A multicenter prospective follow-up study design was conducted from November 11 to December 7, 2021.

This study was conducted in the neonatal intensive care unit (NICU) of five referral hospitals in northwest Ethiopia. We have included all 5 referral hospitals in Northwest Ethiopia namely; the University of Gondar Comprehensive Specialized Hospital, Debre Tabor Comprehensive Specialized Hospital, Debre Markos Comprehensive Specialized Hospital, Felege Hiwot Comprehensive Specialized Hospital, and Tibebe Ghion Specialized Hospital (13).

All neonates with a diagnosis of neonatal sepsis admitted to the Neonatal Intensive Care Unit (NICU) in referral hospitals of North West Ethiopia during the study period were the study population. All neonates with neonatal sepsis admitted to neonatal intensive care units and start treatment were included in this study and neonates with major congenital abnormalities were excluded from the study. The sample size of the study was calculated using the following assumptions; the probability of an event from the previous study (14.7%), the incidence of mortality among admitted neonates with sepsis from a study conducted at Mizan Tepi University Teaching Hospital, Southwest Ethiopia (14). Probability of type 1 error or alpha 0.05, power of the study 80%, 95% confidence level, and 10% withdrawal probability from the study. The sample size was also calculated considering different hazard ratios (HR) of significant predictors from the previous studies including birth weight neonatal age at admission, maternal infection, and length of hospital stay and the largest sample size was considered as a final sample size.E=4(Zα//2+Zβ)2(lnHR)2E=4(1.96+0.84)2(ln2.13)2E=55Where E is the number of required event for the study.

Then the required total number of individuals was found from the formula:N=EP(E)N=550.147N=374Where, N is calculated sample size and P(E) is the probability of an event from the previous study. Finally considering a 10% withdrawal probability, we use a final sample size of 412.

A stratified random sampling was employed to select neonates admitted with sepsis in the selected referral hospitals. The sample size was allocated proportionally to each hospital based on the hospital’s patient flow (average number of neonates admitted in each hospital). Then study subjects were selected using systematic random sampling techniques in which every other neonate admitted with sepsis were included in the study during the study period.

Time to death; is calculated from the time of the initiation of treatment till the death of the neonate.

Censored: neonates who did not experience death during the follow-up period including lost follow-up, referred to other health institution, discharged with improvement, or stayed with admission beyond 28 days of neonatal age.

Neonate: newborn from birth to 28 days old (15).

Neonatal sepsis: sepsis diagnosed clinically by professionals or attending physicians during admission of the neonate (16).

Low birth weight: a birth weight of less than 2,500 g at birth (17).

A pretested and structured Amharic-version questionnaire was used to collect the data. Maternal sociodemographic characteristics were collected using face-to-face interviews, and obstetrics and baseline clinical characteristics of the neonate were recorded during admission. Then the neonates were followed until discharged, died, and were lost to follow-up. One MSc nurse supervisor and two BSc data collectors were recruited in each referral hospital to collect the data.

The data collectors and supervisors were trained for 1 day on the objective of the study, how to extract the treatment outcome from a chart, and the basics of interviewing before the actual data collection. A structured questionnaire was first prepared in English and then translated to Amharic and back to English by language experts to check consistency and conceptual similarity. The data collection process was closely monitored by the principal investigator and the supervisors throughout the data collection period. Filled questionnaires were checked regularly for completeness of information, and any problems were immediately discussed with the data collectors. A pre-test was conducted on 5% of the sample size in Debre Markos Referral Hospital before the actual data collection period.

Data were entered into EpiData version 4.6.0.6 and exported into STATA version 14 for analysis. Frequency tables, percentages, and median were used to describe the data. The hazard of death among neonates admitted with sepsis across different categories of the covariates was compared by using the nelson Aalen curves and log-rank tests. The Cox proportional hazard (PH) assumption was checked graphically by the log-minus-log function and statistically by Schoenfeld residual (global) test and the assumption was fulfilled. Cox PH and three parametric models (Exponential, Weibull, and Gompertz) were fitted to identify the predictors of death among neonates admitted with sepsis. The best model was selected using Akaike Information criteria (AIC), Bayesian information criteria (BIC), and log-likelihood criteria. The exponential regression model was selected as the best model with the lowest AIC and BIC. The Goodness of fit of the model was assessed by using the cox-snell residual technique. The Shared Frailty model was checked to see a random effect, an unobserved multiplicative effect on the hazard rate, for all individuals in the same hospital. But the frailty for the hospital was not statistically significant (theta = 0.24, x² = 0.11, P-value = 0.372).

A bivariable and multivariable exponential regression model was fitted to identify predictors of mortality. Variables with a P-value of 0.2 in the bivariable analysis were entered into the multivariable analysis. The adjusted hazard ratio (AHR) with 95% CI was calculated, and statistical significance was declared at P-value of 0.05 in the multivariable analysis.

A total of 406 neonates were included in the study, providing a response rate of 95.5%. The mean age of the mothers was 29 years with a standard deviation of 4.7 years. More than two-thirds 276 (67.98%) of the mothers of the neonates were between the ages of 25 and 34 years; 384 (94.58%) were married; 123 (30.30%) were unable to read and write, and 198 (48.77%) were housewives (Table 1).

Out of 406 neonates enrolled, 234 (57.64%) were males, and the majority of the neonates, 392 (96.55%), were less than 1 week old at admission. About 217 (43.35%) neonates had a birth weight of less than 2,500 g. More than half of the neonates, 230 (56.65%), were initiated breastfeeding, and about 91 (22.41%) of the neonates had RDS and 13 (3.20%) had PNA at admission. Twenty-five (6.16%) of the neonates had a history of previous admission, and more than half (63.55% and 60.34%) had a normal WBC count and platelet count, respectively. Almost all (99%) of the neonates initiated their treatments with ampicillin and gentamicin. The first antibiotic was changed for 57 (14.04%) of the neonates, and 48 (84.21%) of them were due to a poor response to the first antibiotic (Table 2).

More than half of the mothers (54.93%) received antenatal care. Thirty-five (8.62%) of the mothers had a history of stillbirth, and 47 (11.58%) had a history of abortion. Sixty-five (16.01%) of the mothers had hypertensive disorders of pregnancy during the current pregnancy; of them, 61 (93.8%) had pre-eclampsia and four (6.22%) had eclampsia. More than three-fourths (82.02%) of the neonates were delivered at hospitals, and more than half (66.75%) were delivered spontaneously. Nearly half (47.04%) of the women had a history of premature rupture of membrane (PROM), of them 16 (8.38%) had a latency period of eighteen and more hours (Table 3).

A total of 406 neonates admitted with sepsis were followed for a median follow-up period of 6 days (IQR: 3–8 days), with minimum and maximum follow-up periods of half a day and 26 days, respectively. Of the total study subjects, 75 (18.47%, 95% CI: 14.82, 22.60) neonates died during the study period. The incidence rate of death was 28 (95% CI, 22, 35) per 1,000 person-days of observation, with a total follow-up time of 2,677 person-days of observation. The cumulative hazard of death among neonates admitted with sepsis on the first, third, sixth, eighth, and 26th day of admission were 3.53% (95% CI: 2.09, 5.86), 9.68% (95% CI: 7.07, 13.19), 18.08% (95% CI: 14.22, 2.283), 22.57% (95% CI: 17.96, 28.15), and 35.51% (95% CI: 26.77, 46.07), respectively (Figure 1).

The Nelson-Aalen curve shows that the hazard of death among neonates admitted with sepsis differs across different baseline clinical characteristics of the neonates. Preterm neonates, low birth weight neonates, and neonates with less than 90% oxygen saturation at admission had a higher hazard of mortality as compared to their counterparts (Figure 2).

Cox PH and three parametric models (Exponential, Weibull, and Gompertz) were fitted to identify the predictors of death among neonates admitted with sepsis. The exponential regression model was selected as the best model with the lowest AIC and BIC. The Cox proportional hazard (PH) assumption was fulfilled, as shown both graphically by the log-minus-log function and statistically by the Schoenfeld residual test (global test, X² = 20.80, P-value = 0.7035).

The overall goodness of fit of the exponential regression model was tested by applying the Cox-Snell residual plots. As shown in Figure 3, the line associated with the Cox-Snell residual of the exponential Cox regression model was closer to 45° straight lines of the origin which indicate that model is well-fitted.

On the multivariable exponential regression; duration of labor, breastfeeding status of the neonate, gestational age, birth weight, baseline oxygen saturation, and experiencing RDS at admission were significant predictors of death among neonates admitted with sepsis.

The hazard of neonatal death was two times (AHR = 2.06, 95% CI: 1.02, 4.15) higher among preterm neonates as compared to term neonates. Neonates who were delivered with greater than 24 h of labor duration had four times (AHR = 3.89, 95% CI: 1.38, 11.01) higher risk of death than those who were delivered with 6–12 h of labor duration. The hazard of neonatal death was 2 times (AHR = 2.12, 95% CI: 1.01, 4.43) higher for neonates whose weight was less than 2,500 g compared to normal birth weight neonates. The hazard of death among breastfed neonates was reduced by 57% (AHR = 0.43, 95% CI: 0.23, 0.80) as compared to neonates who were not breastfed. The hazard of neonatal death was two times (AHR = 1.77, 95% CI: 1.02, 306) higher among neonates with respiratory distress syndrome as compared to their counterparts. Neonates whose oxygen saturation was less than 90% at admission had two times (AHR = 2.23, 95% CI: 1.07, 4.61) higher hazard of death than neonates with oxygen saturation ≥90% (Table 4).