Time to Death and Its Predictors Among Neonates Admitted With Neonatal Sepsis at Public Referral Hospitals of Bahir Dar City, Northwest Ethiopia, 2021: Retrospective Cohort Study

Background: In spite of recent advances in health care, neonatal sepsis is still one of the major causes of morbidity and mortality in neonates and is an ongoing major global public health challenge. Its incidence of mortality varies from health institution to health institution and within the same health institution at varied times and depends on predisposing factors. So the aim of this study was to determine the time to death and its predictors among neonates admitted with neonatal sepsis. Methods: An institutional based retrospective cohort study was conducted among 500 randomly selected neonatal charts. The data was entered using Epi data version 3.1, and exported to and analyzed at STATA version 14. Bivariable and multivariable cox regression analysis were conducted to identify predictors of mortality. Association was summarized using adjusted hazard ratio (AHR) and statistical signicances were declared at 95% CI and P-value <0.05. Proportionality assumption was tested by global test based on Schoenfeld residuals analysis. Results: During the follow up time, 58 (11.6%) neonates were died with neonatal sepsis while the rest were censored. In this study the overall death incidence rate was 20.5 per 1000 neonate days. Comorbidity (AHR: 1.81, 95%, CI: 1.04, 3.17), late initiation of exclusive breast feeding over one hour (AHR: 2.29, 95%CI: history of intra-partum fever(AHR: 7.37 95%CI: 2.28,23.79), Birth weight (AHR: 3.37, 95%CI: 1.54, 7.34) and place of delivery(AHR: 3.83, 95%, CI: 1.24, 11.83) were found potential independent predictors of mortality among neonates with neonatal sepsis. Conclusion: In this study, high death incidence rate was observed. The mean time to death among neonates was 4.41 days. With regard to predictors: birth weight, comorbidity, late initiation of exclusive breast feeding over one hour, history of intra partum fever and place of delivery were signicant predictors of mortality among neonates admitted with neonatal sepsis. Cox proportional hazard regression model was tted and used to identify predictors of mortality in neonates with neonatal sepsis. In bivariate Cox proportional hazard regression: neonatal sex, gestational age, weight at admission, history of PROM, place of delivery, history of liquor, history of fever, comorbidity, cry immediately at birth, resuscitated at birth, EBF initiated within one hour, history of previous hospitalization, rst minute APGAR score and fth minute APGAR score were signicant predictors of mortality among neonates with neonatal sepsis with p-value<0.2. In a multivariable proportional hazard model: admission weight, comorbidity, late initiation of EBF over one hour, history of fever and place of delivery remains potential independent predictors of mortality among neonates with neonatal sepsis at (p value<0.05).


Background
Neonatal sepsis is a clinical syndrome in an infant 28 days of life or younger, manifested by systemic signs of infection and isolation of a bacterial pathogen from the bloodstream (1,2). According to the international pediatric sepsis consensus conference of 2005, neonatal sepsis can also be de ned as systemic in ammatory response syndrome in the presence of or as a result of suspected or proven infection in a neonate which presents with clinical features of temperature instability, respiratory problems, feeding intolerance, and isolation of bacteria or other pathogens from the bloodstream(3).
Neonatal sepsis occurs when pathogenic bacteria gain access into the blood stream causing overwhelming general infection. The most common isolated bacteria were gram-negative (4).
Neonatal sepsis is classi ed into two major categories based on the time of onset: Early-onset neonatal sepsis (EONS) and Late-onset neonatal sepsis (LONS). Early onset neonatal sepsis appears within the rst seven days of life and is mainly due to bacteria acquired before and during delivery and most cases appear with in the rst 24 hours (5). According to the World Health Organization (WHO) report in 2006, out of the 130 million live births every year, 4 million die within the rst four weeks of life. Of these deaths, 99% occur in developing countries (6). Neonatal sepsis is accountable for 15% of global neonatal deaths and 30-50% of neonatal deaths in developing countries (7). In sub Saharan Africa, 17% of neonatal deaths are due to neonatal sepsis (8). Around 37% of Ethiopian neonates also died due to neonatal sepsis, which accounts for more than one third of neonatal deaths (9).
Sepsis is still one of the major causes of morbidity and mortality in neonates. In spite of recent advances in health care units, it is an ongoing major global public health challenge leaving the survivors vulnerable to short and long-term neurodevelopmental morbidity (10). Among those who survive, impaired quality of life, increased dependency, re-hospitalization, increase healthcare consumption, and along with increased mortality, all contribute the humanistic and economic burden of sepsis (11). Neonates could also face long-term survival with post infectious hydrocephalus (PIH), which could result in childhood mortality or survival with PIH with resulting neurological impairment and this long-term survival from sepsis with neurodevelopmental impairment could result in cerebral palsy or lifetime neurodevelopmental impairment without cerebral palsy; or uncomplicated recovery following an acute 2-week period of neonatal sepsis (12).
Different studies have been conducted in developed and developing countries to determine incidence density rate of neonatal mortality with sepsis. A study done in Brazil showed that the incidence and the incidence density of sepsis was 22% and 18.6 per 1000 central venous catheter-days, respectively. Considering very low birth weight neonates (VLBW) neonates as the reference group, the hazard ratio (HR) for time to death was 4.06 from day 0 to 60 and for time to the rst episode of sepsis was 1.76 from day 0 to 36. Having the heavier neonates group as reference, the HR for time to the rst episode of sepsis was 2.94 from day 37 to 90 (13).
A study conducted in Tanzania revealed a mortality rate of 19% in neonates with sepsis and high mortality is seen in early onset neonatal sepsis and showed a survival rate of 47% (14). Another study conducted in Uganda shows that mortality rate due to neonatal sepsis is 18.1% and most of the deaths (84%) occurred in the rst 48 hours of admission (15). In Nigeria a study indicate that mortality rate due to sepsis was 15.7%. In this study, low birth weight was 7.2 times more likely to die compared to those who were appropriate for birth weight. Babies that had PROM were 2.5 times more likely to die(16). Another study done in Johannesburg, showed the overall mortality rate for neonatal sepsis was 1.7% (17).
A study done in Sub-Saharan Africa to estimate the incidence density rate of neonatal mortality among neonates the incidence density rate of mortality was signi cantly high. And the hazard of neonatal mortality in NICUs was increased by comorbidities like sepsis with the hazard of death of neonates with sepsis was 1.94 times than neonates free from sepsis (18).
A study in southern Ethiopia revealed that the incidence rate of mortality among neonates admitted with neonatal sepsis was 14.57 per 1000 neonate days. The cumulative proportion of surviving at the end of the fourth day was 99.5%, and it was 98.2% at the end of the fth day. In addition, it was 96.6%, 93.5%, and 91.1% at the end of the sixth, seventh, and eighth day, respectively (1).
Prolonged rupture of membrane(>48hrs), maternal fever, and chorioamnionitis were found signi cant predictors of mortality of neonates with sepsis in a retrospective study conducted in Pakistan (19).
Efforts to reduce hospital acquired late onset infections require much closer attention to appropriate hand washing, infection control, and proper techniques for placement and management of central catheters (20). Antenatal corticosteroid treatment for accelerating fetal lung maturation for women at risk of preterm birth has been shown to reduce birth complications including the risk of neonatal sepsis (21).
Even though the impact of neonatal sepsis remains a public health problem in resource limited settings like Ethiopia, limited evidences exist to show the time to death and predictors of this serious public health problem in different parts of Ethiopia particularly in the study setting. Thus, determining the time to death and identifying its predictors among neonates with neonatal sepsis is very important for optimizing neonatal care.
Though Ethiopia has made considerable achievement in the reduction of under-ve mortality rate, the neonatal mortality burden has not experienced the same reduction, which may be attributed to neonatal sepsis. Neonatal sepsis is a relevant public health issue because it consistently emerges as one of the main causes of neonatal morbidity and mortality. Therefore, evidence on time to death of neonates with sepsis and factors that predicts the death of neonates with sepsis in the area will provide knowledge for health professionals in early identi cation of high-risk neonates with sepsis and timely intervention for their better survival. It also enhances the current neonatal sepsis management strategies by making sure the various determinants contribute to neonatal sepsis. This study will also be used for future researchers as reference. Also identifying the predictors of neonatal sepsis will be helpful in designing strategies to prevent and/or treat neonatal sepsis. Therefore, the aim of this study is to determine time to death and identifying predicting factors for death among neonates admitted with sepsis at public hospitals of Bahir Dar city.

Study design, period, and settings
An institutional-based retrospective cohort study design was carried out from February 1, 2021 to March 1, 2021 among neonates with neonatal sepsis at Bahir Dar city public referral hospitals admitted from January 1, 2019 to December 31, 2020. This study was conducted in Bahir Dar city public referral hospitals namely: Felege Hiwot comprehensive specialized hospital and Tibebe Ghion specialized hospital. Bahir Dar city is the capital city of Amhara regional state located in the Northwest Ethiopia which is found 565 km away from Addis Ababa, the capital city of Ethiopia.
Study participants, sample size and sampling technique All neonates admitted with neonatal sepsis were the study population and those neonates age less than or equal to 28 days at admission and were admitted during January 1, 2019 to December 31, 2020 were included in the study. On the contrary neonates with incomplete chart records were excluded from the study. The minimum sample size required was determined with Epi info version 7 statistical packages using major predictor variables from previous study (1), by considering the following statistical assumptions: CI =95%, and power= 80%. Accordingly, onset of labor was an independent variable which gives maximum sample size of 454 as mentioned in the table below. Then by adding 10% contingency, the nal sample size was 500 Sampling procedure First, medical registration numbers of all neonates with neonatal sepsis admitted for the last two years (January 1, 2019 to December 31, 2020) were obtained from the hospitals' registration logbooks. Then samples were selected by simple random sampling technique using computer generating system, after having lists of the medical registration number of the neonatal charts with neonatal sepsis from both hospitals.

Data extraction procedures
A data extraction tool was adapted from the Ethiopian federal ministry of health, national standardized neonatal and delivery registration book and prepared in English as structured checklist to extract required data from selected charts. The checklist include; socio demographic factors of both maternal and neonatal, neonatal factors and maternal factors. The data was collected from the 1st date of admission to death or censored (lost to follow up, improved, referred or against medical treatment) or until the end of follow up period (28 days of age).The data was extracted with four BSc. nurses who had work experience in NICU and two MSc. nurses were also recruited for supervision throughout the data collection period. Training was given for data collectors and supervisors two days before data collection. Pretest was done in 5% of neonatal charts with neonatal sepsis in FHCSH 5 days prior to data collection period and inconsistencies were checked and corrected. A close supervision was carried out by the supervisors during data collection period. Finally, all the collected data was checked by investigator for its completeness and consistency.

Operational de nitions
Neonate: A new born from birth until 28 days old Neonatal sepsis: Sepsis diagnosed and con rmed by physician among neonates.
Early-onset neonatal sepsis: Sepsis diagnosed and con rmed by physician among neonates age less than seven days.
Late-onset neonatal sepsis: Sepsis diagnosed and con rmed by physician among neonates age seven days or more.
Event: Death of neonate admitted with neonatal sepsis. Censored: Neonates with other than event/death (Improved and discharged, referred to other health facility, against medical treatment) Time to death: The time in days from admission to death of neonates with neonatal sepsis Follow up time: Neonatal age period (neonate with neonatal sepsis from admission until either event or censorship occurs).

Data processing and analysis
Data was cleaned, edited and coded by using Epi data version 3.1. Consistency of data was also checked before analysis to avoid errors during data entry. After that, it was exported to STATA 14 statistical software. Descriptive statistics (mean, standard deviation, frequency and percentages) were computed depending on the nature of variable. Outcome of each participant was dichotomized into censored or death. Incidence density rate (IDR) was calculated for the entire study period. Failure probability table and Kaplan-Meir (KM) failure function was used to estimate the mean time to death and cumulative probability of failure. KM plot and log-rank tests were used to compare failure curves. Before performing The Cox-proportional hazard regression, model, goodness-of-t was checked by Cox Snell residuals and assumptions were also checked by using Schoenfeld residual test. For each independent predictor bivariable cox proportional Hazard regression was performed. Then the variables with p-vale <0.2 were included in multivariable cox proportional hazard regression. AHR with 95% con dence interval and p-values<0.05 was used to measure the strength of association and identify statistical signi cant predictors.

Socio-demographic characteristics of the mothers and neonates
In this study, 500 neonatal charts were reviewed. More than half 296(59.2%) were males which accounted for 70.69% of dead neonates as compared with female neonates. The mean (±SD) age of the neonates was 4.41 ± 6.75 days. About 82.6% of the neonates had age less than seven days. Among the neonates who died with neonatal sepsis, 37(77.08%) of them died within the rst seven days of life. Concerning to mothers' socio demographic characteristics, the mean (±SD) age of mothers was 27.44 ± 5.32 years range from 15 to 44 years (Table 1).  (Table 3). Neonatal related characteristics of neonates with neonatal sepsis From 500 neonates admitted with neonatal sepsis, the mean (±SD) weight of neonates at admission was 2715.4 ± 643.8 grams and more than half, 37 (63.79%) of death was observed among neonates with admission age less than 2500 grams. Regarding gestational age, the mean (±SD) gestational age was 38.1±2.7 weeks ranging from 29-45 weeks. Among dead neonates majority, 49 (84.48) of them were neonates who had APGAR score less than seven at the rst minute and 37(63.79%) of dead neonates had APGAR score less than seven at fth minute (Table 3). In this study since the maximum observation time was censored the median time to death was not determined therefore, mean time to death was the appropriate measure of central tendency. Accordingly, the mean time to death among neonates died was 4.41 days (95% CI: 3.62, 5.21).
The probability of death in neonates with neonatal sepsis during the follow-up time was also presented by the failure curve. Accordingly, during the rst seven days, the graph went up increasingly showing a higher probability of death. Also, between days 7 and 13, the probability of death continued and the graph moved upward. In the remaining days of the follow-up period, the graph became straight indicating the likelihood of death in neonates with neonatal sepsis remained stable with almost no deaths occurring ( gure 2).
Log-rank estimate of mortality among neonates with neonatal sepsis across predictors As shown in the table below the log-rank test estimate between categories of different predictor variables revealed that the survival pattern among neonates with neonatal sepsis was signi cantly different as there is a highly signi cant difference among survival curves. The Kaplan-meier together with the log-rank test shows the effect of each predictor on the neonatal mortality with neonatal sepsis (Table 4). Testing overall tness of the model Model goodness of t was checked by using the Cox-Snell residual test. The residuals had a standard censored exponential distribution with hazard ratio. The jagged line with the reference line (Cox Snell residual line) follows the 45˚ line closely. Hence, the overall cox-regression model ts to the data ( Figure  3).

Cox proportional hazard assumption test
For each covariate, Cox proportional hazard assumption was performed individually and simultaneously (Globally).The test showed that the p-value for each covariate and the whole covariates simultaneously were greater than 0.05 which showed that there were no time-varying covariates in the model (Table 5). 1.04, 3.17). Neonates who were born from mothers who had a history of fever were 7.37 times more likely to die as compared to those neonates born from mothers with no history of fever (AHR: 7.37 95%CI: 2.28,23.79). Neonates who did not get initiated to exclusive breast feeding within one hour showed a 2.44 times higher hazard of death than neonates who did get initiated to exclusive breast feeding within one hour(AHR: 2.29, 95%CI: 1.13,4.63) ( Table 6).

Discussion
This study showed that the overall incidence rate of mortality was 20.5 per 1000 person days observation which was a little higher than the study conducted at Arba Minch, Southern, Ethiopia which was 14.57 per thousand person-days observation (1). This might be because of the difference in the follow up time that is, in our study we followed the neonates for the entire neonatal period but in the other study the neonates were followed for 14 days.
In this study, neonates with admission weight <2500 grams had 3.37 times higher risk of death compared to neonates with admission weight > 2500 grams among neonates with neonatal sepsis and this was in line with a study done in India which had reported that neonates with admission weight < 2500 grams to have an increased risk of mortality with sepsis (22). A similar study done in Indonesia had also reported that admission weight < 2500 grams was signi cant predictor of mortality in neonates with neonatal sepsis (23). This might be due to low birth weight neonates over normal birth weight neonates have de ciencies in humeral and cellular immunity as well as high tendencies for prolonged hospitalization, which increases the risk of nosocomial infection.
In the current study, neonates who were delivered from mothers having history of intra partum fever had 7.3 times higher risk of mortality compared to neonates who were delivered from mothers who had no history of intra partum fever. This study is supported by a study done in Arba-minch, Ethiopia which had reported that neonates who were delivered from mothers having history of intra partum fever had 14 times higher risk of mortality compared to neonates who were delivered from mothers who had no history of intra partum. This study was also supported by a study conducted in Iraq which reported that neonates with neonatal sepsis who were born from mothers having history of fever were more likely to die than neonates with neonatal sepsis who were born from mothers who had no history of fever (24). Our study was also in line with a study conducted in India which reported that intra partum fever was signi cant predictor of mortality related to neonatal sepsis (25). This might be because the fact that the diseasecausing agent of maternal infection can ascend to the baby via circulation and during passage through the birth canal.
Another predictor which was found signi cant in this study was place of delivery. Accordingly, consistent with a study result done at the university of Gondar comprehensive specialized hospital, neonates with neonatal sepsis who were delivered at home had 3.31 times higher risk of death compared to neonates who were delivered at health institution(26). The reason might be due to a lack of basic newborn care and environmental factors when the neonates were delivered at home.
Moreover in this study, the hazard of mortality among neonates with neonatal sepsis having late initiation of exclusive breastfeeding over an hour was 2.4 times higher compared to those neonates with neonatal sepsis who did get initiated to exclusive breast feeding within one hour. This nding is in line with to the studies conducted in India (27) and developing countries(28). This might be the fact that the many defense factors of the mother's milk include large amounts of secretory Immunoglobulin A antibodies produced by lymphocytes which have migrated from the mother's gut to the mammary glands.
Thus breastfeeding modulates the early exposure of the neonate's intestinal mucosa to microbes and limits bacterial translocation through the gut mucosa and neonates who did not get initiated to exclusive breast feeding within one hour could not get this bene t from mother's milk.
The last predictor variable that predicted the neonatal mortality with sepsis was having comorbidities. In this regard our nding revealed that the hazard of mortality among neonates with neonatal sepsis having comorbidities was 1.8 times higher compared to those neonates without comorbidities. And this result is consistent with a study done in Iraq (24) which revealed that the presence of the likes of hypothermia would increase the likelihood of death among neonates with neonatal sepsis. This might be the fact that the presence of comorbidities would expose the neonate for further overwhelming of the immature immune.

Conclusion
In this study, high death incidence rate was observed. The mean time to death among neonates was 4.41 days. With regard to predictors: admission weight <2500 grams, comorbidity, late initiation of EBF over one hour, having history of intra partum fever and home delivery were found statistically signi cant predictors of mortality among neonates admitted with neonatal sepsis. We recommend future researchers to use longitudinal prospective cohort study design to avoid the drawbacks of retrospective cohort design and so as to not to miss important socio demographic predictors.