In Uganda, about 5% of children discharged after hospitalization for a serious infection will die in the weeks after returning home. Doctors and parents are often unaware of this period of vulnerability and are poorly equipped to identify or handle this critical situation. This project builds on past work to develop and evaluate models and technology to predict, before discharge, an individual child's risk of recurrent illness, as well as to provide additional post-discharge support to at-risk children. This study seeks to evaluate the effect of a novel "Smart Discharges" approach on childhood mortality and health seeking behaviour.

PURPOSE:

Our purpose is to conduct an interventional cohort study to evaluate the impact of the Smart Discharges approach to discharge care on pediatric post-discharge mortality.

HYPOTHESIS:

Our Smart Discharges approach will improve post-discharge health seeking and reduce post-discharge mortality.

JUSTIFICATION:

With an improved understanding of risk, and an ability to determine risk at the bedside, a Smart Discharge program will ensure optimized resource allocation, focusing on children most in need of limited resources. Such programs in precision public health can not only save lives and resources, but are much more likely to be scalable in economically strained environments.

OBJECTIVES:

The objective of this study is to determine whether a individualized, risk-based approach to improving pediatric discharges will reduce 6-month post-discharge mortality among children admitted with suspected sepsis

The study has two objectives, each corresponding to a phase:

1. Phase I: An observational period with enrolled patients acting as controls. Furthermore, this cohort will be used to update, refine, and validate previously developed models for post-discharge mortality to be used during Phase II.
2. Phase II: An interventional period to evaluate the effectiveness of a Smart Discharge program on mortality and post-discharge health seeking.

RESEARCH DESIGN:

This will be a two-phase study: Phase I is a multi-site prospective observational cohort study, while Phase II is a multi-site prospective interventional cohort study. This prospective study will be conducted from March 2017 to January 2024. The study will enroll 5700 children under five years of age (2700 <6 months of age, 3000 6-60m of age) in Phase I (non-interventional) and an equal number (5700) of children during phase II (interventional phase), for a total of 11,600 children.

STATISTICAL ANALYSIS:

Our prior work has shown that the 6-month post-discharge mortality rate is 5%. Our preliminary work has also suggested that our expected mortality benefit will be between 25% and 30% relative risk reduction. Assuming a relative risk reduction of 22.5%, we would need to enroll 5250 children per arm. We thus will conservatively aim to enroll 5700 per arm to account for losses to follow-up.

All analyses will be conducted using R 4.2.2 (Vienna, Austria; http://www.R-project.org). External model validation will be conducted using the Phase I cohort on the previously developed Smart Discharge Model. The Smart Discharge Model will then be updated to include data from the Phase I cohort. Final prediction models will be developed separately for children <6m of age and for children 6m - 5 years of age, with an area under the ROC curve analysis used to assess the overall performance of the final models. For the final model in older children (6m - 5y), the risk cut-off will be chosen based on the sensitivity and specificity, ensuring a sensitivity of >80% (initial derived model sensitivity was 82%). The final sensitivity and specificity will be reported, along with positive and negative predictive values (based on the overall mortality rate, and the mortality rates of each site). For the model in younger children, the same approach will be used, but ensuring the sensitivity is at least 85%, due to an expected higher mortality rate among younger children. The separation of ages has been determined to be the optimal approach for these models.

To evaluate the effectiveness of the intervention, a cox-proportional hazards regression on the time to post-discharge mortality will be used, including the year of discharge as a covariate to account for potential trends in mortality unrelated to the intervention. Additional potentially confounding variables will be identified based on a combination of pre-existing and expert knowledge, and univariate analysis of potential confounders on outcomes. A final multivariable model will be used to determine the adjusted effect of the Smart Discharge intervention. Interrupted time series, which use segmented regression modeling to determine the effect of the intervention after controlling for pre- and post-intervention time trends, will also be used to account for potential pre-intervention trends in mortality.

Inclusion Criteria:

• Children under five years of age
• Admission with a proven or suspected infection
• Provide written informed consent

Exclusion Criteria:

• Refusal to participate
• Previous enrolment in the study
• Outside of hospital catchment
• Language barrier
• Direct admission following birth without having been discharged

At each stage of the analysis and data preparation all of the study data will be prepared for public distribution. We will make every effort to prevent re-identification of subjects by coding data that has the potential of being identifiable. For example we will convert all dates into meaningful decimal numbers (date of birth into days since birth and date of recruitment will be reduced to month of recruitment) and all locations will coded into data that is useful but not specific (such as address converted to distance and direction from facility). We will ensure that data elements with small numbers of subjects (less than 10) will be coded or lumped to avoid identification. The study data will be made publically available using a reputable data hosting service (e.g. INDEPTH Data Repository, Dataverse etc.).

During the data analysis stage, data lacking patient identifiers will be accessed from REDCap by team members involved in the statistical analysis.

Data will be deposited to an open access repository with moderated access within 2 years of study completion

Moderated access on a case-by-case basis.