Improving Care and Outcomes for Pediatric Musculoskeletal Infections

The QI initiative took place at a large tertiary children’s health care organization in the Midwestern United States with 2 freestanding hospitals (“hospital A” and “hospital B”) managing ∼90 000 emergency department (ED) encounters and 14 000 inpatient admissions annually. Available pediatric specialists include emergency medicine, hospital medicine, orthopedic surgery, ID, and radiology.

In April 2016, a multidisciplinary MSKI working group including representatives from ED, ID, orthopedic surgery, radiology, pharmacy, and hospital medicine was formed. They used the Model for Improvement²⁴  to develop a key driver diagram, specific aims, measurements, and proposed interventions. An early limitation was a lack of readily obtainable baseline data. The working group noted anecdotally that a patient with suspected MSKI was evaluated in the ED and admitted to hospital medicine at that same hospital. In stable patients, empirical antibiotics were held until obtaining intraoperative cultures and ID was consulted. There was perceived provider variation in the transition to oral antibiotics and use of PICC lines.

The MSKI working group identified 4 key interventions (structured as plan-do-study-act [PDSA] cycles 1–4) through the driver diagram (Fig 1). The first cycle (PDSA 1; October 17, 2016) was education to the ID provider group in the form of a didactic presentation reviewing key literature (eg, IV to oral antibiotic transition) followed by consensus-building discussion regarding potential algorithm recommendations. The second cycle (PDSA 2; November 2016) was centralizing admission location through an intrafacility transfer to group patients onto 1 hospital campus (ie, transfer from hospital A to hospital B). In the third cycle (PDSA 3, December 2016), the working group developed a process with orthopedics, radiology, and anesthesiology to arrange MRI and subsequent operating room (OR) procedures (if indicated) under a single sedation. This process was accomplished by a dedicated orthopedic infection MRI slot and direct radiologist-orthopedist discussion at MRI to assure sufficient imaging was obtained to make operative decisions. In the fourth cycle (PDSA 4; March 2017), the working group reviewed existing literature and developed an evidence-based clinical algorithm and order set for suspected MSKI that was embedded in the electronic health record. The algorithm incorporates local resources and antimicrobial resistance patterns, encourages limiting use of PICCs and empirical vancomycin, and sets expectations for provider communication. The PDSA interventions were reviewed in staff meetings for nurse managers, pharmacy, orthopedics, ID, hospital medicine, and ED from October 2016 to March 2017. Iterative feedback from provider experiences with transfer and communication interventions informed the algorithm. The MSKI working group met during implementation to review challenges and results. The working group leader performed ad hoc case review and provided feedback to individual physicians on 5 occasions. In addition, the working group made minor algorithm revisions clarifying communication expectations and provided additional ED and hospital medicine staff education updates 6 months into the implementation period.

Eleven physicians and 1 pharmacist participated in the working group, with meetings approximately every 4 to 6 weeks for 12 months. Additional time was dedicated for literature and data review and in facilitating work from the organization’s guideline team. Incentive for physician participation may have been financial (organizational QI bonus) and/or American Board of Pediatrics Maintenance of Certification credits.

We conducted a prospective QI initiative with a formal retrospective cohort analysis of patient data to assess whether the specific aims were attained. Under the Model for Improvement,²⁴  4 PDSA cycles were conducted, with every cycle initiating the rollout of each intervention. The “study” portion of these cycles was limited by the lack of readily obtainable real-time data, so feedback to interventions was based on anecdotal reports and informal statistics from providers and hospital staff (eg, nursing, pharmacy) and discussed among the MSKI working group. For example, if there was a delay in consultation of orthopedics, the orthopedic team contacted the MSKI working group leader who reviewed the case and provided feedback to the providers involved. On 2 occasions, this also prompted broader reminders to ED and hospital medicine staff via departmental newsletters and in staff meetings.

To conduct more formal study 24 months after the interventions, retrospective data were electronically abstracted from the health record and manually chart reviewed by 1 physician and 1 advanced practice provider. We included patients 6 months to 18 years of age admitted to either hospital campus from January 1, 2015, to April 30, 2019, with an International Classification of Disease discharge diagnosis code (editions 9 and 10) for acute osteomyelitis, septic arthritis, or pyomyositis (Supplemental Information). Patients with complex chronic conditions,²⁵  admission directly to the ICU, or lack of general hospital consent for research were excluded. Patients found on chart review or with diagnostic codes for algorithm exclusion criteria (≥14 days of symptoms; major trauma; postoperative infection, skull, vertebral, hand, and/or foot infection; Lyme disease; concern for necrotizing fasciitis or unusual organism [eg, tuberculosis]; or viral myositis) were also excluded (Supplemental Information; Supplemental Table 6). Assuming a 50% baseline rate of PICC use (estimated via working group consensus and confirmed on baseline analysis), we a priori determined that a sample size of 200 patients would have 96% power to detect a 50% reduction (ie, 25 percentage point decrease) in PICC use.

Patients were grouped into 2 general periods: baseline (January 2015–October 17, 2016) and implementation (October 18, 2016–April 2019). The primary outcomes were the percentage of patients who had a PICC placed, number of sedations (defined as any documented sedation by hospital sedation team, ICU physician, and/or anesthetist) per patient encounter, and percentage of patients receiving empirical vancomycin (administered in the first 24 hours of admission without a positive culture result for Gram-positive cocci at time of admission or a documented cephalosporin allergy). Secondary outcome measures included total hospital length of stay (LOS) and charges (inflation adjusted to April 2019 US dollars by using the Consumer Price Index²⁶ ). Charge categories are available in Supplemental Table 1.

Process measures included the percentage of eligible (ie, had sedated MRI before an OR procedure) patients undergoing combined MRI-OR sedation and the percentage of total patients discharged from hospital B. Balancing measures included related-cause ED and hospital readmission within 30 days of discharge.

To assure the cohort was comparable over the course of the baseline and intervention periods, we compared demographics, clinical features, and management using 2 proportions test for dichotomous variables and 2 sample t tests for continuous variables. We used Shewhart statistical process control (SPC) charts to evaluate if and when changes in outcome, process, and balancing measures occurred.²⁴  P charts (for dichotomous measures) and charts (for numeric measures) used sequential groups of 10 patients. For p charts, the normal approximation to the binomial distribution cannot be applied to the small sequential group sizes, so control limits were calculated by using exact binomial probabilities corresponding to 1, 2, and 3 SDs from the mean under Gaussian variation (eg, 15.9, 2.3, and 0.1 percentiles for lower limits, respectively). Because we wanted to focus on the changes of central tendency as opposed to changes in the variation, we did not display s charts with charts. Control limits were calculated from the baseline period, and the centerline was shifted when special-cause variation was indicated via the Western Electric rules.²⁷  We set the significance level at 5% and used Stata (Stata Corp, College Station, TX), R, and/or SAS (SAS Institute, Inc, Cary, NC) for analyses.

We compared age, race and/or ethnicity, sex, and insurance status pre-post interventions to assure there were no disparities evident in the application of the algorithm. This study was approved by the organization’s institutional review board.

We identified 468 potential patients during the study period, of which 224 patients (93 baseline and 131 implementation) were used for analysis (Fig 2). The mean age was 6.1 years. Patients in baseline and intervention periods were generally comparable with respect to demographic and infection characteristics (Table 1), including rates of MRSA (baseline 4.3% versus implementation 6.1%; P = .77) and bacteremia (baseline 36.6% versus implementation 40.5%; P = .56).

For the primary outcome of PICC use, there was special-cause variation after PDSA cycles 1 to 3 and around the time of implementation of the algorithm and order set with a centerline shift from 54% to 11% (Fig 3) or a 81% relative reduction (95% confidence interval 70–92).

For the number of sedations per patient, special-cause variation occurred after PDSA cycles 1 to 2 with a 33% relative reduction (95% confidence interval 21–46) from 1.8 during baseline to 1.2 after special-cause variation (Fig 4).

We observed no special-cause variation for empirical vancomycin use (centerline 20%; see Fig 5). To better understand the use of vancomycin, we also completed a pre-post analysis of all vancomycin use (ie, receipt at any point in hospitalization) and found no significant change between the study periods (46% baseline versus 38% implementation; P = .23). Of patients who received vancomycin within the first 24 hours, there was no difference in those meeting indications of cephalosporin allergy or known positive culture result in baseline and implementation periods (16% [4 of 25] and 10% [3 of 30] respectively; P = .45).

Special-cause variation was met for average total facility charges with a centerline shift from $67 000 to $52 000 (Fig 6; see also Supplemental Table 1 for charge categories), but criteria for special-cause variation was not met for average LOS (centerline 6.6 days; see Supplemental Fig 7). Combined MRI-OR sedations among eligible patients remained in control over the implementation period with a centerline at 22% (Supplemental Fig 8). Although the percentage of total patients discharged from hospital B remained in control (centerline 83%; see Supplemental Fig 9), a pre-post subanalysis of 61 patients who were initially evaluated at hospital A revealed a large increase in transfers to hospital B from baseline to implementation (26% [6 of 23] to 84% [32 of 38]; P < .001).

The balancing measure of MSKI-related 30-day revisit remained in control (centerline 12%; see Supplemental Fig 10), and PICC-related complication was the most common reason for revisit, which was seen in 7 patients all in the baseline period (see Supplemental Table 2). Supplemental pre-post analysis for patient management and resource use in the baseline versus implementation periods are available in Supplemental Tables 3 and 4.

We describe a QI initiative to improve management of patients with MSKI at a large tertiary children’s hospital. We achieved relative reduction in PICC placement and number of sedations per patient. Total facility charges also decreased despite no difference in LOS. There was no significant change in empirical vancomycin use, rate of combined MRI-OR procedures, and revisits and/or readmissions.

Through implementation of our QI initiative, we found that our early PDSA cycle interventions (ID provider education and centralized admission location) were key strategies for improvement. Although previous MSKI studies were primarily focused on using clinical algorithms to reduce charges, PICCs, LOS, days of IV antibiotics, time to MRI, and readmissions,^18–22,28  we found that improvements related to PICCs and sedations may have begun before algorithm implementation. This distinction is important to highlight because focused stakeholder education and consensus building can occur with little additional resource or financial investment. In planning our initiative, conversations and education with key stakeholders regarding the algorithm occurred during the development phase and likely led to shifts in clinical practice before official PDSA cycles. It is suggested in our study that this multidisciplinary collaboration is a key step. Our implementation of clinical decision support tools may serve to sustain the improvements by providing guidance for new providers or staff. Electronic health record tools such as templated education, order sets, and clinical algorithms have been found in previous studies to be both feasible and effective for sustaining change.^29–32 

We observed a reduction in the number of sedations per patient despite a lack of change in combined MRI and OR procedures. Although we hypothesized that the number of sedations per patient would decrease as a result of combining procedures (PDSA 3), this was not the case. Rather, the reduction in sedations was likely due to reduced PICC procedures, which often require sedation. The lack of change in combined procedures might be attributed to an unexpected high rate of combined procedures among eligible patients at baseline; these data were not available to the working group before the QI initiative. The intervention could have had an impact on other measures such as time to MRI or OR, staff satisfaction with communication, or amount of empty OR time (as a balancing measure). These measures should be considered in future work using this intervention.

Despite our interventions, we did not observe a reduction in empirical vancomycin use.

Because of low local MRSA prevalence (5.4% in our study), vancomycin was recommended only in limited circumstances (eg, cephalosporin allergy or sepsis) and was preferred to clindamycin given local MRSA resistance. Although we elected not to use diagnosis codes to identify patients with sepsis because of concerns for potential misclassification, we note that the frequency of empirical vancomycin use of 20% found in our study is modestly higher than published rates of sepsis with MSKI (9.2%–17.6%).^33,34  Future studies with prospective sepsis definitions may provide more accurate benchmarks for appropriate vancomycin use.

In addition to our specific aims, examination of charge data revealed that total facility charges decreased significantly although less than found in a previous study.²²  We suspect this may be related to less real-time interventions (eg, real-time case review) and fewer organizational QI resources dedicated to our MSKI initiative. Organizations with similar QI resource constraints may find reassurance in the improvements in our study with modest QI investment.

Notably, facility charges measured in our study did not include costs related to interfacility transfers between our 2 hospital campuses, which occurred in ∼25% of the patients in the implementation period. Organizations should also consider that charges, costs, and reimbursements for transfers vary depending on distance, type of transport used, geographic region, payer policies, and so forth. A further limitation to the use of transfers as an intervention is related to satisfaction. In this study, patient and/or family satisfaction related to transfer was not measured and could be considered for future work; however, balancing measures of not transferring patients (eg, delays to OR, staff satisfaction) should also be considered.

With our study, we add to existing QI literature that is concentrated on MSKI in children.^18–22,28  In previous studies, most authors have used pre-post analysis, which cannot account for secular trends. In our study, we use SPC charts, which can identify that the improvements were related to the QI interventions. In a study of 57 patients with osteomyelitis at a children’s hospital with robust QI culture, the authors used SPC charts with n-of-1 testing.²²  We applied similar methodology but used a broader MSKI definition to account for a range of possible diagnoses when patients present with suspected MSKI and did not exclude outliers. This resulted in a sample size that is 5 times larger and may be more clinically applicable to patients who present with undifferentiated MSKI.

The experience of a single center may not be generalizable to other organizations. Although centralizing admission location pertains less to single-center health care facilities, organizations might consider centralizing within their hospital to concentrate clinical expertise. The relative rarity of MSKI in children makes random assignment less feasible, and our study design is susceptible to confounding or bias. However, there were no substantive differences in patient demographics or clinical characteristics in pre-post analysis, and SPC charts revealed no clear preceding trends. Our sample size may have had insufficient power to reveal a decrease in some measures, such as empirical vancomycin. Lack of granular baseline and real-time data for targeted audit and feedback may have affected the ability to perform clear prospective PDSA cycles.³⁵  These challenges may be overcome with more mature QI data analytics infrastructure. Additional barriers included gathering stakeholders and maintaining clear lines of communication.

At our tertiary children’s hospital, the implementation of multidisciplinary MSKI QI interventions was associated with a significant decrease in PICC use, sedations, and charges but not empirical vancomycin. Because multiple centers have revealed improved cared delivery through QI interventions, development of national pediatric MSKI guidelines may be warranted.

We thank Steve Grapentine, PhD; Emily Chapman, MD; William Pomputius, MD; Deepa Pai, MD; Paul Zenker, MD; William Mize, MD; Elizabeth Weber, MD; Stephen Sundberg, MD; and Caitlin Hiniker, PA; for their contributions to the QI initiative.

ED

emergency department

ID

infectious disease

IV

intravenous

LOS

length of stay

MRSA

methicillin-resistant Staphylococcus aureus

MSKI

musculoskeletal infection

OR

operating room

PDSA

plan-do-study-act

PICC

peripherally inserted central catheter

QI

quality improvement

SPC

statistical process control