Abstract
BACKGROUND:
To evaluate the associations between prenatal and perinatal factors and CP risk using a statewide sibling-comparison design.
METHODS:
We established a cohort of over 4 million singleton births in California during 2007–2015, and we identified families with outcome-discordant siblings of 1213 CP and 1544 non-CP. We estimated odds ratio (OR) and 95% confidence interval (CI) for CP according to perinatal factors including preterm birth (PTB), small for gestational age, low Apgar score, and prenatal factors including maternal pregnancy complications (perinatal infection, gestational diabetes, preeclampsia) and lifestyle-related factors (cigarette smoking, pre-pregnancy overweight).
RESULTS:
The perinatal factors remained strongly associated with CP using sibling design, although the point estimates were smaller for PTB (cohort OR = 4.72, 95%CI 4.42–5.04, sibling OR = 3.49, 95%CI 2.74–4.46) and low Apgar score (cohort OR = 19.62, 95%CI 17.99–21.41, sibling OR = 8.79, 95%CI 5.49–14.08). In sibling design, the associations between maternal pregnancy complications or pre-pregnancy overweight and CP risk were attenuated to null. We observed stronger effects between maternal cigarette smoking and CP in the sibling design, however sensitivity tests indicated possible bias from carryover effects.
CONCLUSION:
Adverse perinatal factors remained strongly associated with childhood CP, while uncontrolled confounding bias required considerations for pregnancy complications and CP development.
INTRODUCTION
Cerebral palsy (CP) is the most prevalent neuromotor disorder diagnosed in early childhood and persists through the full lifespan.1 Despite medical advancements in prenatal and perinatal care, the prevalence of CP has remained stable for decades and affects at about 2 per 1000 live births in developed countries.2–4 Although the etiology of most CP cases is complex and largely unknown,5 most cases has congenital origins thus encouraging investigations focused on prenatal and perinatal factors. At present, several perinatal factors such as preterm delivery,3,6 birth asphyxia,7 and antenatal infections are considered established risk factors for CP.8–10 Certain prenatal factors, such as maternal pregnancy complications, obstetric complications, and lifestyle factors, have also been suspected.9,11–15 However, a variety of preceding factors, such as maternal socio-demographic and economic factors, genetic susceptibility, access to healthcare, and environmental/household factors16–18 can affect these perinatal and prenatal conditions leaving uncontrolled confounding bias a concern.
The traditional cohort design with confounding bias adjustment at the analytic stage is vulnerable to uncontrolled confounding bias when one or more important confounders were left out in the model. The sibling comparison design offers as an alternative research strategy to control for all shared factors between the siblings by comparing the exposure and outcome status only among siblings born to the same mothers.19 Although sibling comparison design has gained popularity in the field of perinatal epidemiology, it has not been applied in cerebral palsy research. Furthermore, the sibling comparison design does operate on several strong study assumptions that need attention.20,21 A careful implementation of the sibling design with sensitivity tests are needed to guide result interpretation.
The aims of our study were to evaluate a priori identified prenatal and perinatal risk factors for CP using a sibling comparison design. We focused on perinatal factors including preterm birth,3,6 small for gestational age (SGA),7,22 and low Apgar score22,23 that have been consistently associated with CP in the traditional cohort studies. Additionally, we investigated several prenatal factors including maternal pregnancy complications of perinatal infection,24 gestational diabetes,9,10,25 preeclampsia,9,11–13,26 and lifestyle-related factors14,15,27,28 of cigarette smoking and pre-pregnancy body mass, which were suggested to affect CP risk but the overall evidence has not been conclusive. Finally, we examined the robustness of sibling design over different analytical approaches and assessed the potential carryover biases and lack of generalizability through sensitivity analyses.
METHODS
This study protocol was approved by the official institutional review board in Yale University (IRB#2000028297) and the California Committee for the Protection of Human Subjects (Project#12-10-0861) and was exempted from informed consent requirement because there was no contact with human subject. The paper has been reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.
Study population
We analyzed a statewide study for CP in California that included 4,511,704 singleton births in 2007–2015 ascertained from the birth records.29 A total of 4563 CP cases was ascertained using the diagnostic records from the Department of Developmental Services (DDS) up to 2021 to allow at least 5 years of follow up for CP diagnosis. The birth records and the DDS records were linked through a probabilistic linkage using the LinkPlus software30 based on maternal, paternal (when available), and child information including first and last name, date of birth, and child’s sex with successful rate of 93%.29 In the DDS, CP was defined as a group of nonprogressive lesions in the brain that manifest in early childhood with movement or posture controls.31 Details regarding the CP diagnostic validity have been described previously.18,29
From the full cohort, we first identified a “sibling cohort” that contains 2,019,739 siblings born to the 926,873 biological mothers. Given that only siblings with discordant outcome and exposure status would contribute to the effect estimates in the sibling design, we focused on the outcome-discordant siblings as our main study sample, which includes 1213 CP cases and their 1544 unaffected siblings (Fig. 1). The sibling identification was also based on a probability linkage approach similar as above, using child’s last name and maternal and paternal (when available) information on first name, last name, and date of birth.32
Fig. 1. Flowchart of the study population.
CP cerebral palsy.
Perinatal and prenatal factors
The evaluated perinatal and prenatal factors were selected based on current literature and the data availability in the California birth records. For perinatal factors, we studied preterm birth (PTB) with definition of gestational age<37 weeks,3,6 calculated from the self-reported last menstrual period and the date of birth. To capture the non-linear relationship between gestational age and CP,33 we estimated the associations for the moderate to late preterm births (mPTB, defined as gestational week of 32–37) and very preterm birth (vPTB, defined as gestational week <32), separately. Other perinatal factors studied include small for gestational age (SGA), defined as infant birth weight<10th percentile at each gestational age,7,22 and 5-minute low Apgar score, defined as the infant Apgar score<7 (range of 0–10) at 5 minutes after delivery.22,23
For prenatal factors, we studied several maternal complications including gestational diabetes, preeclampsia, and maternal chorioamnionitis,34 an acute inflammation of intrauterine structures related to bacterial infection that may induce offspring CP.24 We also investigated several maternal lifestyle-related factors, including the self-reported maternal cigarette smoking during pregnancy,14,32 and the maternal pre-pregnancy obesity15 or overweight27,28 based on calculated BMI using pre-pregnancy weights (in kilograms) and height (in meters), defined as obese (≥30), overweight (25 to <30), normal weight (18.5 to <25), and underweight (<18.5).
Covariates
We extracted maternal individual socio-demographic characteristics from the birth records, including age at delivery (≤18, 19–25, 26–30, 31–35, >35 years), self-reported race/ethnicity (African American/Black, Asian, Hispanic/Latinx of any race, non-Hispanic White, others), educational attainment (<12th grade, high school or diploma, college and above), receipt of prenatal care (yes, no), primary payment source for prenatal care (government, private, others), parity (1, 2, 3+), receipt of Women, Infant, and Children (WIC) Nutrition Program (yes, no), and residential region in California (northern, central, southern California). We also evaluated the census-tract level characteristics captured by the Center for Disease Control and Prevention’s social vulnerability index (SVI), including domains of socioeconomic status, household composition and disability, minority status and language, and housing type and transportation.35
Statistical analysis
Analysis of the full cohort.
We utilized logistic regression models to estimate the odds ratio (OR) and 95% confidence intervals (CIs) for CP, according to each prenatal and perinatal factors. The confounding control strategy considered the temporality of these factors. Specifically, when studying the perinatal risk factors, we adjusted for all socio-demographic covariates and maternal pregnancy factors since they occurred before birth. For maternal pregnancy complications and cigarette smoking, we adjusted for all socio-demographic covariates and maternal pre-pregnancy BMI. For maternal pre-pregnancy BMI, we adjusted for socio-demographic covariates, except for prenatal care and WIC program that occurred during pregnancy. Covariates with missing values (<5% of the complete dataset) were analyzed by including a separate category that represents missingness. In all models, we controlled for child’s sex since CP risk and the occurrence of adverse perinatal factors varied by child’s sex. We also included the exact birth year and the season of conception (winter, spring, summer, fall) to adjust for temporal and seasonal trends.29
Sibling analysis.
Using the outcome-discordant siblings, we conducted conditional logistic regression with strata on family IDs to estimate the OR and 95% CI for CP. Among siblings, we observed that the distribution for maternal race/ethnicity, educational attainment, and the residential regions were no longer different between the CP cases and their unaffected siblings (see Results). We therefore only included covariates that were likely to be time-varying in the conditional logistic models. Further, we evaluated the consistency by restricting to the same-sex siblings only and stratified by sex to examine any sex differences.
Sensitivity analyses.
First, we examined the robustness of our sibling findings by performing two additional analytical approaches, given that different strategies have been used to analyze sibling data in the literature.36,37 Specifically, we fitted the conditional logistic regression using the sibling cohort (not limiting to the discordant siblings only), and we implemented the between-within regression model (instead of logistic regression, see Supplemental methods) on the sibling cohort.36,37 Second, we limited our discordant siblings to family size of two given that families with larger size might contribute more to the effect estimates. Third, sibling analysis is susceptible to amplified unmeasured time-varying confounding bias.38 We aimed to reduce such influence from time-varying factors by excluding siblings with inter-pregnancy intervals greater than 5 years, or restricting to siblings with the same delivery hospitals to limit changes in hospital care. Fourth, we examined the potential carryover effects that the exposure or outcome of older siblings might affect younger siblings (Supplemental methods).21 Fifth, we examined the generalizability of effect estimates from sibling analyses to the full cohort by applying a transportability model with inverse probability selection weights (IPSW) (Supplemental methods).39,40 Briefly, we created a weighted sibling sample that had a similar demographic characteristics to the full cohort and then we compared the effect estimates from the weighted sibling sample to the original (unweighted) sibling cohort and the full cohort. Statistical analyses were conducted in SAS version 9.4 (SAS Institute) between January 2023 and October 2023.
RESULTS
Among the full cohort, CP cases were more common among children born to mothers who were older (e.g., >35 years old), self-identified as African American or Hispanic, had education level of high school or below, or lived in more socially vulnerable neighborhoods (Table 1). We also showed that 1213 (26.6%) out of all 4563 CP cases were included in the discordant sibling analyses. While their characteristics were mostly comparable, the CP cases included in the discordant siblings were more likely to be born to mothers with higher education attainment and used private health insurance compared with CP cases in the full cohort (Table 1). The distribution for several maternal demographic factors became similar or nearly identical comparing the CP cases and their unaffected siblings, but differences still existed for factors that can change over time (Table 1).
Table 1.
Characteristics of all singleton-live births and cerebral palsy (CP) discordant siblings in California, 2007–2015.
| California birth cohort | Discordant siblingsa | |||
|---|---|---|---|---|
| All singleton births | All CP cases | |||
| N = 4,511,704 (%b) | N = 4563 (%b) | N = 1544 (%b) | N = 1213 (%b) | |
| Maternal age at delivery | ||||
| <=18 | 207,293 (4.6) | 264 (5.8) | 42 (2.7) | 47 (3.9) |
| 19–25 | 1,302,695 (28.9) | 1315 (28.8) | 449 (29.1) | 330 (27.2) |
| 26–30 | 1,241,839 (27.5) | 1175 (25.8) | 466 (30.2) | 375 (30.9) |
| 31–35 | 1,110,545 (24.6) | 1037 (22.7) | 356 (23.1) | 304 (25.1) |
| >35 | 648,770 (14.4) | 772 (16.9) | 231 (15.0) | 157 (12.9) |
| Missing/Unknown | 562 (<0.1) | 0 | 0 | 0 |
| Race/ethnicity | ||||
| African American/Black | 254,880 (5.6) | 370 (8.1) | 75 (4.9) | 55 (4.5) |
| Asian | 597,299 (13.2) | 389 (8.5) | 153 (9.9) | 118 (9.7) |
| Latinx/Hispanic of any race | 2,270,730 (50.3) | 2528 (55.4) | 816 (52.8) | 638 (52.6) |
| Non-Hispanic White | 1,258,517 (27.9) | 1140 (25.0) | 456 (29.5) | 360 (29.7) |
| Others | 47,083 (1.0) | 52 (1.1) | 20 (1.3) | 15 (1.2) |
| Missing/Unknown | 83,195 (1.8) | 84 (1.8) | 24 (1.6) | 27 (2.2) |
| Educational attainment | ||||
| <12th grade | 969,422 (21.5) | 1177 (25.8) | 340 (22.0) | 263 (21.7) |
| High school or diploma | 2,225,344 (49.3) | 2392 (52.4) | 796 (51.6) | 613 (50.5) |
| College and above | 1,149,021 (25.5) | 834 (18.3) | 362 (23.4) | 291 (24.0) |
| Missing/Unknown | 167,917 (3.7) | 160 (3.5) | 46 (3.0) | 46 (3.8) |
| Receipt of prenatal care | ||||
| Yes | 4,382,436 (97.1) | 4366 (95.7) | 1495 (96.8) | 1166 (96.1) |
| No | 129,268 (2.9) | 197 (4.3) | 49 (3.2) | 47 (3.9) |
| Payment source for prenatal care | ||||
| Government care | 2,208,714 (49.0) | 2514 (55.1) | 818 (53.0) | 588 (48.5) |
| Private | 2,086,940 (46.3) | 1828 (40.1) | 688 (44.6) | 582 (48.0) |
| Others | 205,666 (4.6) | 204 (4.5) | 34 (2.2) | 39 (3.2) |
| Missing/Unknown | 10,384 (0.2) | 17 (0.4) | 4 (0.3) | 4 (0.3) |
| Parity | ||||
| 1 | 1,781,486 (39.5) | 1923 (42.1) | 395 (25.6) | 398 (32.8) |
| 2 | 1,419,110 (31.5) | 1233 (27.0) | 557 (36.1) | 426 (35.1) |
| 3+ | 1,311,108 (29.1) | 1407 (30.8) | 592 (38.3) | 389 (32.1) |
| Receipt of WIC food supplements | ||||
| Yes | 2,350,611 (52.1) | 2685 (58.8) | 885 (57.3) | 646 (53.5) |
| No | 2,161,093 (47.9) | 1878 (41.2) | 659 (42.7) | 567 (46.7) |
| Residential region at birth | ||||
| North | 1,363,767 (30.2) | 1030 (22.6) | 257 (16.6) | 207 (17.1) |
| Central | 544,316 (12.1) | 407 (8.9) | 171 (11.1) | 131 (10.8) |
| Southern | 2,603,621 (57.7) | 3126 (68.5) | 1116 (72.3) | 875 (72.1) |
| Census-tract level social vulnerability index (range 0–1) | ||||
| Socioeconomic status | 0.57 (0.29) | 0.63 (0.27) | 0.61 (0.28) | 0.60 (0.28) |
| Household composition | 0.56 (0.29) | 0.60 (0.27) | 0.60 (0.28) | 0.59 (0.28) |
| Minority status | 0.58 (0.27) | 0.62 (0.27) | 0.60 (0.28) | 0.60 (0.28) |
| Transportation | 0.55 (0.28) | 0.57 (0.28) | 0.57 (0.28) | 0.56 (0.27) |
| Birth year | ||||
| 2007–2010 | 2,091,282 (46.4) | 2315 (50.7) | 677 (43.8) | 605 (49.9) |
| 2011–2015 | 2,420,422 (53.6) | 2248 (49.3) | 867 (56.2) | 608 (50.1) |
| Season of conception | ||||
| Winter | 1,081,911 (24.0) | 1106 (24.2) | 384 (24.9) | 271 (22.3) |
| Spring | 1,074,637 (23.8) | 1134 (24.9) | 349 (22.6) | 303 (25.0) |
| Summer | 1,076,766 (23.9) | 1025 (22.5) | 361 (23.4) | 298 (24.6) |
| Fall | 1,176,900 (26.1) | 1169 (25.6) | 403 (26.1) | 299 (24.6) |
| Missing/Unknown | 101,490 (2.2) | 129 (2.8) | 47 (3.0) | 42 (3.5) |
| Sex of children | ||||
| Male | 2,312,441 (51.3) | 2526 (55.4) | 812 (52.6) | 671 (55.3) |
| Female | 2,199,263 (48.7) | 2037 (44.6) | 732 (47.4) | 542 (44.7) |
Siblings with discordant cerebral palsy (CP) status.
Percentage may not add up to 100% due to rounding.
In the full cohort, all perinatal factors that we evaluated were strongly associated with a higher risk for CP, including PTB (OR = 4.72, 95% CI: 4.42–5.04), mPTB (OR = 2.58, 95% CI: 2.37–2.82), vPTB(OR = 19.04, 95% CI: 17.51–20.71), SGA (OR = 2.49, 95% CI: 2.31–2.68), and 5-minute Apgar score <7 (OR = 19.62, 95% CI: 17.99–21.41) (Table 2). These positive associations remained in the sibling analysis while the point effect estimates were smaller despite overlapping confidence intervals comparing with the full cohort analyses (PTB, OR = 3.49, 95% CI: 2.74–4.46; mPTB, OR = 2.09, 95% CI: 1.59–2.75; vPTB, OR=15.17, 95% CI: 8.98–25.62; 5-minute Apgar score <7, OR = 8.79, 95% CI:5.49–14.08) (Table 2). Sibling results were also consistent when limited to siblings with the same biologic sex (Supplemental Table S1). When stratified by sex, no differences were found for the associations between PTB or SGA with CP, but the effect size of low Apgar score was more pronounced among females (Supplemental Table S1).
Table 2.
Estimated odds ratio and 95% CI for cerebral palsy (CP) according to neonatal adverse birth outcomes, among all singleton births or the outcome-discordant siblings in California, 2007–2015.
| California birth cohort | Discordant siblingsa | |||||
|---|---|---|---|---|---|---|
| CP/Cohort (4563/4,507,141) | Crude OR (95% CI) | Adjusted ORb (95% CI) | CP/Non-CP (1213/1544) | Crude OR (95% CI) | Adjusted ORc (95% CI) | |
| Adverse birth outcomes | ||||||
| Preterm birthd | 1365/361,964 | 4.98 (4.67, 5.31) | 4.72 (4.42, 5.04) | 337/197 | 3.33 (2.63, 4.22) | 3.49 (2.74, 4.46) |
| Preterm birth (finer classification) | ||||||
| Very preterm (GW < 32) | 725/51,082 | 18.73 (17.27, 20.32) | 19.04 (17.51, 20.71) | 175/40 | 14.48 (8.66, 24.21) | 15.17 (8.98, 25.62) |
| Moderate and late preterm (32 ≤ GW < 37) | 640/310,882 | 2.72 (2.50, 2.96) | 2.58 (2.37, 2.82) | 162/157 | 1.96 (1.50, 2.55) | 2.09 (1.59, 2.75) |
| Small for gestational aged | 942/438,423 | 2.46 (2.29, 2.64) | 2.49 (2.31, 2.68) | 229/164 | 2.29 (1.78, 2.94) | 2.31 (1.79, 2.99) |
| 5-min Apgar Score < 7 | 647/44,264 | 17.10 (15.73, 18.60) | 19.62 (17.99, 21.41) | 151/28 | 8.94 (5.63, 14.19) | 8.79 (5.49, 14.08) |
Siblings with discordant cerebral palsy (CP) status.
Logistic regressions adjusted for birth year, season of conception, maternal age at delivery, race/ethnicity, education, prenatal care, primary payment source for prenatal care, receipt of WIC nutrition program, parity, region of birth (northern, southern, central California), child’s sex, four domain-specific social vulnerability indexes, maternal pre-pregnancy body mass index, cigarette smoking during pregnancy, gestational diabetes, and gestational preeclampsia.
Conditional logistic regressions adjusted for covariates that varied across siblings, including birth year, season of conception, maternal age at delivery prenatal care, primary payment source for prenatal care, receipt of WIC nutrition program, parity, child’s sex, maternal pre-pregnancy body mass index, cigarette smoking during pregnancy, gestational diabetes, and gestational preeclampsia.
Preterm birth is defined as gestational week (GW) < 37, small for gestational age is defined as birth weight < 10th percentile for each gestational week (GW).
Maternal chorioamnionitis was associated with CP in the full cohort (OR = 1.51, 95% CI: 1.24–1.85), but the point estimate was smaller with wider confidence limits among discordant siblings (OR = 1.37, 95% CI: 0.75–2.49) (Table 3). No associations were observed between maternal gestational diabetes and offspring CP in either study designs. Maternal gestational preeclampsia was only associated with offspring CP in the full cohort (OR = 1.34, 95% CI: 1.14–1.56), but not in the sibling analysis (OR = 0.93, 95% CI: 0.55–1.54) (Table 3). Similarly, maternal overweight and obesity were only associated with offspring CP risk in the full cohort but not in the sibling analysis. Conversely, a larger effect estimate was observed between maternal cigarette smoking during pregnancy and offspring CP in the sibling design (OR = 2.83, 95% CI:1.28–6.26) than in the full cohort (OR = 1.50, 95% CI: 1.27–1.79) (Table 3).
Table 3.
Estimated odds ratio and 95% CI for cerebral palsy (CP) according to maternal pregnancy factors, among all singleton births or the outcome-discordant siblings in California, 2007–2015.
| California birth cohort | Discordant sibling subseta | |||||
|---|---|---|---|---|---|---|
| CP/Cohort (4563/4,507,141) | Crude OR (95% CI) | Adjusted OR (95% CI) | CP/Non-CP (1213/1544) | Crude OR (95% CI) | Adjusted OR (95% CI) | |
| Maternal pregnancy complications and lifestyle-related factors | ||||||
| Chorioamnionitisb | 102/72,553 | 1.40 (1.15, 1.71) | 1.51 (1.24, 1.85) | 26/21 | 1.48 (0.83, 2.65) | 1.37 (0.75, 2.49) |
| Gestational diabetesc | 215/194,568 | 1.10 (0.96, 1.26) | 1.10 (0.95, 1.26) | 42/73 | 0.69 (0.44, 1.07) | 0.76 (0.49, 1.20) |
| Gestational pre-eclampsiac | 174/117,988 | 1.48 (1.27, 1.72) | 1.34 (1.14, 1.56) | 37/47 | 0.89 (0.54, 1.47) | 0.93 (0.55, 1.54) |
| Cigarette smoking during pregnancyc | 150/99,259 | 1.52 (1.28, 1.98) | 1.50 (1.27, 1.79) | 32/26 | 2.72 (1.24, 5.97) | 2.83 (1.28, 6.26) |
| Pre-pregnancy BMId | ||||||
| <18.5 | 152/171,048 | 1.04 (0.88, 1.22) | 1.05 (0.88, 1.24) | 37/43 | 1.05 (0.56, 1.97) | 0.91 (0.48, 1.74) |
| 18.5 to <25 | 1722/2,066,395 | Ref | Ref | 511/632 | Ref | Ref |
| 25.0 to <30.0 | 1155/1,086,957 | 1.24 (1.15, 1.34) | 1.17 (1.08, 1.26) | 307/388 | 0.97 (0.75, 1.24) | 1.04 (0.81, 1.35) |
| ≥30.0 | 1128/885,550 | 1.49 (1.38, 1.60) | 1.36 (1.26, 1.48) | 280/359 | 0.94 (0.67, 1.32) | 1.10 (0.77, 1.57) |
Siblings with discordant cerebral palsy (CP) status.
Logistic regressions for birth cohort, adjusted for birth year, season of conception, maternal age at delivery, race/ethnicity, education, prenatal care, primary payment source for prenatal care, receipt of WIC nutrition program, parity, region of birth (northern, southern, central California), child’s sex, four domain-specific social vulnerability indexes, maternal pre-pregnancy body mass index (BMI), cigarette smoking during pregnancy, gestational diabetes, and gestational preeclampsia. Conditional logistic regressions for sibling design, adjusted for abovementioned covariates except for maternal race/ethnicity, education, and region of birth that were well-balanced across siblings.
Logistic regressions for birth cohort, adjusted for birth year, season of conception, maternal age at delivery, race/ethnicity, education, prenatal care, primary payment source for prenatal care, receipt of WIC nutrition program, parity, region of birth (northern, southern, central California), child’s sex, four domain-specific social vulnerability indexes, maternal pre-pregnancy body mass index (BMI). Conditional logistic regressions for sibling design, adjusted for abovementioned covariates except for maternal race/ethnicity, education, and region of birth that were well-balanced across siblings.
Logistic regression for birth cohort, adjusted for birth year, season of conception, maternal age at delivery, race/ethnicity, education, primary payment source for prenatal care, parity, region of birth (northern, southern, central California), child's sex, four domain-specific social vulnearability indexes. Conditional logistic regressions for sibling design, adjusted for abovementioned covariates except for maternal race/ethnicity, education, and region of birth that were well-balanced across siblings.
In sensitivity analyses, we confirmed that the main results from the outcome-discordant siblings were identical when applying the conditional logistic regression using all siblings, given that only discordant pairs would contribute to the effect estimates (Supplemental Table S2). When applying the between-within model on all siblings, the results were not identical but very similar (Supplemental Table S2). Findings for the discordant siblings were robust when we restricted to family size of two (Supplemental Table S3), excluded siblings with inter-pregnancy intervals>5 years (Supplemental Table S4), or limited to siblings with the same hospital delivery (Supplemental Table S5). When assessing the potential carryover effects, we observed similar associations comparing either the older or the younger sibling with adverse birth outcomes, while the statistical power was low concerning maternal pregnancy factors (Supplemental Table S6). Further, elevated ORs were apparent for smoking and obesity only when the older sibling being affected but not the younger, indicating that the carryover effects cannot be ruled out for these lifestyle-related factors (Supplemental Table S6). Finally, while the original sibling cohort has a higher representation of mothers with higher education level and who used private insurance compared to the full cohort (Supplemental Table S7), the distributions of these socio-demographic factors were comparable in the IPSW-weighted sibling sample and the full cohort (Supplemental Table S7). Our analyses showed that the effect estimates obtained from the weighted sibling sample were more comparable to the original unweighted sibling cohort than the full cohort (Supplemental Table S8). The results collectively suggested that the sociodemographic differences we examined was unlikely to explain the differences found in the sibling design and the full cohort design.
DISCUSSION
Preterm birth, small for gestational age, and low Apgar score at birth were strongly associated with higher risk of childhood CP in our statewide sibling study, although the effect sizes were smaller than those obtained from the full cohort. Maternal pregnancy complications and pre-pregnancy BMI were not associated with CP using sibling design. Potential bias from carryover effect cannot be ruled out regarding the observed association between maternal smoking and CP among siblings.
The etiology of CP is multifactorial and complex.5 In the literature, it has been documented that CP is more common among infants with adverse neonatal outcomes.3,6,7 Low Apgar scores, an indication of depressed vitality of newborns, may reflect prenatal or at-delivery hypoxia or depletion of oxygen that can cause brain injury;41 however, only a small portion of all CP cases were affected by hypoxia at birth thus other etiological factors of CP need to be explored.22 Premature delivery and intrauterine growth restriction are two other strong clinical predictors of fetal brain injury and CP risk.3,5,7 In our study, we showed that all three of the adverse birth outcome conditions remained strongly associated with CP risk after implementing a sibling matched design. Adverse birth outcomes may serve as markers indicating impaired placental or fetal development in pregnancy that are predictive of CP risk.42 In fact, the observed smaller estimated associations in the sibling analyses compared to the full cohort may suggest that some “upstream” factors that influence adverse birth outcomes need to be further investigated in their roles on affecting CP development.
There are ongoing research interests in investigating prenatal risk factors of CP, but research evidence regarding maternal prenatal complications and offspring CP risk has been inconclusive. Maternal perinatal inflammation or infection has been hypothesized as a major pathological mechanism of CP, possibly through the increased levels of cytokines that may predispose the developing brain to oxygen deprivations,8 and that inflammatory and immune responses could cause fetal white matter damage.24 In our study, maternal chorioamnionitis was associated with an increased risk of CP in the full cohort, while the positive point estimate was smaller using the sibling design and the result became imprecise. Current evidence regarding maternal preeclampsia9,11–13,26 or maternal pre-pregnancy obesity15,27,28 and offspring CP risk was inconsistent. The positive associations for maternal preeclampsia and maternal pre-pregnancy obesity that were found in our full cohort were attenuated towards null in the sibling analyses. One explanation for such differences is that some unobserved maternal and/or familiar factors that shared by siblings confounded the associations found in the full cohort. However, the largely diminished statistical power in the sibling comparison design by analyzing the discordant pairs only (~0.06% of the full cohort) may also explain the null findings.
CP is a lifelong condition with no cure available, making early risk identification and prevention of CP a public health priority. CP is more common among families with a lower socioeconomic status, and caring for a CP patient would add additional economic burdens. Using sibling design, our study confirmed that several adverse perinatal events remain robustly associated with childhood CP. Therefore, continuous endeavors in identifying the mechanism and modifiable risk factors that lead to these adverse perinatal events are paramount to inform prevention. While preterm birth and other adverse perinatal events are strongly associated with CP occurrence, the majority of the CP cases were born without these conditions, suggesting other unidentified risk factors are involved in CP etiology that remain to be discovered.
Like any study designs the sibling comparison approach also operates on assumptions and limitations that need to be carefully considered.19–21 In our study, we found bias due to carryover effect may disproportionally influence sibling analyses of maternal lifestyle-related factors such as smoking and pre-pregnancy BMI. Specifically, when the exposure or CP status of the prior pregnancy influence behavioral changes of the mothers in the next pregnancy, the causal relationship relying on comparing the discordant siblings may be distorted.21 One strategy to address this issue is to study only the younger sibling being exposed to the pregnancy factor, but other issues such as confounding by time-trend and parity may arise. In practice, the statistical precision would be further diminished when breaking down siblings according to the order of exposure occurrence, highlighting trade-offs between confounding control and statistical efficiency using the sibling comparison design. In addition, results from sibling analyses might represent effect estimates for a different study population, known as lack of generalizability. However, in this study, our weighted analyses suggested that the measured sociodemographic differences between the sibling and the full cohort data did not explain away the differences in results. Finally, careful implementation is warranted when using different statistical approaches to analyze sibling data. While the between-within modelling approach36,37 and the conditional logistic regression derived similar effect estimates, the between-within modelling approach is computationally intensive and has a higher chance for model misspecification.
Strengths and limitations
Our statewide sibling design included a considerable large number of CP cases and their unaffected siblings. Our record-based study limits any influence from recall bias and self-selection participation. We were able to evaluate several prenatal and perinatal factors defined a priori from literature reviews that may influence CP development. Furthermore, we performed a range of several sensitivity analyses, including those that interrogate carryover bias or generalizability issues.
Our study has limitations. First, exposure misclassification bias is possible when using information extracted from the birth records.43 For maternal pregnancy complications, we lack details of these conditions including their severity or treatment effects. Second, the California DDS might under-capture mild CP cases that are less in need of services,16 however, the outcome specificity is likely high.18,29 Third, we cannot rule out the influence from non-shared or time-varying confounding factors that are not controlled for in our models. The direction and magnitude of these time-varying confounding are hard to predict, and they could be specific to the different perinatal exposure factor studied. Fourth, we focused on several major neonatal and pregnancy factors suspected to affect CP. Many other prenatal and perinatal factors such as maternal pre-pregnancy or pregnancy health conditions, social and environmental risks factors,18,44 might also play a casual role in CP etiology and required further investigations. Fifth, our study focused on overall CP which included heterogenous subtypes. Most of the CP cases analyzed were spastic CP (>70%), while other subtypes, including ataxic, hypertonic and mixed CP, represented a smaller subset. Therefore, our results were most applicable for spastic CP, and we lack power to examine other CP subtypes, especially in the discordant sibling analyses. Sixth, our generalizability test was only applicable to the weighted demographic characteristics, therefore other unknown factors that may differ between siblings and the full cohort may still contribute to non-generalizability of findings from the sibling data to the full cohort.
CONCLUSION
Using a statewide sibling design, we showed that preterm birth, small for gestational age, and low Apgar score at birth remain strongly associated with CP risk. Uncontrolled confounding may influence the associations between pregnancy complications and CP development. Our research also showed that while sibling design can provide valuable information to triangulate research evidence, a careful implementation and interpretation of findings admitting the study assumptions and limitations is warranted.
Supplementary Material
Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41390-025-04055-4.
IMPACT:
We conducted a population-based sibling comparison study to evaluate the associations between several prenatal and perinatal factors and cerebral palsy (CP). Preterm birth, small for gestational age, and low Apgar score at birth remained strongly associated with CP using the sibling comparison design.
The associations between several maternal pregnancy complications and CP were close to null in the sibling comparison design, raising the possibility of uncontrolled confounding bias in the cohort analyses.
We demonstrated that a sibling comparison design can provide valuable information to triangulate research evidence for CP etiology, but a careful implementation and interpretation of findings is warranted.
FUNDING
The study is supported by the National Institute of Environmental Health Sciences (NIEHS R01ES034756 and R21ES035485).
Footnotes
COMPETING INTERESTS
The authors declare no competing interests.
INFORMED CONSENT
This study protocol was approved by the official institutional review board in Yale University (IRB#2000028297) and the California Committee for the Protection of Human Subjects (Project#12-10-0861) and was exempted from informed consent requirement because there was no contact with human subject.
DATA AVAILABILITY
The data underlying this article cannot be shared to guarantee the anonymity of subjects. Code is available upon request.
REFERENCES
- 1.Colver A, Fairhurst C & Pharoah PO Cerebral palsy. Lancet (Lond., Engl.) 383, 1240–1249 (2014). Apr 5. [DOI] [PubMed] [Google Scholar]
- 2.Maenner MJ et al. Prevalence of cerebral palsy and intellectual disability among children identified in two U.S. National Surveys, 2011–2013. Ann. Epidemiol 26, 222–226 (2016). Mar. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Chen R et al. Impact of gestational age on risk of cerebral palsy: unravelling the role of neonatal morbidity. Int. J. Epidemiol 50, 1852–1863 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Ahmed A, Rosella LC, Oskoui M, Watson T & Yang S Prevalence and temporal trends of cerebral palsy in children born from 2002 to 2017 in Ontario, Canada: Population-based cohort study. Developmental Med. Child Neurol 65, 243–253 (2023). [DOI] [PubMed] [Google Scholar]
- 5.Korzeniewski SJ, Slaughter J, Lenski M, Haak P & Paneth N The complex aetiology of cerebral palsy. Nat. Rev. Neurol 14, 528–543 (2018). [DOI] [PubMed] [Google Scholar]
- 6.Oskoui M, Coutinho F, Dykeman J, Jetté N & Pringsheim T An update on the prevalence of cerebral palsy: a systematic review and meta-analysis. Dev. Med Child Neurol 55, 509–519 (2013). [DOI] [PubMed] [Google Scholar]
- 7.Reddihough DS & Collins KJ The epidemiology and causes of cerebral palsy. Aust. J. Physiother 49, 7–12 (2003). [DOI] [PubMed] [Google Scholar]
- 8.Bear JJ & Wu YW Maternal Infections During Pregnancy and Cerebral Palsy in the Child. Pediatr. Neurol 57, 74–79 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Razaz N et al. Pre-pregnancy and pregnancy disorders, pre-term birth and the risk of cerebral palsy: a population-based study. Int. J. Epidemiol 10.1093/ije/dyad106 (2023) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Strøm MS et al. Maternal Chronic Conditions and Risk of Cerebral Palsy in Offspring: A National Cohort Study. Pediatrics. 147 10.1542/peds.2020-1137 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Sun BZ, Moster D, Harmon QE & Wilcox AJ Association of Preeclampsia in Term Births With Neurodevelopmental Disorders in Offspring. JAMA Psychiatry 77, 823–829 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Strand KM et al. Mediators of the association between pre-eclampsia and cerebral palsy: population based cohort study. BMJ : Br. Med. J 347, f4089 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Badagionis M, et al. Preeclampsia and Cerebral Palsy in Offspring. Children (Basel). Mar 9 9 10.3390/children9030385 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Streja E et al. Congenital cerebral palsy and prenatal exposure to self-reported maternal infections, fever, or smoking. Am. J. Obstet. Gynecol 209, 332.e1–332.e10 (2013). 2013/10/01/. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Zhang J et al. Maternal obesity and risk of cerebral palsy in children: a systematic review and meta-analysis. Developmental Med. Child Neurol 61, 31–38 (2019). [DOI] [PubMed] [Google Scholar]
- 16.Wu YW et al. Racial, Ethnic, and Socioeconomic Disparities in the Prevalence of Cerebral Palsy. Pediatrics 127, e674–e681 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Tollånes MC, Wilcox AJ, Lie RT & Moster D Familial risk of cerebral palsy: population based cohort study. BMJ (Clin. Res. ed.) 349, g4294 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Liew Z, et al. Ambient Exposure to Agricultural Pesticides during Pregnancy and Risk of Cerebral Palsy: A Population-Based Study in California. Toxics. ;8 10.3390/toxics8030052 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Sjölander A, Frisell T & Öberg S Sibling Comparison Studies. Annu. Rev. Stat. Its Application 9, 71–94 (2022). [Google Scholar]
- 20.Keyes KM & Susser E Uses and misuses of sibling designs. Int. J. Epidemiol 52, 336–341 (2022). [DOI] [PubMed] [Google Scholar]
- 21.Sjölander A, Frisell T, Kuja-Halkola R, Öberg S & Zetterqvist J Carryover Effects in Sibling Comparison Designs. Epidemiology 27, 852–858 (2016). [DOI] [PubMed] [Google Scholar]
- 22.Nelson KB & Blair E Prenatal Factors in Singletons with Cerebral Palsy Born at or near Term. N. Engl. J. Med 373, 946–953 (2015). [DOI] [PubMed] [Google Scholar]
- 23.Pappas A & Korzeniewski SJ Long-Term Cognitive Outcomes of Birth Asphyxia and the Contribution of Identified Perinatal Asphyxia to Cerebral Palsy. Clin. Perinatol 43, 559–572 (2016). [DOI] [PubMed] [Google Scholar]
- 24.Shi Z, et al. Chorioamnionitis in the Development of Cerebral Palsy: A Meta-analysis and Systematic Review. Pediatrics. 139 10.1542/peds.2016-3781 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Ahmed A, Rosella LC, Oskoui M, Watson T & Yang S In utero Exposure to Maternal Diabetes and the Risk of Cerebral Palsy: A Population-based Cohort Study. Epidemiology 34, 247–258 (2023). [DOI] [PubMed] [Google Scholar]
- 26.Trønnes H, Wilcox AJ, Lie RT, Markestad T & Moster D Risk of cerebral palsy in relation to pregnancy disorders and preterm birth: a national cohort study. Dev. Med Child Neurol 56, 779–785 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Villamor E et al. Association Between Maternal Body Mass Index in Early Pregnancy and Incidence of Cerebral Palsy. JAMA 317, 925–936 (2017). [DOI] [PubMed] [Google Scholar]
- 28.Forthun I, et al. Maternal Prepregnancy BMI and Risk of Cerebral Palsy in Offspring. Pediatrics. 138 10.1542/peds.2016-0874 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Zhuo H, Ritz B, Warren JL & Liew Z Season of Conception and Risk of Cerebral Palsy. JAMA Netw. Open 6, e2335164 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Centers for Disesae Control and Prevention. National Program of Cancer Registries. Accessed https://www.cdc.gov/national-program-cancer-registries/registry-plus/link-plus.html (2023).
- 31.Department of Developmental Services. Client Developmental Evaluation Report, Diagnotic Element. https://www.dds.ca.gov/transparency/cder/(2015).
- 32.von Ehrenstein OS, Cui X, Yan Q, Aralis H & Ritz B Maternal Prenatal Smoking and Autism Spectrum Disorder in Offspring: A California Statewide Cohort and Sibling Study. Am. J. Epidemiol 190, 728–737 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Chen R et al. Impact of gestational age on risk of cerebral palsy: unravelling the role of neonatal morbidity. Int. J. Epidemiol 50, 1852–1863 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Tita AT & Andrews WW Diagnosis and management of clinical chorioamnionitis. Clin. Perinatol 37, 339–354 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Social Vulnearability Index. Agency for Toxic Substances and Distance Registry, Centers for Disease Control and Preventino (CDC/ATSDR). Accessed https://www.atsdr.cdc.gov/placeandhealth/svi/index.html (2022). [Google Scholar]
- 36.Sjölander A, Frisell T & Öberg S Causal interpretation of between-within models for twin research. Epidemiol. Methods 1, 217–237 (2012). [Google Scholar]
- 37.Begg MD & Parides MK Separation of individual-level and cluster-level covariate effects in regression analysis of correlated data. Stat. Med 22, 2591–2602 (2003). [DOI] [PubMed] [Google Scholar]
- 38.Esen B, Ehrenstein V, Petersen I, Sørensen HT, Pedersen L Understanding the impact of non-shared unmeasured confounding on the sibling comparison analysis. International journal of epidemiology. Feb 1;53 10.1093/ije/dyad179 (2024). [DOI] [PubMed] [Google Scholar]
- 39.Westreich D, Edwards JK, Lesko CR, Stuart E & Cole SR Transportability of Trial Results Using Inverse Odds of Sampling Weights. Am. J. Epidemiol 186, 1010–1014 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Webster-Clark MA, Sanoff HK, Stürmer T, Peacock Hinton S & Lund JL Diagnostic Assessment of Assumptions for External Validity: An Example Using Data in Metastatic Colorectal Cancer. Epidemiology 30, 103–111 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Persson M, Razaz N, Tedroff K, Joseph KS & Cnattingius S Five and 10 min Apgar scores and risks of cerebral palsy and epilepsy: population based cohort study in Sweden. BMJ (Clin. Res. ed.) 360, k207 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Goldenberg RL, Culhane JF, Iams JD & Romero R Epidemiology and causes of preterm birth. Lancet (Lond., Engl.) 371, 75–84 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Frisell T, Öberg S, Kuja-Halkola R & Sjölander A Sibling comparison designs: bias from non-shared confounders and measurement error. Epidemiology 23, 713–720 (2012). [DOI] [PubMed] [Google Scholar]
- 44.Petridou E et al. Diet during pregnancy and the risk of cerebral palsy. Br. J. Nutr 79, 407–412 (1998). [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data underlying this article cannot be shared to guarantee the anonymity of subjects. Code is available upon request.