Predictors of postdischarge mortality among hospitalised children aged 5−14 years at tertiary hospitals in Tanzania: a prospective observational cohort study

Abstract

Introduction

Postdischarge mortality is an important but under-recognised outcome in paediatric care, particularly in resource-limited settings where reported rates vary widely.

Objectives

To determine the proportion of postdischarge mortality and its predictors among children aged 5−14 years admitted to paediatric wards at Muhimbili National Hospital (MNH), Jakaya Kikwete Cardiac Institute (JKCI) and Muhimbili Orthopedic Institute (MOI).

Methods and analysis

A prospective cohort study was conducted over 10 months among children aged 5–14 years admitted to paediatric wards at MNH, JKCI and MOI. Data were collected using a structured questionnaire. Participants were followed up by telephone at 1, 2 and 3 months after discharge. IBM SPSS Statistics V.25 was used for data analysis. Cox proportional hazards regression was used to identify factors associated with postdischarge mortality at a 5% significance level, and survival was estimated using Kaplan-Meier methods. Verbal autopsy was conducted to determine probable causes of death.

Results

A total of 786 participants were enrolled in the study. The median age of the participants was 7 years (IQR 5–9 years). 467 (59.4%) were male. During the 3-month follow-up, 22 of 744 (2.9%) participants with known postdischarge outcomes died.

Factors associated with postdischarge mortality included HIV infection (HR 9.83, 95% CI 2.3 to 41.5), chronic kidney disease (HR 18.26, 95% CI 1.3 to 39.26), duration of symptoms >2 weeks (HR 4.7, 95% CI 1.5 to 14.73), leukocytosis (HR 7.94, 95% CI 1.97 to 32.03) and discharge against medical advice (HR 55.4, 95% CI 4.46 to 393.11).

More than half of the deaths occurred within the first month after discharge, and 13 (59.1%) occurred at home. The most common probable causes were pneumonia and malignancies, each accounting for 18.2% of deaths.

Conclusion

Postdischarge mortality was relatively low but higher in patients with chronic conditions and those discharged against medical advice. Most deaths occurred within the first month.

Clinical trial number

Not applicable.

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• This study summarises the burden, associated factors and probable causes of PDM among children aged 5−14 years from three tertiary hospitals in Tanzania.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE AND POLICY

• This study provides data to inform postdischarge follow-up by helping identify children at risk and the appropriate timing for follow-up.

Background

Postdischarge mortality (PDM) is an emerging concern in paediatrics, particularly among children aged <5 years, following global reductions in under-5 in-hospital mortality.^1–4

In contrast, older children have received comparatively less attention, particularly regarding postdischarge outcomes.^5–7 Their lower mortality rates compared with children aged <5 years may have contributed to the assumption that postdischarge outcomes are less important.⁴

Global estimates of the burden, causes and predictors of paediatric PDM (PPDM) remain limited because of differences in definitions of the postdischarge period and study populations.⁸

Available data are mostly from low and middle-income countries, where paediatric mortality is high.^8–11 Reported PPDM ranges from 1% to 18%, mainly among studies of children aged <5 years, with a few studies including children up to 18 years of age. This variation may reflect differences in study populations and follow-up duration.^{9 10}

In a recent systematic review of PPDM in developing countries, only 4 of 24 studies included children aged >5 years.¹⁰ One of these, conducted in Kenya with a cohort of 14 971 children followed for 1 year, reported that 88/535 (16%) of PDM occurred among children aged 5–14 years.¹² The other three studies in the review did not provide age-specific data.

Further evidence on PDM in children aged >5 years comes from three additional studies in developing countries. The first, conducted in Mozambique, followed a cohort of 18 000 children aged <15 years for 3 months, reporting a PDM of 2.2% among children aged >5 years.¹³ The second study, in Kenya, included 2997 children aged 5–12 years followed for 1 year, with a PDM of 3%.² The difference in PDM in these studies is likely attributable to the variations in follow-up duration.

The third study, conducted in Tanzania by Hau et al, followed a cohort of children aged 2–12 years for 1-year postdischarge. Among 161 children aged 5–12 years, 16% died within the follow-up period.¹⁴ The higher PDM may reflect the longer 12-month follow-up period, which could have captured deaths unrelated to the index admission. Data on the burden of PDM in children aged 5–14 years remain limited in Tanzania, as available estimates are often combined with those for children aged <5 years.^{14 15}

Evidence on predictors of PDM among children aged >5 years is also limited, as most studies report risk factors across broad age groups without age-specific estimates.^{8 10} Commonly reported risk factors, mainly from studies in children aged <5 years and a few that enrolled children aged >5 years, include unplanned hospital discharge, disease severity at admission, comorbidities (including HIV infection) and malnutrition.^{9 16 17} Severe illness may lead to multiorgan dysfunction, while malnutrition and HIV weaken immunity, increasing vulnerability to postdischarge complications in low-income settings. Unplanned discharge likely reflects incomplete treatment.

Hospitalisation has also been associated with an increased risk of PDM, with deaths occurring across different admission diagnoses.¹⁸ Despite treatment for the acute illness, some children may remain vulnerable after discharge because of underlying clinical and social factors.¹⁶ Those discharged to socioeconomically disadvantaged families may struggle to adhere to follow-up care or to reaccess health services because of financial constraints or distance, increasing their vulnerability after discharge.¹⁷

There are limited data on the specific causes of PDM, largely because most occur at home in low-income settings, making cause determination difficult. Verbal autopsy has been used in several studies to determine the causes of death, despite its limitations.^{15 19 20} For example, a study by Lyimo et al sought to identify causes of PDM among children aged 1 month to 14 years but could not determine the exact causes, instead reporting broad categories and acknowledging the limitations of verbal autopsy in establishing precise causes.¹⁵

In this prospective cohort study, we aimed to determine the burden, causes and predictors of PDM among children aged 5–14 years admitted to tertiary hospitals in Tanzania.

Methodology/design

Study design, aim and duration

We conducted a 10-month prospective cohort study (May 2023−February 2024) of children aged 5−14 years discharged from the paediatric wards of Muhimbili National Hospital (MNH), Jakaya Kikwete Cardiac Institute (JKCI) and Muhimbili Orthopedic Institute (MOI) to determine the proportion and predictors of PDM. The protocol of this study has been published previously.²¹

Study setting

The study was conducted at the three national health institutes, MNH, JKCI and MOI, located in Dar-Es-Salaam. Together, these facilities provide specialised medical, cardiac and surgical care and receive paediatric admissions from across the country.

Further details about the study settings are provided in the previously published protocol.²¹

Study population and sample size

We consecutively enrolled children aged 5–14 years admitted to the paediatric wards of the three hospitals, regardless of diagnosis, after obtaining caregiver consent and assent from children aged ≥7 years. Children admitted directly to the ICU were excluded, as post-ICU mortality has been reported in a separate study.¹⁵ Participants were followed up for 3 months after discharge by telephone. Detailed recruitment procedures and sample size determination (n=786) have been described in the published study protocol.²¹

Variables

Candidate variables for predicting PDM included demographics, socioeconomic factors and clinical characteristics.

The primary outcome was PDM at 3 months. Secondary outcomes included readmission and unplanned outpatient hospital visits.

Data collection

Three research assistants and the principal investigator collected data using a structured questionnaire. Participants were followed up by telephone at 1, 2 and 3 months after discharge.

A modified WHO verbal autopsy tool was used to determine the probable causes of death.²² Probable causes were assigned by consensus between the principal investigator and an independent physician.

Definition of terms

PDM was defined as any death after discharge from the index admission and within 3 months of follow-up, and discharge against medical advice was defined as discharge requested by the parent/guardian against the clinician’s advice. Abscondment was defined as leaving the hospital without written documentation of discharge.

Patient involvement statement

Patients or the public were not involved.

Data management

Paper-based data collection tools were kept in a locked cupboard accessible only to personnel involved in the research. Research assistants, together with the principal investigator, transferred the information from the paper forms into SPSS software V.25. The datasets were subsequently merged for cleaning and analysis. Data were kept confidential, and no personally identifying information was reported.

Data analysis

Data were analysed using SPSS V.25. Categorical variables were described as proportions (percentages), and continuous variables as means with SD. Univariable and multivariable Cox regression analyses were used to identify factors associated with PDM. All factors that were significantly associated with PDM in the univariable analysis were included in the multivariable analysis. Participants with missing outcome data were excluded from the analysis.

PDM was analysed as the proportion of deaths during the 3-month follow-up. Time to death was calculated from the date of discharge to the date of death among participants with confirmed outcomes.

The survival rate was displayed using the Kaplan-Meier curve. HRs and their 95% CIs were reported with a significance level of 5%.

Ethical approval

Ethical approval (reference number 04–2023-1662) was granted by the MUHAS Institutional Review Board. Permission was obtained from the management of the three participating hospitals.

Informed consent was obtained from parents or guardians before recruitment, and children aged ≥7 years were asked to provide assent.

Results

Study enrolment

A total of 786 participants were enrolled, out of which 647 (82.3%) were from MNH, 48 (6.1%) and 91 (11.6%) were from JKCI and MOI, respectively.

Of the 786 enrolled participants, 32 (4.1%) died during hospitalisation and were, therefore, not eligible for postdischarge follow-up. Of the remaining 754 participants, 6 (0.8%) absconded and 4 (0.5%) were lost to follow-up. Thus, 744 participants had known postdischarge outcomes and are included in the analysis (figure 1).

Figure 1. Flowchart illustrating participants’ enrolment. JKCI, Jakaya Kikwete Cardiac Institute; MNH, Muhimbili National Hospital; MOI, Muhimbili Orthopedic Institute.

General characteristics of the study participants

Among the study participants, 59.4% were males. The median age of the study participants was 7 years (IQR: 5–9 years). More than half (55.2%) of the participants had health insurance. Nearly half of the caregivers had a primary level of education (48.7%), and 13.2% were formally employed (table 1).

Table 1. Sociodemographic characteristics of the study participants.

Variable		Frequency (n)	Per cent
Age groups	5–9 years	592	75.3
	10–14 years	194	24.6
Sex	Male	467	59.4
	Female	319	40.6
Health insurance	Yes	434	55.2
	No	352	44.8
Caregiver’s education status	No formal education	51	6.5
	Primary education	383	48.7
	Secondary education	261	33.2
	College/university	91	11.6
Caregiver’s employment status	Formal employment	104	13.2
	Other occupations	682	86.8

Clinical characteristics of the study participants

Fever was the most commonly reported symptom at admission, followed by pain, respiratory symptoms, body swelling and symptoms suggestive of anaemia.

More than half of the participants had symptoms for >2 weeks before admission, while 48.9% were hospitalised for >1 week. The median length of hospital stay was 7 days (IQR: 4–17 days).

On admission, the most common signs observed were tachycardia, followed by tachypnoea, pallor and higher temperature >37.5. Anaemia was prevalent in over half of the participants, with most of the children experiencing moderate anaemia, while a smaller proportion had mild or severe anaemia.

Leukocytosis was observed in over a third of the participants, with neutrophils being the predominant white cell type. Electrolyte imbalances were also present, with hyponatraemia, hypokalaemia and hyperkalaemia being the most common, affecting different proportions of the participants. Sickle cell anaemia was the most common diagnosis (table 2). The individual diagnoses included under the category ‘others’ are provided in online supplemental table 1.

Table 2. Clinical characteristics of the study participants^*.

Variable		Frequency (n)	Per cent
Admission symptoms	Fever	210	26.8
	Respiratory symptoms	177	22.5
	Pain	180	23
	Body swelling	142	18
	Symptoms of anaemia	139	17.7
	Neurological symptoms	84	10.7
	Other complaints	463	59
Admission signs	Tachycardia Tachypnoea Temperature>37.5°C Pallor Others	499 232 116 168 97	63.5 29.5 14.8 21.4 12.3
Laboratory values	Anaemia Leukocytosis Hypokalaemia Hyponatraemia Abnormal renal functions	416 276 83 199 131	52.9 35.1 14.6 29.1 1.7
Diagnoses	Sickle cell anaemia Acute leukaemia Lymphoma HIV infection Chronic kidney disease (CKD) Others	116 64 54 13 9 372	14.8 8.1 6.9 1.7 1.2 47.3
Duration of symptoms	<2 weeks	346	44
	>2 weeks	440	56
Duration of hospital stay	≤1 week	402	51.1
	>1 week	384	48.9
Discharge status	Absconded	6	0.8
	Against advice	4	0.5
	Normal discharge	744	94.7
	Deceased	32	4

^*Admission symptoms included multiple responses. ‘Others’ includes diagnoses that were individually uncommon; a detailed breakdown is provided in online supplemental table 1.

Outcomes of the study participants and the probable causes of PDM

Overall mortality was 6.9% (54/786), comprising 32 in-hospital deaths and 22 postdischarge deaths. Twenty-two deaths occurred after hospital discharge, of which 13 (59.1%) occurred at home. The proportion of PDM, excluding 10 losses to follow-up and 32 in-hospital deaths, was 2.9% (22/744). The median time to death after discharge was approximately 20 days. Pneumonia and malignancies were the most common probable causes of death as determined by the verbal autopsy, each contributing 18.2% of the PDM (table 3).

Table 3. Outcomes of the study participants and probable causes of postdischarge mortality.

Variable	Frequency (n)	Per cent
Postdischarge deaths	22	2.9
Place of death
Died at home	13	59.1
Readmission
Re-admitted after discharge	58	7.8
Frequency of readmission
Once	42	72.4
More than once	16	27.6
Reasons for readmission
Fever	16	28
Pain	12	21
Difficulty in breathing	7	12.1
Unplanned outpatient visits
Unplanned outpatient visit	109	14.7
Reasons for unplanned outpatient visits
Fever	32	29.4
Convulsions	6	5.5
Body malaise	15	13.8
Probable causes of death
Pneumonia	4	18.2
Malignancies	4	18.2
Severe anaemia	1	4.5
Shock Undetermined	3 8	13.6 36.4

Factors associated with PDM

The factors associated with PDM in this study were noted to be duration of symptoms of more than 2 weeks (HR 4.7 (95% CI 1.5 to 14.73), p=0.008), discharge against medical advice (HR 55.4 (95% CI 4.4 to 393.11), p=0.002), leukocytosis (HR 7.9 (95% CI 1.97 to 32.03), p=0.004), HIV infection (HR 9.8 (95% CI 2.3 to 41.54), p=0.002) and chronic kidney disease (CKD) (HR 18.26 (1.3–39.27), p=0.027) (table 4).

Table 4. Univariate and multivariate analysis of factors associated with postdischarge mortality.

Variable		Univariate Cox regression			Multivariable Cox regression
		Crude HR	95% CI	P value	Adjusted HR	95% CI	P value
Age	5–9 years	1.28	0.49 to 3.03	0.6
	10–14 years	Ref
Sex	Male	1.33	0.56 to 3.13	0.5
	Female	Ref
Health insurance	Yes	2.57	1.04 to 6.36	0.003	2.6	0.9 to 7.3	0.07
	No	Ref
Duration of symptoms	>2 weeks	3.25	1.094 to 9.66	0.034	4.7	1.5 to 14.7	0.008
	<2 weeks	Ref
Discharge type	Against advice	28.47	3.79 to 113.11	0.001	55.4	4.5 to 393.1	0.002
	Normal discharge	Ref
Respiratory symptoms	Yes	2.97	1.23 to 7.17	0.015	2.5	0.8 to 8.8	0.13
	No	Ref
Oedema	Yes	2.92	1.21 to 7.05	0.017	1.5	0.2 to 9.4	0.68
	No	Ref
White cell count	Normal	Ref
	High	11.48	2.87 to 45.89	0.001	7.947	1.97 to 32.0	0.004
	Low	5.10	1.60 to 16.26	0.006
Neutrophil count	Normal	Ref
	High	4.90	1.38 to 17.37	0.014	1.2	0.02 to 2.7	0.23
	Low	2.80	0.94 to 8.3	0.06
Anaemia	Moderate	7.92	2.21 to 28.40	0.001	8.26	0.02 to 14.61	0.91
	Severe	10.47	2.71 to 40.50	0.001	7.64	0.19 to 11.21	0.7
	No anaemia	Ref
Renal functions	Abnormal	1.98	1.04 to 3.75	0.037	1.02	0.7 to 10.23	0.19
	Normal	Ref
HIV infection	Yes	15.23	4.48 to 51.76	0.000	9.83	2.3 to 41.54	0.002
	No	Ref
CKD	No	Ref
	Yes	13.15	3.1 to 46.51	0.001	18.26	1.3 to 39.27	0.027
Readmission	Yes	3.30	1.1 to 9.93	0.034	2.045	0.07 to 5.15	0.66
	No	Ref

CKD, chronic kidney disease.

The Kaplan-Meier curve in figure 2 represents the survival probability of all discharged children over 3 months of follow-up. Most deaths occurred within the first month after discharge, with a cumulative survival of 97% at 90 days postdischarge.

Figure 2. Kaplan-Meier curve showing the survival of the study participants 3 months postdischarge.

Discussion

We conducted a prospective cohort study of postdischarge outcomes among children aged 5–14 years in Tanzania. Of 786 enrolled participants, 744 had known postdischarge outcomes after 3 months of follow-up. Forty-two were excluded, including 32 in-hospital deaths and 10 lost to follow-up. Fever was the most common symptom at admission, tachycardia the most frequent sign and sickle cell anaemia the most prevalent diagnosis. The high frequency of fever aligns with studies in Kenya and Tanzania and reflects the importance of infections as a leading cause of hospitalisation in this age group in Africa.^{2 14} Similarly, the high frequency of tachycardia may be related to the burden of anaemia and the proportion of children with sickle cell anaemia in this cohort, consistent with the recognised burden of sickle cell disease (SCD) in Tanzania.²³

PDM was 2.9%, slightly higher than the 2.2% reported by Madrid et al in Mozambique among children aged >5 years followed for 3 months and similar to the 3% reported in a Kenyan study involving a comparable age group and follow-up period.^{2 13} The 3-month follow-up period was selected to capture early postdischarge vulnerability following hospitalisation, while limiting the likelihood of including deaths unrelated to the index admission. The lower PDM reported by Madrid et al, despite a similar follow-up period, may reflect differences in study design and mortality ascertainment methods, including potential underascertainment of deaths in retrospective demographic surveillance. Although overall PDM was relatively low in this cohort, higher mortality was found in certain subgroups, suggesting a higher risk among vulnerable children.

Clinical characteristics at admission, including chronic illnesses (HIV and CKD), prolonged symptom duration (>2 weeks) and leukocytosis, as well as discharge-related factors such as discharge against medical advice, were associated with higher PDM. The link between chronic conditions and PDM was also emphasised in the Tanzanian study by Hau et al, where the authors suggested that these patients may have experienced multiple sequelae, increasing their vulnerability to further complications.¹⁴ Leukocytosis, on the other hand, may indicate underlying disease severity or an acute inflammatory response rather than represent an independent mechanism contributing to mortality.

Additionally, children with HIV are prone to opportunistic infections due to immunosuppression, while those with CKD may be particularly vulnerable because of progressive disease and limited access to renal replacement therapy. In older children, CKD is often associated with infections and SCD, both of which are prevalent in Africa and potentially treatable. HIV infection, often acquired vertically, can be prevented, identified early and treated. These observations highlight the potential importance of primary prevention and early detection. Discharge against medical advice may reflect interruption of treatment during a clinically vulnerable period and has also been identified as a risk factor for PDM in studies conducted in Mozambique and Guinea, where deaths often occurred early.^{13 16} In this study, most deaths among children discharged against medical advice occurred within 2 weeks of discharge. However, given the small number of postdischarge deaths (n=22), these findings should be interpreted cautiously. The limited number of outcome events constrained the complexity of the multivariable model, and large HRs with wide CIs suggest limited precision and possible model instability. Consequently, the multivariable analysis should be regarded as exploratory, and the identified variables should be interpreted as potential associations rather than definitive independent predictors. Validation in larger, independent cohorts with a greater number of outcome events is needed before these associations can be considered robust predictors of PDM.

More than half of the deaths occurred within 1-month postdischarge, with a median time to death of approximately 20 days, similar to findings reported in several other studies.^{2 13 16} This highlights the vulnerability of children during the early transition from hospital to home. These findings emphasise the need for early postdischarge follow-up, particularly within the first 2 weeks for at-risk patients. Patient vulnerability during the early postdischarge period is often attributed to stressors experienced during hospitalisation, described in the literature as ‘post-hospital syndrome’.²⁴ Targeted interventions during this period may reduce PDM, as suggested by a Ugandan study focusing on children aged <5 years, which demonstrated the effectiveness of postdischarge measures such as detailed follow-up referrals and postdischarge kits with educational materials given in the local language.²⁵ Most of the deaths occurred at home, aligning with findings from other studies.^{6 7} Delays in seeking care and socioeconomic challenges prevalent in these settings likely contribute to this trend, reflecting broader postdischarge vulnerabilities. However, data on these factors were not collected in sufficient detail for quantitative analysis in this study; they represent important contextual determinants that may influence postdischarge outcomes in similar settings. The probable causes of PDM were respiratory diseases and malignancies. Although several deaths lacked an established cause due to limitations of verbal autopsy methods, similar challenges have been reported in previous studies. While verbal autopsy provided important contextual information, recall bias and misclassification remain possible, and these findings should, therefore, be interpreted as probable rather than definitive causes of death.^{15 19} Future research should explore alternative methods to ascertain causes of death more accurately.

The readmission rate within 3-month postdischarge was 7.8%, consistent with the literature’s wide range across various disease conditions.^26–28 Fever was a common symptom leading to readmission, highlighting the importance of comprehensive treatment during the initial admission.

Unscheduled outpatient visits were reported by 14.7% of participants, with fever being the most common reason. These visits may reflect persistent or recurrent illness, although this could not be determined from the available data.

The strengths of the study included a relatively large sample size providing good population representation, monthly follow-up, which may have minimised recall bias, and a low loss to follow-up supported by an active tracking strategy. Although loss to follow-up was low, the impact of missing postdischarge outcomes on estimates cannot be completely excluded. Limitations included challenges in establishing causes of death, partly due to incomplete interviews due to emotional distress during verbal autopsies and unclear symptom descriptions, which complicated accurate interpretation of clinical signs. Additionally, children requiring intensive care were excluded from this study; therefore, the findings cannot be generalised to critically ill paediatric populations, including those requiring intensive care support.

Conclusion and recommendations

The findings highlight the potential value of improved postdischarge follow-up for children aged 5–14 years, particularly those with chronic conditions or discharged against medical advice. Most postdischarge deaths occurred within the first month, with more than half occurring at home. These findings support consideration of early follow-up, particularly during the first 2 weeks after discharge, for children at increased risk.

Data availability statement

Data are available upon reasonable request.