Study design

PPDM is a prospective, observational study being conducted at Muhimbili National Hospital in Dar es Salaam, Tanzania and John F. Kennedy Medical Center in Monrovia, Liberia. The study protocol was developed by collaborators in Tanzania (Muhimbili University of Health and Allied Sciences and Muhimbili National Hospital), Liberia (John F. Kennedy Medical Center) and the USA (Boston Children’s Hospital, Harvard T.H. Chan School of Public Health and Emory University School of Medicine). The study diagram is included in figure 1. Enrolment began on 1 October 2019. Patient follow-up data collection will conclude on 31 January 2022. This protocol follows the Strengthening the Reporting of Observational Studies in Epidemiology guidelines for cohort studies (online supplemental file 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

The Predicting Post-Discharge Mortality study design. (NICU: neonatal intensive care unit).

The combined study population between Liberia and Tanzania will be approximately 4000 participants (2000 neonates, 2000 children aged 1–59 months) who are discharged from the two sites. Each site will enrol 1000 neonates and 1000 children aged 1–59 months. Enrolled neonates and children will be followed through phone calls after discharge at 7, 14, 30, 45 and 60 days to assess postdischarge outcomes and health-seeking behaviours.

Patient and public involvement

Caregivers, patients and the public were not involved in the development of the study protocol.

Primary and secondary objectives

The primary objective of the study is to derive and internally validate two, age population-specific, clinical prediction rules to identify neonates and children aged 1–59 months at risk of postdischarge mortality within 60 days of discharge from the NICU or paediatric ward at two sites in sub-Saharan Africa. Our secondary objectives are to (1) externally validate existing clinical prediction rules for postdischarge mortality developed from studies in Uganda and Mozambique,9 10 (2) determine clinicians’ ability to accurately identify neonates and children at risk of postdischarge mortality and hospital readmission at the time of hospital discharge, (3) describe factors associated with hospital readmission and unplanned clinic visits after discharge and (4) describe health-seeking behaviours in the postdischarge period among children who die as well as those who survive beyond 60 days.

Study setting

Muhimbili National Hospital and John F. Kennedy Medical Center are large, referral, training hospitals for Tanzania and Liberia, respectively. Both hospitals are government hospitals supported by each country’s Ministry of Health. Both hospitals are located in urban areas with surrounding populations of 4 million for Muhimbili National Hospital and 1.5 million for John F. Kennedy Medical Center. Both hospitals have approximately 140–150 paediatric discharges and approximately 170 NICU discharges per month.

Eligibility

We will include participants aged 0–59 months with all diagnoses who are discharged from the NICU or the paediatric ward. Participants will be included if their caregivers consent to have their hospital admission data gathered and to receive follow-up phone calls. We will exclude children who (1) die during their initial hospitalisation, (2) are older than 59 months of age or (3) have non-consenting caregivers. We will not enrol participants who are transferred to other hospitals; however, both hospitals are large referral hospitals and, therefore, transfer out is rare.

Sample size

Previous studies at Muhimbili National Hospital evaluating patients with malnutrition and abdominal pain suggest postdischarge mortality rates as high as 13%–15%.12 13 Since there are no previous studies describing postdischarge mortality at John F. Kennedy Medical Center and postdischarge mortality rates among all discharged patients at Muhimbili National Hospital may be lower than among specific populations, we conservatively estimate a 5% postdischarge mortality rate. We also anticipate a lost-to-follow-up rate of 3% in each cohort. Therefore, we will enrol 1000 patients in each cohort (NICU and paediatric wards) at each site (n=4000 total) to derive and validate two, age population-specific, clinical prediction rules to predict postdischarge mortality (2000 in each population). We expect at least 100 deaths and 150 hospital readmissions in each cohort. Using the rule of 10 events per item in a clinical prediction rule and assuming a 5% mortality rate, we will be able to include at least five variables in each clinical prediction rule for postdischarge mortality.14

Standard of care

All enrolled children will receive standards of care per Tanzanian Ministry of Health guidelines at Muhimbili National Hospital and standards of care set forth by the John F. Kennedy Medical Center and the Liberian Ministry of Health. This care will be provided by the clinical care teams and not by this study’s staff.

Recruitment

Caregivers of participants will be approached by research assistants at the time of hospital discharge and asked to consent to the study. Consenting providers will provide informed consent in their own language (written in Kiswahili or witnessed thumbprint if not literate in Tanzania and verbal consent in English in Liberia). We will collect data at the time of discharge using a secure case report form on password-protected tablets or computers. Research assistants at each site will undergo rigorous training regarding study enrolment, data collection, retention and follow-up.

Screening and enrollment procedures

Research assistants will identify consecutive patients eligible for study enrolment in the days preceding discharge. During that time, research assistants will confirm inclusion and exclusion criteria for each potential participant who is admitted to the NICU or the paediatric ward. For consenting participants, key clinical findings and laboratory data will be documented in case report forms for factors present at admission and discharge (online supplemental files 2, 3). If no contraindication to study participation is found, a research assistant will approach the primary caregiver of the child and describe the study procedures. If the child is eligible and the caregiver consents to participation, informed consent will be obtained. The parent or guardian must provide informed consent prior to enrolment.

Research assistants will fill out case report forms on password-protected, encrypted software (Microsoft SQL in Tanzania and Kobo Toolbox in Liberia) with variables collected as part of the study. We will collect patient anthropometrics as measured during hospitalisation using standardised scales for weight, height/length as measured using available measuring boards and middle-upper arm circumference if documented by clinicians caring for children during their hospitalisation. We will calculate weight for height z-scores for age. We will collect data on documented nutritional status for the child during the hospitalisation (eg, severe acute malnutrition, kwashiorkor, etc).

Primary and secondary diagnoses at the time of admission and discharge will be collected by research assistants as documented. We will also collect information on any reported comorbidities (eg, HIV, tuberculosis, malignancies, etc) documented by the clinical care team. We will document the duration of hospital admission as documented in patients’ charts. Research assistants will ask caregivers about recent hospitalisations and document responses. Research assistants will record results from labs and imaging done during the hospitalisation, receipt of therapies including oxygen (ie, nasal cannula, face mask, etc), intravenous fluids and mode of receipt (ie, bolus, maintenance fluids, etc), medications (both type and duration) received during hospitalisation and procedures or surgeries performed.

As this is an observational study, we anticipate there will be some missing variables in clinical documentation. Research assistants will attempt to fill in any missing variables prior to discharge. Variables for individual participants still missing after the research assistants’ attempt to complete data will be accounted for with multiple imputations or excluded from analyses.

Research assistants will collect detailed contact information from the caregivers using a standardised patient locator form, including at least two working cell phone numbers for the caregiver, and a physical description of the whereabouts of the child’s home. If a caregiver does not have a mobile phone, we will obtain the mobile phone information of a household member or friend. We expect fewer than 5% of participants to be without cell phone access.

Follow-up visits

Research assistants will make phone calls to each caregiver of children at 7, 14, 30, 45 and 60 days after hospital discharge to ascertain the child’s vital status and to ask about health-seeking behaviours (online supplemental file 4). The equivalent of US$1 will be transferred via local phone companies to each participant for each follow-up phone call. Text messages will be sent to caregivers who do not answer the phone.15 If caregivers are still unreachable, research assistants will conduct home visits. Participants whose caregivers are unreachable despite three phone calls, two text messages and a home visit will be classified as lost-to-follow-up.

At each telephone or in-person contact, a research assistant will ask the caregiver about the following and record the responses:

1. The child’s status (ie, well, ill, hospitalised, dead).

2. Any intercurrent illnesses.

3. Any clinic or emergency department visits since hospitalisation.

4. The caregivers’ seeking of other forms of treatment (ie, herbs, pharmacy, etc).

5. Reason for returning to clinic (eg, fever, vomiting, etc).

Caregivers will be asked by study staff to take their child to a clinic or hospital if they feel their children are severely ill between phone calls. For patients who are found to have died at home or in hospitals other than Muhimbili National Hospital and John F. Kennedy Medical Center prior to follow-up phone calls, research assistants will conduct home visits to use the WHO’s 2016 Verbal Autopsy Tool16 (online supplemental file 5) used to describe the death of a child outside of the formal healthcare system.17 18 Home visits will be arranged at a time convenient for caregivers and with adequate time after the child’s death to allow the family to mourn as is culturally appropriate. Research assistants will bring a contextually appropriate condolence gift (eg, a bag of rice) to caregivers of the deceased child. All research staff who will conduct home visits will undergo rigorous training in use of the WHO’s 2016 Verbal Autopsy Tool.

Quality assurance

A study coordinator will be employed to assure that training of the field staff is high quality and rigorous. A data manager will generate weekly data summaries showing recruitment progress, follow-up and event rates to ensure high-quality and complete data reporting. These quality assurance approaches will be reinforced by each site’s principal investigator and by collaborators from the USA. Data cleaning will be performed using consistency and range checks. Data quality checks will also be applied weekly, and feedback will be provided to the study coordinators at both sites.

Outcome measures

Our primary outcome of postdischarge mortality and secondary outcomes including hospital readmissions, unplanned clinic visits and other care-seeking behaviours will be measured by caregiver report during follow-up calls and in-person home visits.

Statistical analysis

All candidate variables will be selected a priori based on prior studies,7 19–22 modified Delphi studies of factors thought to be associated with postdischarge mortality in sub-Saharan Africa,23 24 and experience among the research team. For the derivation of the clinical prediction rules, we will use elastic net regression,25 including all candidate variables to assess the strength of the association of each candidate variable on postdischarge mortality. Associations with 95% CIs for adjusted ORs that do not cross one will be considered significant. To determine the weighted points assigned to each candidate variable, we will calculate the adjusted log coefficient of each candidate variable from the multivariable model, round it to the nearest 0.5, and then double the rounded log coefficients to form an integer.26–29 We will assess the performance of the clinical prediction rules through internal validation using bootstrapping methodology with repetitions and calculate the area under the receiver operating characteristic (ROC) curve.30–33

We will also externally validate the clinical prediction rules derived in Uganda9 and in Mozambique10 by calculating the sensitivity, specificity and positive and negative likelihood ratios of each score at ≥ each specified cut point as defined in these rules. We will also create ROC curves for each existing score. We will assess the association of clinicians’ estimated probability of postdischarge outcomes and actual outcomes using ROC curves. We will use multivariable logistic regression to identify factors independently associated with 60-day hospital readmission and unplanned clinic visits among neonates and children. As previous studies suggest that healthcare seeking is infrequent among children who experience postdischarge mortality,34 35 we will review data from follow-up calls and verbal autopsies to calculate descriptive statistics of the proportion of caregivers who seek care in the postdischarge period and compare rates of care-seeking between children who die in the postdischarge period to those who survived.