The Impact of the Auckland Cellulitis Pathway on Length of Hospital Stay, Mortality Readmission Rate, and Antibiotic Stewardship

Abstract

Background

The Dundee classification of cellulitis severity, previously shown to predict disease outcomes, provides an opportunity to improve the management of patients with cellulitis.

Methods

We developed and implemented a pathway to guide the management of adults with cellulitis based on their Dundee severity class, and measured its effect on patient outcomes. We compared the outcomes in patients admitted to Auckland City Hospital (ACH) between July 2014 and July 2015 (the baseline cohort) with those in patients admitted between June 2017 and June 2018 (the intervention cohort).

Results

The median length of stay was shorter in the intervention cohort (0.7 days, interquartile range (IQR) 0.1 to 3.0 days) than in the baseline cohort (1.8 days, IQR 0.1 to 4.4 days; P < .001). The 30-day mortality rate declined from 1.8% (19/1092) in the baseline cohort to 0.7% (10/1362; P = .02) in the intervention cohort. The 30-day cellulitis readmission rate increased from 6% in the baseline cohort to 11% (P < .001) in the intervention cohort. Adherence to the ACH cellulitis antibiotic guideline improved from 38% to 48% (P < .01) and was independently associated with reduced length of stay.

Conclusions

The implementation of the Auckland cellulitis pathway, readily generalizable to other settings, improved the outcomes in patients with cellulitis, and resulted in an annual saving of approximately 1000 bed days.

Cellulitis is one of the leading infectious causes of hospital admission. We have recently shown that, in adult patients presenting to Auckland City Hospital with a clinical diagnosis of cellulitis, a modified version of a cellulitis severity score, validated in Dundee reliably predicted bacteremia, length of hospital stay, 30-day mortality, and 30-day readmission rate [1, 2]. We concluded that this modified Dundee severity score could be used to guide the management of patients with cellulitis. The Dundee severity score adapted the simple clinical criteria proposed by the Clinical Resource Efficiency Support Team (CREST) to categorize patients with cellulitis into 4 disease severity classes: Class 1—systemically well with no risk factors for failure of oral antibiotic treatment; Class 2—systemically well but with risk factors for failure of oral antibiotic treatment (eg, morbid obesity, severe venous insufficiency, or arterial disease); Class 3—systemically unwell; and Class 4—severely unwell and/or suspicion of necrotizing fasciitis [2].

During our validation of the Dundee Severity Score in our setting, we found that 94% of 806 Class 1 patients were treated with intravenous (IV) antibiotics, likely contributing to prolonged lengths of hospital stays [1]. Thus, we developed a pathway to guide the management of adult patients with cellulitis, based on their Dundee Severity class at presentation, introduced this pathway in our hospital, and measured its effect on patient outcomes.

MATERIALS AND METHODS

Setting and cellulitis pathway development

Auckland City Hospital (ACH) is a 710-bed hospital in Auckland, New Zealand (NZ), that provides secondary and tertiary care for a population of approximately 545 640 people (https://www.health.govt.nz/new-zealand-health-system/my-dhb/auckland-dhb/population-auckland-dhb). In December 2016, an initiative was launched to improve the management of patients presenting to ACH with cellulitis. This was led by clinicians from primary care, the emergency department (ED), medical and surgical inpatient departments, pharmacists, quality improvement experts, and hospital managers; and resulted in the deployment of the Auckland cellulitis pathway, for use in primary care and in secondary care, in June 2017.

Ethics

As a retrospective audit of standard clinical practice, this study was exempt from formal ethics approval by the New Zealand Health and Disability Ethics Committee.

Key pathway features

The Auckland cellulitis pathway allocates patients to 1 of 4 classes based on their early warning score (EWS), but uses the NZ EWS instead of the UK EWS [3]. The NZ EWS utilizes the same clinical assessments (temperature, respiratory rate, blood pressure, heart rate, oxygen saturation, and level of consciousness) as the UK EWS, plus one further assessment (requirement for supplemental oxygen) (Supplementary Table 1). The pathway also takes into account the presence of features shown to be associated either with an increased risk of failure of oral treatment (morbid obesity/body mass index (BMI) > 40, symptomatic venous insufficiency, symptomatic peripheral vascular disease, >9% body surface area affected), or an increased risk of cellulitis related mortality (eg, chronic obstructive pulmonary disease (COPD), end-stage renal failure (ESRF), decompensated congestive heart failure (CHF), decompensated cirrhosis, severe immune suppression). The Auckland cellulitis pathway recommends: oral antibiotic treatment in the community for Class 1 patients, intravenous (IV) antibiotic treatment either in the community or in hospital for Class 2 patients, and IV treatment in hospital for Class 3 and 4 patients (Table 1). The pathway encourages clinicians to discharge patients from hospital once systemic illness has resolved, with oral or IV antibiotic treatment to be continued in the community. The pathway recommends that the total duration of antibiotic treatment in most patients should be 5 days.

Pathway implementation

Implementation of the pathway was facilitated by educational sessions held throughout the hospital, hospital staff communications, and by the participation of a team of nurse specialists already involved in facilitating hospital discharge for patients without cellulitis. This team of nurses reviewed patients with cellulitis in ED and on inpatient wards whenever their workload allowed.

Participants and data collection

The impact of the pathway was determined by comparing the management and outcomes for patients in the baseline cohort, who presented to ACH between 1 July 2014 and 30 June 2015, with the management and outcomes for patients in the intervention cohort, who presented to hospital during the 12 months following the deployment of the pathway (12 June 2017 to 11 June 2018). In order to ensure there were no unintended harms created by the pathway we compared 30-day all-cause mortality and 30-day cellulitis specific readmission rates between the baseline and intervention cohorts.

Patients were identified using the same 2 methods in each time period. First, adult (>15 years) inpatients with cellulitis were identified from discharge coding using all codes in the ICD-10 L03 group. Second, a word search of electronic discharge summaries for diagnoses of “cellulitis” was performed for patients discharged within 3 hours of presentation, who do not receive formal discharge coding.

Demographic data were provided by the information management system. Clinical data were obtained by reviewing all patients’ clinical records. Patients were excluded if the coding or diagnosis was incorrect. The discharge services were identified as: ED, general medicine, medical specialties (eg, renal medicine, oncology), ophthalmology, otorhinolaryngology, orthopedics, and surgery (neurosurgery, vascular surgery, and general surgery).

Modified dundee classification

The modified Dundee scoring system used to classify our patients during the intervention period is shown in Table 1. There was a change in the EWS used at ACH between the baseline period and the intervention period. Prior to 2016 the ACH EWS was used to classify the patients, while after 2016 the NZEWS score was used (Supplementary Table 1). The ACH EWS scores used to classify patients in the baseline cohort were: Classes 1 and 2: <5, Class 3: 5–10, Class 4: >10; and the NZ EWS scores used to classify patients in the intervention cohort were: Classes 1 and 2: <5, Class 3: 5–10, Class 4: >10.

Analysis

Antibiotic adherence was defined as treatment with the correct antibiotic, in the correct dose, delivered by the correct route recommended by our antibiotic guideline. Antibiotic treatment was defined as “correct” if the guideline-recommended antibiotic was used, even if the dose or route was not concordant with our antibiotic guideline. Patients for whom a partially or fully completed pathway form was present in their medical records were considered to have documented evidence of pathway use. Because the ED used an electronic version of the pathway, we were unable to determine the proportion of ED patients for whom the pathway had been used.

Comparisons of the demographic and clinical factors between the baseline and intervention cohorts were performed using Chi-squared tests. Because the distribution of length of stay (LoS) data was highly skewed, the univariate comparison between cohorts for LoS was performed using Mann-Whitney tests. Logarithmic transformation of LoS data was required for multivariate analysis using multiple linear regression, to determine whether LoS was independently associated with baseline and intervention cohorts. The impact that each significant factor had on LoS is presented as a per cent change in comparison with the reference group, due to the requirement for logarithmic transformation. Binomial logistic regression was used to determine whether the 30-day cellulitis readmission rate was independently associated with each cohort. All statistical analyses were performed using SPSS v25 (IBM, Armonk, NY, USA).

RESULTS

Demographic and clinical information

The baseline cohort, recruited during 2014–2015, comprised 1092 patients, and the intervention cohort, recruited during 2017–2018, comprised 1433 patients. The demographic features of each cohort were similar, although women and people of NZ European ethnicity comprised lower proportions of the intervention cohort (Table 2).

The proportion of patients discharged directly from the emergency department was 26% in the baseline cohort, and 36% in the intervention cohort; and the proportion of patients discharged from the orthopedic service was 13% in the baseline cohort and 5% in the intervention cohort (P < .01).

Pathway impact on provision of guideline adherent treatment

Documented evidence of pathway use was present for only 113/866 (13%) of the cases in the intervention cohort who were discharged from services other than the ED. The rate of documented evidence of pathway use was 103/509 (20%) for patients cared for by general medicine, and 10/357 (3%) for patients cared for by all other departments.

The proportion of patients who received guideline adherent antibiotic treatment increased from 38% in the baseline cohort to 48% in the intervention cohort (P < .01). The increase in provision of guideline adherent treatment was most marked for Class 1 patients—in the baseline cohort only 102/628 (16%) patients received adherent treatment compared with 253/799 (32%) in the intervention cohort (P < .01).

Pathway impact on length of stay

The median LoS was significantly lower in the intervention cohort (0.7 days, IQR 0.1 to 3.0 days) than the baseline cohort (1·8 days, IQR 0·1 to 4·4 days; Mann-Whitney U test, P < .001). The mean LoS also differed by 1.1 days between the 2 cohorts. The impact of the pathway on LoS was not able to be assessed independently of the discharge service—there was a significant interaction between cohort and discharge service (P < .001). LoS was independently associated with age (0.4% increase for each additional year of age), with Māori or Pacific ethnicity (9% and 7% increase, respectively, in comparison with people of NZ European ethnicity), with cellulitis of the face or head (18% increase in comparison with cellulitis of the lower limb), with Dundee Class (65% increase in Class 4 in comparison with Class 1), and with antibiotic treatment that was not guideline adherent (14% increase for non-adherent treatment in comparison with adherent treatment) (Table 3).

Due to the significant interaction between cohort and discharge service, we examined the factors associated with LoS separately for patients discharged from ED, General Medicine, and all other services combined (Table 4). In all 3 groups, longer LoS was associated with increasing age, and with treatment that was not guideline adherent. The median LoS for ED patients treated with oral antibiotics (0·10 days, IQR 0.06) was significantly shorter than for patients treated with IV antibiotics (0.13 days, IQR 0.09; Mann Whitney P < .001).

Impact of the pathway on 30-day mortality and rate of hospital readmission

The proportion of patients who died within 30 days of hospital admission was 1.8% (19/1092) in the baseline cohort and 0.7% (10/1362) in the intervention cohort (P = .02). The proportion of patients who were readmitted with cellulitis within 30 days of discharge was 6% (60/1092) in the baseline cohort and 11% (144/1362) in the intervention cohort (Chi-squared P < .001). Binomial logistic regression did not identify any factors, including cohort, that were independently associated with readmission with cellulitis within 30 days of discharge.

LoS and readmission rate in relation to route of administration of flucloxacillin

Overall, 62%(1525/2454) patients received flucloxacillin alone as treatment for their cellulitis; 63% (692/1092) in the baseline cohort and 61%(833/1362) in the intervention cohort (P = .3) The proportion of patients treated with only oral flucloxacillin increased from 18%(123/692) in the baseline cohort to 29% (238/833) in the intervention cohort (Fisher’s exact test P < .001). Not surprisingly, the median LoS was significantly lower for patients treated with oral flucloxacillin (0.2 days, IQR 2.2) in comparison with patients who were initially administered intravenous flucloxacillin (2.1 days, IQR 4.0; Table 5). However, the 30 day rate of readmission with cellulitis was not significantly different for patients treated with only oral flucloxacillin and patients in whom flucloxacillin was initially administered intravenously.

DISCUSSION

This large cohort study found that the introduction of a pathway for the management of adults with cellulitis was followed by a substantial improvement in antibiotic guideline adherence (from 38% to 48%), which was associated with a reduction in median LoS of more than 1 day. There was a reduction in 30-day mortality (from 1.8% to 0.7%), but an increase in the rate of readmission with cellulitis within 30 days of discharge (from 6% to 11%). A reduction in median LoS of 1.1 days across cellulitis admissions of one year would be expected to save approximately 1200 bed days at our institution. The median LoS of patients readmitted to hospital within 30 days of discharge was relatively short (2.1 days), and taking approximately 80 excess readmissions into account, the pathway saved approximately 1000 bed days each year. The pathway also increased the proportion of patients discharged from the ED by 10%, and these avoided admissions almost certainly resulted in a substantial number of additional bed days saved.

These effects were consistent with the goals of the pathway: to encourage treatment with oral flucloxacillin and care in the community for patients with Class 1 disease, to encourage early discharge and transition to treatment with oral flucloxacillin for patients with Class 2 disease, to encourage intravenous flucloxacillin treatment for patients with Class 3 disease, and to encourage more intensive treatment of patients with Class 4 disease. We consider these changes in management facilitated our goal to reduce the rate and duration of hospital admissions in patients with Class 1 and Class 2 disease.

The main impact of the pathway on LoS was seen for patients discharged from inpatient services—the median LoS in the ED was 2.8 hours, and an 18% increase in LoS in the ED corresponded to approximately 30 min. Patients in the ED, who were treated with IV antibiotics, stayed for an hour longer than patients who were treated with oral antibiotics. It is likely that further gains could be made if the use of IV antibiotics reduced in Class 1 patients, who made up the majority (76%) of the ED patients treated with IV antibiotics.

We were concerned that increased use of oral flucloxacillin to treat systemically well patients with cellulitis might have contributed to the increased 30-day readmission rate. None of the factors we examined were independently associated with readmission, and the readmission rate for patients treated with oral flucloxacillin was not different to that of patients treated with IV flucloxacillin. While this comparison between oral and IV flucloxacillin should be interpreted carefully due to the many biases inherent with selection of antibiotic therapy, we found that oral flucloxacillin treatment was satisfactory for a large number of systemically well patients with lower limb cellulitis.

The pathway form was not often used as we intended. It was designed to complement the clinical notes and we expected clinicians to fill out a form for each patient. We think that a large number of forms were used for patient care, because 2 print runs of 500 forms were performed during the intervention period. However, we found that the records of only a minority (13%) of inpatients contained evidence that the pathway form was used. It is likely that pathway forms were added to patients’ notes and referred to for clinical decision making, but many of the forms were not filled out (even partially) and were not kept in the clinical record after discharge.

Other studies have also demonstrated the benefits of clinical pathways for the management of cellulitis. Adherence to antimicrobial guidelines for the treatment of cellulitis has been associated with an reduced risk of a composite of death, admission to ICU, or surgical procedure 5 days or more after initiation of antibiotics [4]. The introduction of a clinical pathway or guideline for the management of cellulitis has also been shown to reduce use of unnecessarily broad-spectrum antibiotics, antibiotic duration, unnecessary investigations, and cost of care without increasing risk of treatment failure [5–8]. However, these studies are limited by small numbers of participants and use of different interventions limiting their adoption in other settings. In contrast, our large study adds to an increasing number of publications evaluating the Dundee severity classification [1, 2, 9–12]. The modified Dundee classification used in our study has been validated at our institution and is very similar to the original (uses NZ EWS rather than SEWS). A recent prospective, observational study from Norway evaluated a different modification of the Dundee classification, including alternative criteria in addition to an elevated early warning score for the definition of class IV: a quick sequential organ failure assessment (qSOFA) of greater than 1 or clinical suspicion of a necrotizing soft tissue infection (NSTI) [11]. Their modification of the Dundee classification was found to be slightly more sensitive (83% vs 75%) than the original Dundee classification for a composite outcome of “complicated course” in class III and IV patients. The Norwegian study population was skewed towards severe disease: 54/219 (25%) were Class IV patients, including 19 (9%) with NSTI. The finding of a higher proportion of class IV patients than in our study (25% vs 3%) might be explained by the inclusion of multiple criteria for class IV criteria, but also by the fact we did not include NSTI ICD-10 codes in our study. Regardless, our study, together with the Norwegian study, has demonstrated the successful introduction of the Dundee Classification into a clinical pathway that is almost certainly generalizable to other healthcare settings. If the Dundee classification were to be more widely used in future cellulitis research, it might help to align research efforts to answer important clinical questions, highlighted in a recent review of the cellulitis management [13].

In addition to the points already mentioned, our study is limited in the same manner as described in the pre-intervention cohort, although our retrospective identification of cellulitis cases was improved by the use of free-text search of discharge documentation in addition to clinical coding.

CONCLUSION

The introduction of a clinical pathway for cellulitis management at Auckland City Hospital, based on modified Dundee Classification, was associated with improved antibiotic guideline adherence, reduced median length of hospital stay, and reduced 30-day mortality, despite limited measurable use of the Clinical Pathway document. These findings are consistent with other published accounts of the benefits of clinical care pathways on cellulitis management and clinical outcomes. An increase in 30-day readmission was observed, but readmission was not independently associated with the route of flucloxacillin administration or any of the other factors we measured. To the best of our knowledge, this study represents the first “real-world” demonstration of the impact and safety of incorporating the Dundee classification into routine clinical care for patients with cellulitis. The Dundee classification should now be considered sufficiently validated to warrant wider use, including as a standardized tool for severity assessment in future trials of cellulitis management.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. The authors thank Bret Vykopal and Paul Birch from Performance Improvement at ADHB for coordinating the development of the ADHB Cellulitis Clinical Pathway. They thank all members of the multidisciplinary team who helped to develop the pathway, as well as the nurse specialists who helped to implement the pathway in addition to their usual workload, and to Business Intelligence at Auckland City Hospital for their assistance with data collection.

Financial support. This work was supported by the A+ Trust who funded Summer studentships to HW and RG. The funder had no involvement with the planning, design, analysis, or writing of this study.

Potential conflicts of interest. The authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest.

References