Kawasaki Disease
Conditions
Keywords
Kawasaki Disease, Machine-learning, Artificial Intelligence (AI), Decision support, Point-of-care
Brief summary
Evaluating the impact of a machine-learning clinical decision support tool on provider practice when evaluating febrile patients with Kawasaki Disease (KD) and non-KD illnesses.
Detailed description
Following laboratory evaluation, providers will be randomized to treat patients according to usual practice/standard of care vs. receiving clinical decision support from the Kawasaki MATCH tool - a previously validated machine-learning clinical decision support tool to identify Kawasaki Disease. The study aim is to evaluate the accuracy of Kawasaki MATCH prospectively when used at the point of care, as well as how this tool impacts clinical decisions including additional evaluation and hospital admission.
Interventions
OTHERKawasaki MATCH
Providers access the Kawasaki MATCH decision support tool. Patient information is entered into the tool and a risk score is indicated to the provider. Kawasaki MATCH is a previously validated machine-learning decision support tool for the diagnosis of Kawasaki Disease. This tool utilizes patient age, 18 laboratory features, and 5 clinical features to formulate a risk score.
Sponsors
Gordon and Marilyn Macklin Foundation
Rady Children's Hospital, San Diego
University of California, San Diego
Study design
Allocation
RANDOMIZED
Intervention model
PARALLEL
Primary purpose
DIAGNOSTIC
Masking
NONE
Eligibility
Sex/Gender
ALL
Age
30 Days to 17 Years
Healthy volunteers
No
Inclusion criteria
* Measured or subjective fever for \>= 3 calendar days * One measured fever \>= 38.0 C (home or in ED) * One or more clinical feature of Kawasaki Disease including: * Rash * Conjunctival injection * Oropharyngeal changes * Extremity changes (erythema, edema, desquamation) * Cervical adenopathy (\>=1.5cm) * Infants \< 6 months of age with \>= 7 days of fever eligible even if none of the above clinical features * Requires IV/phlebotomy for clinical evaluation
Exclusion criteria
* Congenital or Acquired Immune function * Genetic disorders * Current systemic steroid, immunosuppression, or chemotherapy treatment (not including inhaled steroids)
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Time to Kawasaki Disease treatment (KD patients only) | 90 days | Time in days from initial ED evaluation to initial IVIG treatment in patients ultimately diagnosed with Kawasaki Disease |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Hospital Admission Rate | Day 1 (day of enrollment) | Proportion of patients admitted to the hospital from the Emergency Department (ED). Admitted patients = 1 Discharged patients = 0 |
| Kawasaki Disease consultation rate | Day 1 (day of enrollment) | Proportion of patients receiving Kawasaki service consultation |
| ED return visit | 7 days | Return to Emergency Department |
| Kawasaki MATCH score | Day 1 (day of enrollment) | MATCH score determined by entering patient clinical and laboratory information into the Kawasaki MATCH algorithm. Algorithm output is a decimal between 0 and 1. Scores less than 0.4 will be deemed a prediction of not Kawasaki Disease while scores of 0.4 or higher will be deemed Kawasaki Disease |
| Provider assessment of KD likelihood - baseline | Day 1 (day of enrollment) | Visual analog scale of suspected likelihood of Kawasaki Disease. Scale between 0 (Definitely not Kawasaki Disease) and 100 (Definitely Kawasaki Disease) with higher values indicating higher provider suspicion for Kawasaki Disease. This outcome will be assessed on providers randomized to both study arms |
| Provider assessment of KD likelihood - after algorithm | Day 1 (day of enrollment) | Visual analog scale of suspected likelihood of Kawasaki Disease. Scale between 0 (Definitely not Kawasaki Disease) and 100 (Definitely Kawasaki Disease) with higher values indicating higher provider suspicion for Kawasaki Disease. This outcome only assessed for providers in the experimental arm. |
| Algorithm helpfulness | Day 1 (day of enrollment) | Visual analog scale of the perceived helpfulness of the Kawasaki MATCH algorithm. Scale between 0 (Not helpful) and 100 (Extremely helpful) with higher values indicating increased helpfulness. This outcome only assessed for providers in the experimental arm. |
| Additional interventions | Day 1 (day of enrollment) | Additional ED orders for laboratory/imaging studies or additional procedures after the time of randomization |
Countries
United States
Contacts
Primary ContactMichael A Gardiner, MD
Outcome results
None listed