The Influence of Prenatal Depression on Language Development in Infants

The aim of the study

This study researches whether prenatal depression influences language development in infants at 6 months of age. In its research it also controls postnatal depression and mother’s age.

Research Questions

  • Do higher scores on EPDS prenatally influence language scores at 6 months of age?

  • Do higher scores on EPDS prenatally influence language scores at 6 months of age together with other factors, such as postnatal depression and age?

Hypothesis:

H1: Higher average score on EPDS prenatally results in lower language scores at 6 months of age for infants.

H1: The negative effect on language scores due to higher prenatal EPDS scores is moderated by postnatal EPDS scores, prenatal stress or maternal age, such that the negative effect decreases as maternal age increases.

Statistical Hypothesis:

H0_1: There is no effect of prenatal EPDS score on the language score.

H0_2: There is no interaction effect between prenatal EPDS scores, prenatal stress or maternal age on language scores at 6 months of age.

Instruments

EPDS : The Edinburgh Postnatal Depression Scale

The written answers to EPDS tests taken prenatally and postnatally were converted into numeric scores. EPDS test was taken once at 34 SSW and at least once more (up to 6 times all together) prior to birth. An average for all the tests taken prenatally was taken as the final score (D1), which was later used in the statistics. The first three tests taken postnatally (once at 2 weeks, 8 weeks, and 6 months) were also processed and converted into an average score for the tests taken postnatally (D2). The individual EPDS tests scores showed high within-subject variance and low to moderate between-subject variance.

Bayley III: Bayley Scales of Infant and Toddler Development, 3rd Edition. The test was taken at 6 months of age.

Data Cleaning:

Several subjects had to be erased from the dataset due to missing values for the language score (Y) or for both, D2 (postnatal score) and Y. The erased subjects are F_069, F_070, F_071, F_072, F_062, F_044, F_020, F_017, F_008.

The variables

  • D1 - average score on EPDS questionnaire taken prenatally (there is great variability in the number of tests taken prenatally)

  • D2 - average score on EPDS questionnaire taken postnatally (at 2 weeks, 8 weeks, and 6 months postnatally)

  • Y - language score on Bayley III test (language subtest) taken at 6 months of age for infants

  • M_ALT - age of mother at birth

Statistics

{r}library(car) {r}library(GFD) {r}library(ggplot2) {r}library(pscl) {r}library(plotly) {r}library(rgl) {r}open3d() {r}library(lmtest)

getwd() # Path where the file will be downloaded

url <- "https://www.dropbox.com/scl/fi/dwseh1vfk26bb6y6q1fc6/data_prenatal-depression_language.txt?rlkey=ioj8tdad21rl16ylbei95dwrz&st=nkl5dk8y&dl=1"

download.file(url, "my_file.txt")

data <- read.table("my_file.txt", header = TRUE, dec = ",", sep = "\t", na.strings = c("n/a", "na", " ", ""))

Descriptive Statistics

Sample

Mothers

  • mean age: 33,8; median age: 33,75; min age: 23,41; max age: 45,82

{r}summary(data$M_ALT) boxplot(data$M_ALT) {r}dens=density(data$M_ALT) {r}plot(dens$x,length(data)\*dens\$y,type="l",xlab="Age of Mother",ylab="Count estimate")

{r}library(ggplot2)}

{r}p_M <- ggplot(data, aes(x = M_ALT, y = Y, color = "cyl", size = "cyl")) + geom_point(alpha = 0.7) + theme_minimal() + theme(legend.position = "none")}

{r}ggMarginal(p_M, type = "boxplot")}

{r}ggMarginal(p_M, margins = "x", type = "densigram", col = "blue", fill = "pink")}

Infants

Weight at birth (in grams)

  • mean: 3343; median: 3300; min: 2170; max: 4740

{r}summary(data$Geburtsgewicht) boxplot(data$Geburtsgewicht)

Infant weight at birth
Infant weight at birth

Size at birth (in cm):

  • mean: 51,56; median: 51,00; min: 48,00; max: 58,00

{r}summary(data$Geburtgrosse) boxplot(data$Geburtgrosse)

Apgar score:

  • mean: 9,181; median: 9,667; min: 3,667; max: 10

{r}summary(data$APGAR, rm.na = TRUE)

boxplot(data$APGAR, main="Apgar Score", xlab="Apgar score", ylab="Count estimate", rm.na = TRUE)

INDEPENDENT VARIABLE: D1 - prenatal depression

The questionnaire EPDS has a range of values from 0 to 30. It screens for depressive symptoms, and is not a diagnostic tool. It reports the likelihood for depression - values from 0 - 9 reflect a low risk of depression, values from 10 - 12 reflect a moderate risk, and values from 13-30 a heightened risk for underlying depression.

The sample group has values that reflect a low risk for depression overall.

  • mean value = 3.87, median = 3,25

{r}summary(data$D1, na.rm = TRUE) {r}var(data$D1, na.rm = TRUE)

  • confidence interval (CI) for D1 = 3,228 - 4,516

{r}resultd1 <- t.test(data$D1) confidence_interval_d1 <- resultd1$conf.int {r}confidence_interval_d1

  • IQR - half of all subjects have a value of 3,25 or over. The “average 50%” are between 2,2 and 5,67.

{r}quantile(data$D1, probs = c(0.5, 0.25, 0.75), type = 1, na.rm = TRUE)

  • standard deviation: 95% of all subjects scored 2*2,56 over or below the median. This suggests little variance in the data for EPDS scores prenatally.

{r}sd(data$D1, na.rm = TRUE)

  • line histogram: the histogram does not display normal distribution

{r}dens=density(data$D1) plot(dens$x,length(data)*dens$y,type="l",xlab="average EPDS scores prenatally",ylab="Count estimate")

  • boxplot

{r}boxplot(data$D1)

{r}p_D1 <- ggplot(data, aes(x = D1, y = Y, color = "cyl", size = "cyl")) + geom_point(alpha = 0.7) + theme_minimal() + theme(legend.position = "none")}

{r}ggMarginal(p_D1, type = "boxplot")}

{r}ggMarginal(p_D1, type = "boxplot", xparams = list(binwidth = 1, fill = "orange"))}

  • Non-normality confirmed by Shapiro-Wilk Normality test (p-value = .010)

{r}shapiro.test(data$D1)

CONTROL VARIABLE: D2 - postnatal depression

  • mean value D2 = 4.32, median = 3,67 {r}summary(data$D2, na.rm = TRUE)

  • IQR - half of all subjects have a value of 3,67 or over. The “average 50%” are between 2,0 and 6,0

{r}quantile(data$D2, probs = c(0.5, 0.25, 0.75), type = 1, na.rm = TRUE)

  • Standard deviation (SD): 95% of all subjects scored 2*2,87 over or below than the median. There is some variance in the EPDS scores postnatally.

{r}sd(data$D2, na.rm = TRUE)

  • Boxplot

{r}boxplot(data$D2)

{r}p_D2 <- ggplot(data, aes(x = D2, y = Y, color = "cyl", size = "cyl")) + geom_point(alpha = 0.7) + theme_minimal() + theme(legend.position = "none")}

{r}ggMarginal(p_D2, type = "boxplot")}

{r}ggMarginal(p_D2, type = "boxplot", xparams = list(binwidth = 1, fill = "yellow"))}

  • Line histogram: the histogram displays a non-normal distribution of data. Non-normality was confirmed by Shapiro-Wilk Normality test in the next step.

{r}dens2=density(data$D2) plot(dens2$x,length(data)\*dens2\$y,type="l",xlab="D2",ylab="Count estimate")

  • Shapiro-Wilk Normality test - confirms non-normal distribution of values for D2 (p-value = .011) {r}shapiro.test(data\$D2)

DEPENDENT VARIABLE: Y - Language Score

  • mean value Y = 115,13, median = 114

{r}summary(data$Y, na.rm = TRUE)

  • confidence interval for Y = 112,487 - 117,767

{r}resultY \<- t.test(data$Y) confidence_interval_y <- resultY$conf.int confidence_interval_y

  • IQR - half of all subjects have a value of 114 or over. The “average 50%” are between 109 and 123

{r}quantile(data\$Y, probs = c(0.5, 0.25, 0.75), type = 1, na.rm = TRUE)

  • standard deviation: 95% of all subjects scored 2*10,48 over or below than the median

{r}sd(data\$Y, na.rm = TRUE)

  • line histogram - data appears normally distributed,normality was confirmed in the next step by Shapiro-Wilk Normality test

{r}densY=density(data$Y) plot(densY$x,length(data)\*densY\$y,type="l",xlab="Y",ylab="Count estimate")

{r}p_D1 <- ggplot(data, aes(x = D1, y = Y, color = "cyl", size = "cyl")) + geom_point(alpha = 0.7) + theme_minimal() + theme(legend.position = "none")}

{r}ggMarginal(p_D1, type = "boxplot")}

{r}ggMarginal(p_D1, type = "boxplot", yparams = list(binwidth = 1, fill = "red"))}

  • Shapiro-Wilk Normality test - confirms normal distribution of values for Y (p-value = .209

{r}shapiro.test(data\$Y)

CONTROL VARIABLE: M_ALT - Age of mother at birth

{r}summary(data$M_ALT)}

{r}sd(data$M_ALT)}

{r}shapiro.test(data$M_ALT)}

{r}p_M <- ggplot(data, aes(x = M_ALT, y = Y, color = "cyl", size = "cyl")) + geom_point(alpha = 0.7) + theme_minimal() + theme(legend.position = "none") ggMarginal(p_M, type = "boxplot")}

{r}ggMarginal(p_M, margins = "x", type = "densigram", col = "blue", fill = "pink")'}

Inferential Statistics

GENERALIZED LINEAR MODEL (GLM)

Generalized linear model was chosen to analyze independent variable Y (normally distributed) and D1, D2, SVF.NEG (non-normally distributed) and M_ALT. The analysis was done with D2, SVF_NEG and M_ALT as a way to control the variables that could potentially interact with D1.

Checklist for GLM:

{r}model2 <- glm(Y ~ D1*M_ALT, data = data)}

  1. Linearity: the relationship between independent variable and dependent variables should be linear

  2. Independence: observations should be independent from one another

    {r}dw_test_result <- dwtest(model)}

    {r}print(dw_test_result)}

  3. Homoscedasticity: the residuals should have constant variance at all levels of the independent variables

    {r}bptest(model)}

  4. Normality of residuals

    {r}residuals <- residuals(model)}

    {r}qqnorm(residuals)}

    {r}qqline(residuals, col = "red")}

  5. No multicollinearity

    {r}vif(model)}

Multiple linear regression model 1 - Y, D1, M_ALT

with control for age of mother

{r}model1 <- glm(Y ~ D1*M_ALT, data = data)}

{r}vif(model1)}

Due to the high multicolinearity between the independent variables, the variables needed to be centred:

{r}data$D1_centered <- data$D1 - mean_D1}

{r}data$M_ALT_centered <- data$M_ALT - mean_M_ALT}

{r}model_centered <- glm(Y ~ data$D1_centered * data$M_ALT_centered, data = data)}

{r}summary(model_centered)}

{r}pR2(model_centered)['McFadden']}

{r}confint(model_centered)}

There is a light interaction effect between prenatal depression, age of mother and language (p = .036, pseudo-R2 = .015, Std. Error = .12, 95% CI [0.024, 0.508])

Interaction Effect between D1 and M_ALT:

For a mother who is one year younger than the mean age, each additional EPDS point decreases the child’s language score by approximately 0.26607 points. If the mother’s depression score is two points above the mean, the interaction effect adds approximately -0.53214 units to the language score. Vice versa for mothers above the mean age, where an additional EPDS point increases language score.

Multiple linear regression model 2 - Y, D1, D2

with control for EPDS postnatally

{r}model2 <- glm(Y ~ D1*D2, data = data)}

{r}summary(model2)}

{r}confint(model2)}

There is no significant influence of prenatal depression on infant’s language development in combination with postnatal depression (p = .800; pseudo-R2 = .006, Std. Error = .14, 95% CI [-0.300, 0.231].)

Linear regression model 4- Y and D1

{r}model3 <- glm(Y ~ D1, data = data)}

{r}summary(model3)}

{r}confint(model3)}

There is no significant influence of prenatal EPDS scores on infant’s language development independently (p = .125; pseudo-R2 = .005, Std. Error = 0.51, 95% CI [-0.208, 1.810] ).

VISUALIZATION

  1. Plot - No interaction effects between EPDS prenatally, EPDS postnatally and language development

{r}plot3d(model2, plane.col='red')}

{r}title3d("Multiple Linear Regression 3D Plot")}

{r}play3d(spin3d(axis = c(0, 0, 1)), duration = 30)}

GNU Octave:

  1. Plot - Interactions between EPDS prenatally, age of mother, and language

{r}plot3d(model1, plane.col='red')}

{r}title3d("Multiple Linear Regression 3D Plot")}

{r}play3d(spin3d(axis = c(0, 0, 1)), duration = 30)}

Light interaction effects between prenatal depression, age of mother and language development for infants.

GNU Octave:

  1. Plot - No Interaction between EPDS scores prenatally and language development independently

{r}plot3d(model3, plane.col='red')}

{r}title3d("Multiple Linear Regression 3D Plot")}

{r}play3d(spin3d(axis = c(0, 0, 1)), duration = 30)}

GNU Octave:

Conclusion

The interaction reveals that the effect of prenatal depression on a child’s language ability is moderated by the age of the mother.

  • Infants of mothers below the mean age are negatively impacted by higher EPDS scores.

  • Infants of mothers at the mean age or around the mean age are not impacted by EPDS scores

  • Infants of mothers above the mean age are positively impacted by higher EPDS scores.

The effect, however, should be interpreted cautiously.

Limitations: Participants exhibited low scores on EPDS questionnaire; the outcome of the study is thus limited in its conclusion. Further studies with a more diverse sample are required.

Literature

Apgar, V. (1953). A proposal for a new method of evaluation of the newborn infant. Current Researches in Anesthesia and Analgesia, 32(4), 260-267. https://doi.org/10.1213/00000539-195301000-00041

Bayley, N (2005). Bayley Scales of Infant and Toddler Development, Third Edition (Bayley--III®) [Database record]. APA PsycTests. https://doi.org/10.1037/t14978

Eaton J.W., Bateman D., Hauberg S., & Wehbring R. (2019). GNU Octave version 5.2.0 manual: a high-level interactive language for numerical computations. https://www.gnu.org/software/octave/doc/v5.2.0/

Evans, J., Melotti, R., Heron, J., Ramchandani, P., Wiles, N., Murray, L., & Stein, A. (2011). The timing of maternal depressive symptoms and child cognitive development: a longitudinal study. Journal of Child Psychology and Psychiatry, 53(6), 632–640. https://doi.org/10.1111/j.1469-7610.2011.02513.x

Goodman, S. H., Rouse, M. H., Long, Q., Ji, S., & Brand, S. R. (2011). Deconstructing antenatal depression: What is it that matters for neonatal behavioral functioning? Infant Mental Health Journal, 32(3), 339–361. https://doi.org/10.1002/imhj.20300

Ibanez, G., Bernard, J. Y., Rondet, C., Peyre, H., Forhan, A., Kaminski, M., & Saurel-Cubizolles, M.-J. (2015). Effects of Antenatal Maternal Depression and Anxiety on Children’s Early Cognitive Development: A Prospective Cohort Study. PLOS ONE, 10(8), e0135849. https://doi.org/10.1371/journal.pone.0135849

O’Leary, N., Jairaj, C., Molloy, E. J., McAuliffe, F. M., Nixon, E., & O’Keane, V. (2019). Antenatal depression and the impact on infant cognitive, language and motor development at six and twelve months postpartum. Early Human Development, 134, 41–46. https://doi.org/10.1016/j.earlhumdev.2019.05.021

Rogers, A., Obst, S., Teague, S. J., Rossen, L., Spry, E. A., Macdonald, J. A., Sunderland, M., Olsson, C. A., Youssef, G., & Hutchinson, D. (2020). Association Between Maternal Perinatal Depression and Anxiety and Child and Adolescent Development. JAMA Pediatrics, 174(11). https://doi.org/10.1001/jamapediatrics.2020.2910

Sohr-Preston, S. L., & Scaramella, L. V. (2006). Implications of Timing of Maternal Depressive Symptoms for Early Cognitive and Language Development. Clinical Child and Family Psychology Review, 9(1), 65–83. https://doi.org/10.1007/s10567-006-0004-2

RStudio Team (2020). RStudio: Integrated Development for R. RStudio, PBC, Boston, MA, http://www.rstudio.com