Predictors of neonatal mortality in Iranian cases of preterm birth: a retrospective cohort study

Abstract

Introduction

Although the risk factors of neonatal mortality are currently known, it was necessary to know the risk factors of mortality in neonatal preterm birth. The present study was conducted to find and investigate the potential maternal, fetal and neonatal risk factors of neonatal mortality among the cases of preterm birth.

Methods

A retrospective cohort study was conducted with a secondary analysis of existing data of a national registry. All premature infants born in Iran from March 21^st 2019 to March 21^st 2020 were included in the study using census. Mixed model logistic regression was used to identify the potential risk and protecting factors from a selected groups of the variables of the national registry with reporting odds ratio (OR) and 95% confidence interval (CI).

Results

A total of 120,688 cases of preterm labor were studied with GA in the range of 22–36 weeks. A total of 5458 neonatal deaths (4.52% of total cases) were recorded. The most effective protecting factor was increased gestational age (GA) (adjusted OR = 0.664 per week, 95% CI: 0.658–0.669). The risk factors remained in the multivariable modeling were maternal hypertension, eclampsia / preeclampsia, addiction, placental abruption, meconium staining, impaired fetal heart rate, previous neonatal death, intrauterine growth restriction, lack of corticosteroid administration, rural and nomadic residency, cesarean delivery, malformation, public hospital (vs. private), and out-of-hospital delivery (P < 0.1). The protecting factors remained in the multivariable modeling were gestational diabetes mellitus, premature rupture of membrane for more than 18 h, female sex, GA, and mother age (P < 0.1).

Conclusion

The present study showed that many maternal and neonatal related variables were potential risk and protecting factors of neonatal death among the preterm birth neonates. The main predictor of the outcome was GA, and using corticosteroid before birth was a modifiable protecting factor.

Introduction

Neonatal mortality is widely recognized as an indicator of a country’s development. It is inversely related to socioeconomic and health system indicators, being considerably lower in developed nations and higher in developing regions [1]. Late fetal and early infancy are periods of life with the highest mortality compared to other stages [2]. Even though the global number of infant mortalities has decreased from 5 million in 1990 to 2.4 million in 2019, children are still at the greatest risk of mortality during the first 28 days of life [3]. The causes of neonatal mortality is generally divided into two categories including biological and non-biological factors [4]. Although biological factors such as prematurity, infections, and asphyxia at birth are known causes of neonatal mortality [5], non-biological causes including economic, social, gender, and maternal literacy are important factors [6].

Neonatal mortality is influenced by various factors and conditions, and the identification and modification of these factors can be effective in mortality prevention [7]. Preterm birth is one of the major causes of perinatal death throughout the globe [8]. Preterm neonates are at a greater risk of mortality throughout the neonatal period until the third decade of life, in comparison to mature neonates [9]. In Iran, a meta-analysis showed that the prevalence of preterm labor was 10% in the range of 5.4% to 19.85% [10]. A study in Shiraz showed that the overall mortality among the preterm neonate cases was about 10% [11]. A study in Sari, Iran, showed the mortality among the preterm birth cases as 9.3%. The most important factor associated with neonatal mortality was gestational age [12]. In addition to mortality, the economic burden of preterm birth in Iran was high. This burden included hoteling, neonatal intensive care unit (NICU) admission, physician visit and consultation, medicine and medical equipment, nursing, and so on [13].

In spite of conducting a lot of research on the role and impact of preterm birth, it was necessary to find potential risk factors among the preterm birth cases. In addition, there is no report in this regard in Iran. Therefore, the present census-based retrospective cohort study was conducted to find and investigate the potential maternal, fetal and neonatal risk factors of neonatal mortality specifically among the cases of preterm birth.

Materials and methods

Study design

A retrospective cohort study was conducted using strengthening the reporting of observational studies in epidemiology (STROBE) statement with a secondary analysis on the collected data of a national registry. The study was approved by the ethics committee of Iran University of Medical Sciences with registration number IR.IUMS.REC.1398.834. All the ethical issues such as secrecy of the data and Declaration of Helsinki were respected the participants gave informed consent for registration in Iranian Maternal and Neonatal System and any secondary research use of the data.

Setting and participants

All premature infants born in Iran from March 21st 2019 to March 21st 2020 were included in the study using census. In addition, all out-of-hospital deliveries were subjected. The exclusion criteria were gestational age (GA) 37 weeks or more, not being Iranian (not having national ID), still birth cases and cases with influential missing data. The follow up time (retrospectively for research aim) was up to the time of discharge, death before day 28 or 28 days of hospital stay.

Variables

The main tool used in recording information in infant birth registration systems and infant death cases included two separate forms with over 200 variables. About 50 variables were used in this study. The birth registration system (Iranian Maternal and Neonatal System [IMaN]) consisted of two parts; the first included the characteristics of the mother and the second the characteristics of the baby. The outcome variable was neonatal death defined as death before day 28. Province of residence of the mothers was considered as the random effects variable.

The section for maternal profiling contained demographic and gestational information of the mother including maternal hypertension, cardiovascular disease, autoimmune disease, diabetes mellitus, hypothyroidism, mental disorders, hepatitis B virus infection, syphilis infection, chorioamnionitis, gestational pyelonephritis, gestational diabetes mellitus, eclampsia / preeclampsia, hemoglobin < 7 g/dl, tobacco smoking, addiction, psychedelic use, alcohol use, premature rupture of membrane for more than 18 h, placental abruption, meconium staining, placental adhesion, residency status, multiple pregnancy, participation in childbirth preparation class, previous neonatal death, previous intrauterine fetal death, gestational corticosteroid administration, using assisted reproduction techniques (ART), GA, delivery type, having insurance, mother education, mother age, hospital type and out-of-hospital delivery.

In the newborn profile section, information included head circumference, length, weight, impaired fetal heart rate, intrauterine growth restriction (IUGR), sex, malformation, resuscitation need, Apgar score < 5 at first and fifth minute, stay day at hospital and neonatal intensive care unit admission (NICU).

Data sources

The information was extracted from the birth registration system of the Ministry of Health (IMaN), which holds data related to all deliveries throughout the country and deaths that occur in the delivery room during the first two hours of birth. Some of the information about infant deaths after the first two hours of birth was extracted from the NICU system, which records infant deaths that occur outside the delivery room. The desired information was extracted from the two mentioned systems and linked to each other through the national ID of mother. For out-of-hospital births, if this information is not recorded, a birth certificate will not be issued. Such cases were included in the study.

Risk of bias

The main potential risk of bias was the missing data that might be not at random and also different operators for filling the birth forms over the country.

Statistical analysis

Study size was based on census. Descriptive statistics including frequency, mean and standard deviation (SD) was used to describe the variables in survived and death groups. Simple logistic regression was performed for crude comparison of the demographic variables and suspected risk factors between survived and death groups. Then a group of variables were selected for mixed model multiple logistic regression considering province as the random intercept of the model. Variance inflation factor (VIF) was used to check multi-collinearity. Statistical analyses were performed in Stata 17 (Stata Corp. LLC, TX, US).

Results

Participants and outcome

After cleaning the dataset, a total of 120,688 cases of preterm labor (out of 1,221,274 births of the year) were studied with GA in the range of 22–36 weeks. A total of 5458 neonatal deaths (4.52% of total cases) were recorded in which 1852 cases (1.53% of total cases) were death during first 24 h, 2264 cases (1.88% of total cases) were death from the end of first 24 h to the end of first six days and 1342 cases (1.11% of total cases) were death from day seven up to day 28.

Main results

The means and prevalence rates of the demographic variables and suspected risk factor are shown (Table 1) among the survived and dead neonates. The comparisons of these means and prevalence rates between the groups of survived and dead using simple logistic regression are also shown (Table 1). Briefly, maternal hypertension, chorioamnionitis, preeclampsia, addiction, premature rupture of membrane, placental abruption, meconium staining, impaired fetal heart rate, previous neonatal death, previous intrauterine fetal death, IUGR, ART, rural and nomadic residency status, ambiguous sex, school literacy, malformation, resuscitation, admission in public hospital, Apgar score < 5, out-of-hospital delivery, and NICU admission were associated with increased neonatal mortality (P < 0.001). Gestational diabetes mellitus, head circumference, body length, participation in child birth preparation class, using corticosteroid, weight, GA, academic literacy, and mother age were associated with decreased neonatal mortality (P < 0.001).

Table 1.

Mean and prevalence of demographic variable and suspected risk factors among survived and dead neonates

Variables	Prevalence / mean (SD)		P value
	survived	dead
Maternal hypertension	0.0422	0.0489	< 0.001*
Cardiovascular disease	0.0125	0.0117	0.935
Autoimmune disease	0.0050	0.0059	0.379
Diabetes mellitus	0.0123	0.0107	0.724
Hypothyroidism	0.1254	0.1068	0.068
Mental disorders	0.0112	0.0097	0.858
Hepatitis B infection	0.0017	0.0013	0.233
Syphilis infection	0.0002	0.0000	> 0.999
Chorioamnionitis	0.0027	0.0082	< 0.001*
Gestational pyelonephritis	0.0009	0.0019	0.013
Gestational diabetes mellitus	0.0964	0.0623	< 0.001*
Eclampsia / preeclampsia	0.0953	0.1139	< 0.001*
Hemoglobin < 7 g/dl	0.0008	0.0010	0.04
Tobacco smoking	0.0012	0.0019	0.179
Addiction	0.0076	0.0103	0.001*
Psychedelic use	0.0002	0.0008	0.011
Alcohol use	0.0000	0.0002	0.091
Premature rupture of membrane	0.0469	0.0604	< 0.001*
Placental abruption	0.0218	0.0529	< 0.001*
Meconium staining	0.0133	0.0273	< 0.001*
Placental adhesion	0.0075	0.0097	0.062
Impaired fetal heart rate	0.0370	0.0621	< 0.001*
Head circumference (cm)	32.031 (3.111)	27.300 (4.433)	< 0.001*
Length (cm)	45.385 (4.730)	37.845 (6.715)	< 0.001*
Multiple birth	0.2619	0.2172	0.680
Participation in childbirth preparation class	0.0817	0.0598	< 0.001*
Previous neonatal death	0.0305	0.0556	< 0.001*
Previous intrauterine fetal death	0.0214	0.0334	< 0.001*
IUGR	0.0455	0.0822	< 0.001*
Corticosteroids	0.5550	0.4729	< 0.001*
ART	0.0380	0.0399	< 0.001*
Residency status
Urban	0.7950	0.7440	Reference
Rural	0.2041	0.2530	< 0.001*
Nomadic	0.0009	0.0029	< 0.001*
Sex of neonate
Male	0.5581	0.5564	Reference
Female	0.4410	0.4394	0.022
Ambiguous	0.0009	0.0042	< 0.001*
Weight (kg)	2258.17 (652.55)	1298.81 (689.17)	< 0.001*
GA (week)	33.95 (2.46)	29.07 (3.64)	< 0.001*
Delivery type (cesarean)	0.7130	0.6295	< 0.001*
Having insurance	0.9180	0.9026	0.026
Mother education
Illiterate	0.0308	0.0468	Reference
School literacy	0.6893	0.7348	< 0.001*
Academic literacy	0.2798	0.2184	< 0.001*
Mother age	30.33 (6.35)	29.85 (6.49)	< 0.001*
Malformation	0.0115	0.0648	< 0.001*
Resuscitation	0.1845	0.6039	< 0.001*
Hospital type (public)	0.8630	0.9394	< 0.001*
Apgar score < 5 at first minute	0.0690	0.4885	< 0.001*
Apgar score < 5 at fifth minute	0.0149	0.2870	< 0.001*
Stay day at hospital	8.23 (10.60)	4.75 (5.67)	< 0.001*
Out-of-hospital delivery	0.0104	0.0237	< 0.001*
NICU	0.5713	0.9754	< 0.001*

* Significant at 0.001 (significant at 0.05 after Bonferroni correction) based on simple logistic regression

The variables with P < 0.3 and enough frequency (> 0.001 based on researcher choice) were subjected for mixed model multiple logistic regression. Variables NICU, stay day at hospital, resuscitation and Apgar score variables were excluded from multivariable modeling to prevent over-controlled bias as they have intermediate roles. From the variables GA, weight, height and head circumference, only GA was subjected for multivariable modeling due to collinearity (r > 0.7). Finally, a backward approach based on P values was used up to the time that there was no covariate with P > 0.1. Random intercept of provinces was added to the model.

According to the multivariable modeling (Table 2), the largest effect size was for malformation (OR = 6.281, 95% CI: 5.484–7.193). As expected, GA showed a protecting effect (OR = 0.664 per week, 95% CI: 0.658–0.669). Comparing with the crude results, the association of PROM was changed from risk to protective and the association of delivery type (cesarean) was changed from protective to risk (Table 2). No multi-collinearity was detected on the basis of putative linear regressions (largest VIF = 1.13, mean VIF = 1.04). A good performance with AUC = 0.894 was obtained for this model (Fig. 1). The final variables of this model are shown in Table 2. According to the protective role of female sex and risk role of rural and nomadic residency and their importance, the association GA with predicted probabilities was reported divided by female sex (Fig. 2) and residency (Fig. 3). Accordingly, nomadic residency showed higher probabilities of death in lower GAs, while it showed lower probabilities of death in higher GAs (Fig. 3).

Table 2.

Mixed effects logistic regression for prediction of neonatal death

Variables	Adjusted OR (95% CI)	P value	Crude OR (95% CI)	Adjusted association	Crude association
Maternal hypertension	1.199 (1.036–1.387)	0.015	1.272 (1.121–1.443)	Risk	Risk
Gestational diabetes mellitus	0.837 (0.737–0.949)	0.006	0.693 (0.619–0.776)	Protective	Protective
Eclampsia / preeclampsia	1.103 (0.996–1.221)	0.060	1.445 (1.324–1.576)	Risk	Risk
Addiction	1.565 (1.147–2.135)	0.005	1.543 (1.183–2.013)	Risk	Risk
Premature rupture of membrane	0.836 (0.734–0.952)	0.007	1.659 (1.483–1.857)	Protective*	Risk
Placental abruption	1.480 (1.276–1.716)	< 0.001	3.012 (2.653–3.420)	Risk	Risk
Meconium staining	1.783 (1.449–12.193)	< 0.001	2.129 (1.791–2.531)	Risk	Risk
Impaired fetal heart rate	1.573 (1.375-1.800)	< 0.001	2.013 (1.792–2.292)	Risk	Risk
Previous neonatal death	1.240 (1.076–1.428)	0.003	1.988 (1.768–2.236)	Risk	Risk
Intrauterine growth restriction	1.824 (1.616–2.058)	< 0.001	2.363 (2.135–2.617)	Risk	Risk
Corticosteroids**	0.777 (0.728–0.828)	< 0.001	0.719 (0.681–0.759)	Protective	Protective
Residency status
Urban	1	Reference	1
Rural	1.260 (1.168–1.360)	< 0.001	1.306 (1.227–1.390)	Risk	Risk
Nomadic	2.870 (1.485–5.548)	0.002	2.967 (1.787–4.923)	Risk	Risk
Female sex	0.909 (0.854–0.968)	0.003	0.933 (0.883–0.985)	Protective	Protective
GA	0.664 (0.658–0.669)	< 0.001	0.681 (0.676–0.686)	Protective	Protective
Delivery type (cesarean)	1.563 (1.451–1.684)	< 0.001	0.803 (0.757–0.849)	Risk*	Protective
Mother age	0.994 (0.989–0.999)	0.025	0.991 (0.986–0.995)	Protective	Protective
Malformation	6.281 (5.484–7.193)	< 0.001	6.125 (5.468–3.267)	Risk	Risk
Hospital type (public)	1.844 (1.618–2.101)	< 0.001	2.911 (2.594–3.267)	Risk	Risk
Out-of-hospital delivery**	1.390 (1.137-1.700)	0.001	1.512 (1.288–1.776)	Risk	Risk
Constant	3710.6 (2465.1-5585.3)	< 0.001
Random part
Province
Variance(constant)	0.134 (0.077–0.233)

* Changing effect direction after adjustment

** Modifiable

Fig. 1.

Post-estimation ROC curve for model performance according to the model presented in Table 2

Fig. 2.

Sigmoid prediction of the outcome by GA divided by sex

Fig. 3.

Sigmoid prediction of the outcome by GA divided by residency

Discussion

The aim of the present study was to identify the potential risk factors of neonatal death among the preterm birth cases. Data of a national registry was used to reach this aim. From the investigated about 50 variables (including subgroups of categorical variables), 20 variables (including subgroups of categorical variables) remained for mixed model logistic regression after the stepwise process. From them, two shifting directions of association was achieved; (1) premature rupture of membrane for more than 18 h was changed from risk factor to protecting factor in multivariable modeling, (2) cesarean type of delivery was changed from protecting factor to risk factor in multivariable modeling.

The protecting factors in multivariable modeling were gestational diabetes mellitus, premature rupture of membrane for more than 18 h, corticosteroid, female sex, GA and mother age. From them, GA was the most effective protecting factor as one week increase in GA was along with 0.664 folds decrease in the odds of death as expected. Among the variables of multivariable modeling, only prenatal corticosteroid use was completely modifiable. In this regard, although choosing hospital type was modifiable, it isnot possible to encourage mothers to choose private instead of public hospital. However, it could be interpreted that this variable was indicator of the quality of healthcare service. Hence, it seemed that enhancing the healthcare facilities (like private hospitals) was associated with reduction of mortality. Regarding out-of-hospital delivery, prevention of this condition might be helpful to reduce mortality.

Hug et al. (2021) analyzed global and regional datasets and confirmed that neonatal mortality is strongly influenced by maternal education and healthcare accessibility, consistent with our findings [1]. Similarly, Blencowe et al. (2023) reported that preterm birth rates remain high in developing regions, highlighting persistent disparities in neonatal outcomes despite global progress [14]. Congenital anomalies remain among the leading causes of neonatal deaths globally, as confirmed by Liu et al. (2019), consistent with our finding that malformation had the largest effect size [15]. Kozuki et al. (2015) in Nepal studied on 685 cases of preterm birth. They found female sex as a protecting factor similar to our study [16]. In a 33-year data study in Brazil on 19.625 cases of preterm birth, it was found that low income and lower birth weights were associated neonatal mortality. They could not find a significant association for death [17].

Although there were a lot of studies regarding the epidemiology of neonatal mortality and preterm birth, there was not a census-based study investigating the risk factors. In addition, the present study investigated many maternal conditions. Using mixed multivariable modeling was another strength of the study. The most important limitation of this study was secondary design that cause variation in the quality of data recording. However, large study size along with considering the random effects of the provinces helped us to overcome this limitation. Another limitation was the retrospective design could influence generalizability of the results outside the national and regional context. Prospective follow-up is suggested for future studies. Also, missing data was another important limitation.

Conclusion

The present study showed that many maternal and neonatal related variables were potential risk and protecting factors of neonatal death among the preterm birth neonates. The main predictor of the outcome was GA, and using corticosteroid before birth was a modifiable protecting factor. Clinicians should focus on the modifiable factors. Prevention of out-of-hospital delivery and increasing healthcare facilities should be studied in future. To improve the prediction model of neonatal mortality, prenatal care should be increased, such as timely use of effective drugs such as corticosteroids, preventing births in unequipped settings such as home, and increasing access to health centers, especially in rural areas. It is also important to strengthen the quality of hospital services and better educate mothers about care and risks during pregnancy. Identifying and caring for mothers with specific diseases or infants with growth problems or abnormalities will help reduce neonatal deaths. Improving data collection and recording methods will also make the prediction model more accurate and useful. All of these will help reduce neonatal deaths.

Abbreviations

AUC

Area Under the Curve

ART

Assisted Reproductive Techniques

CI

Confidence Interval

GA

Gestational Age

IUGR

Intrauterine Growth Restriction

IMaN

Iranian Maternal and Neonatal System

NICU

Neonatal Intensive Care Unit

OR

Odds Ratio

PROM

Premature Rupture of Membrane

SD

Standard Deviation

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

VIF

Variance Inflation Factor

Authors’ contributions

MJ: data collection, draftingSAYA: formal analysis and interpretation, draftingNK: methodology, conceptualization, critical revisionMH: methodology, conceptualization, critical revisionAH: methodology, conceptualization, critical revisionAT: methodology, conceptualization, critical revision.All the authors approved the final version of the manuscript.

Funding

Iran University of Medical Sciences with grant number [15604].

Data availability

The datasets generated and/or analysed during the current study are not publicly available due to security reasons as they are national level data; but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the ethics committee of Iran University of Medical Sciences with registration number IR.IUMS.REC.1398.834. All the ethical issues such as secrecy of the data and Declaration of Helsinki were observed. The participants gave informed consent for registration in Iranian Maternal and Neonatal System and any secondary research use of the data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.