Mortality during and following hospital admission among school-aged children: a cohort study

Background: Far less is known about the reasons for hospitalization or mortality during and after hospitalization among school-aged children than among under-fives in low- and 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 within the Kilifi Health Demographic Surveillance System (KHDSS). Children discharged alive were followed up for one year by quarterly census. 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 outside hospital. Post-discharge mortality was positively associated with weak pulse, tachypnoea, severe anaemia, HIV infection and severe wasting and negatively associated with malaria. Conclusions: Reasons for admissions are markedly different from those reported in under-fives. There was significant post-discharge mortality, suggesting hospitalisation is a marker of risk in this population. Our findings inform guideline development to include risk stratification, targeted post-discharge care and facilitate access to healthcare to improve survival in the early months post-discharge in school-aged children.


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 to hospital 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 to 12 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 . Two of the four studies from Tanzania, used same study participants to evaluate overall one-year post-discharge mortality based on admission diagnosis 3 and on levels of haemoglobin 16 respectively.
In this retrospective cohort study, 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. At KCH, approximately 60% of paediatric admissions are aged 1 to 59 months, who are mostly admitted and treated for pneumonia and diarrhoea 15,17,18 . 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

Amendments from Version 1
Introduction, clarified on paragraph three: two studies referenced in the background used same participants from Tanzania to evaluate effect of admission diagnosis and levels of hemoglobin on postdischarge mortality respectively.
Methods, Study settings: Paragraph 1 and 2; Added details of the common admission diagnosis for children 1 to 59 admitted at KCH. Provided more details of enumeration in the KHDSS and measures for improving data quality. Replaced Figure 2 with a new one that includes 95% confidence intervals.
Discussion, Paragraph 1: explained the results on association of epilepsy/convulsion with inpatient mortality.
Paragraph 2: Explained the post-discharge mortality rate in reference to background mortality within KHDSS and causes of deaths estimated from verbal autopsies in KHDSS.
Last paragraph: Added unavailable causes of deaths among those children who died in community or other health facilities as a study limitation.
Any further responses from the reviewers can be found at the end of the article REVISED and in-or out-migration 19 . During each enumeration round, data collectors move from one household to another using GIS-derived maps and ask pre-specificized questions to the head of household and other members as described elsewhere 19 . Data on admissions to KCH are linked to the KHDSS population database by matching individual child using unique ID number through predesigned database query programme. Both the KCH admissions and KHDSS databases are programmed with validation checks to improve data quality. Community enumerators within the KHDSS, clinicians and clinical assistants collecting data in the KHDSS and KCH respectively are regularly trained.

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 20 . 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) which were regularly checked for consistency. 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 outpatient 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 21 . 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. 22 . 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) 23 . Anthropometric measurements and systematically collected laboratory tests were regarded as missing not at random (Extended data Table S1). To ensure all children were included in the multivariable regression models, we used categorical 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 22 , is less affected by dehydration than weight-based measures 24 , 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 25 .

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 S4. Being severely wasted was associated with inpatient death in the univariable model, but the effect was attenuated in the multivariable model (Extended data Table S4). 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 S5).
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 post-discharge 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  Table S6).

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 26-28 . 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. Surprisingly, epilepsy/convulsion appeared 'protective' compared to other reasons for admission in this hospitalised population, but would not be expected to be protective if compared to children in the community. The 3.0% one-year post-discharge mortality, with almost three-quarters occurring within three months, broadly concurs with that observed in rural Mozambique three months after hospital-discharge (2.2%) among children 5 to 15 years old 14 , but was much lower than one-year post-discharge mortality (16%) observed among children 5 to 12 years old in Tanzania 3 . However, the Tanzanian study did not stratify results by age above or below 5 years and it seems likely that a higher prevalence of non-communicable diseases including cancer and heart disease reported in that population may have contributed to the higher risk of post-discharge death. However, the mortality rate of 31.2 deaths/1000 child-years observed was >34 fold higher than 0.91 deaths/1000 child-years within KHDSS among this age-group (unpublished data). The leading cause of death among the 26 children who died during readmission at KCH was malaria, similar to the leading cause assigned through verbal autopsies to children 1 to 4 years old who died in the community, but differing from that in infants among whom pneumonia was the leading causes of death in KHDSS 29 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 which would suggest some children may be discharged with ongoing unstable vital signs that may lead to post-discharge deaths. 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 3,7,14,15,17,30 . 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 22 .
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 3,14,16,31-34 . 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 than other reasons for admission 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 apparent 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 12 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 35 , and current guidelines recommend potentially nephrotoxic empiric gentamicin, without capacity for monitoring levels 36 . 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 mortality occurs outside hospital even among school-age children 7,11,17 . 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 37,38 .
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. We did not have access to causes of deaths among children who died at home or in other health facilities. 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.

Data availability
Underlying data • 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:
• 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 Over5years_multipleadmissions.dta with the over-5years_khdss.dta, computes time under follow-up, postdischarge deaths, mortality rates and factors associated with post-discharge deaths).

University of British Columbia, Vancouver, Canada
This is another excellent study by the Kilifi team and their colleagues which further highlights the importance of pediatric post-discharge mortality in resource limited settings, this time among a new population of children that has to date been largely neglected -school age children. I applaud this group for this new focus and believe that as evidence such as this continues to accumulate, that policy makers and others involved in the development and implementation of guidelines and recommendations, will increasingly see the provision of improved discharge and post-discharge care as a major priority to further improve child health outcomes.
This study is well justified and the reasons are well outlined in the introduction. The methodology follows a similar strategy as other previously published work by this team and is clear and concisely recorded.

Methods:
The methodology follows a similar strategy as other previously published work by this team and is clear and concisely recorded. Given that this study is retrospective in nature, and utilized data routinely collected by health workers during routine care, some mention of data quality should be considered, both in terms of standardized training, as well as quality assessment.
In terms of missing data, I see some minor statements at the bottom of Table 1, but this does not comprehensively state missingness in its totality. Would it perhaps be better to have another column in this table outlining the % missing for each for the variables of potential interest? For example, how many were missing "clinical features", how many were missing temp, how many were missing a diagnosis, etc. I recognize that some variables are defined by their presence (lethargy, decreased skin turger, etc.), thus missingness is impossible to determine, but for those that are not subject to this issue, a report of missingness will give the readers an overall picture of the quality of the data from which these analyses are conducted. I do note that nearly 1 in 4 are missing WAZ, which does seem quite high.
For children who were below 155 months, but would be older than 155 months prior to the completion of 12 month follow-up, did these children remain in the cohort beyond the 155 month upper age limit? If yes, did the KHDSS properly capture outcomes among these older children?

Results:
The results are adequately presented given the analyses conducted. Perhaps more details could be presented on missingness in the tables.
Did the authors examine discharge disposition as an associated factor with death (i.e. discharge against medical advice).
I note that the authors mention that 45% of deaths did not occur during readmission to KCH, but does this also mean that they died outside of ANY health facility?
For those that died during a readmission, can these causes of death be briefly described?
Was any comparative exploration done on early vs late post-discharge deaths?
Was any exploration on readmissions in general done?

Discussion
The discussion touches on the main points needed for this topical area of research. I have no further comments.

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

Moses Ngari, KEMRI/Wellcome Trust Research Programme, P.O Box 230 -80108, Kilifi, Kenya
We thank the reviewer for his comments. This has provided an opportunity to improve the manuscript. Below find point-by-point responses in italics: This study is well justified and the reasons are well outlined in the introduction. The methodology follows a similar strategy as other previously published work by this team and is clear and concisely recorded.
Thank for the compliment.

Methods:
The methodology follows a similar strategy as other previously published work by this team and is clear and concisely recorded. Given that this study is retrospective in nature, and utilized data routinely collected by health workers during routine care, some mention of data quality should be considered, both in terms of standardized training, as well as quality assessment.

We added details about training, validation checks in the databases and measurements equipment consistency checks in the methods (study settings and Data sources/measurement)
In terms of missing data, I see some minor statements at the bottom of Table 1, but this does not comprehensively state missingness in its totality. Would it perhaps be better to have another column in this table outlining the % missing for each for the variables of potential interest? For example, how many were missing "clinical features", how many were missing temp, how many were missing a diagnosis, etc. I recognize that some variables are defined by their presence (lethargy, decreased skin turger, etc.), thus missingness is impossible to determine, but for those that are not subject to this issue, a report of missingness will give the readers an overall picture of the quality of the data from which these analyses are conducted. I do note that nearly 1 in 4 are missing WAZ, which does seem quite high. For children who were below 155 months, but would be older than 155 months prior to the completion of 12 month follow-up, did these children remain in the cohort beyond the 155 month upper age limit? If yes, did the KHDSS properly capture outcomes among these older children?

We already had a table of numbers missing anthropometry (Extended data
Yes, a child older than 155 after 12 months follow-up would still remain in study. The KHDSS census enumerates the entire population including adults. So the one-year outcome within the KHDSS was properly captured.

Results:
The results are adequately presented given the analyses conducted. Perhaps more details could be presented on missingness in the tables.

We have added an extra table with list of variables with missing values (Extended data Table s1)
Did the authors examine discharge disposition as an associated factor with death (i.e. discharge against medical advice).
We have added a sentence in the results section (under post-discharge mortality section). 13 (0.43%) children absconded from hospital and all of them were alive after one-year postdischarge.
I note that the authors mention that 45% of deaths did not occur during readmission to KCH, but does this also mean that they died outside of ANY health facility?
We have provided more details in the results section. Briefly, 40 deaths occurred at home, 26 during re-admission at KCH and 23 in other health facilities.
For those that died during a readmission, can these causes of death be briefly described?
Yes, we have provided the estimated causes if death for the 26 deaths during readmission at KCH (Extended data Table S6).
Was any comparative exploration done on early vs late post-discharge deaths?
No. We did not consider any exploration because of the few post-discharge deaths (63 Vs 26 deaths).

Was any exploration on readmissions in general done?
No. Our pre-specified endpoints were inpatient and post-discharge deaths. Hospital readmission can be examined as a separate publication.

Discussion
The discussion touches on the main points needed for this topical area of research. I have no further comments.
Thank you.   It is interesting to note that some signs at admission you found in your study (i.e.: elevated RR, weak pulse) were associated with post-discharge death. These were also variables that were associated with in-patient death. The assumption that children are discharged from the hospital with stable vital signs (normal RR and normal pulse strength) would make me think these variables (unstable vital signs or concerning physical examination findings) would only be factors for in-patient mortality. I think in resource limited settings, some children are perhaps discharged with unstable vital signs which can lead to death post-discharge.
○ We agree, it is possible that children are discharged with unstable vital signs. However, most studies have no access to discharge clinical signs and thus not able to evaluate their effects on post-discharge outcomes. We have also highlighted this is in paragraph 3 of discussion.
The authors mentioned 45% deaths occurred without readmission to KCH. Could these children have been admitted to another health care facility in the area? I don't think we can assume they died "outside hospital" unless that was verified. This would lead back to explaining in the methods section how families were contacted postdischarge and what questions were asked. In paragraph #7 when you state "our findings add to previous data in under-fives suggesting..", shouldn't it be "over-five" since your study is on school aged children?
○ As we mentioned in the background, we found no study reporting post-discharge mortality specifically on children over-five years. Therefore, the study findings extend what has been observed in under-fives. We have made the statement clearer in the discussion (paragraph 6).
one admission. Can you clarify? Are you referring to the lack of data on subsequent clinic attendance when children are admitted in other hospitals (possibly outside the HDSS) after being discharged from KCH?
One important result is that 45% of deaths occurred without readmission to KCH. Among these, some children could have been admitted in other hospitals or in health facilities (and die before being referred to KCH). Based on info from the HDSS (e.g. verbal autopsies), is it possible to know the share of deaths that occurred outside of health facilities? ○ Can you add more details about how the patient data are linked to the HDSS database? Is this record linkage deterministic based on some kind of ID number or performed on-site through searches in the HDSS database?
○ The abstract states that reasons for admissions were markedly different from those reported in under-fives, but what are these? These are only mentioned very quickly in the discussion, with only two reasons (pneumonia and diarrhoea). More details should be provided on these, ideally from KCH.
○ Second paragraph of the main text, "children may be admitted", please specify in the hospital or/health facilities. ○ Page 3, "5 to 12" instead of '5 to12'.

○
In Table 1, I don't understand why 29% of children of the sample of 3907 children were admitted prior to age 5 years as these 3907 admissions are supposed to be of children aged 5-12 years (from the description of the sample in the Results section). Can you clarify? ○ In Figure 2, can you add confidence intervals around the cumulative hazard curves? ○ Is the work clearly and accurately presented and does it cite the current literature? Yes

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Partly
Thank for the question. We are referring to subsequent outpatient clinic attendance for management of chronic illness like HIV and TB. We have added the word "outpatient" to make the sentence clearer.
○ One important result is that 45% of deaths occurred without readmission to KCH. Among these, some children could have been admitted in other hospitals or in health facilities (and die before being referred to KCH). Based on info from the HDSS (e.g. verbal autopsies), is it possible to know the share of deaths that occurred outside of health facilities? We have added more details under the study settings in the methods section.
○ The abstract states that reasons for admissions were markedly different from those reported in under-fives, but what are these? These are only mentioned very quickly in the discussion, with only two reasons (pneumonia and diarrhoea). More details should be provided on these, ideally from KCH.
We added a sentence in the methods section: study settings to the effects that ~60% of admissions at KCH are children 1 to 59 months old and majority are admitted with pneumonia and diarrhoea and provided references.
○ Second paragraph of the main text, "children may be admitted", please specify in the hospital or/health facilities.
Thank you, we have corrected.
○ In Table 1, I don't understand why 29% of children of the sample of 3907 children were admitted prior to age 5 years as these 3907 admissions are supposed to be of children aged 5-12 years (from the description of the sample in the Results section). Can you clarify?