antimicrobial susceptibility results were submitted for the given organism-drug-specimen combinations.
Selective AST Reporting in the U.S. Microbiology Spectrum specialists. Further, surveillance software vendors who are responsible for data extraction and submission for surveillance purposes are often unaware of SR/CR. These combined factors may make it challenging to produce accurate CAs. The main purpose of CR is to reduce unnecessary use of broad-spectrum antimicrobials. However, using partial AST results for CAs could lead to adverse consequences. Because isolates that are nonsusceptible to a primary drug (e.g., cefazolin) would be more likely to be nonsusceptible to a secondary drug in the same class (e.g., ceftriaxone) than those isolates that are susceptible, a CAs constructed using only the isolates that were nonsusceptible to the primary agents for the secondary drug is likely to appear less susceptible. Therefore, prescribers could be mistakenly encouraged by such biased CAs toward empirical use of broader-spectrum antimicrobials (e.g., carbapenems). Also, CAs based on SR data can only reflect the CAs among corresponding patient populations. For example, CAs results for an antibiotic that is mainly reported for outpatients, such as macrolides for S. aureus in this study, may not be representative of CAs results for inpatients. We also found that several isolate and patient factors were associated with reporting of AST results for selected organism-drug combinations among high SR/CR hospitals (Table S2 and 3). This suggests that differences in %S may be due to selection biases. Additionally, the wider distributions of %S among SR/CR hospitals for some drugs (e.g., cephalosporins and fluoroquinolones for E. coli) might be attributed to variations on data suppression algorithms ( Fig. S1 and 2). Therefore, CAs are not comparable between SR/CR and non-SR/CR hospitals or among SR/CR hospitals with different data suppression algorithms.
Many studies have shown that SR/CR can effectively influence antimicrobial prescribing behaviors (Table S4). SR/CR has been adopted in many clinical laboratories across the world (8,9). Our study suggests routine reporting of fluoroquinolone and carbapenem results is common in the United States. Our study consists of an indirect evaluation of SR and CR via data reported for AR surveillance; however, even if, as in our sensitivity analysis, one removes all hospitals without evidence of AST suppression across commonly tested agents (i.e., hospitals likely submitting unsuppressed data to the NHSN) from the number of facilities routinely reporting, still nearly 56% (blood) and 48% (urine) of hospitals routinely reported ciprofloxacin/levofloxacin results, and 50% (blood) and 39% (urine) of hospitals routinely reported carbapenem results for E. coli isolates. Similarly, for S. aureus, after removing those with no data suppression, vancomycin was still routinely reported in at least 53% (blood) and 59% (lower respiratory tract) of hospitals, despite more than half of isolates being MSSA. Fluoroquinolones and vancomycin were among the most frequently prescribed antimicrobials in acute care hospitals (10). As E. coli and S. aureus are two of the most commonly identified clinical pathogens, better use of CR could lead to improved antibiotic stewardship (5,11,12).
One possible explanation for why some hospitals do not adopt SR/CR is the relative lack of guidance regarding best practices, particularly with regard to safe implementation (8). Indiscriminate use of SR/CR without prescriber awareness may lead to potentially harmful consequences (13). For example, unexpected nonsusceptible results should always be reported to clinicians who might otherwise assume that a drug result that is not reported is uniformly susceptible (4), potentially delaying appropriate treatment and infection control procedures (13,14). On the other hand, although the CLSI drug groups drugs provide some guidance on AST reporting, given the diversity and complexity of clinical scenarios, implementing reporting algorithms still require active participation of prescribers, pharmacists, and antimicrobial stewardship programs, as well as information system technical support. Other barriers to the implementation of SR/CR could include lack of human resources in the laboratory to guarantee timely release of suppressed results 24 h/day and 7 days a week and lack of information technology support (8).
This study has several limitations. First, we analyzed AST reporting patterns to identify SR/CR; however, we were unable to know the underlying data suppression algorithms that were applied for each hospital. The wide-ranged CAs shown among those hospitals with a lower percentage of isolates with AST reported suggests that there could be a mix of reporting algorithms across hospitals. Second, the cut-off 20% difference in AST reporting for identifying hospitals with high SR/CR is arbitrary for presentation convenience. In a further sensitivity analysis, when we changed the definition of high SR from 20% to a 10% shortfall in % reported results, with only few exceptions, the proportions of high SR/CR hospitals remained similar (changes #5%, results not shown). Third, a small number of hospitals with no results reported could have been misclassified as not testing if the drug was routinely tested but all results were suppressed. Conversely, a drug that was not routinely tested but later tested per clinician's request could have been misclassified as SR/CR. To mitigate this potential bias, we only analyzed drugs that are commonly tested in the United States. Fourth, our data could not differentiate hospitals that used SR/CR but submitted complete, presuppressed data to NHSN from hospitals routinely reporting all drugs. However, our sensitivity analyses suggested successful submission of presuppressed data to NHSN occurred infrequently. Also, we do not know if local programs report their antibiograms using selective reporting, even if that is the way the results are reported into NHSN. Finally, our data reflect the AST reporting patterns of 750 hospitals voluntarily submitting data to the NHSN, only about 12% of the 6180 acute-care hospitals in the United States that are mandated by Centers for Medicare and Medicaid Services payment policies to report health care associated infection data to the NHSN. Incidentally, compared to the general distribution of hospitals that participate in NHSN, AR reporting hospitals were more likely to be larger (percentage of hospital with more than 200 beds: 42% versus 19%, P , 0.0001) and teaching hospitals (percentage of teaching hospitals: 67% versus 51%, P , 0.0001; data not shown).
SR/CR is a powerful tool to enhance antimicrobial stewardship. A significant proportion of the hospitals that participated in the NHSN AR Option appear to have adopted SR/CR for clinical isolates of E. coli and S. aureus. However, the reporting practices vary across hospitals and agents, and CLSI's SR/CR suggestions are used inconsistently. Meanwhile, CAs produced using incomplete results should be interpreted with caution. Moving forward, technical solutions that can obtain complete unsuppressed AST data, by consolidating data from multilevel hospital information systems, are needed for reporting electronic data to national AR surveillance.

MATERIALS AND METHODS
Data source. We analyzed AST results that hospitals voluntarily submitted to the Centers for Disease Control and Prevention's (CDC's) National Healthcare Safety Network (NHSN) Antimicrobial Resistance (AR) Option. For our analysis, we used AST data that hospitals submitted for specimens collected from April 1, 2020 to March 31, 2021 and reported to NHSN by June 30, 2021. The AR Option protocol calls for hospital participants to submit the isolate-level final AST interpretations (susceptible, intermediate, resistant, etc.) generated for deduplicated isolates of specific pathogens identified in cerebrospinal fluid, blood, lower respiratory tract, and urine cultures. Hospitals are encouraged to report any AST data meeting the NHSN requirements, which include all the drugs in the CLSI M100 group A, B, C, and U for the corresponding organisms, regardless of whether those data are suppressed from clinical end users. However, because NHSN could not mandate the submission of AST that were suppressed from end users during the study period, those suppressed results could be reported as not tested in the data stream from the hospitals lacking the capability of extracting presuppression data. These hospital AST results are captured and processed electronically from their LIS or EHR system and submitted to NHSN using a data exchange standard known as Clinical Document Architecture. Only the isolates with at least one antimicrobial agent, whether the drug is on the surveillance protocol, was tested are eligible for reporting (i.e., isolates that are clinically considered contamination are not eligible). The details of data collection and isolate deduplication rules for the NHSN AR Option are specified in the surveillance protocol (https://www.cdc.gov/nhsn/pdfs/pscmanual/11pscaurcurrent.pdf).
AST reporting pattern assessment. We analyzed the patterns of AST reporting for each of four isolate-specimen type combinations: E. coli isolates from blood, E. coli from urine, S. aureus from blood, and S. aureus from sputum or other lower respiratory tract specimens. For each hospital, we determined the percentage of isolates with AST results reported among the isolates for which isolate identifications were submitted to NHSN (% reported). Isolates from different specimen sources were analyzed separately. First, we excluded hospital isolate-specimen type combinations without at least one antimicrobial reported for $90% of isolates over concerns of data quality. Routine reporting of a particular antimicrobial was defined by the most frequently reported AST results for the isolate-specimen type combination. If the percentages of reporting are equal for multiple drugs that are most frequently reported for the organism, a hospital can have multiple drugs determined to be routine reporting. (Fig. 3).
Second, among the hospitals that that reported AST results for at least 30 isolates of a given isolatespecimen type (a subset of hospitals with at least one antimicrobial reported for $90% of isolates), to categorize hospitals with different patterns of AST reporting if the drug did not meet the criteria of routine reporting, we applied the following definitions of high SR/CR, low SR/CR, and nontesting. For each antimicrobial, we calculated the difference in the percentage reported between the agent of interest and the routinely reported agent(s) by hospital. High SR/CR for a given drug-organism (and specimen type) combination was defined by at least one AST result being reported (percentage reported > 0%) and the difference of the percentage reported being $20% compared to the routinely reported drug(s); low SR/CR was defined as the difference of the percentage reported being ,20% compared to the routinely reported drug; and nontesting was defined as no AST result being reported. (Fig. 4) The cutoff of 20% was selected by the authors for the purpose of identifying those with larger differences of reporting across antimicrobials. We calculated the number and proportion of hospitals in each category for different drugs. Antimicrobials are presented using the CLSI M100 Table 1A groupings.
Sensitivity analysis. The methods described above compare the percentage of reporting across drugs within each hospital. However, because the above definitions are based on surveillance reporting, and we did not observe actual clinical reporting within hospitals, we could not differentiate hospitals that implemented SR/CR for clinical use but submitted presuppressed, complete AST data to NHSN, from those that did not implement SR/CR for clinical reporting. To understand the extent of this potential bias, we conducted sensitivity analyses by calculating the proportion of hospitals that consistently submitted $90% AST results (suggesting either no data suppression or complete data were extracted and submitted to the NHSN) for all drugs that are commonly tested. For E. coli this included an aminoglycoside (gentamicin or tobramycin), a narrow-spectrum beta-lactam (ampicillin or cefazolin), a third-generation cephalosporin (ceftriaxone or cefotaxime), a narrow-spectrum beta-lactam combination agent (ampicillin-sulbactam or amoxicillin-clavulanate), any carbapenem, piperacillin-tazobactam, cefepime, a quinolone (ciprofloxacin or levofloxacin), and trimethoprim-sulfamethoxazole. For S. aureus this included oxacillin or cefoxitin, a tetracycline (doxycycline, minocycline, or tetracycline), a macrolide (azithromycin, clarithromycin, erythromycin), vancomycin, trimethoprimsulfamethoxazole, rifampin, clindamycin, linezolid, and daptomycin (blood only).
Assessing impact of selective or cascade reporting on cumulative antibiograms. We produced CAs for the hospitals that submitted AST results for $30 isolates from eligible isolate-specimen types (3). To calculate percent susceptible (%S) for each organism in the CAs, we summed the isolates reported to be susceptible to a specified agent and divided that sum by the total isolates with a reported AST result for the same agent. Between high SR/CR and combined routine reporting and low SR/CR hospitals, we compared the  Microbiology Spectrum medians of hospital-level percent susceptible (%S) using Moods tests and the empirical distributions of hospitallevel %S using a nonparametric two-Sample Kuiper test for the antimicrobials with $20 hospitals in both groups. We analyzed the factors associated with reporting AST results among isolates in high SR/CR hospitals for selected group B agents using log binomial regression models. All statistical analyses were performed using SAS version 9.4 software (SAS Institute Inc, Cary, North Carolina).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 0.6 MB.

ACKNOWLEDGMENTS
The findings and conclusions in this manuscript are those of the author(s) and do not necessarily represent the views of the Centers for Disease Control and Prevention.
None of the authors have conflicts of interest to declare.