The challenge

Though the development of AXR as a “utilization metric” is one of its greatest strengths, this also creates challenges in effective interpretation by frontline clinicians. The AXR does not measure the appropriateness of antibiotic prescriptions, as its denominator includes diagnoses for which antibiotics are “sometimes” (Tier 2) or “always” (Tier 1) indicated; that is to say, the target performance will never be 0%. As a result, the absolute rate is less immediately actionable for clinicians, as it does not specify for what diagnoses their prescribing may be above average. The absolute rate is also less informative to clinicians than their performance relative to peers—as the AXR metric can readily identify outlier clinicians whose performance deviates significantly from colleagues in their service line.

In addition, higher-prescribing physicians may seek explanations for why they are an outlier compared to peers. For example, physicians working in an internal medicine clinic may feel that higher rates of antibiotic utilization are justified relative to peers in family medicine clinics. For this reason, it is particularly important to ensure clinicians are aware that the AXR is benchmarked by service line—ie internal medicine clinics are compared only to other internal medicine clinic settings—so that meaningful comparisons with peer facilities can be made. In addition, clinicians often fall victim to the fallacy of believing that their patients are sicker and more likely to require antibiotics than peers’ patients. This misconception is more challenging to address (especially as no risk adjustment is made between clinics within a service line), though emphasizing that the AXR metric excludes patients with certain comorbid conditions (as defined within the HEDIS^® Comorbid Conditions Value Set, who may be perceived as higher acuity or higher risk) may alleviate this concern. ^{12,Reference Stenehjem, Wallin and Willis13} However, given that the AXR code is proprietary, some organizations have developed internally derived AXR measurements—which, importantly, may not exclude identical (or in some cases, any) comorbidities.

Tactics

To optimally utilize the AXR metric while still delivering meaningful and accessible feedback to clinicians, we suggest co-reporting of AXR alongside more concrete, complementary metrics (like URI and AAB). This approach allows clinicians to target their improvement efforts on appropriateness of antibiotics in a specific condition (ie bronchitis) and observe, in parallel, its impact on AXR. If clinicians still find the AXR metric opaque and difficult to interpret, an alternative approach is to monitor AXR at an organization-level while reporting only concrete, disease-specific metrics to clinicians (and providing individual AXR values to requesting clinicians only).

Even for clinicians who lack a nuanced understanding of the AXR measure, providing their AXR trend over time can still serve as powerful feedback. Increases in AXR over time may prompt clinicians to self-reflect and institute changes in their prescribing practices; in contrast, decreases in AXR over time may serve as positive reinforcement for high-performing clinicians who are striving for improvement.

For clinicians who are particularly invested in understanding their antibiotic utilization, providing patient-level data for Tier 3 encounters included in the AXR metric may also be a meaningful intervention. Access to this data enables clinicians to self-audit cases where antibiotics were prescribed to assess the appropriateness of their prescribing and may inform behavior change.

Encouraging organizations to create their own AXR measure with internally derived electronic health record data allows for easier manipulation of the data to co-report complimentary metrics. In addition, it increases the sample size by not limiting the AXR measure to one payor type and makes the measure accessible to a global audience that uses ICD-10 coding.

The Challenge

Because AXR includes a wide array of respiratory tract diagnoses, it is unclear what the “optimal” or “target” rate should be. In contrast to AAB (which includes exclusively Tier 3 conditions), the AXR rate will always be greater than zero. In addition, it is important to recognize that AXR will vary in response to changes in case mix in the population—and that the “optimal” rate may thus likewise change over time. For example, the AXR would be expected to decrease in the setting of a respiratory viral pandemic, while the AXR would necessarily increase in the setting of high community rates of streptococcal pharyngitis (during which antibiotic prescribing will necessarily rise). ^{Reference Hersh, Stenehjem and Fino14,Reference Seibert, Stenehjem and Wallin15} In both examples, other metrics would inform interpretation of AXR changes; for example, in the setting of a community outbreak of streptococcal pharyngitis, CWP rates and percent positivity of test results would shed light on the underlying impetus behind changes in AXR. ^{Reference Hersh, Stenehjem and Fino14} Consequently, the “optimal” AXR rate will be fluid and prone to change in response to community infection rates, and it is crucial for payors to have timely access to disease-specific data to enable accurate interpretation of AXR trends. This fluidity in response to community infection rates also highlights a challenge with comparison across systems with highly discrepant patient populations (in terms of acuity) or in different geographical locations (where infection outbreaks and case rates may differ).

Tactics

One approach is to set the first-year targets at the median rate and subsequently decrease the goal by a pre-determined, incremental amount in subsequent years. This strategy has been adopted by Intermountain Health, which has identified a goal of reducing AXR by 5% per year. It is important to ensure that the median used to inform the target rate is comprised of hospital systems with a similar patient population and case mix to maximize its impact, as highlighted above.

An alternative approach to address this challenge is to not set a target and instead use the metric to identify outlier clinicians, clinic sites, and regions in larger health systems. Notably, providing data on antimicrobial utilization may function as a meaningful intervention even without providing an external goal or target, as outlying high-prescribing clinicians, clinics, and regions may be incentivized to reduce their prescribing based on peer comparison alone, while low-prescribing clinicians may desire to maintain their high-performance.

In addition, AXR trajectory may serve as a harbinger for the durability of stewardship interventions—with rises signaling a potential need to reassess the sustainability of previously-instituted interventions—and serve as an early signal of changes in prescribing culture.

The challenge

While inappropriate prescribing for respiratory conditions is a key area of antibiotic overuse, it is worth noting that there is substantial overprescribing for non-respiratory conditions captured by neither the AXR metric nor the pre-existing URI, AAB, and CWP metrics (including asymptomatic bacteriuria and skin conditions incorrectly diagnosed as cellulitis). Furthermore, AXR captures the dichotomous decision to prescribe an antibiotic but does not reflect whether the right antibiotic, correct dose, or appropriate duration were prescribed. It is well-established that reducing broad-spectrum antimicrobial use and overall duration of therapy are highly effective stewardship interventions—but improvements in either of these areas would not be reflected in the AXR rate, as the overall rate of prescribing would be unchanged. ^{Reference Hersh, Stenehjem and Fino14}

Tactics

Given the important gaps in AXR described above, co-reporting alongside additional modified rate measures is of particular importance. For example, reporting days of therapy per diagnosis (that is, utilizing the AXR metric and multiplying by the number of antibiotic days) would allow for a more comprehensive assessment of overall antibiotic utilization in outpatient settings. This approach has the additional benefit of aligning well with inpatient antibiotic use reporting mechanisms, allowing comparison of antibiotic utilization across systems with integrated outpatient and inpatient systems. Likewise, co-reporting of broad-spectrum antibiotic use rates (for example, reporting the AXR rate with only broad-spectrum or fluoroquinolone use in the numerator) would also allow for more granular understanding of antibiotic selection trends over time, even when overall prescribing rates are stable. ^{Reference Seibert, Schenk and Buckel16,Reference Yogo, Shihadeh and Young17}

The challenge

AXR is the fourth antibiotic stewardship-related HEDIS^® measure, and it may feel daunting to health plan leaders to identify the human and IT resources to report, monitor, and develop interventions to impact AXR. Likewise, clinicians may feel bandwidth is limited to focus on AXR while maintaining ongoing improvement efforts in URI, AAB, and CWP.

Tactics

Though each of the disease-specific HEDIS^® metrics provide distinct diagnostic and antibiotic stewardship data, the cases included in each of these measures are also included in the AXR denominator—and as such, performance improvements in any of these three domains will also lead to improvement in AXR. It may therefore be beneficial to focus limited resources on impacting the disease-specific metric with the greatest clinical impact (based on frequency of clinical encounters in each setting) or lowest overall performance, as quality improvement interventions in individual metrics will yield improved AXR rates, as well.

The challenge

Like other HEDIS^® metrics, a final key limitation of AXR is its inability to account for differences in antibiotic prescribing rates for respiratory conditions between patient groups (including race, ethnicity, preferred language, and rural vs urban environment) served by a given healthcare organization. That is to say, AXR does not provide information about why a difference in antibiotic prescription rates between patient groups exists—rather, only that one does. A critical evaluation of the differences among patient groups will thus be required to understand why disparities exist and what can be done to eliminate them.

Tactics

Since AXR does not yield specific insight into where disparities in care exist, additional data infrastructure is required by health systems to effectively leverage AXR to promote equity in care. To achieve this aim, health systems must prioritize collection and reporting of race, language, and ethnicity data in internal dashboards and reports monitoring antimicrobial use. Monitoring trends in these data alongside AXR following a stewardship intervention or change in care delivery may help clarify if there is differing impact of the change on specific patient groups and guide further evaluation of inequities. Secondly, AXR can be leveraged to engage key patient, community, service line, and organizational stakeholders to understand the potential reasons for variability between patient groups and resolve gaps when inequities are identified.