Trends in inpatient and post-discharge mortality among young infants admitted to Kilifi County Hospital, Kenya: a retrospective cohort study

Objectives To describe admission trends and estimate inpatient and post-discharge mortality and its associated exposures, among young infants (YI) admitted to a county hospital in Kenya. Design Retrospective cohort study. Setting Secondary level hospital. Participants YI aged less than 60 days admitted to hospital from January 2009 to December 2019: 12 271 admissions in 11 877 individuals. YI who were resident within a Kilifi Health and Demographic Surveillance System (KHDSS): n=3625 with 4421 admissions were followed-up for 1 year after discharge. Primary and secondary outcome measures Inpatient and 1-year post-discharge mortality, the latter in KHDSS residents. Results Of 12 271 YI admissions, 4421 (36%) were KHDSS-resident. Neonatal sepsis, preterm complications and birth asphyxia accounted for 83% of the admissions. The proportion of YI among under-5s admissions increased from 19% in 2009 to 34% in 2019 (Ptrend=0.02). Inpatient case fatality was 16%, with 66% of the deaths occurring within 48 hours of admission. The introduction of free maternity care in 2013 was not associated with a change in admissions or inpatient mortality among YI. During 1-year post-discharge, 208/3625 (5.7%) YI died, 64.3 (95% CI 56.2 to 73.7) per 1000 infant-years. 49% of the post-discharge deaths occurred within 1 month of discharge, and 49% of post-discharge deaths occurred at home. Both inpatient and post-discharge deaths were associated with low admission weight. Inpatient mortality was associated with clinical signs of disease severity, while post-discharge mortality was associated with the length of hospitalisation, leaving against advice and referral to a specialised hospital. Conclusions YIs accounted for an increasing proportion of paediatric admissions and their overall mortality remains high. Post-discharge mortality accounts for a lower proportion of deaths but mortality rate is higher than among children aged 2–59 months. Services to address post-discharge mortality are needed and should focus on infants at higher risk.


BACKGROUND
The United Nations Sustainable Development Goal 3 aims to ensure healthy living and promote well-being for all ages, with all countries aiming to reduce neonatal and under-5 mortality to below 12 and 25 per 1000 live births by 2030, respectively. In sub-Saharan Africa, child mortality has declined by ~58% in the last 30 years. However, the estimated neonatal and under-5 mortality rates in sub-Saharan Africa remained high in 2019 (27 and 76 per 1000 live births, respectively) with a similar neonatal mortality rate of 27 per 1000 live births in Kenya. 1 Combined neonatal and post-neonatal infant mortality accounts for over three-quarters of all under-5 deaths in Kenyan children. 2 Young infants (YI) aged <60 days old comprise around half of the hospital admissions in sub-Saharan Africa and continue to face high risk of in-hospital mortality and long-term neurodisability. [3][4][5][6] Post-discharge mortality is emerging as a major problem in children in low-income and middle-income countries (LMICs), 7 however, there are limited data among YI. A systematic review of paediatric post-discharge mortality in developing countries included 24 studies published up to July 2017 with 19 from Africa. 8 Four studies included YI. Although young age was

STRENGTHS AND LIMITATIONS OF THIS STUDY
⇒ Large sample size with systematic data collection. ⇒ Linkage of hospital admissions to a well-established demographic surveillance system, with low loss to follow-up. ⇒ Lack of accurate gestational age estimation or birth weight of most participants. ⇒ Data are from a single hospital and only the population covered by demographic surveillance.

Open access
reported as a risk factor of mortality, no studies specifically identified deaths among infants aged <60 days. We have previously demonstrated excess post-discharge mortality among all hospitalised children, suggesting that hospitalisation itself selects vulnerable children with a sustained increased risk of dying over the longer-term. 7 9 Better understanding of YI deaths occurring during hospitalisation and after discharge from hospital is vital for development and use of targeted interventions aimed at improving survival.
This analysis aimed to describe admission trends and measure inpatient and post-discharge mortality and its associated exposures, including the introduction of free maternity care, among YI admitted to Kilifi County Hospital (KCH), Kenya, and followed-up through the Kilifi Health and Demographic Surveillance System (KHDSS).

Study participants and design KCH is a secondary-level referral hospital situated in Kilifi
County along the Kenyan coast. It serves a rural and periurban population. It has a maternity unit with approximately 6000 deliveries per year, a general paediatric ward with a newborn unit for babies aged less than 1 month and a paediatric High Dependency Unit (HDU) that also admits YIs. The year 2009 was selected as a starting point, because a previous analysis of mortality among YI covered admissions from 1990 to 2008. 10 Free maternity care was introduced by the Kenyan government on 1 June 2013 and led to a marked increase in health facility births. 11 The KHDSS, established in 2002, covers a population of 279 158 within an area of 900 km 2 centred on KCH. 12 Census rounds visit each household every 4 months to ascertain vital status and migration in and out of the hospital catchment area.
We conducted a retrospective cohort study of YIs resident within the KHDSS who were admitted to KCH between 1 January 2009 and 31 December 2019. Children discharged alive and followed-up in KHDSS census rounds until March 2021 were eligible for analyses of post-discharge mortality. During the study period, there were nine health workers' strikes with the last nurses' strike lasting for 150 days (5 June 2017 to 2 November 2017) 13 (online supplemental table S1). For comparison, we also examined admissions aged 60 days to 59 months during the same period.

Procedures
At admission, standardised medical history and clinical examination, including anthropometric measurements were obtained by trained clinical staff. Blood samples were systematically taken for complete blood count, slide for malaria microscopy and clinical chemistry, HIV antibody test and blood culture at hospital admission, as described previously. 14 A lumbar puncture for cerebrospinal fluid analysis was done at admission in infants in whom sepsis was suspected and deferred in those seriously ill or with other contraindications. Clinical and laboratory data were recorded in real time on a ward surveillance database linked to the KHDSS database. Empiric antibiotics were initiated according to national guidelines 15 with ampicillin/benzylpenicillin plus gentamicin as first-line intravenous therapy. Second-line and subsequent antimicrobial therapy was guided by blood culture results and clinical progress. Mechanical ventilation was not available at KCH.

Study variables
Outcomes of interest were death in hospital and during 1 year after discharge. Exposures of interest were demographic, nutritional, clinical features and haematological, biochemical and microbiological findings at the time of admission. De-identified study data were deposited in the Harvard Dataverse depository. 16 Weight at admission and mid-upper arm circumference (MUAC) were categorised as shown in table 1. Because approximately 40% of the YI were underweight (<2.5 kg), and 60% were aged ≤2 days at admission, YI's admission weights rather than anthropometric Z scores using WHO standards were reported. Furthermore, most YI who were born at home or in other hospitals and referred to KCH were missing gestational age estimates and birth weight to be able to estimate gestational age at birth using the INTERGROWTH 21st Newborn Size Standards (INSS).
Prematurity was defined as gestation age <37 weeks and low birth weight as birth weight <2500 g for YIs born at KCH. Admission blood glucose was categorised into <2.6, 2.6-7.0 and ≥7.0 mmol/L representing low, normal and high levels, respectively. 15 Missing data were not assumed to be missing at random. We, therefore, created categorical variables and added a missing category which was included in the regression analysis.
Demographic, anthropometric and clinical data are presented as frequencies and proportions for categorical variables and means (SD) or median (IQR) for continuous variables depending on the underlying distribution. Proportions of missing data for each variable are shown in online supplemental table S2.
Monthly admissions and case fatality were plotted against time (month of admission) to visually inspect the trend from 2009 to 2019 and the predicted trend line superimposed on the curves. We used the Augmented Dickey Fuller test to test if the time series were stationary (no trend or seasonal effects). We also presented annual absolute admissions, proportion of YI among all admissions <60 months and case fatality. Monthly admissions and case fatality were tested for annual linear trend using an extension of the Wilcoxon rank-sum test of trend across ordered groups. 17 We used interrupted time series analysis to estimate the level and trend changes before and after introduction of free maternity care (1 June 2013). We created a time month variable coded sequentially from January 2009 to

Open access
December 2019 and a binary variable coded as 0 and 1 for admissions before and after June 2013, respectively, to represent introduction of free maternity care. We defined seasonal effect variables using month of the year modelled on harmonic terms using the Fourier code in Stata. To measure the effect of free maternity care, we used the negative binomial regression model because of the presence of overdispersion in the trends and reported regression coefficients transformed into incidence rate ratios (IRR). All the negative binomial regression models included the following independent variables: the time month variable, the binary pre and post free maternity care variable and the seasonal effect variable. Since YIs could be admitted more than once while <60 days old, we included multiple admissions using unique IDs and adjusted for clustering by individual with robust SEs. To identify exposures associated with inpatient death, we treated being discharged alive as a competing event and fitted the proportional subdistribution hazard model using the Fine-Gray competing risk model. 18 The measure of effect reported from the model was the subdistribution HRs (SHR) and their respective 95% CIs. To build the multivariable regression model, a backward stepwise approach was used where all the independent variables assessed in the univariate models were included in the model and only those with a p value<0.1 retained in the final multivariable model.
For the post-discharge analysis, only data from those YI discharged alive and resident within the KHDSS were analysed. Time at risk was defined from the date of discharge to 365 days later or censure at the date of death or outmigration from the KHDSS. We performed a 'multiple discharges' analysis where YI with multiple admissions had their follow-up time reset at each successive discharge date. Exposures associated with postdischarge were assessed using a gamma distribution shared frailty Cox proportional hazards regression model accounting for YI with multiple discharges. The proportional hazards assumption was assessed using the scaled Schoenfeld residuals test (online supplemental tables S3 and S4). All exposures assessed in the univariate models were considered for inclusion in the multivariable Cox proportional hazards regression model using a backward stepwise approach similar to the inpatient analysis. Both the inpatient and post-discharge multivariable regression models' discrimination performance were assessed using bootstrapped area under receiver operating characteristic curves (AUC) replicated 1000 times.
As sensitivity analysis, we assessed the YI born at KCH and enrolled to the Kilifi Perinatal and Maternal Research Project, which had collected comprehensive birth data including birth weight and gestational age (weeks). 19 We estimated their birth weight Z scores using the INSS and ran the regression models replacing admission weight with birth weight Z score. 20 Statistical significance was evaluated using 95% CI and a two-tailed p value<0.05. Statistical analyses were conducted using Stata V.17.0 (College Station, Texas, USA).

Study size
We used all available eligible YI data from 2009 to 2019 (4421 for inpatient and 3625 for post-discharge analyses) regardless of sample size.

Ethical considerations
Written consent was provided by the caregivers of all the surveillance study participants.

Patient and public involvement
There was no patient and public involvement in the planning or execution of this retrospective cohort study.

Baseline characteristics
During the study period, there were 42 742 paediatric admissions to KCH, of which 12 271 (29%) admission events among 11 877 individuals were aged <60 days. Of the 12 271 YI admission events, 4421 (36%) were resident in the KHDSS and included in the analysis (figure 1). This comprised 4272 individual YI: 4131 with one admission, 133 with two admissions and 8 with three admissions within the first 60 days of life.
Among the 4421 KHDSS-resident YI admissions, median (IQR) time to death was 2 (1-4) days, while the survivors were admitted for 5 (3-8) days. A total of 423/645 (66%) deaths occurred within the first 48 hours following admission. Forty-one YI left against medical advice, and 55 were referred to other hospitals for further care.

Admission diagnosis and case fatality ratio
The most common reasons for hospital admission were neonatal sepsis (47%), preterm complications (20%) and birth asphyxia (16%) accounting for 83% of all YI admissions (table 2). The case fatality ratios for YI with  Open access respiratory distress syndrome, preterm complications and birth asphyxia were 52%, 29% and 28%, respectively (table 2).
One hundred and one (49%) of the 208 post-discharge deaths occurred at home without hospital readmission, 67 (32%) occurred during readmission to KCH and 40 (19%) occurred at other health facilities. The five leading assigned causes of deaths for the 67 deaths at KCH were: neonatal sepsis (24%), preterm complications (22%), congenital heart disease (15%), neonatal jaundice (7.4%) and meningitis (7.4%) which were similar to index admission diagnosis (online supplemental table S7). Causes of other deaths were unknown.
Exposures assessed for association with post-discharge mortality are shown in online supplemental table S8. In the multivariable Cox regression model, born outside KCH, log days of hospital admission, leaving against advice and referral to more specialised hospital were positively associated with post-discharge mortality. Other exposures associated with post-discharge mortality were low admission weight, fever and weak pulse (table 3). The multivariable model bootstrapped AUC was 0.76 (95% CI 0.73 to 0.80).

DISCUSSION Trends in admissions and proportions of YI
During the study period, we observed a marked increase in YI admissions and the proportion of YI among admissions in under-5s increased from around one-fifth in 2009, to more than one-third in 2019. However, this did not seem to be associated with the introduction of free maternity care in 2013. Lack of observable effect may be due to challenges faced during policy implementation arising from inadequate expansion of quality healthcare facilities and resources. Several authors reported an increase in mothers attending Kenyan health facilities for antenatal care and delivery, 11 21 however our results suggest this occurred in the context of a general trend which we previously observed during 1990-2008. 10 Conversely, the number of admitted children older than 60 days decreased alongside a reduction in local malaria transmission, 22 introduction of routine childhood pneumococcal conjugate and rotavirus immunisation 23 and expansion in numbers of health facilities in Kilifi County. 24 Variation in annual admissions over the years was due to multiple health workers' strikes. 13 During these periods, the general paediatric ward was closed and only the sickest children were admitted to the paediatric HDU due to limited staffing and bed capacity. The time series analysis indicated an increase in inpatient mortality during strikes (figure 3C).
The leading diagnoses at admission in our analysis were neonatal sepsis, preterm complications and birth asphyxia, similar to the period 1990-2008. 10 Over onethird of the admissions from KCH maternity were preterm and the hospital also received referrals of preterm and very low birthweight infants from subcounty hospitals and local health centres. There are few African published data sets of neonatal or YI inpatient diagnoses; in a network of seven Nigerian and Kenyan hospitals, prematurity accounted for over half (52%) and birth asphyxia almost a quarter (24%) of neonatal admissions. 25 The leading bacterial isolates from blood cultures in our study (K. pneumoniae, E. coli, S. aureus) were similar to those among YI in rural settings of Tanzania and Burkina Faso. 26 Kenya attained elimination status of maternal and neonatal tetanus in 2018, following immunisation campaigns in high-risk regions. 27 Compared with 1990-2008, 10 neonatal tetanus was uncommon at our centre with only 10 cases in 11 years.

Inpatient deaths
The WHO has reported that in 2019, '47% of all under-5 deaths occurred in the newborn period with about onethird dying on the day of birth and close to three-quarters dying within the first week of life'. 28 Delivery by a skilled health worker has been shown to be effective in reducing perinatal mortality. 29 We did not collect data on delivery by a skilled birth attendant but in 2018/2019 69% of the  Open access births in Kilifi County were reported to be attended by skilled health personnel which is slightly higher than the national average. 30 We found YI accounted for more than 60% of under-5s inpatient deaths, similar to a retrospective study of 16 Kenyan public hospitals in which neonatal deaths comprised 66% of the inpatient paediatric deaths. 5 We found respiratory distress syndrome, birth asphyxia and preterm complications had the highest inpatient mortality. Mechanical ventilation was not available in KCH. Improvements in peripartum care of mothers and infants together with appropriate technology such as non-invasive ventilation for management of respiratory complications of preterm birth are priorities for reduction in neonatal mortality in hospitals in LMICs. 5 Post-discharge deaths Less than a quarter (24%) of all deaths during 1 year of follow-up occurred post-discharge. This reflects a high inpatient (16%) case fatality rate with many very early inpatient deaths compared with 6.6% in children aged ≥60 days. 7 Nevertheless, the post-discharge YI mortality rate (64.3 per 1000 child/years) was more than twice that of a cohort of children aged 2-59 months admitted to KCH between 2007 and 2015. 31 This reflects postdischarge mortality rates being highest in younger age groups, such as in Tanzania among under 1 year olds: 72 per 1000 child/years (95% CI 67.2 to 77.2) falling to 6.9 (95% CI 5.5 to 8.7) per 1000 child/years in 4 to <5 years old. 32 A greater proportion of YI post-discharge deaths occurred in hospital than among older children, 7 implying that caregivers may be more likely to seek readmission for YI or may live closer to KCH. About half of the post-discharge deaths occurred within the first month, highlighting the need for formal 'down-referral' for continuity of care after discharge in high-risk YI.
Analysis of exposures revealed that some were common for both inpatient and post-discharge mortality: low admission weight, axillary temperature and respiratory rate. Birth weight was not available for most YI but low admission weight <2.5 kg was common (40%) in our participants. In YI it is difficult to distinguish low birth weight from malnutrition, but we have reported the higher case fatality rates in the lower admission weight categories (online supplemental tables S3 and S4). Of known causes of post-discharge deaths, leading ones were related to problems in the early neonatal period.
Strengths and limitations of the study Strengths of this study are large sample size, systematic collection of data and linkage to a well-established demographic surveillance system, with few losses to follow-up. Limitations are lack of accurate gestational age estimation, unknown birth weight of most participants and that individual socioeconomic data were unavailable. We did not have clinical data collected at discharge, which may be of value in taking a risk-based approach to post-discharge care. This analysis is from a single hospital and excludes residents outside KHDSS who may have different exposures and risks.

CONCLUSIONS
Neonatal and YI admissions account for an increasing proportion of inpatient paediatric admissions, and their overall mortality rate remains high. Post-discharge mortality accounts for a lower proportion of all deaths than hospital admissions aged 2-59 months but the postdischarge mortality rate among YI is higher. 31 33 This is likely because of the predominance of fatal neonatal conditions such as extreme prematurity or birth asphyxia. Services to address post-discharge mortality are needed and should focus on infants at higher risk.

Twitter Alison Talbert @AlisonWendyT and James Berkley @jayberkley
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.
Open access This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/ licenses/by/4.0/.