Predictors of Infection after Cesarean Delivery

Abstract

Objective

Postoperative infections remain a significant complication of cesarean delivery (CD), with rates ranging from 3 to 30%. These infections increase healthcare costs, prolong recovery, and negatively impact maternal outcomes. Identifying risk factors for infection might help guide preventative strategies to mitigate this burden. The primary aim of this study was to develop and validate a predictive model for infection after CD using preoperative and perioperative characteristics.

Study Design

This study was a secondary analysis of a multicenter randomized trial comparing tranexamic acid versus placebo to prevent postpartum hemorrhage in individuals undergoing either scheduled or unscheduled CD. The primary outcome of this analysis was a composite of surgical site infection, endometritis, or pelvic abscess diagnosed within 6 weeks postpartum. Univariable and multivariable bootstrapped logistic regression models with stepwise selection were used to identify predictors of infection. Model performance was evaluated using the receiver operator curve area under the curve (AUC).

Results

Of the 10,995 participants, 287 (2.6%) developed an infection. Significant predictors included tobacco use (OR: 1.67, 95% CI: 1.21–2.31), BMI at delivery ≥ 30 kg/m² (OR: 1.41, 95% CI: 1.04–1.91), labor before cesarean (OR: 1.71, 95% CI: 1.33–2.18), longer surgical duration (OR: 1.01 per minute, 95% CI: 1.01–1.02), uterine incision extension (OR: 1.54, 95% CI: 1.02–2.34), and the use of uterotonics other than oxytocin (OR: 1.48, 95% CI: 1.09–2.02). The predictive model demonstrated modest discrimination with an AUC of 0.64 (95% CI: 0.61–0.68).

Conclusion

Multiple modifiable and nonmodifiable factors influence infection after CD. This predictive model offers a framework for assessing individualized risk, though its modest performance indicates that further refinement is necessary before it can be confidently applied in clinical decision-making. Future research should aim to enhance predictive accuracy and explore whether risk stratification meaningfully informs prevention or patient counseling strategies.

Introduction

Cesarean delivery (CD) is one of the most common surgical procedures performed in the United States, accounting for a third of all births in 2021.¹ While CD can be lifesaving for both the mother and the neonate, it is not without complications. Among these, infection remains the most prevalent postoperative complication, significantly impacting maternal health and healthcare systems.^2–4

The prevalence of infection after CD varies widely, with estimates ranging from 3 to 20% in the general population and climbing as high as 30% in high-risk groups such as individuals with morbid obesity.^5,6 Such infections can include surgical site infections (SSIs), endometritis, and other serious complications that often require additional medical interventions. These complications lead to increased healthcare costs due to extended hospital stays, readmissions, reoperations, and the need for home healthcare services.^7–9

Efforts to reduce the incidence of infections have focused on implementing evidence-based preventive strategies. Guidelines from the Centers for Disease Control and Prevention (CDC) and recommendations endorsed by the American College of Obstetricians and Gynecologists have emphasized the effectiveness of prophylactic antibiotics administered within 30 to 60 minutes of incision and the use of antiseptic solutions for preoperative skin preparation.^10,11 While these measures have contributed to a decline in infection rates, infections following CD remain a substantial and persistent issue.

In parallel, surgical quality initiatives and enhanced recovery pathways (e.g., ACS NSQIP-supported efforts and ERAS for CD bundles) have emphasized standardized perioperative processes such as weight-based antibiotic dosing, adjunctive azithromycin when indicated, chlorhexidine-alcohol skin preparation, vaginal antiseptic preparation, and appropriate hair removal. While these interventions can reduce infectious morbidity, their implementation varies across centers and over time, and this variability may influence observed infection rates.

Emerging data indicate that multidisciplinary initiatives aimed at addressing preoperative, perioperative, and postoperative factors have the potential to significantly reduce morbidity from infection, achieving more than 60% reduction in SSI rates following the implementation of comprehensive, evidence-based infection control interventions.^12–14 Despite such successes, infection continues to complicate cesarean deliveries. Identifying which individuals are at the highest risk is critical to optimizing preventive strategies. By utilizing preoperative and perioperative characteristics, such a model could allow clinicians to implement tailored prevention strategies, potentially improving outcomes.

Risk estimation may also support shared decision-making for individuals considering CD without labor. Although mode of delivery decisions involve multiple outcomes (e.g., pelvic floor morbidity and neonatal considerations), counseling should acknowledge that CD carries a baseline infectious risk that can increase in the presence of modifiable factors such as smoking and obesity.

This study aims to develop and internally validate a prediction model for infection in a contemporary cohort of individuals undergoing CD. This model leverages a well-characterized dataset to identify key risk factors and improve understanding of how commonly available perioperative data may inform infection risk assessment.

Materials and Methods

Study Design and Data Collection

This is a secondary analysis of a multicenter randomized controlled trial that evaluated the use of tranexamic acid (TXA) versus placebo to prevent postpartum hemorrhage after CD (ClinicalTrials.gov number: NCT03364491).¹⁵ The double-blinded trial, which was conducted at 31 hospitals participating in the Eunice Kennedy Shriver National Institute of Child Health and Human Development Maternal-Fetal Medicine Units Network found that prophylactic administration of TXA during CD did not reduce maternal death or the use of packed red blood cell transfusion.

Patients were eligible for the clinical trial if they were undergoing a scheduled or unscheduled CD and were carrying a singleton or twin gestation. Exclusion criteria were maternal age less than 18 years, transfusion or planned transfusion of any blood products during the current admission, contraindications to TXA (e.g., hypersensitivity to TXA or its ingredients, history of seizure disorder, kidney disease, thromboembolic disease, or medical conditions or use of treatments that convey a high risk of thrombosis), patient refusal of blood products, or a plan to administer prophylactic antifibrinolytics such as open-label TXA or uterotonics other than oxytocin postoperatively. Trained and certified research staff members, blinded to trial treatment assignment, abstracted information from medical records, including demographics, medical history, and outcomes data.

For this secondary analysis, the study population included all participants randomized in the original trial. The primary outcome was a composite measure of infection, defined as the presence of SSI per CDC criteria, pelvic abscess, or endometritis diagnosed within 6 weeks postpartum. Diagnoses of pelvic abscess and endometritis were based on clinical criteria, including fever, fundal tenderness, and leukocytosis, as recorded in the medical record.

The parent trial did not mandate a uniform infection-prevention protocol. Standard perioperative practices (including prophylactic antibiotics prior to skin incision) were performed per local policy at each site. Data collection captured whether prophylactic antibiotics were administered prior to skin incision but did not include details on antibiotic agent(s), dose, timing, adjunctive prophylaxis (such as azithromycin or metronidazole), skin preparation solution, vaginal preparation, or hair-removal method.

Diagnoses of endometritis and pelvic abscess were based on the treating clinician’s diagnosis as recorded in the medical record. Endometritis was typically diagnosed in the setting of postpartum fever with uterine or fundal tenderness and/or leukocytosis, prompting antibiotic therapy. Pelvic abscess was diagnosed when a pelvic collection consistent with abscess was documented by the treating team (often supported by imaging and/or procedural drainage).

In the trial dataset, uterotonics other than oxytocin reflected any administration within 48 hours postpartum and could include commonly used agents such as methylergonovine, carboprost, and misoprostol; individual drug names and doses were not uniformly captured for this secondary analysis.

Predictor Variables

Candidate predictors were identified a priori and included demographic, medical, and operative characteristics recorded during the study. These encompassed preoperative factors such as maternal age, body mass index (BMI), tobacco use, and comorbidities; perioperative factors including labor status before cesarean, type and duration of surgery, estimated blood loss, transfusion, and TXA group assignment; and intraoperative complications such as uterine incision extension or need for additional uterotonics. A complete list of candidate predictors is shown in Table 1.

Table 1.

Univariable analysis of baseline and perioperative characteristics

	No infection (n = 10,708)	Infection (n = 287)a	p-Valueb
Maternal age (y)	30.2 ± 5.8	29.2 ± 5.5	0.014
Race and ethnicity			0.346
Non-Hispanic White	4,210 (39)	119 (41)
Non-Hispanic Black	2,565 (24)	79 (28)
Hispanic	3,205 (30)	73 (25)
Asian	403 (4)	8 (3)
Other, unknown, or more than one race or ethnic group	325 (3)	8 (3)
Married/living with partner	6,615 (62)	148 (52)	<0.001
Government-assisted insurance	6,076 (57)	179 (62)	0.058
TXA group assignment	5,363 (50)	162 (56)	0.033
Cigarette use during pregnancy	1,114 (10)	47 (16)	0.001
Most recent weight (kg)	93.5 ± 23.7	100.5 ± 29.4	<0.001
BMI ≥ 30 kg/m2	7,897 (74)	233 (81)	0.006
Aspirin	1,582 (15)	33 (12)	0.122
Confirmed or suspected chorioamnionitis	392 (4)	16 (6)	0.091
Diabetes			0.029
Pregestational (T1 and T2)	567 (5)	25 (9)
Gestational	1,184 (11)	35 (12)
Hypertensive disorders
Chronic hypertension [only]	758 (7)	25 (9)
Without severe features [GHTN, PE without severe, superimposed PE without severe]	1,240 (12)	42 (15)	0.001
With severe features [PE with severe, superimposed PE with severe]	613 (6)	29 (10)
Antibiotic for chorioamnionitis prior to skin incision	409 (4)	16 (6)	0.128
Antibiotic for prophylaxis prior to skin incision	10,573 (99)	285 (99)	0.589
Placenta previa	193 (2)	3 (1)	0.495
Placental abruption	82 (1)	5 (2)	0.077
Placenta accreta	32 (0.3)	1 (0.4)	1.0
Gestational age at delivery (wk)
<34	280 (3)	14 (5)
34–36	1,340 (13)	42 (15)	0.029
>37	9,088 (85)	230 (80)
Type of labor
Spontaneous/augmented/induced	3,870 (36)	142 (49)	<0.001
None	6,838 (64)	145 (51)
Estimated blood loss greater than 1 L	672/8,980 (8)	35/234 (15)	<0.001
Length of surgery (min)	56 ± 22	65 ± 35	<0.001
Time from skin incision to transfer (min)	66 ± 26	76 ± 39	<0.001
Bowel injury	7 (0.1)	0	1.0
Bladder/ureter injury	32 (0.3)	2 (0.7)	0.222
Uterine extension	517 (5)	29 (10)	<0.001
Treatments and interventions in response to bleeding and related complications			<0.001
Surgical or radiologic interventionsc	440 (4)	24 (8)
Uterotonics (other than oxytocin)d	1,324 (12)	57 (20)
Open-label TXAc	211 (2)	6 (2)	<0.001
Transfusion of any blood productc	416 (4)	27 (9)
			0.885
			<0.001
Hemoglobin change g/dLe	−1.8 ± 1.1	−2.0 ± 1.2	0.002

Abbreviations: GHTN, gestational hypertension; PE, preeclampsia; TXA, tranexamic acid.

Note: Data are shown as n (%) or mean standard ± deviation. Missing data: BMI missing on n = 33, estimated blood loss missing on n = 1,781, surgery length missing on n = 52, time from skin incision to OR transfer missing on n = 17, hemoglobin change missing on n = 80.

^aDefined as a composite of surgical site infection (CDC definition), endometritis, or pelvic abscess.

^bChi-squared or Fisher's exact test (for categorical variables) and the Wilcoxon test (for continuous variables).

^cBy 7 days postpartum.

^dBy 48 hours postpartum.

^eFrom the most recent value < 4 weeks before delivery, to the lowest measurement in the first 48 hours postpartum.

Model Development

Candidate predictors were compared between those with and without the outcome of infection using the Wilcoxon rank sum test for continuous variables and chi-square or Fisher’s exact test for categorical variables, as appropriate. Those significant at < 0.05 in the univariable analysis and nonmissing on at least 95% of the sample were eligible for model selection. A bootstrap resampling method was used to develop and internally validate logistic prediction models.¹⁶ Backward elimination was performed for each of the 1,000 bootstrap samples, and candidate models were developed based on the frequency with which each candidate factor was selected in these models. The final predictor model was based on goodness of fit (Hosmer–Lemeshow test) and predictive ability (area under the curve [AUC]) with the goal of obtaining a parsimonious model without overfitting.

Receiver operator curve analysis was conducted to evaluate the predictive accuracy of the final model. The AUC was calculated to quantify the model’s discrimination. An AUC of 0.5 indicates no predictive ability, while an AUC of 1.0 represents perfect prediction. A nomogram was developed for easier visualization of the risk factors and their contributions to the outcome.

All statistical analyses were conducted using SAS 9.4. No imputation for missing data was performed. p-Value < 0.05 was used to define statistical significance in the final model, and all tests were two-tailed.

Results

The trial enrolled 10,995 participants, of whom 287 (2.6%) experienced the composite infection outcome (Fig. 1). The components of this composite outcome were SSI, occurring in 185/287 participants (64.5%), followed by endometritis in 96/287 (33.5%), and pelvic abscess in 10/287 (3.5%; Fig. 1).

Fig. 1.

Flow diagram.

Predictive Model Development

In the univariable analysis, multiple factors were significantly associated with the risk of infection, including maternal age, marital status, TXA randomization assignment, tobacco use, diabetes, hypertensive disorders, spontaneous or induced labor before CD, longer surgical duration, uterine incision extension, BMI at delivery, gestational age, surgical or radiologic interventions, transfusion of any blood product, and use of uterotonics other than oxytocin within the first 48 hours postpartum (Table 1). To achieve parsimony, variables that were significant in 60% or more of the bootstrapped models were selected for the final model. These included BMI (≥30 kg/m²), labor type, uterotonics (other than oxytocin) in the first 48 hours, tobacco use, uterine incision extension, and length of surgery. After adjustment, the final multivariable logistic regression model identified these variables as predictors of infection (Table 2). The final model had a Hosmer–Lemeshow p-value of 0.997 on 12 degrees of freedom, indicating an adequate fit to the data.

Table 2.

Summary of the final model

Predictor	% of times entered in the bootstrapped models	AUC for each variable addition	ORa	95% CI
Labor spontaneous/augmented/induced	93	0.57 (0.54, 0.60)	1.71	1.33, 2.18
Length of surgery (min)	91	0.62 (0.59, 0.66)	1.01	1.01, 1.02
Uterine incision extension	72	0.63 (0.60, 0.66)	1.54	1.02, 2.34
Tobacco use	68	0.63 (0.60, 0.67)	1.67	1.21, 2.31
Uterotonics (other than oxytocin) in the first 48 h	68	0.64 (0.61, 0.67)	1.48	1.09, 2.02
BMI at delivery ≥ 30 kg/m2	60	0.64 (0.61, 0.68)	1.41	1.04, 1.91

Abbreviations: AUC, area under the curve; BMI, body mass index; CI, confidence interval; OR, odds ratio.

^aPer 1-minute increase.

Model Performance

The final predictive model demonstrated modest discrimination with an ROC AUC of 0.64 (95% CI: 0.61–0.68; Fig. 2). The inclusion of multiple perioperative and demographic factors provided a framework for identifying patients at risk of postpartum infection, although the moderate AUC indicates room for further improvement in model accuracy.

Fig. 2.

Receiver operating characteristic (ROC) curve.

The analysis identified several significant predictors of infection, highlighting the multifactorial risk factors for postoperative infections (Table 2). In multivariable analysis, both maternal and operative factors were independently associated with an increased risk of infection following CD. Tobacco use, elevated maternal BMI, and labor prior to cesarean were among the strongest maternal predictors. Operative characteristics, including longer surgical duration and uterine incision extension, were also linked to higher infection risk. In addition, women who required uterotonics other than oxytocin within the first 48 hours postpartum had significantly greater odds of developing an infection. Together, these findings highlight how both baseline patient characteristics and intraoperative or early postpartum interventions contribute to infectious morbidity.

To visually assess the contribution of each predictor, a nomogram is shown in Fig. 3. To use the nomogram, locate the value of each variable and draw a vertical line to see the corresponding number of points. After summing the points for each variable, locate the total on the line called “total points” and draw a line to see the corresponding risk of infection.

Fig. 3.

Nomogram for predicting infection.

Discussion

Principal Findings

This study developed and internally validated a predictive model for postpartum infection after CD. Significant predictors included tobacco use, obesity, labor before CD, longer surgical duration, uterine incision extension, and use of uterotonics other than oxytocin. The model demonstrated modest predictive accuracy, with an AUC of 0.64. These findings highlight the multifactorial risk factors for infection and provide a starting point for developing and refining models that may 1 day support individualized risk assessment.

Results in the Context of What Is Known

To our knowledge, this work leverages one of the largest contemporary, prospectively collected multicenter datasets used to derive an infection risk model after CD, and it integrates several intraoperative factors not consistently available in administrative datasets. Our findings align with prior research that identifies elevated BMI and tobacco use as risk factors for infection.¹⁷ Studies have consistently shown that obesity increases the likelihood of postoperative infections due to altered wound healing and higher rates of comorbidities.^6,18–20 Similarly, prior investigations have identified preterm delivery and prolonged labor as risk factors for infection,^21,22 likely due to prolonged exposure to pathogens and maternal immune dysregulation. While our AUC was modest, it is comparable to other widely used predictive models, such as the Framingham cardiovascular risk score²³ and vaginal birth after cesarean prediction tools,²⁴ which have concordance indices around 0.7.

Clinical Implications

This predictive model enables clinicians to assess individualized risk for postpartum infection based on a combination of patient-specific and surgical factors. The model may be useful for informing discussions about relative risk by showing how commonly collected maternal and operative factors combine to influence infection outcomes. Given its modest discrimination and the absence of several established infection-prevention variables in the dataset, the model should not be used to direct antibiotic selection, alter standard prophylaxis, or change clinical management. While it cannot, at this stage, guide changes in antibiotic use, surveillance intensity, or other interventions, it highlights patient groups who appear to be at increased risk and may therefore warrant closer attention in future research. Importantly, these findings should be viewed as hypothesis-generating, and further studies are needed to determine whether individualized risk prediction can meaningfully inform prevention strategies or clinical management.

Research Implications

Future research should aim to externally validate this model in diverse populations and clinical settings. Additional studies are needed to determine whether individualized risk prediction meaningfully informs prevention or patient counseling strategies. Beyond validation, further work could also explore integrating novel biomarkers or leveraging machine learning approaches to improve predictive accuracy and clinical utility of infection risk models.

Strengths and Limitations

A major strength of this study is its large, multicenter dataset, which included over 10,000 individuals who underwent CD with robust documentation of perioperative variables. The model integrates multiple risk factors into a single, interpretable prediction tool, making it practical for clinical use. However, its moderate AUC reflects room for improvement in predictive performance, and if applied for individual classification, the model’s modest discrimination would be expected to yield substantial false-positive and false-negative results. Limitations include the fact that internal validation was performed using bootstrap resampling to reduce optimism bias, but external validation in independent populations and calibration assessment across settings remain necessary, which may limit its generalizability to other populations. However, because the data were collected prospectively as part of a randomized trial, they are likely to be more complete and reliable than those obtained from retrospective EMR abstraction alone. Additionally, the composite outcome grouped various infection types, potentially masking unique predictors specific to individual infections (e.g., adjunctive prophylaxis such as azithromycin or metronidazole, skin and vaginal preparation, and hair removal method), which could confound observed associations and partly explain the model’s performance. Furthermore, it included grouped infections with differing pathophysiology; however, the majority of events were SSIs, with fewer endometritis diagnoses and rare pelvic abscesses, limiting the power for component-specific modeling in this dataset. Finally, some predictors, such as surgical or radiologic interventions, transfusion of blood products, or the use of uterotonics other than oxytocin, were included without the ability to confirm temporality relative to the onset of infection, which may limit their interpretability as causal risk factors. Despite these limitations, the model represents a step toward better characterization of infection risk following CD.

Conclusion

Several pre- and perioperative characteristics were associated with post-cesarean infection. A multivariable model including some of these characteristics provides modest predictive ability for post-cesarean infection.

Key Points.

• This study aimed to create a model to predict postpartum infection after cesarean delivery.

• Several factors were linked to higher infection risk after cesarean delivery. These included tobacco use, obesity, labor before cesarean, longer surgery, uterine incision extension, and uterotonic use.

• The model had moderate accuracy, with an AUC of 0.6.

• Common perioperative factors may help predict infection risk after cesarean delivery.