Maternal genitourinary infection and risk of cerebral palsy
This research was funded by grant no. R40MC06636-01-00 from the Health Resources and Services Administration.
Dr Joshua R Mann at Department of Family and Preventive Medicine, University of South Carolina School of Medicine, 3209 Colonial Drive, Columbia, SC 29203, USA. E-mail: Joshua.firstname.lastname@example.org
Aim To examine the association between genitourinary infection during pregnancy and cerebral palsy (CP) in children.
Method Medicaid and birth certificate data were obtained for 135 835 pregnant women with singleton births paid for by Medicaid from 1996 to 2002. Linked Medicaid billing data were obtained for their children in 2007. The association between maternal genitourinary infection and CP was modeled using generalized estimating equations.
Results Maternal genitourinary infection was significantly associated with CP (odds ratio [OR]=1.27, p=0.007). Additional analyses revealed that the association was strongly significant for preterm or low birthweight infants when maternal infection was diagnosed in the first two trimesters of pregnancy (OR=1.62, p<0.001). This association remained (OR=1.72, p<0.001) when the model was limited to cases of CP diagnosed by at least two different clinicians. Infection was not significantly associated with CP in term or normal-birthweight infants.
Interpretation Maternal genitourinary infection occurring in the first two trimesters was associated with increased risk of CP in preterm or low-birthweight children. Additional research is needed to determine whether this association is affected by antimicrobial treatment.
List of abbreviations
International Classification of Diseases, version 9
Cerebral palsy (CP) is the most common cause of motor disability in childhood, with a prevalence of 1.5 to 3 cases per 1000 live births.1–3 One of the strongest risk factors for CP is preterm birth. A meta-analysis of 26 studies demonstrated that the prevalence of CP is 14.6% among infants born between 22 and 27 weeks’ gestation, 6.2% among infants born between 28 and 31 weeks, 0.7% among those born between 32 and 36 weeks, and 0.1% in term infants.4 Approximately half of all cases of CP occur in infants born preterm.5
Given the strong association between preterm birth and CP, it stands to reason that risk factors for preterm birth may also be associated with increased risk of CP. One important risk factor for preterm birth is maternal genitourinary infection. Infections such as bacterial vaginosis,6 trichomoniasis,7 and Chlamydia trachomatis8 are associated with increased risk of preterm birth. Intrauterine infection has been documented in 25 to 40% of preterm births, but the actual proportion of preterm births at least partly attributable to maternal infection is probably even greater.9–12
There is substantial evidence that clinical chorioamnionitis (symptomatic intrauterine infection) is a risk factor for CP.13–15 A meta-analysis of 19 studies15 found that clinical chorioamnionitis was significantly related to increased risk of CP in term (relative risk [RR] = 4.7) and preterm infants (RR = 1.9). The association between asymptomatic histologic chorioamnionitis and CP was not significant in preterm infants (RR = 1.6), but one study demonstrated a markedly elevated risk of CP in term infants born to women with histologic chorioamnionitis (RR = 8.9).16 It is believed that the inflammatory response to intrauterine infection (cytokines including interleukin [IL]-1, IL-6, and tumor necrosis factor alpha) can cause periventricular leukomalacia, the most common brain lesion in infants with CP.17–20
It is reasonable to hypothesize that common genitourinary infections may be associated with CP, either indirectly by increasing the risk of preterm birth or directly via microbial toxins or initiation of an inflammatory response that extends to the chorioamnion. Nelson and Ellenberg analysed data from the National Collaborative Perinatal Project (from 1959 to 1966) and reported that urinary tract infection was not a significant risk factor for CP.21 However, a more recent study reported that urinary tract infection at admission for delivery was significantly associated with greater risk of periventricular leukomalacia in very preterm infants (OR = 5.71).22 Additional research on the role of urinary tract infection and other common genitourinary infections as potential risk factors for CP is needed. The purpose of this study was to test the hypothesis that maternal genitourinary infection is associated with increased risk of CP.
The project was granted exempt status by the University of South Carolina Institutional Review Board. We obtained de-identified South Carolina Medicaid billing records for all Medicaid-eligible women pregnant from 1996 to 2002, plus linked birth certificates and Medicaid billing records for children. Because we were interested in the role of genitourinary infection as a risk factor for unexplained CP, we excluded children who were diagnosed with a genetic or chromosomal abnormality, central nervous system anomaly, central nervous system infection, or traumatic injury that would substantially increase the risk of CP. We also limited the analysis to singleton births. Two thousand five hundred and thirty-seven children were removed because of high-risk diagnoses, and 2559 children were not singleton births.
Cases of CP were defined as children diagnosed with International Classification of Diseases, version 9 (ICD-9) code 343 in the Medicaid data. Maternal infections were ascertained using ICD-9 codes indicating a diagnosis of one or more of the following conditions during pregnancy: trichomoniasis; gonorrhea; Chlamydia trachomatis/non-gonococcal urethritis (referred to throughout the rest of the paper as chlamydia); vulvovaginal candidiasis; urinary tract infection; vaginitis; cervicitis; upper reproductive tract infections; and unspecified ‘infections of the genitourinary tract during pregnancy’. We determined that a women was pregnant at the time of an infection diagnosis by counting backward from the date of delivery, using the gestational age reported on the birth certificate. For example, an infection occurring 35 weeks before delivery of a 36-week infant was counted as occurring during pregnancy; had the infection been diagnosed 37 weeks before delivery it would not have been included in the analyses.
Because some women had more than one pregnancy during the study period, observations were not completely independent. Therefore, we modeled the outcome of a CP diagnosis using generalized estimating equations modeling (the GENMOD procedure in SAS 9.1), with the binomial distribution and the logit link function where the denominator in the model was the number of pregnancies per mother, and the mother was the independent unit of analysis. We assumed an exchangeable correlation structure, in which the degree of correlation between pregnancies in the same mother remained constant over time. We first modeled CP with any genitourinary infection as the key independent variable (regardless of specific diagnosis or timing). However, since CP is far more likely in children born preterm, we hypothesized that infections occurring earlier in pregnancy would be more strongly associated with increased risk than infections occurring late in pregnancy. Therefore, we also analysed infection in the first two trimesters (before 27 weeks’ gestation) separately.
We controlled for year of birth, maternal age and race, child’s sex, and maternal education level (12 or more years versus less than 12 years) as reported on birth certificates. We wanted to obtain an overall estimate of the association of infection with CP (including any effect mediated by preterm birth/low birthweight), and to evaluate whether infection exerted an effect on CP risk independent of gestational age and birthweight. Therefore, we modeled the outcome of CP first without including gestational age, birthweight, and small for gestational age (birthweight below the 10th centile for gestational age).23 Then we repeated the analysis, adding gestational age, birthweight, and small for gestational age to the models.
Since we were unable to clinically verify CP diagnoses, we repeated the modeling using a more restrictive case definition. In this step, we considered children to have ‘confirmed’ CP if they were diagnosed with ICD-9 code 343 by more than one healthcare provider. Children diagnosed with CP by only one provider were excluded from these models. The results of both approaches are presented.
After completing these analyses, we conducted an additional analysis of the effect of antimicrobial treatment in women with infection, using data from Medicaid outpatient pharmacy billing records. This step was purely hypothesis-generating, since there may be many reasons why some women filled a prescription for appropriate treatment and others did not, and these unmeasured factors could confound the association between treatment and CP. We limited this step to infections that were identified as having a significant association with CP in the primary models. Women with the diagnosis of interest were evaluated for the receipt of appropriate treatment (based on CDC Sexually Transmitted Disease Treatment Guidelines in effect from 1996 to 2007 and medical texts) within 30 days of any diagnosis with the condition. ‘Treated’ women were compared with those who were not appropriately treated for a condition within 30 days.
After limiting the data to singleton births and excluding the high-risk medical conditions, there were 135 835 mother−child pairs for analysis. Descriptive statistics are shown in Table I. Mean gestational age and birthweight were in the expected ranges. The maternal racial distribution was roughly evenly split between white and black women. Six hundred and six children (0.45%) were diagnosed with CP. Three hundred and sixty-two children (0.27%) were diagnosed with CP by more than one clinician. The median number of times children with any CP received the 343 ICD-9 code was 9. For children diagnosed with CP by more than one provider, the 343 ICD-9 code was used a median of 36 times.
Table I. Descriptive statistics (n = 135 835 mother−child pairs)
|Maternal age, y||22.96 (5.25)|| || |
|Gestational age, wks||38.67 (2.21)|| || |
|Birthweight, g||3195.67 (590.73)|| || |
|White ethnic origin|| ||64 095||47.19|
|Black ethnic origin|| ||69 852||51.42|
|Other ethnic origin|| ||1 888||1.39|
|Males|| ||69 268||50.99|
|Females|| ||66 565||49.01|
|Preterm (<37wks)|| ||13 424||9.88|
|Low birthweight (<2500g)|| ||12 798||9.42|
|Small for gestational age|| ||8 985||6.61|
|Cerebral palsy (any)|| ||606||0.45|
|Cerebral palsy (confirmed)|| ||362||0.27|
Data on maternal genitourinary infections are summarized in Table II. Thirty-eight percent of women were diagnosed with at least one infection. The most frequent infection was urinary tract infection, diagnosed in 18% of women. The most common specific infection diagnosed was vulvovaginal candidiasis (4%), followed by trichomoniasis (3%).
Table II. Infection frequencies
|Any infection|| ||51 637||38.11|
|Urinary tract infection||590, 590.0, 590.00, 590.01, 590.1, 590.10, 590.11, 590.2, 590.3, 590.8, 590.80, 590.81, 590.9, 595, 595.0, 595.2, 595.9, 599.0||24 361||17.98|
|Trichomoniasis||131, 131.0, 131.00, 131.01, 131.02, 131.09, 131.9||4 049||2.99|
|Gonorrhea||098, 098.0, 098.1, 098.10, 098.11, 098.15, 098.16, 098.17, 098.19, 098.2, 098.3, 098.30, 098.31, 098.35, 098.36, 098.37, 098.39, 647.1||1 013||0.75|
|Chlamydia trachomatis/ Non-gonococcal urethritis||099.41, 099.5, 099.50, 099.53 099.54, 099.55 099.56, 099.59, 647.2||884||0.65|
|Candidiasis||112, 112.1, 112.2||5 582||4.12|
|Ascending reproductive tract infection (pelvic inflammatory disease, chorioamnionitis)||614, 614.0-614.9, 615, 615.0, 615.1, 615.9, 658.4||2 370||1.74|
|Non-specific genitourinary infections||616, 616.0, 616.1, 616.10, 616.9, 646.6, 646.60-646.64||29 375||21.63|
The regression model predicting CP by infection, regardless of infection timing, is summarized in Table III. Infection was a significant risk factor for CP, whether using the less restrictive or more restrictive case definition. In both models, the odds of CP were increased by 27% in children of women with a genitourinary infection. Sex and maternal education were significant covariates in both models, while maternal age and race were significant only for the less restrictive case definition. More recent birth year was associated with reduced likelihood of being diagnosed with CP in both models.
Table III. Any genitourinary infection and cerebral palsy (CP)
|Any infection||1.27||1.07, 1.51||0.007||1.27||1.02, 1.58||0.032|
|Maternal age||1.03||1.02, 1.05||<0.001||1.02||0.996, 1.04||0.120|
|≥12y maternal education||0.71||0.59, 0.84||<0.001||0.74||0.59, 0.93||0.009|
|White race||0.83||0.69, 0.98||0.031||0.81||0.65, 1.01||0.057|
|Females||0.69||0.58, 0.82||<0.001||0.69||0.56, 0.86||<0.001|
|Birth year||0.88||0.84, 0.92||<0.001||0.88||0.84, 0.93||<0.001|
Table IV summarizes the findings of the additional regression analyses. As hypothesized, infections occurring in the first two trimesters were more strongly associated with CP (OR =1.45, p<0.001 for any CP; OR=1.52, p<0.001 for ‘confirmed’ CP). Third-trimester infection was not included in the initial modeling of early infection, since we were not controlling for gestational age and birthweight (women who remained pregnant long enough to acquire a third- trimester infection were de facto less likely to deliver an infant with CP). We added third- trimester infection to the models and controlled for small for gestational age, gestational age, and birthweight. Early infection remained significant for both case definitions, but third-trimester infection was not significant for either. As expected, increasing gestational age (OR=0.88/wk of gestation age, p<0.001 for any CP) and birthweight (OR=0.50/kg, p<0.001 for any CP) were strongly protective.
Table IV. Genitourinary infection and cerebral palsy (CP): sub-analyses
|All births||First two trimesters||1.45||<0.001||1.52||<0.001|
|All births, controlling for SGA, GA, BW, third- trimester infection||First two trimesters||1.28||0.007||1.32||0.017|
|All births, controlling for SGA, GA, BW, infection in first two trimesters||Third trimester||1.14||0.208||0.98||0.865|
|GA ≥37wks; BW≥2500g||Any infection||1.11||0.336||0.99||0.935|
|GA ≥37wks; BW≥2500g||First two trimesters||1.16||0.229||1.10||0.572|
|GA ≥37wks; BW≥2500g, controlling for SGA, GA, BW, third-trimester infection||First two trimesters||1.13||0.333||1.13||0.493|
|GA ≥37wks; BW≥2500g, controlling for SGA, GA, BW, infection in first two trimesters||Third trimester||1.10||0.474||0.79||0.251|
|GA <37wks OR BW <2500g||Any infection||1.36||0.016||1.42||0.022|
|GA <37wks OR BW <2500g||First two trimesters||1.62||<0.001||1.72||<0.001|
|SGA||First two trimesters||1.73||0.020||1.51||0.172|
When limiting the modeling to 117 514 children who were born at term (≥37 weeks’ gestation) and had normal birthweight (≥2500g), maternal genitourinary infection was not a significant risk factor for CP. Among preterm or low-birthweight children, infection in the first two trimesters was a strongly significant risk factor (OR=1.62, p=0.002 for the less restrictive case definition; OR=1.72, p<0.001 for the more restrictive definition). Infection was similarly associated with increased risk in children born small for gestational age, though the association was statistically significant only for the less restrictive case definition of CP.
When we examined specific infection diagnoses, chlamydia was most strongly associated with CP. Children of women with chlamydia had twice the odds of any CP diagnosis (OR=2.07, p=0.045) and an even greater increase in the odds of confirmed CP (OR=2.58, p=0.025). Trichomoniasis and urinary tract infection were significantly associated with any diagnosis of CP but not confirmed CP.
We considered that severe ascending reproductive tract infections could account for the associations observed, given that chorioamnionitis is an established risk factor for CP. We repeated the multivariable model with infection in the first two trimesters as the key independent variable, excluding those diagnosed with upper reproductive tract infections (pelvic inflammatory disease and chorioamnionitis). The association remained strongly significant (OR=1.47, p<0.001 for any CP; OR=1.53, p<0.001 for confirmed CP).
We also noted that some women were diagnosed with infections up to and even on their delivery date. We were concerned that some infections may have been diagnosed because physicians were looking for an explanation for preterm labor or another pregnancy complication; such differential ascertainment could bias the observed association. To account for this possibility, we modeled the impact of infection in the first two trimesters, removing children born before 29 weeks’ gestation. Thus the least amount of time between the infection diagnosis and delivery was two weeks. The results changed very little (OR=1.40, p<0.001 for any CP; OR0=1.48, p=0.002 for ‘confirmed’ CP).
As stated in the Method section, our modeling approach assumed that all children born to a particular mother were equally correlated with one another. This assumption may not be entirely accurate. Health status and health-related behaviors may change over time, and different children born to the same mother do not necessarily have the same father. These and other changes over time may produce unequal correlation between siblings over time. A conservative way to deal with this uncertainty is to limit the modeling to only one child per mother, so that there is complete independence of observations. The outcome of CP can then be modeled using standard logistic regression. We tested the effect of this approach, limiting the observations to the first child born to each women during the study period, and modeling the association of infection in the first two trimesters with any diagnosis of CP. We controlled for maternal age, race and education, child’s sex, and year of birth. The strongly significant association between early infection and CP remained (OR=1.51, p<0.001). The same was true for the outcome of confirmed CP (OR=1.57, p<0.001). Therefore, we do not believe the findings were substantially impacted by the assumptions of the repeated-measures (GEE) modeling.
Finally, we explored the risk of CP in children of treated versus untreated women diagnosed with chlamydia, trichomoniasis, or urinary tract infection. There were only eight cases of CP diagnosed in children of women with chlamydia. Because of the small cell size, we did not model the impact of treatment for this condition. There were 33 cases of CP in children of women with trichomoniasis and 131 in children of women with urinary tract infection. We modeled the outcome of any CP diagnosis in children of women with each condition, with treatment status as the key independent variable. We controlled for maternal age, race and education, and child’s sex and year of birth. A diagnosis of CP was more likely in children of women who were treated for trichomoniasis, but the association was not statistically significant (OR=1.18, p=0.644). The impact of treatment was similarly not significant for women with urinary tract infection (OR=1.16, p=0.419).
Genitourinary infection in the first two trimesters of pregnancy was significantly associated with increased risk of CP in children. Adjusting for birthweight, gestational age, and small for gestational age reduced but did not eliminate the association. Stratification by preterm or low birthweight status revealed that the association was very strong in children born before 37 weeks and weighing less than 2500g. It was not significant in term or normal-birthweight infants. The findings remained when limiting ‘cases’ of CP to children who were diagnosed by more than one healthcare provider.
When we examined specific infections, chlamydia, trichomoniasis, and urinary tract infection appeared to be associated with CP while gonorrhea and candidiasis were not. Only chlamydia was significantly associated with increased risk of confirmed CP. The results relating to specific infections should be considered with caution, because of small cell sizes for some of the infections and the fact that many women (15%) were diagnosed with a non-specific infectious condition such as vaginitis or cervicitis, without identification of a specific causal organism. Since we did not have access to clinical records or laboratory reports, we were not able to ascertain the causative agents for these women.
Our reliance on billing data for identifying women with genitourinary infections almost certainly resulted in incomplete exposure ascertainment. Some infections were less prevalent in our cohort than would be expected on the basis of research in which pregnant women were screened for infections.24 Trichomoniasis and chlamydia in particular appear to have been underdiagnosed. Some of the ‘missing’ cases of trichomoniasis and chlamydia may be accounted for by those diagnosed with non-specific infections. Other cases of trichomoniasis and chlamydia probably went undiagnosed, as females with these infections are frequently asymptomatic. We do not have enough information to speculate whether the findings of this study can be extrapolated to women with undiagnosed, asymptomatic infections.
Billing data are also imperfect for identifying children with CP, though given the potential stigma of the condition we do not believe clinicians would diagnosis CP without reasonable certainty. Recent surveillance research identified CP in 0.41% of low-to-middle income children at age 8.24 After exclusions, the overall prevalence of any CP diagnosis in our cohort was 0.45% and the prevalence of ‘confirmed’ CP was 0.27%. The prevalence of CP in children whose oldest age in the Medicaid file was 8 years was 0.85% before exclusions, and 0.62% after exclusions. However, using the more restrictive case definition, requiring a diagnosis by at least two providers, the prevalence was 0.54% before exclusions and 0.38% after exclusions. Thus, the prevalence of ‘confirmed’ CP is similar to what would be expected. We do not know whether the additional CP diagnoses, made by only one provider, represent actual cases of CP or some other neurological condition. Importantly, association between early infection and CP changed very little (generally becoming slightly stronger) when using the more restrictive case definition, which argues that the findings are not substantially biased by outcome misclassification. To be doubly certain, we restricted the case definition even further, requiring diagnosis with CP by at least three different providers. This reduced the prevalence among 8-year-old children in the cohort to 0.46% before exclusions, and 0.22% after exclusions. Early infection remained a strongly significant risk factor with this even more restrictive case definition (OR=1.61, p<0.001).
Our study population was limited to women receiving Medicaid coverage for their pregnancy care (and their children). Medicaid primarily insures low-income women and children (the income limit in South Carolina for pregnant women is 185% of poverty as defined by the United States Government [http://aspe.hhs.gov/poverty/08poverty.shtml]). Approximately 38% of South Carolina births from 1996 to 2002 were paid for by Medicaid. Consistent with socio-economic trends in South Carolina, our sample of Medicaid clients included a large proportion of young women and racial minorities. We cannot be certain that the association between genitourinary infection and CP is generalizable to other populations of pregnant women. Though we do not believe there is any biological reason to expect the effects of maternal infection to differ on the basis of demographic and economic factors, it is possible that unmeasured factors associated with age, race, and socio-economic status may have affected the results.
A final limitation is that our data do not provide information about the mechanism(s) through which genitourinary infection may increase the risk of CP. On the basis of the literature reviewed in the first part of the text, we believe that the most likely explanation is initiation of an inflammatory response that extends to the chorioamnion. Additional research, perhaps with animal models, is needed to evaluate the extent to which lower reproductive tract and urinary tract infections affect the intrauterine environment.
In spite of the above limitations, this study provides substantial evidence that maternal genitourinary infection is associated with increased risk of CP in children. This finding has particular clinical relevance if antimicrobial treatment is found to mitigate the effects of infection. There is a paucity of studies investigating the effects of maternal antimicrobial treatment on child neurological outcomes. Kenyon et al. recently reported that antimicrobial treatment of women with spontaneous preterm labor (irrespective of documented infection status) was actually associated with increased risk of CP in children.25 Our analysis did not find a significant protective or harmful effect of treatment for trichomoniasis or urinary tract infections. However, this finding must be viewed with caution since the effect of filling a prescription for appropriate antimicrobial therapy may be confounded by a number of issues including the severity of symptoms, method and precision of the diagnosis (presumptive versus confirmed diagnosis), direct provision of treatment in a hospital, emergency room, or physician’s office, length of time between the initial infection and the diagnosis, and the women’s general level of self-care and adherence to medical advice.
In summary, we found that maternal genitourinary infection in the first two trimesters of pregnancy was associated with increased risk of CP. The association is strong and robust for preterm or low-birthweight infants but not for term or normal birthweight infants. Additional research using clinical records is needed to verify the findings and more thoroughly evaluate the impact of appropriate antimicrobial treatment.