Discussion

The primary finding of this study is that antibiotic use is strongly associated with a discharge diagnosis of infection, factors associated with infection (chief complaint of fever, elevated WBC counts, higher proportion of neutrophils, bacteriuria), and duration of hospitalization. The model was excellent, with a C-statistic of 0.907, and the model, although not linear, had the lowest and highest 2 deciles of risk of 4.3% and 94.4%, respectively. This model might be adopted to compare antibiotic use in various settings.

However, several predictors in this model such as the discharge diagnosis of infection at discharge and length of stay categories might be influenced by coding and downstream clinical decisions, so we included models that considered only variables available on admission. The model predicting antibiotic use was good, and there was also a graded risk, with the lowest and highest 2 deciles of risk of 10.1% and 89.2%, respectively. The model identified 1 of the 3 internal medicine departments with a lower odd of using antibiotics.

There are limited attempts to predict antibiotic use in the hospital, but the results of studies with differing methodology and inclusion criteria are mostly consistent with our results. A study in Norwegian hospitals found that the hospital’s antibiotic use was associated with short and long hospital stays, ICD-10 infectious codes, and more nurse availability. ^{Reference Haug, Berild, Walberg and Reikvam1} A study across 576 US hospitals found that 65% of patients received antibiotics, which was associated with geographic location, a bacterial infection diagnosis, and younger age. ^{Reference Goodman, Cosgrove and Pineles8} A study in Israel in 2003–2004 of 26 internal medicine departments reported that the only variables significantly contributing to predicting the defined daily doses (DDD) per 100 bed-days were hospital affiliation, which explained 43% of the variance, and the rate of 1-day hospitalizations, which explained only 7.4% of the variance. ^{Reference Shalit, Low and Levy9} Variables not contributing to the model included age, gender, diagnosis of infectious diseases, neutropenia, respiratory support, and the Charlson score. We found that those hospitalized for up to 48 hours had a lower prevalence of antibiotic therapy, consistent with an increased wait-and-see approach in patients who might not have required hospitalization. Poran et al found that antibiotic use after 48 hours from admission, prescribed in 8.1% of patients, was associated with male gender, poor functional capacity, prophylactic antibiotic use, transplantation, urinary catheterization, the use of a central venous catheter, poor mental status, and low albumin. The model was only fair with a C-statistic of 0.72. ^{Reference Poran, Elbaz and Turjeman3} Only 0.4% of patients in this study had antibiotic treatment starting more than 48 hours after admission, and although male gender, dementia, low albumin, and permanent catheter were associated with antibiotic therapy, they did not add to the multivariate analysis. Beck et al studied patients from mostly hematological, oncological, and cardiothoracic surgery departments selected with planned or high likelihood of receiving systemic antibiotic treatment for ≥5 days within the first 10 days of their stay; 52.8% received antibiotics, and they found that key predictors included hematologic malignancies, immunosuppressive medication, and past hospitalizations with a C-statistic of 0.74. ^{Reference Beck, Koll and Dumpis4} We also found that previous hospitalization in the 28 days predicted the use of antibiotics that are limited but had a lower odds ratio for the association with any antibiotic therapy. In a systematic review, Lim et al found that patients seen in the emergency department with normal C-reactive protein levels and positive influenza tests were less likely to receive antibiotic treatment. ^{Reference Lim, Htun, Ong, Guo and Chow5} Nonclinical factors associated with antibiotic use were longer waiting time and perceived patient desire for antibiotics. Patients attended by internal medicine physicians co-managed by house staff or who visited an Emergency Department that provided education to healthcare providers on antibiotic use were less likely to receive antibiotics. We did not study those variables. Clearly, studies of internal medicine patients in other settings are needed to confirm our findings.

A secondary finding of our study was that nearly a quarter of the total antibiotic days were in patients with short hospital stays, antibiotic treatment over 7 days, and in those without a primary diagnosis of infection. Selecting those outcomes might aid in comparing internal medicine departments and focus audits and feedback interventions. The models predicting those outcomes were not as strong as for associations with antibiotic use, but nevertheless identified differences between the Departments that might aid in focusing attempts to change antibiotic prescribing behavior.

Short hospital stays of under 48 or 72 hours may suggest that a patient could have been cared for in an alternative setting, such as a high-acuity clinic. ^{Reference Conley, O’Brien, Leff, Bolen and Zulman6,Reference Cram10,Reference Lynch, Fitzgerald, Corcoran, Buckley, Healy and Browne11} Also delayed antibiotic prescription, using a wait-and-see approach with re-evaluation of the patient, has been reported as a useful tactic to help reduce antibiotic use, particularly for respiratory tract infections. ^{Reference Stuart, Hounkpatin and Becque12} A before-and-after trial compared universal early antibiotic therapy (aggressive approach) to a second period when only hypotensive patients received immediate antibiotic therapy. ^{Reference Hranjec, Rosenberger and Swenson13} After the change in policy, patients underwent therapy only after microbiological confirmation of infection and had a higher rate of initial appropriate therapy, shorter duration of therapy, and lower mortality. Further studies are warranted to determine when postponing treatment is safe and can reduce hospital antibiotic use. Chart reviews are required to determine if in-hospital systemic antibiotic therapy might be reduced in this group of patients.

Of the total drug days, 13.6% were in patients with over 7 days of antibiotic therapy, and 3.4% of the drug days occurred after day 7 that might have been excessive. The proportion of inappropriate antibiotic therapy days is unclear, but for most infections, treatment should be limited to 5 days or less. For example, the American College of Physicians recommends that treatment for chronic obstructive lung disease exacerbations and community-acquired pneumonia be limited to 5 days. ^{Reference Lee, Centor and Humphrey14} In US hospitals, unsupported treatment was found for community-acquired pneumonia in 79.5%, most commonly for excess duration. ^{Reference Magill, O’Leary and Ray15} Chart review is required to determine the proportion of days that were not indicated in that group and those receiving antibiotics without a primary diagnosis of a bacterial infection.

This study has several strengths. We selected a homogeneous cohort of internal medicine patients, excluding elective admissions and hospitalization in intensive care, cardiology, and surgical departments, increasing internal validity. Antibiotic use was based on consumption rather than prescription. The exploratory model was confirmed during the subsequent year, and the final model was excellent.

There are also limitations. The study was done in a single hospital, and extrapolation to other settings should be done with caution. Even for internal medicine departments, studies are needed to confirm our results because it is unclear if variable usage rates will affect the observed associations. However, we found the rate of total drug days of treatment to be 51.8 DOT per 100 HD, consistent with the findings in 5 internal medicine departments in Canada with usage rates that ranged from 45.4 to 58.1 DDD per 100 HD. ^{Reference Langford, Brown, Chan and Downing16} Also, a systematic review of antibiotic consumption in acute care hospital medical departments reported use of 59.3 per 100 hospital days in studies where collection methods were clearly reported. ^{Reference Bitterman, Hussein, Leibovici, Carmeli and Paul17} Still in that review, methods of data collection were significantly associated with consumption rates. The factors included data sources (dispensing vs purchase vs usage data), the counting of admission and discharge days, and inclusion of low-consumption departments. They concluded that major heterogeneity precluded defining acceptable antibiotic consumption ranges in acute care hospitals. There is also significant international variation in the consumption of antibiotics with limited indications that may be an important consideration when extrapolating our results to other settings. For example, in our hospital, only 4.1% of patients were treated with piperacillin/tazobactam, and 1.1% with anti-MRSA(methacillin-resistant Staphylococcus aureus) treatment, whereas in US hospitals, it was 14.5% and 23.9%, respectively. ^{Reference Goodman, Cosgrove and Pineles8} Second, we did not do chart reviews, so we cannot determine acceptable antibiotic consumption ranges in internal medicine departments. Also, the definition of an infectious disease did not include those with other diseases that might be considered infectious like chronic obstructive lung disease and asthma. However, those cases made up only around 5% of the cohort and, if included, did not change the models significantly (results not shown). Finally, the model intended for retrospective risk adjustment for setting comparisons should not be extrapolated to attempts to predict antibiotic use at admission. Nevertheless, the model that only included admission variables was also good (C-statistic > 0.8).

We conclude that adjustments for non-modifiable factors, such as the ones we have established, may increase confidence in observed surveillance results. Furthermore, nearly a quarter of the total antibiotic days were in patients with short hospital stays, antibiotic treatment over 7 days, and in those without a primary diagnosis of infection. Selecting those outcomes might aid in comparing internal medicine departments and focus audits and feedback interventions. The results need to be confirmed in other settings.