Mortality during and following hospital admission among school-aged children: a cohort study [version 1; peer review: awaiting peer review]

Background: Far less is known about the reasons for hospitalization or mortality during and after hospitalization among school-aged children than under-fives in lowand middle-income countries. This study aimed to describe common types of illness causing hospitalisation; inpatient mortality and post-discharge mortality among school-age children at Kilifi County Hospital (KCH), Kenya. Methods: A retrospective cohort study of children 5−12 years old admitted at KCH, 2007 to 2016, and resident of the Kilifi Health Demographic Surveillance System (KHDSS). Children discharged alive were followed up for one year by quarterly census. Main outcomes were inpatient and one-year post-discharge mortality. Results: We included 3,907 admissions among 3,196 children with a median age of 7 years 8 months (IQR 74−116 months). Severe anaemia (792, 20%), malaria (749, 19%), sickle cell disease (408, 10%), trauma (408, 10%), and severe pneumonia (340, 8.7%) were the commonest reasons for admission. Comorbidities included 623 (16%) with severe wasting, 386 (10%) with severe stunting, 90 (2.3%) with oedematous malnutrition and 194 (5.0%) with HIV infection. 132 (3.4%) children died during hospitalisation. Inpatient death was associated with signs of disease severity, age, bacteraemia, HIV infection and severe stunting. After discharge, 89/2,997 (3.0%) children died within one year during 2,853 child-years observed (31.2 deaths [95%CI, 25.3−38.4] per 1,000 child-years). 63/89 (71%) of post-discharge deaths occurred within three months and 45% of deaths occurred Open Peer Review Reviewer Status AWAITING PEER REVIEW Any reports and responses or comments on the article can be found at the end of the article. Page 1 of 11 Wellcome Open Research 2020, 5:234 Last updated: 08 OCT 2020


Introduction
Despite a remarkable decline in global child mortality, more than 6 million children died in 2018, of which 0.9 million (15%) deaths occurred among children aged 5 to 14 years, mostly in low-and middle-income countries (LMICs) 1,2 . Public health efforts to improve survival are generally directed towards children <5 years old [2][3][4] . Despite there being no specific new health interventions targeting children aged 5 to 14 years, their mortality risk from 1990 to 2016 declined by 51% globally 4 . However, since 2000, the annual rate of mortality reduction in this group (2.7%) has been lower than amongst children <5 years old (4.0%) 4 .
Children ≥5 years old may be admitted with different conditions than younger children and their risks for inpatient or post-discharge mortality may also differ. However, there are only limited data describing reasons for admission and mortality-patterns in LMICs. Among school-aged children admitted to six Kenyan hospitals in 2013, 3.5% of children aged 5 to 9 years and 5.0% of children 10 to 14 years died 5 . Infectious diseases such as malaria were the main reported causes of death in children ≥5 years 2,5,6 .
Post-discharge mortality is increasingly recognized as a significant contributor to the burden of mortality among under-fives in LMICs and adults in resource-rich settings [7][8][9] . The most recent systematic review (2018) of paediatric post-discharge mortality in LMICs did not identify any studies focusing specifically on school-aged children 7,8 . Of 24 studies reviewed, four included children ≥5 years old. A study in Bangladesh with a sample size of 74 children aged 24 to 72 months did not report the number of children who were ≥5 years old 10 . A large study in Kenya among children 0 to 15 years old (N=14,971) reported that 16% (88/535) of all post-hospital discharge deaths were among children aged 5 to 14 years 11 and two studies from Uganda with children aged 2 months to 12 years reported ~1% and 3.8% children dying after discharge, respectively, but were not disaggregated by age group 12,13 . Since the latest systematic review (2018) 7 , four more studies have evaluated paediatric post-discharge mortality in LMICs 3,14-16 . Hau et al. included children aged 2 to 12 years followed-up for 12 months after discharge from two hospitals in Tanzania and found that 16% (26/161) of children aged 5 to12 years died after hospital discharge 3 . Post-discharge mortality risk was reported to be higher than that of children <5 years (hazard ratio 2.44 (95%CI, 1.37 to 4.34) 3 . The commonest diagnoses at admission were malaria and sickle cell disease 3 . The second study from southern Mozambique included 18,023 children aged <15 years, of which 83/3,816 (2.2%) aged ≥5 years died within three months after discharge 14 . The third study was from Kenya, but excluded children ≥5 years old 15 . The fourth study from Tanzania included children aged 2 to 12 years, of whom 47/466 (10%) died one year after hospital discharge, but deaths were not disaggregated by age 16 .
We aimed to describe the reasons for hospitalisation, underlying illnesses, and the clinical characteristics and features associated with mortality during hospitalisation and for one year after discharge among children 5 to 12 years old admitted to a rural hospital in Kenya.

Study setting
The study was conducted at Kilifi County Hospital (KCH), located on the Indian Ocean coast in Kenya. KCH is a secondary care hospital handling approximately 5,000 annual paediatric admissions. Most of the population served by the hospital are rural farmers. Clinical care at the hospital follows Kenyan and World Health Organization (WHO) guidelines.
Systematic data including clinical signs, anthropometry and laboratory investigations at admission have been collected by research clinicians and entered in a database since 1998. At discharge or death, up to two final diagnoses are assigned by the discharging clinician. From 2002, approximately 250,000 residents in an area of 891 km 2 neighbouring KCH have been enumerated every four months by the Kilifi Health and Demographic Surveillance System (KHDSS) for births, deaths and inor out-migration 17 . Data on admissions to KCH are linked to the KHDSS population database.

Study population
Children aged 60 to 155 months admitted to KCH between 2007 and 2016 and resident within the KHDSS were included. The post-discharge analysis included all children discharged alive. Data from the KHDSS up to the 2018 August census round were used to confirm vital status post-discharge.

Study design
We performed a retrospective cohort study. Exposures evaluated were clinical and demographic features, anthropometry and laboratory variables at admission. Outcomes examined were inpatient and one-year post-discharge mortality.
Data sources/measurement Anthropometry, clinical history and examination, complete blood count, HIV antibody test, blood smear for malaria and blood culture were systematically conducted at admission as previously described 18 . Anthropometry was taken by trained clinical assistants: mid-upper arm circumference (MUAC) using a non-stretchable insertion tape (TALC, St Albans, UK), weight using an electronic scale (Seca 825, Birmingham, UK) and height using a stadiometer (Seca 215, Birmingham, UK). HIV antibody testing used two rapid tests (Determine; Inverness Medical, Fl, USA; and Unigold; Trinity Biotech, Bray, Ireland). Caregivers of children with a positive HIV antibody test were counselled and referred to an HIV comprehensive care clinic. Details of subsequent clinic attendance and antiretroviral treatment were not recorded on the database. Children found to have malnutrition, sickle cell disease, tuberculosis, cardiac or neurological conditions were referred to outpatient clinics for continued care.
For this analysis, malaria was defined as a blood smear positive for Plasmodium falciparum and anaemia was defined as moderate (haemoglobin 8 to 11.4 g/dl) or severe (haemoglobin <8g/dl), as per WHO guidelines 19 . Biochemical tests, radiology, sickle cell testing, lumbar puncture and other investigations were done at the discretion of the treating clinician. Meningitis was diagnosed using the cerebrospinal fluid (CSF) examination and culture as follows: positive culture for known pathogen, positive CSF microscopy (Gram stain and/or Indian ink stain), positive antigen test (S. pneumoniae, H. influenzae type B, N. meningitidis, and C. neoformans) and CSF white blood cell count (WBC) ≥10 cells/µl.

Statistical methods
Statistical analysis was performed using STATA (version 15.1; StataCorp, College Station, TX, USA). All eligible admissions and discharges from the population of KCH residents of KHDSS within the study period were included in the study, therefore, no formal sample size estimation was conducted.
MUAC-for-age z-score (MUACZ) were calculated using the method of Mramba et al. 20 . Body mass index-for-age z-score (BMIZ), weight-for-age (WAZ) and height-for-age z-score (HAZ) were calculated using WHO 2007 growth references and classified as normal (≥-2Z), moderate (-3 to -2Z) and severe (<-3Z) 21 . Anthropometric measurements were regarded as missing not at random. To ensure all children were included in the multivariable regression models, we used categorical anthropometry variables and included a 'missing' category in the analysis.
Because the children could be admitted more than once during the study period, for the analysis of inpatient deaths we performed a multiple-admission analysis where each child could contribute more than one admission record using their unique person IDs. To examine features at admission associated with inpatient mortality, we used a backward stepwise log-binomial regression model with robust standard errors to account for multiple admissions retaining variables with a P-value <0.1 and reported adjusted risk ratios for variables with P-value <0.05 in the final multivariable model.
To calculate the post-discharge mortality rate, time at risk was defined as the date of hospital discharge until death, out-migration or 365 days later. We performed a multiple-discharge analysis where children with multiple admissions with live discharge contributed separate person-time periods when there was no overlap during the one-year follow-up. After assessing and confirming the proportional hazard assumption was not violated using the Schoenfeld residuals test, we used a Cox proportional hazard regression model with robust standard errors to account for multiple discharges to examine admission features associated with post-discharge mortality.
For both inpatient and post-discharge regression analysis, individual clinical signs, laboratory tests and final diagnoses assigned by clinical staff were used for diagnoses not captured by syndromic definitions using clinical signs. In the regression models, we used MUACZ to define undernutrition rather than BMIZ because MUACZ predicts mortality as effectively as BMIZ 20 , is less affected by dehydration than weight-based measures 22 , and fewer children were missing MUAC measurements.
We initially excluded biochemical features that were not systematically collected and performed exploratory analyses of these features. We tested if the effects of HIV status, anaemia and malaria were modified by age and if the malaria effects on post-discharge mortality were modified by anaemia or nutritional status using likelihood-ratio tests. Goodness-of-fit of the multivariable regression models was assessed using the area under receiver operating characteristic curves (AUC) and internally validated using the bootstrapping method with 1000 resampling with replacement 23 .

Ethical statement
The Kenya Medical Research Institute (KEMRI) National Ethics Review Committee (SCC 2778) approved the study. Written consent for the children's participation in the original study was provided by their parents/guardians, which included consent for subsequent analyses.

Factors associated with inpatient mortality
In the multivariable model, older age, signs of disease severity (tachypnoea, history of breathing difficulty, weak pulse and impaired consciousness), HIV infection, bacteraemia and severe stunting were positively associated with inpatient mortality, whilst epilepsy/convulsions were negatively associated with inpatient mortality (Table 3). All variables tested in univariable analysis are shown in Extended data Table S3. Being severely wasted was associated with inpatient death in the univariable model, but the effect was attenuated in the multivariable model (Extended data Table S3). There was no evidence that the effect of HIV infection on inpatient mortality was modified by age (P=0.13), severe wasting (P=0.97), malaria (P=0.34) or moderate and severe anaemia (P=0.13). We found no evidence that anaemia (P=0.16), age (P=0.21) or severe wasting (P=0.14) modified the effect of malaria on inpatient mortality.
In exploratory analysis including biochemical variables, hyperkalaemia and elevated creatinine were associated with inpatient mortality (Extended data Table S4).
Post-discharge mortality Among the 3,064 children who were discharged alive, follow-up data were missing for 67 (2.2%) children (Figure 1). We therefore analysed data from 2,997 children who accrued 2,853 child-years of observation. Eighty-nine (3.0%) children died during follow-up; 63 (71%), 80 (90%) and 84 (94%) of postdischarge deaths occurred within three, six and nine months of discharge, respectively. The overall mortality rate was 31.2 (95%CI, 25.3 to 38.4) per 1,000 child-years. During the first three, six and nine months after discharge, mortality rate was 244 (95%CI, 190.3 to 311.8), 89.6 (95%CI, 72.0 to 111.6), and 53.2 (95%CI, 43.0 to 65.9) per 1000 child-years, respectively. Forty (45%) deaths occurred without readmission to KCH. Some signs of disease severity at admission (elevated respiratory rate and presence of weak pulse), HIV infection, severe anaemia and severe wasting were positively associated with post-discharge death (Table 3 and Figure 2). Malaria was negatively associated with post-discharge death (Table 3 and Figure 2). Hospital admission duration was not independently associated with post-discharge death (Extended data Table S3).
There was no evidence that the effect of HIV infection on postdischarge mortality was modified by age (P=0.95), anaemia (P=0.25), malaria (P=0.37) or severe wasting (P=0.88). The effect of malaria on post-discharge mortality was not modified by anaemia (P=0.33) but was modified by severe wasting (P=0.007) and weak evidence of being modified by age (P=0.05). Compared to MUACZ ≥-2 with a negative malaria slide, the groups of MUACZ -3 to -2 with a negative malaria slide (HR 3.60 [95%CI, 1.75 to 7.39]) and MUACz<-3 with a negative malaria slide (HR 6.36 [95%CI, 3.10 to 13.1]) had greater post-discharge mortality. Among children with positive slide results, MUACZ was not associated with post-discharge mortality.
In the exploratory analysis of biochemical variables, hyperkalaemia and elevated creatinine were associated with post-discharge mortality (Extended data Table S5).

Discussion
In this large study of school-aged children admitted to hospital and systematically followed-up during admission and for one year after discharge, we observed that anaemia, malaria, sickle cell disease and trauma were the leading reasons for admission. This profile is unlike that reported among under-fives, in whom pneumonia and diarrhoea are the leading reasons for admission, accounting for more than two-thirds of hospital admission in Kenya and South Africa 24-26 . All-cause inpatient case fatality was 3.4%, which was similar to the 3.5% previously reported in six hospitals in Kenya among children 5 to 17 years old in 2013 5 . Markers of disease severity at admission were the main predictors of inpatient mortality despite following WHO care guidelines, suggesting that current management strategies and resources may be insufficient for the sickest school-aged children.
#Empyema-1, cerebral palsy-1, chromosomal abnormality-1, unclassified disease-3, septicaemia-7, renal failure-3, pulmonary tuberculosis-1, chickenpox-1. Variables are included in this table if they were selected through the backward step-wise approach and have adjusted P<0.05 for either inpatient or post-discharge analysis. All variables considered in the regression models are shown in the univariable analysis models (Table S3). MUAC, mid-upper arm circumference; HAZ, height-for-age z-score; AUC, area under receiver operating characteristics; RR, risk ratio from the logbinomial regression model; HR, hazard ratio from Cox proportion regression model, the global Schoenfeld residuals test for proportional hazard assumption P-value=0.11. *Insufficient numbers.
non-communicable diseases including cancer and heart disease reported in that population may have contributed to the higher risk of post-discharge death.
Tachypnoea, which may be associated with anaemia, hypoxia, sepsis or pneumonia, and weak pulse, a sign of circulatory insufficiency, were independently associated with post-discharge mortality. Undernutrition was associated with post-discharge mortality, similar to most previous studies, which have identified nutritional status as a major risk-factor for post-discharge mortality in under-fives 3,7,14,15,27,28 . We previously showed that MUACZ is valuable in predicting post-discharge mortality among children >5 years in a model only adjusted for age, sex and HIV status 20 .
We found severe anaemia was independently associated with post-discharge mortality; however, prior evidence of the effect of haemoglobin concentration on post-discharge mortality has been inconsistent, and may be influenced by the predominant causes of anaemia and local transfusion policies 3,14,16,[29][30][31][32] .
Interestingly, we found no significant effect modification by anaemia on the effect of malaria on post-discharge mortality. The finding that children with malaria had lower post-discharge mortality is consistent with previous reports from this site among children <5 years of age where malaria parasitaemia had an apparently 'protective' effect on post-discharge mortality 11 . This likely reflects the fact that when appropriately treated, mortality risk in children with malaria may be lower than in children with other causes of a similar severity of illness.
In exploratory analyses, elevated creatine and hyperkalaemia, markers of impaired kidney function, were identified as predictors of both inpatient and post-discharge mortality. One study among children aged 2 to 15 years in Tanzania identified estimated glomerular filtration rate <60 ml/min/1·73m 2 as a predictor of one-year post-discharge mortality 3 . This is an important finding since dehydration and sepsis are common and associated with acute kidney injury 33 , and current guidelines recommend potentially nephrotoxic empiric gentamicin, without capacity for monitoring levels 34 . However, clinical trials that could delineate attributable nephrotoxicity from background risks from renal disease, serious illness and dehydration have not yet been done.
Overall, our findings add to previous data in under-fives suggesting that hospitalisation marks an extended period of vulnerability, and reports suggesting that significant post-discharge deaths occur outside hospital 7,11,27 . The transition from inpatient to home care in the immediate three-month period following hospital discharge appears to be a critical period of risk. Risk stratification and targeted interventions for this population during this period might improve survival 4 . However, current services, such as for the management of malnutrition in the community or those to manage anaemia, largely focus on children <5 years of age and may not operate in a way that fully addresses the mortality risk in this population 35,36 .
Strengths of this study include the large sample size, high rates of follow-up and the detailed longitudinal data available. However, this study also had some limitations. We used data from a single hospital, including only children resident within the nearby KHDSS, thus our results may not be generalizable to other sites or children living further away from the main road. Biochemical features were not systematically collected and could not be included in the primary analysis. We were not able to ascertain whether HIV infected children and malnourished children attended and complied with treatments from comprehensive care and nutrition clinics after discharge from hospital. Moreover, this study did not include data on caregiver or household characteristics, socio-economic situation or access to care.

Conclusion
This study highlights important differences in reasons for hospitalisation among school-aged children compared to younger children, and high rates of inpatient and post-discharge mortality among subgroups of school-aged children. To improve survival, active risk stratification and targeting intervention and follow-up in the early months after hospital discharge are needed, along with expansion of services that normally focus on the underfives to vulnerable over-fives. The large proportion of deaths outside hospital suggests that facilitating access to healthcare among the most vulnerable and priority clinical evaluation if unwell may be important measures.
The project contains the following underlying data: • Over5years_multipleadmissions.csv (contains clinical, anthropometric, CBC, blood & CSF culture at the time of hospital admission for children aged 5 to 12 years from 2007 to 2016, also provided in .dta format).
• over5years_khdss.csv (contains vital status in the community following hospital discharge, also provided in .dta format).
• over5yearschemistry.csv (contains the biochemistry variables that were not systematically collected at admission. This file is used to run a sub-analysis of the chemistry factors associated with both inpatient and post-discharge mortality, also provided in .dta format).
• Data_Dictionary_NgariMM.pdf (contains a list of the variables of data collected at admission and discharge and their description).
• Discharge_diagnosis codes.csv (contain the list of codes for discharge diagnosis). This project contains the following extended data: • Additional data_Mngari et.al 2020.pdf (contains supplementary Tables S1−S5).
• 5older years analysis_v1.do (STATA script used to generate the summary participants characteristics at admission, reasons for admission to hospital and inpatient mortality including factors associated with inpatient deaths).
• post-discharge analysis_over5years.do (STATA script that runs the post-discharge analysis. It merges the Over-5years_multipleadmissions.dta with the over5years_khdss. dta, computes time under follow-up, post-discharge deaths, mortality rates and factors associated with post-discharge deaths).