Predictors of low birth weight deliveries at a primary health care facility: Suhum Government Hospital, Ghana – a retrospective cross-sectional study

Abstract

Introduction

Low birth weight (LBW), defined as weight below 2500 g, remains a major contributor to neonatal morbidity and mortality. Ghana’s LBW prevalence stands at 13.5%, but determinants at the primary health care level remain underexplored. This study examines maternal, health-related and behavioural factors influencing LBW among women delivering at Suhum Government Hospital, Eastern Region, Ghana.

Methods

A retrospective cross-sectional study analysed 413 delivery records from 2021 to 2024. Maternal sociodemographic, health, antenatal care (ANC), immunisation and behavioural data were extracted from antenatal and delivery records. Bivariate analysis used χ² or Fisher’s exact tests, as appropriate. Multivariable logistic regression identified factors independently associated with LBW, with aORs and 95% CIs reported.

Results

Late ANC initiation was associated with higher odds of LBW compared with early initiation (aOR 2.78, 95% CI 1.28 to 6.17; p=0.009). Receipt of fewer than two tetanus–diphtheria (TD) vaccine doses was also associated with higher odds of LBW (aOR 4.07, 95% CI 1.79 to 9.26; p<0.001). Haemoglobin ≥11 g/dL at 36 weeks was associated with lower odds of LBW (aOR 0.097, 95% CI 0.011 to 0.903; p=0.040), as was normal body mass index (BMI) compared with obesity (aOR 0.386, 95% CI 0.156 to 0.953; p=0.039). Below-high-school education, Muslim religion and HIV-positive status were also associated with lower adjusted odds. However, these unexpected associations should be interpreted cautiously given the small number of LBW events and potential residual confounding and model instability.

Conclusion

In this single-facility study, late ANC initiation and fewer than two TD doses were associated with higher odds of LBW, while adequate haemoglobin at 36 weeks and normal BMI were associated with lower odds. Strengthening timely ANC, maternal anaemia monitoring and management and completion of recommended maternal immunisation may be relevant to similar primary healthcare settings; larger multicentre studies are needed to assess generalisability.

WHAT IS ALREADY KNOWN ON THIS TOPIC?

WHAT THIS STUDY ADDS?

• This study identifies delayed antenatal care initiation, inadequate tetanus–diphtheria vaccination, maternal anaemia at 36 weeks of gestation and maternal obesity as independent predictors of low birth weight among women delivering at a primary healthcare facility in Ghana.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY?

• These findings highlight the need to strengthen early antenatal care, improve maternal anaemia prevention and management, promote healthy maternal weight and ensure completion of recommended tetanus–diphtheria vaccination during pregnancy to reduce the burden of low birth weight in Ghana and similar settings.

Background

Low birth weight (LBW), defined as a birth weight less than 2.5 kg, remains one of the most critical indicators of neonatal health and a major contributor to infant morbidity and mortality worldwide.^1,2

Globally, approximately 14.7% of newborns, representing nearly 19.8 million infants born annually with LBW, with over 80% of these cases occurring in low-income and middle-income countries, particularly in South Asia and Sub-Saharan Africa.^3,4 The consequences of LBW extend beyond the neonatal period, predisposing children to malnutrition, impaired cognitive development and an increased risk of non-communicable diseases such as diabetes and cardiovascular disorders later in life.^5,6

In Sub-Saharan Africa, persistent socioeconomic disparities, poor maternal nutrition, infections such as malaria and HIV and limited access to quality antenatal care (ANC) continue to drive the high prevalence of LBW.^7,8 Ghana mirrors these regional challenges, with an estimated 13.5% of infants born with LBW despite nationwide interventions such as the Free Maternal Healthcare Policy under the National Health Insurance Scheme.⁹ However, significant subnational variation persists across districts and regions, reflecting differences in healthcare access, maternal nutrition and service utilisation. Recent Ghanaian studies (2022–2025) have identified maternal anaemia, late ANC initiation and obesity as key determinants of LBW, while highlighting that contextual and facility-level factors influence birth outcomes.^8,10–12

The Eastern Region, home to Suhum Municipality, contributes substantially to Ghana’s LBW burden, reporting over 10 000 cases annually.¹³ Suhum is a semiurban district where healthcare access remains uneven, and adherence to maternal health recommendations, such as early ANC attendance, completion of tetanus–diphtheria (TD) vaccination and adequate iron folic acid (IFA) intake, remains suboptimal.^14,15 Despite being representative of many mid-level districts in Ghana, Suhum has received little research attention, and empirical data on LBW determinants within its primary healthcare facilities are scarce. Understanding the local drivers of LBW in such settings is crucial for tailoring interventions that address service gaps and behavioural risk factors at the district level.

While national surveys have identified broad demographic correlates such as maternal age, education and parity, they often lack detailed clinical and behavioural data, obscuring the specific pathways through which maternal health status and service utilisation influence birth outcomes.^9,16 Key maternal factors, including anaemia, HIV status, timing of ANC visits, malaria prophylaxis adherence and alcohol use, remain understudied within primary healthcare facilities in Ghana, particularly in semiurban settings like Suhum.¹⁷

Therefore, this study aimed to identify maternal, health-related and behavioural factors associated with LBW among deliveries at Suhum Government Hospital in Ghana’s Eastern Region. Specifically, it examined maternal demographic characteristics, health status, healthcare utilisation and behavioural factors in relation to neonatal birth weight, providing facility-level evidence that may inform locally relevant maternal and newborn health interventions.

Methods

Study design

A retrospective cross-sectional study was an appropriate study design to determine the relationship between maternal risk factors and LBW among neonates delivered at Suhum Government Hospital, a primary healthcare facility in the Eastern Region of Ghana. This design was appropriate for exploring health outcomes by identifying predictors of LBW using existing maternal health record information. Additionally, this design was chosen due to the availability of detailed records and the feasibility of gathering data on a relatively large sample without the time and ethical constraints of recruiting participants prospectively. Furthermore, while the retrospective approach meant the study relied on the accuracy of existing records, it allowed for the inclusion of all eligible cases over 4 years, thereby enhancing the study’s statistical power and relevance.

Study area

This study was conducted at Suhum Government Hospital, a primary healthcare referral facility located in the Suhum Municipality of the Eastern Region, Ghana. The hospital offers essential maternal and child health services and maintains comprehensive records of antenatal and delivery care, making it an ideal location for retrospective research. Suhum Municipality covers approximately 450 square kilometres and comprises 166 communities, with a projected population of 133 907 in 2025. The municipality is semiurban, with over 70% of residents engaged in agriculture, which can influence income levels and access to healthcare, possibly affecting maternal nutrition and ANC attendance. Suhum is supported by forty health facilities, including Community-Based Health Planning and Services (CHPS) compounds and private clinics, with CHPS initiatives active in 27 out of 32 zones. However, it still faces gaps in the health system, such as insufficient skilled personnel and limited infrastructure. This setting was ideal for the study because the hospital maintains comprehensive antenatal and delivery records (ensuring data availability), and the population served includes both rural and semiurban communities, which may yield diverse risk factors for LBW.¹⁸

Study population and sample

The study analysed a total of 413 eligible delivery records from Suhum Government Hospital covering the years 2021–2024, representing approximately 82% of total deliveries recorded during this period. This sample was considered adequately representative of the hospital’s delivery population, as it captured a diverse range of births from both Suhum Municipality and its surrounding subdistricts, reflecting the sociodemographic composition of semiurban settings in Ghana. The inclusion of 4 consecutive years of data enhanced the likelihood of capturing an adequate number of LBW births for thorough statistical analysis. While some records were excluded due to missing or incomplete data, this exclusion was essential to maintain data quality and ensure the accuracy of the analysis, even if it introduced minor selection bias. Overall, the selected sample provided a realistic and representative picture of maternal and neonatal outcomes in the study area.

Operational definitions and measurement of variables

All variables were operationally defined and measured based on standardised clinical and public health indicators in line with the Ghana Health Service (GHS): Key Indicators Report and WHO guidelines, as shown in the supporting information (online supplemental SI table 1).

Data collection and quality control

A structured data extraction form was developed in Microsoft Excel with predefined fields corresponding to the study variables. Trained data collectors, supervised by the principal investigator, systematically reviewed routine antenatal and delivery registers; the data were accessed on 26 February 2025, and relevant information was extracted into the form. The extracted data were verified for consistency and completeness, and a double-entry procedure was employed to ensure accuracy. Two independent data clerks entered the same records into separate Excel sheets, and discrepancies were resolved by cross-checking with the original registers. Each delivery was assigned a unique identification (ID) number to prevent duplication, and records appearing in both registers were merged under a single ID. The dataset was subsequently cleaned to correct entry errors and remove implausible values before analysis.

Additional quality assurance measures were implemented to enhance data reliability. Inter-rater consistency was maintained through standardised training and periodic cross-checks among data clerks, while approximately 10% of records were reaudited by the principal investigator to verify accuracy. This proportion was selected in line with established quality control standards in retrospective health record studies, where a 10% recheck is typically sufficient to detect systematic entry errors and ensure inter-rater reliability.¹⁹ Missing or incomplete data were reviewed against source documents, and unresolved cases were excluded. Records with missing information on key variables such as birth weight, maternal age, ANC attendance and haemoglobin level were excluded using listwise deletion. This approach was chosen because the proportion of missing data was minimal (<5%) and unlikely to bias the results; therefore, no imputation procedures were applied. These procedures minimised errors, enhanced data accuracy and ensured the validity and reliability of the dataset used for statistical analysis. The final cleaned dataset was exported to SPSS V.28 for appropriate coding, labelling and statistical analysis.

Data analysis

Data were analysed using descriptive, bivariate and multivariable statistical methods. Descriptive statistics summarised the study population, with continuous variables presented as means and SDs and categorical variables as frequencies and percentages. Bivariate analyses examined associations between maternal characteristics and LBW using the χ² test or Fisher’s exact test, as appropriate (table 1). Before multivariable modelling, multicollinearity among candidate predictor variables was assessed using tolerance and variance inflation factor (VIF) statistics. Tolerance values >0.20 and VIF values <5 were considered indicative of the absence of problematic multicollinearity (online supplemental SI table 2). All candidate predictors that met these criteria were retained for multivariable modelling; no additional screening based on bivariate p values was performed. Binary logistic regression was performed in two stages. First, all 20 candidate predictors that passed the multicollinearity assessment were entered simultaneously into the initial multivariable model using the enter (block entry) method (table 2): maternal age, marital status, occupation, educational level, place of residence, religion, parity, BMI, HIV status, hepatitis B status, previous maternal complications, hypertension, haemoglobin at registration, haemoglobin at 36 weeks, ANC timing, number of ANC visits, sulphadoxine–pyrimethamine doses, TD doses, IFA intake and alcohol use. Second, backward elimination was applied sequentially to remove variables with p≥0.05 while monitoring potential confounding and model stability. The variables removed during this process were occupation, place of residence, parity, hepatitis B status, previous maternal complications, hypertension, haemoglobin at registration, number of ANC visits, sulphadoxine–pyrimethamine doses, IFA intake and alcohol use. The remaining variables maternal age, marital status, educational level, religion, BMI, HIV status, haemoglobin at 36 weeks, ANC timing and TD doses, constituted the final parsimonious model (table 3). Adjusted ORs (aORs) with 95% CIs were reported, and p<0.05 was considered statistically significant. Model fit was assessed for both the initial and final multivariable models using the Hosmer-Lemeshow goodness-of-fit test. A statistically significant result (p<0.05) was interpreted as evidence of lack of fit, whereas a non-significant result indicated no statistical evidence of lack of fit. Cox and Snell R² and Nagelkerke R² statistics were also used to describe model explanatory power. The final parsimonious model was retained as the primary model for inference. Given the 42 LBW events, the potential for sparse-data effects and model instability was considered when interpreting the estimates.

Table 1. Distribution of maternal sociodemographic, health and behavioural characteristics according to birth weight status (<2.5 kg vs ≥2.5 kg).

	Birth weight
Attribute	Birth weight		Birth weight		P value
	<2.5 kg		≥ 2.5 kg
	(N=42) n (%)		(N=371) n (%)
Maternal age					0.002
≤19	10	23.8	28	7.6
20–35	24	57.1	281	75.7
≥36	8	19.1	62	16.7
Mean±SD		28.7±	6.8
Marital status					0.101
Married	15	35.7	182	49.1
Single	27	64.3	189	50.9
Educational level					0.006
Below high school	5	11.9	121	32.6
High school or higher	37	88.1	250	67.4
Occupation					0.027
Employed	27	64.3	294	79.2
Unemployed	15	35.7	77	20.8
Place of residence					0.747
Urban	14	33.3	133	35.8
Rural	28	66.7	238	64.2
Religion					0.006
Christian	30	71.4	323	87.1
Muslim	12	28.6	48	12.9
Parity					0.931
Primiparous	15	35.7	130	35.0
Multiparous	27	64.3	241	65.0
BMI					0.064
Normal	13	31	61	16.4
Overweight	10	23.8	117	31.6
Obese	19	45.2	193	52.0
Mean±SD 30.2±5.4
HIV					0.007
Negative	38	90.5	363	97.8
Positive	4	9.5	8	2.2
Hepatitis B					0.608
Negative	37	88.1	336	90.6
Positive	5	11.9	35	9.4
Previous maternal complication					0.114
Yes	5	11.9	21	5.7
No	37	88.1	350	94.3
Hb at registration					0.189
< 11 g/dL	36	85.7	285	76.8
≥ 11 g/dL	6	14.3	86	23.2
Mean±SD 10.1±1.2
Hb at 36 weeks					0.011
< 11 g/dL	41	97.6	306	82.5
≥ 11 g/dL	1	2.4	65	17.5
Mean±SD 10.2±1.1
Hypertension					0.226
Yes	8	19.0	46	12.4
No	34	81.0	325	87.6
Timing of ANC					< 0.001
Early timing	21	50.0	285	76.8
Late timing	21	50.0	86	23.2
Number of ANC visits					0.228
<8	38	90.5	309	83.3
≥8	4	9.5	62	16.7
Mean±SD 5.4±2.0
SP doses intake					0.11
<3 doses	17	40.5	106	28.6
≥3 doses	25	59.5	265	71.4
TD doses administered					0.053
< 2 doses	15	35.7	191	51.5
≥2 doses	27	64.3	180	48.5
IFA tablets intake					0.113
≥ 180 tablets	36	85.7	277	74.7
<180 tablets	6	14.3	94	25.3
Alcohol use
No	34	81.0	344	92.7	0.009
Yes	8	19.0	27	7.3

Percentages are column percentages calculated using the total number of <2.5 kg (n=42) and ≥2.5 kg (n=371) as the denominators. P values were obtained using Pearson’s χ² test or Fisher’s exact test, as appropriate.

ANC, antenatal care; BMI, body mass index; Hb, haemoglobin; IFA, iron folic acid; SP, sulphadoxine pyramethamine; TD, tetanus–diphtheria.

Table 2. Initial multivariable logistic regression analyses for predictors of low birth weight.

		95% CI
Predictor	aOR	(lower to upper)	P value †
Maternal age
20–35 (Ref)	1
≤19	0.484	0.136 to 1.721	0.262
≥ 36	1.061	0.303 to 3.716	0.926
Marital status
Married (ref)	1
Single	2.703	1.042 to 7.143	0.040
Occupation
Employed (ref)	1
Unemployed	0.954	0.369 to 2.471	0.923
Educational level
High school or higher (ref)	1
Below high school	0.322	0.111 to 0.934	0.037
Place of residence
Urban (ref)	1
Rural	1.655	0.730 to 3.755	0.228
Religion
Christian (ref)	1
Muslim	0.363	0.147 to 0.895	0.028
Parity
Multiparous (ref)	1
Primiparous	0.795	0.330 to 1.913	0.608
BMI ‡
Obese (ref)	1
Normal	0.356	0.136 to 0.930	0.035
Overweight	1.278	0.521 to 3.137	0.592
HIV status
Negative (ref)	1
Positive	0.138	0.025 to 0.766	0.024
Hepatitis B
Negative (ref)	1
Positive	0.729	0.223 to 2.381	0.601
Previous maternal complications
No (ref)	1
Yes	1.142	0.291 to 4.485	0.849
Hypertension
No (ref)	1
Yes	1.653	0.593 to 4.608	0.337
Hb at registration
≥11 g/dL (ref)	1
<11 g/dL	1.811	0.624 to 5.255	0.274
Hb at 36 weeks
<11 g/dL (ref)	1
≥11 g/dL	0.085	0.008 to 0.857	0.037
ANC timing
Early timing (ref)	1
Late timing	2.809	1.116 to 7.092	0.028
ANC visits
≥8 (ref)	1
<8	1.291	0.297 to 5.623	0.733
SP doses
≥3 doses (ref)	1
<3 doses	0.931	0.379 to 2.291	0.877
TD doses
≥2 doses (ref)	1
<2 doses	4.550	1.850 to 11.110	< 0.001
IFA intake
≥180 tablets (ref)	1
<180 tablets	0.988	0.278 to 3.512	0.986
Alcohol use
No (ref)	1
Yes	0.401	0.119 to 1.351	0.140

†P < 0.05 was statistically significant on multivariable.

‡The dataset did not capture mothers with underweight BMI (<18.5 kg/m²), hence this category was excluded from the analysis.

*Model and goodness-of-fit Statistics: Cox and Snell R² = 0.147; Nagelkerke R² = 0.305. Hosmer and Lemeshow Test: χ² = 18.737; df = 8; p = 0.016.

ANC, antenatal care; aOR, adjusted OR; BMI, body mass index; Hb, haemoglobin; IFA, iron folic acid; SP, sulphadoxine pyramethamine; TD, tetanus–diphtheria.

Table 3. Final parsimonious multivariable logistic regression model of predictors of low birth weight.

		95% CI
Predictor	aOR	(lower to upper)	P value †
Maternal age
20–35 (ref)	1
≤19	1.902	0.729 to 4.962	0.189
≥36	1.848	0.472 to 7.231	0.378
Marital status
Married (ref)	1
Single	0.484	0.209 to 1.117	0.089
Educational level
High school and higher (ref)	1
Below high school	0.351	0.127 to 0.971	0.044
Religion
Christian (ref)	1
Muslim	0.370	0.162 to 0.844	0.018
BMI ‡
Obese (ref)	1
Normal	0.386	0.156 to 0.953	0.039
Overweight	1.115	0.472 to 2.638	0.804
HIV status
Negative (ref)	1
Positive	0.119	0.025 to 0.562	0.007
Hb @ 36 weeks
<11 g/dL (Ref)	1
≥11 g/dL	0.097	0.011 to 0.903	0.040
ANC timing
Early timing (ref)	1
Late timing	2.780	1.280 to 6.170	0.009
Tetanus doses
≥2 doses (ref)	1
<2 doses	4.070	1.790 to 9.260	< 0.001

†P < 0.05 was statistically significant on multivariable.

‡The dataset did not capture mothers with underweight BMI (<18.5 kg/m²), hence this category was excluded from the analysis.

^*Final parsimonious multivariable logistic regression model showing adjusted aORs and 95% CIs for factors associated with low birth weight (LBW). The model demonstrated acceptable goodness-of-fit (Hosmer-Lemeshow χ² = 15.006, df = 8, p = 0.059). Cox and Snell R² = 0.131; Nagelkerke R² = 0.272.

.ANC, antenatal care; BMI, body mass index; Hb, haemoglobin.

Ethical consideration

The GHS Directorate and the Suhum Municipal Health Directorate also granted formal written permission to access and use secondary data. Authorisation was further obtained from the management of Suhum Government Hospital to review antenatal and delivery registers. As this study involved retrospective analysis of anonymised secondary data, it posed no risk to participants. Therefore, informed consent from individuals was not required. All personal identifiers were excluded before data extraction, and records were anonymised using unique codes to ensure confidentiality. Access to data was restricted to the principal investigator, and all electronic files were password-protected in accordance with GHS data protection and ethical research guidelines.

Results

Bivariate (χ²) analysis of associations between maternal sociodemographic, health and behavioural factors and LBW outcomes

Table 1 presents the distribution of maternal sociodemographic, health-related and behavioural characteristics according to birth weight status. Compared with mothers of non-LBW infants, mothers of LBW infants were more frequently adolescents aged ≤19 years (23.8% vs 7.6%), single (64.3% vs 50.9%), unemployed (35.7% vs 20.8%), Muslim (28.6% vs 12.9%), HIV-positive (9.5% vs 2.2%), reported alcohol use during pregnancy (19.0% vs 7.3%), had haemoglobin concentrations <11 g/dL at 36 weeks’ gestation (97.6% vs 82.5%) and initiated ANC late (50.0% vs 23.2%). Significant differences between the LBW and non-LBW groups were observed for maternal age, educational level, occupation, religion, HIV status, alcohol use during pregnancy, haemoglobin concentration at 36 weeks and timing of ANC initiation (all p<0.05). No statistically significant differences were observed for marital status, place of residence, parity, body mass index, hepatitis B status, previous maternal complications, haemoglobin concentration at registration, hypertension, number of ANC visits, intermittent preventive treatment doses, tetanus–diphtheria vaccination status or iron folic acid tablet intake (all p>0.05).

Multivariable logistic regression analysis for predictors of LBW following the enter (block entry) method, showing variables with significant adjusted ORs

Initial multivariable logistic regression model of predictors of LBW

In the initial multivariable model (table 2), fewer than two TD doses were associated with higher odds of LBW (aOR=4.550, 95% CI 1.850 to 11.110; p<0.001), as was late ANC initiation (aOR=2.809, 95% CI 1.116 to 7.092; p=0.028). Normal BMI was associated with lower odds of LBW compared with obesity (aOR=0.356, 95% CI 0.136 to 0.930; p=0.035), as was haemoglobin ≥11 g/dL at 36 weeks (aOR=0.085, 95% CI 0.008 to 0.857; p=0.037). Single status, below-high-school education, Muslim religion and HIV-positive status were also statistically significant in the initial model. The initial model showed evidence of lack of fit according to the Hosmer–Lemeshow test (χ²=18.737, df=8, p=0.016).

Following backward elimination, occupation, place of residence, parity, hepatitis B status, previous maternal complications, hypertension, haemoglobin at registration, number of ANC visits, sulphadoxine–pyrimethamine doses, IFA intake and alcohol use were removed because they did not meet the elimination criterion (p≥0.05). The remaining variables formed the final parsimonious model (table 3).

In the final parsimonious model (table 3), fewer than two TD doses were associated with higher odds of LBW (aOR=4.070, 95% CI 1.790 to 9.260; p<0.001), as was late ANC initiation (aOR=2.780, 95% CI 1.280 to 6.170; p=0.009). Haemoglobin ≥11 g/dL at 36 weeks was associated with lower odds of LBW compared with haemoglobin <11 g/dL (aOR=0.097, 95% CI 0.011 to 0.903; p=0.040), and normal BMI was associated with lower odds compared with obesity (aOR=0.386, 95% CI 0.156 to 0.953; p=0.039). Below-high-school education, Muslim religion and HIV-positive status were also associated with lower adjusted odds of LBW, but these findings were based on relatively small subgroups and should be interpreted cautiously. Maternal age and marital status were not statistically significant in the final model.

The final parsimonious model showed no statistical evidence of lack of fit according to the Hosmer–Lemeshow test (χ²=15.006, df=8, p=0.059). The model explained 13.1% and 27.2% of the variation in LBW based on the Cox and Snell R² and Nagelkerke R² statistics, respectively. The final parsimonious model was, therefore, retained as the primary model for inference.

Discussion

This study found that late ANC initiation and receipt of fewer than two TD doses were associated with higher odds of LBW, while haemoglobin ≥11 g/dL at 36 weeks and normal BMI were associated with lower adjusted odds. These findings provide facility-level evidence on potentially modifiable maternal and healthcare-related factors associated with LBW among women delivering at Suhum Government Hospital. Given the retrospective design, the findings should be interpreted as associations rather than causal effects. Late ANC initiation was associated with higher odds of LBW, consistent with evidence from Ghana and other low and middle-income settings.^8,20,21 Timely ANC provides opportunities for early identification and management of maternal conditions and delivery of preventive interventions. However, the present study cannot establish whether delayed ANC itself caused LBW because women who initiate ANC late may differ from those who attend early in ways that were not fully captured in the records.

Receipt of fewer than two TD doses was also associated with higher odds of LBW. This finding is consistent with previous evidence linking maternal immunisation and engagement with antenatal services to pregnancy outcomes.^4,8 However, TD vaccination may also be a marker of broader engagement with maternal healthcare rather than an independent causal determinant. The observational design, therefore, limits attribution of the association to vaccination alone.

Haemoglobin ≥11 g/dL at 36 weeks was associated with lower adjusted odds of LBW, consistent with evidence linking maternal anaemia with adverse birth outcomes.²²

Normal BMI was associated with lower adjusted odds of LBW compared with obesity. This association should be interpreted cautiously because BMI was obtained from routine records and may not reflect prepregnancy nutritional status. In addition, unmeasured maternal and clinical factors may have contributed to the observed association.

The lower adjusted odds observed for HIV-positive status, Muslim religion and below-high-school education were unexpected and differed from the crude pattern for HIV status, where HIV-positive mothers had a higher proportion of LBW in table 1. These associations should not be interpreted as evidence that HIV infection, religion or lower educational attainment is protective. The reversal observed for HIV status may reflect residual confounding, selection effects, sparse data or model instability. Only 12 mothers were HIV-positive, of whom four had LBW, making the estimate particularly sensitive to small changes in the data. Similarly, the associations involving religion and education may reflect unmeasured socioeconomic, behavioural, healthcare-access or other contextual factors. Larger studies are needed to determine whether these associations are reproducible.

The significant Hosmer-Lemeshow test for the initial model (χ²=18.737, df=8, p=0.016) indicates evidence of lack of fit, whereas the final parsimonious model showed no statistical evidence of lack of fit (χ²=15.006, df=8, p=0.059). The improved fit after backward elimination supports use of the final model for primary inference, but the non-significant test does not establish that the model is perfectly calibrated. Given the 42 LBW events and the number of candidate predictors considered, sparse-data effects and overfitting remain possible, and the adjusted estimates should, therefore, be interpreted cautiously.

Overall, the findings support attention to timely ANC, maternal anaemia monitoring and management and completion of recommended maternal immunisation within the study setting. These priorities are consistent with existing maternal health guidance,^15,23 but the present single-facility study does not provide sufficient evidence to recommend national-scale policy changes. Larger multicentre studies are needed to assess whether the observed associations are consistent across different populations and healthcare settings.

Limitations

This study has several limitations. First, it relied on retrospectively collected secondary data from routinely maintained facility records; therefore, data quality and completeness could not be fully controlled, and some potentially important variables were incompletely documented. The retrospective design also limits the ability to establish causal relationships between maternal factors and LBW. In addition, dietary intake, psychosocial stress, environmental exposures and other potentially relevant factors were not assessed, leaving the possibility of residual confounding. Selection bias cannot also be excluded because the study was conducted in a single district hospital. Consequently, the findings should not be considered nationally representative of Ghana.

Second, the study included only 42 LBW events, with small numbers in some predictor categories, including HIV-positive mothers. Sparse data may, therefore, have contributed to imprecise or unstable estimates and increased the risk of overfitting given the number of candidate predictors considered.^24,25 Although multicollinearity was assessed and backward elimination was used to derive a more parsimonious final model, these approaches cannot completely eliminate the possibility of model instability or biased estimates. The significant Hosmer–Lemeshow test for the initial model further indicates that model fit was sensitive to model specification, although the final parsimonious model showed no statistical evidence of lack of fit. The unexpected adjusted associations observed for HIV status, religion and educational attainment should, therefore, be interpreted cautiously and not as causal effects.

Also, it was not possible to distinguish whether LBW resulted from preterm birth or small-for-gestational-age birth because gestational age and the information required to classify birth weight relative to gestational age were insufficient. Larger multicentre prospective studies with adequate numbers of LBW events and more comprehensive measurement of maternal, socioeconomic, clinical and environmental factors are needed to validate these findings.

In addition, the study period (2021–2024) included the COVID-19 pandemic and its aftermath; however, pandemic-specific exposures were not available in the dataset and could not be assessed. Finally, waist circumference data required to calculate BRI were not available. Consequently, BRI could not be retrospectively calculated or compared with BMI in this dataset.

Policy implications

This study provides facility-level evidence on potentially modifiable factors associated with LBW among women delivering at Suhum Government Hospital. The findings suggest that timely ANC initiation, routine haemoglobin monitoring and management of maternal anaemia and completion of recommended TD vaccination may be relevant priorities within the study setting. Community-based health promotion and counselling could support timely ANC attendance, maternal nutrition, anaemia prevention and adherence to recommended maternal immunisation. These approaches may also be relevant to similar primary healthcare facilities serving semiurban and rural populations; however, their applicability to the wider Ghanaian population should be assessed through larger multicentre studies. The findings may inform local service planning within the broader Universal Health Coverage framework, but should not be interpreted as nationally representative evidence.

Conclusions

In this single-facility study, late ANC initiation and receipt of fewer than two TD doses were associated with higher odds of LBW, while haemoglobin ≥11 g/dL at 36 weeks and normal BMI were associated with lower adjusted odds. The unexpected associations observed for HIV status, religion and educational attainment should be interpreted cautiously given the small number of LBW events and potential residual confounding, sparse data and model instability. The findings support attention to timely ANC, maternal anaemia monitoring and management and completion of recommended maternal immunisation within the study setting and potentially similar primary healthcare contexts. However, given the single-centre design and limited number of LBW events, the findings cannot be assumed to represent the wider Ghanaian population. Larger multicentre and longitudinal studies are needed to validate these associations, clarify causal pathways and distinguish preterm birth from small-for-gestational-age birth. Future studies should assess BRI alongside BMI for LBW prediction.

Data availability statement

Data are available upon reasonable request.