Association of Models of Care for Kawasaki Disease With Utilization and Cardiac Outcomes

We conducted a multicenter, retrospective cohort study across children’s hospitals in the United States. First, a survey through the Pediatric Research in Inpatient Settings (PRIS) network was used to identify the model type for children’s hospitals. Second, we conducted a multicenter, retrospective cohort study using the Pediatric Health Information System (PHIS) database (Children’s Hospital Association, Lenexa, KS), with the exposure being model of care type and outcomes involving utilization and cardiac outcomes for hospitalized children with KD. Third, chart review at 6 hospitals was conducted to gather additional clinical data not available in PHIS and validate cardiac outcomes. The local institutional review board deemed the PHIS portion of this de-identified study exempt from formal review and approval was obtained at all sites participating in chart review.

Model of care for hospitals within the PHIS database was assessed with a survey distributed via e-mail to PRIS network site leads or pediatric hospital medicine leaders for PHIS institutions not within the PRIS network. Questions regarding institutional KD model of care and clinical management were drafted by 2 authors (N.M., S.W.) using the Research Electronic Data Capture software tool. The survey was piloted by individuals at 4 institutions and modified for clarity. Surveys were distributed to 52 hospitals between July 2019 and October 2019. A copy of the 39-question survey can be found in the Supplemental Information. Institutions were asked regarding recent model of care changes and assigned to the model practiced during the study period. Using survey responses, hospitals were grouped into 1 of 3 models that were defined a priori: Model 1, hospitalist as primary service with as-needed subspecialty consultation; Model 2, hospitalist as primary service with automatic subspecialty consultation upon diagnosis; and Model 3, subspecialist as primary service. Although pediatric hospital medicine is a recognized subspecialty, we will refer to subspecialists as nonhospitalist subspecialists.

Children aged 0 to 18 with KD hospitalized between January 1, 2017, and December 31, 2018, at PHIS institutions where a survey was completed were included. The PHIS database contains utilization and outcome data from participating US children’s hospitals, including patient demographics, hospital characteristics, diagnoses, procedure coding, and clinical and resource utilization data. Patients were identified based on presence of an International Classification of Diseases, 10th Revision discharge diagnosis code for KD. Patients with ≥1 complex chronic condition²²  or congenital cardiac lesions (Supplemental Table 5) and those who did not receive intravenous immune globulin (IVIG) were excluded because IVIG is standard of care for KD and patients who did not receive IVIG during their hospitalization were likely not newly diagnosed with KD.^23,24  Patients initially admitted to the ICU and interhospital transfers were excluded to isolate a specific model of care and identify “typical” KD admissions. Finally, we excluded patients with diagnoses of asthma or croup to allow for evaluation of steroid use by model type.

Utilization and outcome variables were identified a priori by consensus among investigators. Utilization indices analyzed using the PHIS database included cost of hospitalization, length of stay (LOS), emergency department revisits within 7 days, same and all-cause hospital readmissions within 7 days, laboratory tests per 100 patient days, medications received, and imaging studies including electrocardiogram (EKG), echocardiogram, and advanced cardiac imaging. Total cost was standardized by geographic area.

Laboratory tests were examined in aggregate as the total number of laboratories obtained per 100 patient days. Laboratory panels (eg, complete blood counts) were counted as 1 laboratory rather than counting each component (eg, hemoglobin) independently. Medications were grouped into antibiotics, steroids, and immune modulators, and imaging studies were grouped into echocardiogram, advanced cardiac imaging, and other imaging. Details of utilization classification are presented in Supplemental Tables 6–8.

KD outcomes analyzed via PHIS were modeled after a prior study²⁴  and included coronary artery aneurysms (CAA) during hospitalization and requirement of anticoagulation medications beyond aspirin.

Chart review was conducted for all patients with KD identified through the PHIS database at 6 freestanding children’s hospitals (2 from each model type), which comprised a convenience sample based on coinvestigator availability to conduct chart review. The chart review was conducted to obtain data not available in PHIS, including coronary measurements and KD presentation/course. KD type (complete versus incomplete) was found by reviewing progress notes and defined using the American Heart Association definitions.²⁵  Treatment refractory KD was found by reviewing progress notes and medication administration records and was defined as patients with recrudescent fever after initial treatment with IVIG who received a second IVIG dose or another medication, with the stated indication being refractory KD. Medication variables were found in the medication administration record and included administration timing for IVIG, steroids, and immune modulators.

Initial and subsequent echocardiogram reports up to 12 weeks postdischarge were reviewed and CAAs were defined as mild, moderate, or giant.²⁵  CAAs were evaluated for persistence, improvement (≥20% decrease in diameter), or worsening (≥20% increase) on the basis of previous study definitions.²⁶  We also used these reports to assess follow-up echocardiogram adherence, defined as 2 postdischarge echocardiograms within 12 weeks of discharge, on the basis of American Heart Association recommendations.²⁵  CAAs that developed postdischarge were analyzed.

To standardize Z-score measurements, we collected height, weight, and coronary artery measurements and manually calculated each Z-score.^27,28  PHIS codes for CAAs were also assessed for accuracy using echocardiogram reports as the reference standard.

Descriptive statistics were used to describe model types and model characteristics and compared using χ² or Kruskal-Wallis tests. Generalized linear mixed-effects models using a random intercept term to account for patient clustering within a hospital were used to compare utilization and cardiac outcomes among the 3 models. Continuous variables were log-transformed before modeling. Utilization rates were modeled with a Poisson distribution and binary variables were modeled using binomial distributions. Sensitivity and specificity were calculated, along with 95% confidence intervals (CIs) for validation of CAAs in PHIS. All statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC) and P values < 0.05 were considered statistically significant.

Surveys were completed by 44 of 52 (87%) hospitals, of which 21 (48%) identified as Model 1, 19 (43%) as Model 2, and 4 (9%) as Model 3. For Model 2 institutions, services automatically consulted included cardiology (32%), infectious disease (37%), rheumatology (5%), and multiple services (26%). Model 3 institutions models of care were composed of infectious disease as primary (n = 3, 1 of which is a special “KD team” of 2 infectious disease physicians) and multisubspecialty comanagement (n = 1). Seven hospitals were excluded for survey nonparticipation.

During the study period, 4111 patients were admitted with KD within 44 PHIS-participating hospitals. After applying eligibility criteria, 2080 (51%) patients met criteria for inclusion (Fig 1); the most common reasons for exclusion were the presence of ≥1 complex chronic condition (n = 747, 18%), interhospital transfers (n = 607, 15%), and not receiving IVIG (n = 433, 11%).

Demographic information is found in Table 1. Age distribution was similar among model types and the population exhibited a male predominance (58%). Model 1 had a higher proportion of children with non-Hispanic Black race, government insurance, and hospitalization in the southern United States. A higher proportion of children in Models 2 and 3 had private insurance and were hospitalized in the western United States (P < .001). In terms of the hospital characteristics within models, the annual case volume for KD cases was highest in Model 3 hospitals, with 82% of hospitals with >56 cases per year as compared with 43% and 39% of hospitals in Model 1 and Model 2, respectively (P < .001). The average daily census was also higher in Model 3 hospitals, with 82% of hospitals with >300 patients daily as compared with 44% and 33% of hospitals in Model 1 and 2 (P < .001).

Utilization data for the administrative database cohort can be found in Table 2. Total standardized cost for hospitalization was higher among Model 2 and 3 institutions, but Model 1 institutions ordered more laboratory tests per 100 patient days (both P < .001). Categorized frequency details of ordered laboratory tests are included in Supplemental Table 9. Median LOS did not vary (P = .57) and no difference was found in emergency department revisits and same-cause readmissions within 7 days. Model 3 institutions used more immune modulators, obtained more EKGs, and had a higher percentage of patients who received >1 echocardiogram while inpatient (all P < .001). Antibiotic use, advanced cardiac imaging, and non-KD–related imaging were similar among models (Table 2).

Cardiac outcome data from the administrative database cohort can be found in Table 2. The analysis found CAA rates of 7.7%, 9.3%, and 9.4% for Models 1, 2, and 3, respectively (P = .44). Anticoagulation receipt was similar among models (P = .21).

In the chart review cohort, 6 freestanding children’s hospitals accounted for 485 hospitalizations, or 23% (485 of 2080), of the total cohort. Demographic characteristics for this subset were similar to the administrative data population (Table 3). Rates of incomplete and treatment refractory KD were similar among groups, although day of illness on presentation varied (P = .03), with patients presenting to Model 3 institutions earlier. Additionally, there was no difference in the timing of IVIG dose (P = .48) and Model 3 institutions gave immune modulators earlier, whereas Model 1 institutions gave steroids earlier (Supplemental Table 10).

Patients with at least 1 aneurysm were 27%, 16%, and 23% for Models 1, 2, and 3, respectively (P = .1). Aneurysms were found in nearly a quarter of patients (100 of 485, 21%) in the chart review subset, substantially higher than the administrative data review (177 of 2080, 8.5%). Eighty-five percent (85 of 100) of patients with CAAs had CAA present on the initial echocardiogram, and CAAs at diagnosis did not vary among model types (Table 3). There were no significant differences in improved, worsened, persistent, or new CAAs found at follow-up among model types (Table 4). Giant CAAs comprised 14 of 173 (12%) total CAAs and were found in 6 patients from the same Model 1 institution. One patient had a new giant aneurysm discovered at follow-up that was not present during hospitalization.

The overall sensitivity and specificity of CAA billing codes were 42% (95% CI: 32–52) and 97% (95% CI: 95–99), respectively; these were similar among model types, with sensitivities of 40%, 47%, and 42% and specificities of 97%, 95%, and 99% for Models 1, 2, and 3, respectively.

This is the first large nationwide cohort study to examine the influence of model of care on utilization and patient outcomes for children hospitalized with KD. Our findings further characterize current models of care in the United States. The most common models for managing hospitalized children with KD are pediatric hospitalist as primary, with either as-needed (Model 1) or automatic (Model 2) subspecialty consultation models. Hospital models with subspecialists as the primary service (Model 3) are less common and primarily exist in large children’s hospitals. Patient race, insurance type, and region within the United States varied among model types. Model 1 hospitals were more often located in the southern United States and had a higher proportion of patients with non-Hispanic Black race and government insurance.

Few studies have investigated differences in primary service and consultation practices for KD or other conditions. A survey of pediatric hospitalists found most (95%) were comfortable managing uncomplicated KD without subspecialty consultation, although only 27% routinely did so, and consultation practices varied.²⁰  Our study similarly found <50% of children’s hospitals had hospital medicine as the primary service with as-needed subspecialty consultation and similar comanagement variability for models with automatic consults, with cardiology and infectious diseases being the most common specialties consulted.

Our study found some differences in utilization. Model 1 institutions used more laboratory tests per 100 patient days, whereas Model 3 institutions obtained more EKGs and echocardiograms. We also found differences in adjunctive treatment administration among models. Subspecialist as primary models were more likely to use immune modulators and used them earlier than other models, whereas Model 1 hospitals were more likely to use steroid therapies earlier. These data do not differentiate whether this represented undertesting or overtesting or similarly undertreatment or overtreatment. Subspecialty consultation may lead to improved laboratory stewardship. Conversely, extra laboratory testing might be used by Model 1 providers to avoid unnecessary imaging or immune modulator prescription. Further studies are needed to delineate the impact of subspecialty consultation on utilization of laboratory tests, imaging, and medications. We did not find differences in LOS or readmissions, and differences in cost were modest, with lower cost in Model 1 institutions. This also may reflect geographical differences in treatment preference, and both treatments are beneficial for high-risk or refractory KD cases.^29–32 

Despite differences in utilization, IVIG treatment timing and cardiac outcomes were similar. There was no difference in patient outcomes among models, as represented by CAA frequency/progression and receipt of anticoagulation; however, long-term outcomes for patients with KD are being investigated and difficult to evaluate retrospectively. Additionally, posthospitalization follow-up data from our chart review cohort delineates a low prevalence of new or worsening CAAs across all model types. All giant CAAs in our cohort were from the same Model 1 institution; we postulate this was because patients with giant CAAs at this institution were not necessarily transferred to the PICU, whereas others may have done so.

CAA rates by chart review were higher than those in PHIS. It is unknown whether this difference is because of mild CAAs that were not considered clinically significant or undercoding of CAAs. However, we are reassured that the sensitivity and specificity was similar across models. Future investigations using databases to examine CAAs should take this into consideration. Our results also do not support that automatic subspecialty consultation led to better follow-up as evidenced by similar rates of postdischarge echocardiograms.

Our study has several limitations. First, our results may be impacted by intrinsic, unmeasured-among-hospital differences unrelated to model of care. To minimize this, we chose a single condition (KD) and used the PHIS database, which is comprised mostly of freestanding children’s hospitals. Local contextual factors that influence model types (eg, availability of subspecialists) are unknown and may have impacted outcomes. Additionally, we were unable to adjust outcomes for average daily census or KD annual case volume because these variables were highly associated with Model 3 hospitals. We did not adjust for race, given the small differences in distribution of race across models, and assumed differences in payer and regions to be less impactful to their respective models. Further prospective studies are needed to evaluate if patient or hospital characteristics influence utilization and patient outcomes associated with model types. By using International Classification of Diseases, 10th Revision discharge codes to identify our cohort, we did not capture patients evaluated for, but not diagnosed with, KD, limiting generalizability to this population. Billing data to verify subspecialty consultation is not available in PHIS. As a result, there is a risk of misclassification of Models 1 and 2. However, the objective of this study was to compare general models of care rather than the impact of a specific consultation service’s involvement, which is a point for future studies. Some cardiac outcomes may have been missed if patients followed up outside their admitting hospital system. Statistical comparisons of CAA outcomes by model in the chart review cohort were not possible because of limited sample size and CAA frequency, which hindered our ability to detect small but meaningful differences in CAA outcomes. Lastly, hospitals used for chart review were a convenience sample based on coinvestigator availability. Despite this, the similar percentages of refractory and incomplete KD observed in our study and seen in the literature for other KD populations supports the generalizability of our findings.^33–36 

Hospitalized children with newly diagnosed KD were primarily cared for by pediatric hospitalists with either as-needed or automatic subspecialty consultation. We identified intrinsic differences in patient and hospital characteristics among model types. Evaluation and treatment practices differed; however, outcomes were similar aside from a modest difference in overall cost. Cardiac outcomes were similar in terms of a low proportion of patients with new or worsening CAAs across all models. Future prospective studies are needed to confirm these findings and better isolate model of care impact on utilization and patient outcomes.

We thank the PRIS network for their support and the PRIS site leads that participated in the survey.