Vomiting is one of the most common complications of cancer therapies in pediatric
patients, with substantial negative impacts on quality of life. Vomiting can also reduce
oral intake, worsen nutritional status and lead to hospitalization. Thus, efforts to
control vomiting are crucial. The ability to predict which patients are most likely to
vomit is limited; machine learning (ML) is a promising approach. Preliminary work
completed for this study includes development of an enterprise data warehouse sourced
from Epic suitable for ML named SickKids Enterprise-wide Data in Azure Repository (SEDAR)
and validation of vomiting outcomes in SEDAR. Next, a standardized process for model
training, evaluation and deployment was conducted by the study team. This was implemented
to train a retrospective model to predict vomiting (0-96 hours post prediction time),
which demonstrated satisfactory performance during a prospective silent trial. The care
pathway and patient-specific report to facilitate clinical care based on a positive
prediction has also been created by the study team, expending on a previously developed
antiemetic care pathway based on clinical practice guidelines. The patient-specific
report lists each patient's risk of vomiting (0-96 hours post prediction time), vomiting
prior to prediction time, planned chemotherapy or procedures, current antiemetic orders
and history of vomiting with the most recent admission.
For model deployment, pharmacists will be charged with bringing information about
patients' vomiting risk to the attention of the medical team and implementing care
pathway-consistent interventions. Pharmacists will receive a list of high-risk patients
and the developed tools (care pathway and patient-specific report) each morning. Outcomes
will be evaluated for a one-year period pre- and post-deployment. Primary outcome will be
any vomiting within the 96-hour period post prediction time. Secondary outcomes will be
the number of vomiting episodes within the 96-hour period, care pathway-consistent care,
antiemetic administrations and antiemetic costs.
The study team includes pediatric pharmacists, pediatric oncologists and experts in
machine learning, clinical epidemiology, implementation sciences, care pathway
development and biostatistics.
Vomiting is one of the most distressing aspects of cancer therapy and, with current
approaches, medical management is failing a substantial number of patients. This work
will contribute to precision medicine by identifying patients with the highest need for
individualized review and therapy optimization. This effort is anticipated to improve the
quality of care and quality of life for pediatric cancer patients.