Young Maternal Age and Poor Child Development: Predictive Validity From a Birth Cohort Catherine R.

Chittleborough, Debbie A. Lawlor and John W. Lynch Pediatrics 2011;127;e1436; originally published online May 2, 2011; DOI: 10.1542/peds.2010-3222

The online version of this article, along with updated information and services, is located on the World Wide Web at:
http://pediatrics.aappublications.org/content/127/6/e1436.full.html

PEDIATRICS is the official journal of the American Academy of Pediatrics. A monthly publication, it has been published continuously since 1948. PEDIATRICS is owned, published, and trademarked by the American Academy of Pediatrics, 141 Northwest Point Boulevard, Elk Grove Village, Illinois, 60007. Copyright 2011 by the American Academy of Pediatrics. All rights reserved. Print ISSN: 0031-4005. Online ISSN: 1098-4275.

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

Young Maternal Age and Poor Child Development: Predictive Validity From a Birth Cohort
WHATS KNOWN ON THIS SUBJECT: Teen-aged mothers and their children are targeted by policies and programs aimed at improving child development. Teenage motherhood may be a risk factor for poor childhood development, but it may be an inaccurate criterion for predicting risk of developmental outcomes. WHAT THIS STUDY ADDS: To reach the goal of improving child development outcomes across the population, factors such as maternal education level, nancial difculties, smoking, and depression during pregnancy should be considered in addition to young maternal age when recruiting women to preventive programs.
AUTHORS: Catherine R. Chittleborough, PhD,a,b Debbie A. Lawlor, PhD, MB, ChB,a,c and John W. Lynch, PhDa,b,d
aSchool of Social and Community Medicine, University of Bristol, Bristol, United Kingdom; bSchool of Population Health and Clinical Practice, University of Adelaide, Adelaide, Australia; cMRC Centre for Causal Analysis in Translational Epidemiology, University of Bristol, Bristol, United Kingdom; and dSansom Institute for Health Research, Division of Health Sciences, University of South Australia, Adelaide, Australia

KEY WORDS ALSPAC, child development, maternal age, maternal health services, predictive value of tests ABBREVIATIONS ALSPACAvon Longitudinal Study of Parents and Children ADSALSPAC developmental scale SDQStrengths and Difculties Questionnaire SEASchool Entry Assessment EPDSEdinburgh Postnatal Depression Scale PPVpositive predictive value AUROCarea under the receiver operator characteristic curve All authors contributed to the conceptual development, analysis plan, and interpretation of results. Dr Chittleborough undertook the analyses and wrote the rst draft of the paper. Drs Lawlor and Lynch contributed later drafts. All authors approve the nal version to be published and take responsibility for the analyses of data collected and provided by ALSPAC and act as guarantors. www.pediatrics.org/cgi/doi/10.1542/peds.2010-3222 doi:10.1542/peds.2010-3222 Accepted for publication Feb 1, 2011 Address correspondence to Catherine R. Chittleborough, PhD, School of Social and Community Medicine, University of Bristol, Canynge Hall, Whatley Road, Bristol BS8 2PS United Kingdom. E-mail: catherine.chittleborough@bristol.ac.uk PEDIATRICS (ISSN Numbers: Print, 0031-4005; Online, 1098-4275). Copyright 2011 by the American Academy of Pediatrics FINANCIAL DISCLOSURE: The authors have indicated they have no nancial relationships relevant to this article to disclose.

abstract
OBJECTIVE: We aimed to examine the ability of mothers age, and other factors measured during pregnancy (education, nancial difculties, partner status, smoking, and depression), to predict child development outcomes up to age 5 years. METHODS: Data were obtained from the Avon Longitudinal Study of Parents and Children (ALSPAC). Poor child development was dened as scoring in the worst 10% of a parent-reported ALSPAC developmental scale (ADS) at 18 months (n 7546), the Strengths and Difculties Questionnaire (SDQ) at 47 months (n 8328), or teacher-reported School Entry Assessment (SEA) scores at 4 to 5 years (n 7345). RESULTS: Only a small proportion of children with poor development had mothers aged younger than 20 years at their birth (3.3%, 6.4%, and 9.2%, for the ADS, SDQ, and SEA, respectively). A greater proportion with each measure of poor development would be identied (48.9%, 63.6%, and 74.4%, respectively) if all 6 predictors were used and a woman had at least 1 of these. Model discrimination was poor using maternal age only (area under the receiver operator characteristic curve 0.5 for all 3 outcomes). This improved when all 6 predictors were included in the model (ADS: 0.56; SDQ: 0.66; SEA: 0.67). Calibration also improved with the model including all 6 predictors. CONCLUSIONS: Even if programs targeted at teen-aged mothers are successful in improving child development, they will have little impact on population levels of poor child development if young maternal age is the sole or main means of identifying eligibility for the program. Pediatrics 2011;127:e1436e1444

e1436

CHITTLEBOROUGH et al

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

ARTICLES

The highest priority recommendation in the recent Marmot Review to reduce health inequalities was to give every child the best start in life,1 with actions that are universal but with a scale and intensity that is proportionate to the level of disadvantage1 so that more support goes to those with greater need. This progressive universalism is challenging for early child development services because an accurate method of identifying those most in need has to be balanced against efcient use of limited resources and the risk of stigmatizing mothers and families who may be labeled as poor parents.27 Programs such as the Family Nurse Partnership in the United Kingdom,8 Family Home Visiting in Australia,9 and Nurse Family Partnership in the United States10 offer extended services beyond immediate postnatal contact to vulnerable families often identied by maternal age, with mothers aged 20 years eligible for the program.8,9 Although teen-age motherhood can be an important risk factor for poor childhood development,11,12 it may be an inaccurate predictor of developmental risk in children.13 Other risk factors in combination might improve identication of those at greatest risk of poorer development. If such additional factors were routinely obtainable in the antenatal or early postnatal period, they would increase the ability to effectively target limited program resources to those most likely to benet. The present study examines the predictive validity of young maternal age (20 years) and 5 other factors (maternal education, nancial difculties, partner status, smoking during pregnancy, and depression) in predicting poor development in childhood up to age 5 years.

ALSPAC pregnancies among women resident in Avon area of southwest England with an expected delivery date between April 1, 1991 and December 31, 1992 N = 14 541 Excluded triplet and quadruplet births and women with missing data for age at last menstrual period N = 10 Eligible cohort N = 14 531 No data provided at 18 mo N = 5949 ADS data, 18 mo N = 8582 No data provided at 47 mo N = 5109 SDQ data, 47 mo N = 9422 Linked data unavailable N = 5195 SEA data, 45 y N = 9336

RESPONSE SAMPLE:

Missing data on covariates N = 1036 ADS data, 18 mo N = 7546

Missing data on covariates N = 1094 SDQ data, 47 mo N = 8328

Missing data on covariates N = 1991 SEA data, 45 y N = 7345

ANALYSIS SAMPLE:

Impute data on missing outcomes and covariates IMPUTED SAMPLE:

Impute data on missing outcomes and covariates

Impute data on missing outcomes and covariates

Multiply imputed analysis for participants with data on at least one of the three outcomes N = 12 570

FIGURE 1
Eligible cohort and numbers included for analyses.

study of children born to women resident in the Avon area of southwest England with an expected delivery date between April 1, 1991, and December 31, 1992. Details of the background, methods, recruitment, and response rates have been reported elsewhere (www.bristol.ac.uk/alspac/).14 The core ALSPAC sample consists of 14 541 pregnancies (Fig 1). Ethical approval was obtained from the ALSPAC Law and Ethics committee and local research ethics committees. Child Outcomes Child developmental abilities at 18 months were assessed using the ALSPAC developmental scale (ADS), created using items derived from the Denver Developmental Screening Test shown to be most predictive of developmental abnormality.15 Because many of the Denver items were designed to be observed by trained examiners, the ADS was adapted for parental report after focus group piloting with members of the ALSPAC cohort.

Parents reported whether their child could do 56 activities within 4 developmental domains (gross motor, ne motor, communication, and social skills). The number of passes, indicated by yes, can do well responses, was summed in each of the 4 subscales, and the total development score was summed across subscales. Age for completion of the ADS was restricted to an 8-week window around 18 months given the developmental agespecic nature of the questions.16 The parent version of the Strengths and Difculties Questionnaire (SDQ)17 was completed by the main caregiver (usually the mother) when the child was 47 months old, using a scale from 1 to 3 (does not apply, applies somewhat, denitely applies). The scale consisted of 25 items in 5 subscales (prosocial behavior, hyperactivity, emotional symptoms, conduct problems, and peer problems). A total difculties score was created by summing the scores from the last 4 subscales.
e1437

METHODS
The Avon Longitudinal Study of Parents and Children (ALSPAC) is a prospective, geographically representative
PEDIATRICS Volume 127, Number 6, June 2011

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

The prosocial subscale was excluded because it measures positive aspects of behavior. High scores on the SDQ have been shown to be predictive of psychiatric disorders among children.18 Children were rated by their teacher in the School Entry Assessment (SEA) during the rst half of their rst term in reception class at ages 4 to 5 years.19 This assessment is undertaken in all reception classes in England, and scores for ALSPAC participants were obtained through consented record linkage with data provided by local education authorities. Integer scores between 2 and 7 on each of 4 required scales (language, reading, writing, and mathematics) were summed to provide a total SEA score. Potential Predictors of Childhood Development Problems Age of mother at last menstrual period was obtained for 14 531 women (Fig 1) and dichotomized at younger than 20 years, the cutoff point commonly used to identify mothers eligible for programs.8,9 Highest education level was categorized into O level or higher (where O level is ordinary level examinations most commonly taken at 16 years of age, the legal minimum age for leaving school in the United Kingdom) and less than O level (Certicate of Secondary Education commonly taken at age 16 years by individuals considered to be unable to obtain an O level in that subject, a vocational qualication, or no educational qualications). The nancial difculties factor was assessed using 5 questions asking how difcult the mother found it to afford food, clothing, heating, rent or mortgage, and things she will need for the infant, with a score of 1 (very difcult) to 4 (not difcult) for each response. The algorithm for calculating the overall nancial difculties score was 20 minus the sum of the scores of each of the 5
e1438 CHITTLEBOROUGH et al

items, resulting in an overall score in which 0 represented no nancial difculties and 15 the maximum nancial difculties. Participants scoring 8 were dened as experiencing nancial difculties.20 Partner status at study enrollment (married or cohabitating versus no partner or not living with partner) and whether women had smoked during the rst 3 months of their pregnancy were assessed by using a questionnaire. Ten items that formed the depression scale of the Edinburgh Postnatal Depression Scale (EPDS)21,22 were administered via questionnaire at 18 to 20 weeks gestation. None of the 10 items is specic to the postnatal experience, and this scale has been validated for use postnatally and during pregnancy.2325 Each question had 4 response categories scored from 0 to 3 and referred to the feelings of the mother in the past week. A score 12 is used to indicate probable depressive disorder.21 Analysis Distributions of continuous outcome variables were skewed and so each distribution was dichotomized, with the lower tail (or upper for SDQ) containing 10% of those with the poorest child developmental outcome.16,17 We calculated the proportion of children with poor developmental outcomes whose mothers had each of the individual binary predictive factors, and also whose mothers had at least 1 and at least 2 of the 6 binary predictors. Specicity, positive predictive value (PPV) and likelihood ratio of each binary predictor were calculated (Supplemental Table 5). Univariable and multivariable (with mutual adjustment for all other predictors) logistic regression examined associations of predictors with each child outcome. The predicted probability of poor child development was calculated from these regression models. In clinical

practice, the predictors would likely be used as binary variables, but because calibration statistics cannot be easily interpreted using a single binary predictor, maternal age, nancial difculties, and EPDS score were included as continuous variables in the prediction models. The area under the receiver operator characteristic curve (AUROC) was used to assess the discriminatory capability of the models, or how accurately each model separates mothers into those with and without children with poor outcomes. Model 1 contained only maternal age; model 2 included all 6 predictors. An AUROC of 1 represents a model that perfectly discriminates the outcome; an AUROC of 0.5 represents a prediction tool that is no better than chance at identifying those at risk. AUROCs were also calculated using all binary predictor variables, as would be more commonly used in clinical practice, and these were lower but consistent with predictive models that included continuous variables (data not shown). Calibration of the 2 models, or the agreement between observed and predicted outcomes, was assessed by ranking mothers into deciles of their predicted risk and comparing the predicted to observed proportion within each decile. The Hosmer-Lemeshow goodness-of-t 2 statistic was used to test the accuracy of calibration.26 This statistic tests the null hypothesis that the predicted proportion equals the observed proportion within ranked groupings (deciles) of predicted risk. A high P value suggests good calibration of predicted and observed risk. The integrated discrimination improvement27 for model 2 compared with model 1 was also calculated. This assesses discrimination without relying on cutoff points and compares the average difference in predicted risk for women whose children have poor development with women whose chil-

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

ARTICLES

TABLE 1 Prevalence and Amount of Data Available for Each Child Outcome and Potential Predictor
Measured During Pregnancy
Time of Data Collection Response Samplea N Child outcome ADS SDQ SEA Predictor Age 20 y No partner or not cohabitating Financial difculties Depression Smoking in rst 3 mo of pregnancy Education O level At least 1 of the 6 predictors At least 2 of the 6 predictors
a

Analysis Sample, %b ADS (n 7546) 10.0 3.8 6.4 SDQ (n 8328) 11.4 3.1 6.0 SEA (n 7345) 8.5 4.9 7.5

Imputed Sample, %c (N 12 570)

18 mo 47 mo 45 y Study enrollment Study enrollment

8582 10.0 9422 11.6 9336 10.1 14 531 13 485 6.6 8.8

10.2 12.7 9.5 6.1 8.7

mothers had each predictive factor. A small proportion of children with poor development had mothers aged 20 (3.3%, 6.4%, and 9.2% for ADS, SDQ, and SEA, respectively). High proportions of poor development could be identied if information on all 6 predictors was used and a woman had at least 1 of these predictors (48.9%, 63.6%, and 74.4% for ADS, SDQ, and SEA). The predictor that alone identied the highest proportion of each child outcome was mothers low education. Table 3 shows univariable and multivariable associations between the potential predictors and child developmental outcomes. Associations between the potential predictors and child outcomes using the multiply imputed data set (Supplemental Table 6) were consistent with analyses of complete cases. Table 4 shows calibration and discrimination for both models. Discrimination was poor using model 1 (maternal age only). This nding improved a little for ADS and more so for SDQ and SEA when all 6 predictors were used in model 2. AUROC values calculated using the multiply imputed data set (Supplemental Table 7) were consistent with complete case analyses. The Hosmer-Lemeshow goodness-of-t tests indicated better calibration using model 2 than model 1 for SDQ and SEA, whereas both models showed good calibration for ADS. Model 1 underesti-

32 weeks gestation 1820 weeks gestation 1820 weeks gestation 32 weeks gestation

12 011 10.0 12 177 13.9 13 189 25.1

9.1 12.1 21.8

8.4 11.6 20.6

10.5 13.1 24.0

9.5 13.4 24.9

12 340 30.1 10 955 51.2 10 955 22.8

28.1 50.0 21.0

24.7 46.5 19.0

32.3 54.3 24.5

30.9 53.9 25.1

Response sample is the number who responded to specic questionnaire/assessment for each child outcome or predictor. b Analysis sample includes respondents with complete data on the relevant child outcome and all 6 predictors. c Imputed sample includes data imputed on child outcomes or predictors for participants who provided data on at least 1 of the 3 child outcomes.

dren do not have poor development. The integrated discrimination improvement is greater when the second model correctly assigns individuals to higher or lower probabilities of having the outcome in comparison to the rst model. Missing Data Sensitivity analyses were conducted on an imputed data set to examine the inuence of missing data on the ndings. Multiple imputation by chained equation was used to impute missing data on child outcomes and predictors for respondents who had data on at least 1 child outcome (N 12 570; Fig 1) using the ice command in Stata (Stata Corp, College Station, TX).28 The imputation model included all child outcomes and predictors as well as predictors of missingness (birth weight, parity, social class, ethnicity,
PEDIATRICS Volume 127, Number 6, June 2011

and reaction to pregnancy). We generated 20 data sets and undertook 20 cycles of regression switching.28 Table 1 shows the prevalence and amount of data available for each child outcome and predictor.

RESULTS
Table 2 displays the proportion of children with poor development whose

TABLE 2 Proportion of Child Outcome Cases That Would be Detected With Potential Predictors
Measured During Pregnancy
Predictor Measured During Pregnancy Age 20 y No partner or not cohabitating Financial difculties Depression Smoking Education O level At least 1 of the 6 predictors At least 2 of the 6 predictors ADS 18 Months (N 7546; n cases 755), % 3.3 5.0 9.1 14.3 17.7 29.0 48.9 19.5 SDQ 47 Months (N 8328; n cases 946), % 6.4 9.1 16.0 22.3 31.9 34.5 63.6 35.1 SEA 45 Years (N 7345; n cases 621), % 9.2 12.2 17.7 19.0 34.6 55.4 74.4 44.1

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

e1439

TABLE 3 Univariable and Multivariable Associations of Potential Predictors With Child

Developmental Outcomes
Predictor Measured During Pregnancy ADS 18 Months (N 7546; n cases 755) OR (95% CI) Univariable Age group 20 20 Partner status Married/cohabitating No partner/not cohabitating Financial difculties score 9 9 EPDS score 12 12 Smoked in rst 3 mo of pregnancy No Yes Highest education level O level O level Multivariablea Age group 20 20 Partner status Married/cohabitating No partner/not cohabitating Financial difculties score 9 9 EPDS score 12 12 Smoked in rst 3 mo of pregnancy No Yes Highest education level O level O level P SDQ 47 Months (N 8328; n cases 946) OR (95% CI) P SEA 45 Years (N 7345; n cases 621) OR (95% CI) P

1 0.86 (0.571.31) 1 0.76 (0.541.06)

1 1 .492 2.50 (1.863.36) .001 2.13 (1.592.87) .001 1 1 .110 1.69 (1.322.15) .001 1.82 (1.412.36) .001

1 1.01 (0.781.31) 1 1.25 (1.001.55)

1 1 .929 2.35 (1.932.85) .001 1.98 (1.592.48) .001 1 1 .046 2.53 (2.133.00) .001 1.63 (1.322.02) .001

1 0.76 (0.620.92) 1 1.05 (0.891.24)

1 1 .005 1.99 (1.712.31) .001 1.77 (1.482.11) .001 1 1 .568 1.72 (1.491.99) .001 2.88 (2.443.40) .001

1 0.95 (0.611.48) 1 0.78 (0.550.12)

1 .825 1.72 (1.252.37) 1 .181 0.98 (0.751.28)

1 .001 1.36 (0.981.88) 1 .869 1.11 (0.841.48)

.062

.465

1 1 1 1.02 (0.781.33) 0.887 1.73 (1.412.12) .001 1.54 (1.221.94) .001 1 1 1 1.33 (1.061.66) 0.013 2.02 (1.692.42) .001 1.27 (1.021.59)

.036

1 1 1 0.73 (0.600.90) 0.003 1.55 (1.321.81) .001 1.29 (1.071.56)

.007

1 1 1 1.10 (0.930.31) 0.246 1.44 (1.241.67) .001 2.56 (2.153.04) .001

OR indicates odds ratio; CI, condence interval. a Mutually adjusted for all other variables in table.

mated the likelihood of poor development according to SDQ and SEA among those at highest risk (Table 4 and Fig 2). Integrated discrimination improvements indicated that model 2 resulted in an improvement in calibration over model 1, particularly for SDQ and SEA with 3% of the children being core1440 CHITTLEBOROUGH et al

rectly reclassied by model 2 compared with model 1.

DISCUSSION
The main nding of our study is that programs, which have been shown to be effective at improving child development, will have little impact on child

development outcomes at the population level if young maternal age is used as the sole or main criterion for identifying eligible mothers. This nding is sound because, in general the proportion of births to women aged 15 to 19 years is low, at 6.1% of live births in England and Wales in 2009,29 10.2% of births in the United States in 200630 and 4.2% of births in Australia in 2008.31 Therefore, only a small proportion of children with poor development have teen-aged mothers. Maternal age 20 years identies only 9% of the cases of poor development at 5 years, whereas 74% of these cases would be identied among mothers with 1 of the 6 predictors, and 44% would be identied among mothers with 2 of the 6 predictors. If the 23% of women experiencing 2 of these characteristics could be engaged in programs aimed at supporting the development of their children, 44% of cases of poor child development at school entry could be identied and potentially prevented. Furthermore, a model including all predictors provides better discrimination and calibration for predicting these child outcomes than a model based solely on mothers age. Mothers low education is the single characteristic that accounts for the highest proportion of cases of poor child development, from almost 30% at 18 months to 58% at 4 to 5 years. Low education is more common than teen-age motherhood in the population, and parents with higher levels of education are thought to positively inuence their childrens academic achievement through use of more varied and complex language and reading interactions, and exposing children to increased educational opportunities.32 The strengths of this study are the large sample size and longitudinal design with inclusion of a large number of relevant predictors measured dur-

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

ARTICLES

TABLE 4 Calibration and Discrimination of the 2 Models

Model 1 Observed ADS 18 mo (n 7546) Lowest 10th 2nd 3rd 4th 5th 6th 7th 8th 9th Highest 10th Hosmer-Lemeshow 2 testa AUROC (95% CI) IDI, % (95% CI) SDQ 47 mo (n 8328) Lowest 10th 2nd 3rd 4th 5th 6th 7th 8th 9th Highest 10th Hosmer-Lemeshow 2a AUROC (95% CI) IDI, % (95% CI) SEA 45 y (n 7345) Lowest 10th 2nd 3rd 4th 5th 6th 7th 8th 9th Highest 10th Hosmer-Lemeshow 2a AUROC (95% CI) IDI, % (95% CI) 7.7 8.8 8.1 9.1 10.4 10.9 12.3 12.3 10.8 10.9 Predicted Ratio Observed 7.0 8.6 7.4 6.9 11.5 10.3 12.1 11.8 11.1 13.4 Model 2 Predicted Ratio

8.3 1.07 9.0 1.02 9.3 1.15 9.6 1.05 9.9 0.95 10.3 0.94 10.5 0.85 10.8 0.87 11.2 1.03 12.3 1.13 6.83, P .555 0.5395 (0.51830.5607)

6.8 0.98 8.1 0.94 8.8 1.19 9.3 1.35 9.7 0.85 10.1 0.98 10.6 0.88 11.1 0.95 11.9 1.07 13.6 1.01 12.35, P .262 0.5629 (0.54160.5842), P .020b 0.26 (0.150.37), P .001c 5.1 0.86 6.2 1.21 7.0 1.09 7.9 1.10 8.9 0.94 10.1 0.99 11.6 0.97 13.7 0.98 17.0 0.94 26.0 1.03 5.28, P .872 0.6600 (0.64120.6787), P .001b 3.13 (2.763.51) , P .001c 3.8 1.13 4.3 1.13 4.7 0.93 5.1 1.03 5.8 0.88 6.8 0.94 9.0 1.12 11.3 0.97 14.1 1.01 19.8 0.99 2.94, P .983 0.6732 (0.65090.6955), P .001b 2.84 (2.463.22) , P .001c 3.4 3.8 5.0 5.0 6.5 7.2 8.0 11.6 14.0 20.0 5.9 5.1 6.5 7.2 9.5 10.2 12.0 14.0 18.0 25.1

10.0 8.7 9.6 9.7 9.8 11.0 11.1 11.1 15.0 21.9

7.5 0.75 8.9 1.02 9.8 1.02 10.6 1.09 11.1 1.14 11.7 1.07 12.3 1.11 13.2 1.19 14.5 0.97 16.9 0.77 27.82, P .001 0.5691 (0.54880.5893) 5.9 0.75 6.9 1.05 7.4 1.16 7.9 1.22 8.4 1.02 8.8 1.15 9.2 1.15 9.8 0.89 10.6 1.00 12.1 0.86 16.52, P .036 0.5659 (0.54120.5907)

due to missing data. Societal changes (eg, downward trends in the proportion of births to young mothers,29 smoking during pregnancy,34 and increased participation of young people in higher education35) mean that our ndings do not necessarily generalize to more contemporary populations or those from other countries. However, our conclusion that young maternal age is likely to identify only a small proportion of children with developmental problems is likely to hold across most high-income countries where the prevalence of young maternal age is low. Differences in prediction between the SDQ and SEA, despite being measured at similar ages, reect the fact that children with behavioral difculties (as reported by parents) were not necessarily the same group as those with poor SEA scores (as assessed by teachers). Association of mothers age, and other factors, with child development outcomes have been shown previously11,3638 but this is the rst study, to our knowledge, to demonstrate discrimination, calibration, and sensitivity of these maternal factors in predicting child development. Such analyses may more directly inform targeting of programs to support parenting and child development. An analysis of the UK Millennium Cohort Study found that approximately one quarter of children with poor development and behavior would be identied among the 10% of children with the highest risk, as predicted by a model using many predictors, including depression, smoking during pregnancy, educational qualications, and socioeconomic position.38 Higher AUROC values (0.80) for predicting poor child development at age 5 years than we have found were subsequently reported.39 These prediction models included, in addition to the characteristics we examined, birth weight, gender, and breastfeeding,
e1441

7.8 6.6 6.4 6.4 8.2 7.7 8.0 10.9 10.6 14.1

Model 1, maternal age; Model 2, maternal age, highest education level O level, nancial difculties score, no partner or not cohabitating, smoked in rst 3 months of pregnancy, and EPDS score. IDI, integrated discrimination improvement. a P value tests null hypothesis that the predicted proportion equals the observed proportion within deciles. b P value tests null hypothesis that there is no difference in the AUROC of model 1 and model 2. c P value tests null hypothesis that IDI is not different from 0.

ing pregnancy. Self-reported smoking status may underestimate smoking prevalence among pregnant women,33 but self-reported smoking still contributed to the prediction of poor child development and reects the clinical situation in which pregnant women report their smoking status at antenatal consultations. Given that calibration cannot be assessed with a single
PEDIATRICS Volume 127, Number 6, June 2011

binary predictor, we used the continuous age variable, which may underestimate the poor calibration of maternal age with a cutoff 20 years, as is used in practice. Reduced power from cohort attrition is not a major problem in a study of this size, and analyses using multivariate multiple imputation produced similar results to complete case analyses, suggesting little bias

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

Model 1
.14 observed predicted .12 .12 .14 observed predicted

Model 2

.1

.1

.08

.08

.06 .08 .09 .1 .11 predicted (proportion) .12

.06 .06 .08 .1 predicted (proportion) .12 .14

ALSPAC developmental scale 18m

.25 observed predicted .25 observed predicted

.2

.2

.15

.15

.1

.1

.05 .08 .1 .12 .14 predicted (proportion) .16 .18

.05 .05 .1 .15 predicted (proportion) .2 .25

Strengths and Difficulties Questionnaire 47m

.2 observed predicted .2 observed predicted

.15

.15

.1

.1

.05

.05

.06

.08 .1 predicted (proportion)

.12

.05

.1 .15 predicted (proportion)

.2

School Entry Assessment 4-5y

FIGURE 2
Calibration plots of the observed and predicted probability of poor child developmental outcomes on the ADS, SDQ, and SEA for model 1 (maternal age) and model 2 (maternal age, highest education level O level, nancial difculties score, no partner or not cohabitating, smoked in rst 3 months of pregnancy, and EPDS score).

which we did not include because these factors would be unavailable for selecting women for programs during pregnancy.
e1442 CHITTLEBOROUGH et al

Although a broader range of risk factors may more accurately identify mothers whose children are at high risk of poor development, there are

many issues to consider. First, collection of all of the characteristics would need to be feasible in routine clinical settings and acceptable to pregnant

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

ARTICLES

women. Forms completed by Family Nurse Partnership nurses in the second-wave pilot sites in England had missing data for 8.8% of women on marital status, 9.3% on employment status, and 10.1% on education.40 Although responding to a research questionnaire is different compared with answering questions in a clinic setting, our study suggests that most pregnant women provide information on the characteristics we have examined. It is possible that the 15% to 17% who did not answer questions about education, depression, or nancial difculties are those with children at particular high risk of childhood development problems. Second, a simple tool would be needed for using the collected data and generating a risk score for each individual. This could range from a simple checklist of predictors in which, for example, women with 2 of the binary predictors are considered for interventions, through to computer-based tools that make use of predictive risk algorithms containing continuous variables. The former is likely to be feasible in most settings; the latter is becoming increasingly common, for example, in the prediction of cardiovascular risk. Third, although there is some randomized controlled trial evidence that these interventions improve outcomes for children of teen-aged mothers,4143 for the other predictors that we examined there is little such evidence, and it REFERENCES
1. The Marmot Review. Fair Society, Healthy Lives. London, United Kingdom: UK Department of Health; 2010 2. Lynch JW, Law C, Brinkman S, Chittleborough C, Sawyer M. Inequalities in child healthy development: some challenges for effective implementation. Soc Sci Med. 2010;71(7):1244 1248 3. Hall D, Hall S. The Family-Nurse Partnership: Developing an Instrument for Identication, Assessment and Recruitment of Clients. London, United Kingdom: Depart-

would be important to determine the effectiveness of programs among women identied using a larger number of predictive factors. Fourth, available resources would be needed to provide programs to all families identied as at risk. One advantage of a multiple characteristic prediction score with good calibration is that it allows policy makers to decide the number of people to whom they are able to provide the programs, and a well-calibrated prediction score will identify those at most risk. For example, policy makers might decide to provide programs to the top 10% of those at predicted risk.

search is needed to examine the feasibility of collecting these data in practice, the effectiveness and costeffectiveness of providing programs to these targeted groups in real-world settings, and how this broader range of factors can be used in clinical decisions about which women should be offered preventive programs.

CONCLUSIONS
Programs and services designed for teen-aged mothers may remain necessary to provide for the specic needs of this group.44,45 However, even if programs for teen-aged mothers are successful in improving child outcomes, they will have little impact on improving population levels of poor child development because maternal young age is not an adequate singular predictor, and few children with poor developmental outcomes have teen-aged mothers. If the goal of improving child development outcomes across the population is to be reached, factors such as maternal education level, nancial difculties, smoking, and depression during pregnancy should be considered when recruiting women to preventive programs. Additional re-

ACKNOWLEDGMENTS This research was funded by a grant from the UK Economic and Social Research Council (RES-060-23-0011). The UK Medical Research Council (grant 74882), the Wellcome Trust (grant 076467), and the University of Bristol provide core support for ALSPAC. Prof Lawlor works in a center that receives support from the UK Medical Research Council (G0600705) and the University of Bristol. Prof Lynch is supported by an Australia Fellowship from the National Health and Medical Research Council of Australia. Dr Chittleborough is also supported by funds from the Australia Fellowship awarded to Prof Lynch. The funding bodies had no role in the decision to publish or the content of this article. We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses.

ment for Children, Schools and Families; 2007 4. McDermott E, Graham H, Hamilton V. Experiences of Being a Teenage Mother in the UK: A Report of a Systematic Review of Qualitative Studies. Lancaster, United Kingdom: Institute for Health Research, Lancaster University; 2004 5. Yardley E. Teenage mothers experiences of stigma. J Youth Studies. 2008;11(6):671 684 6. Wilson H, Huntington A. Deviant (m)others:

The construction of teenage motherhood in contemporary discourse. J Soc Policy. 2005; 35(1):59 76 7. Duncan S, Alexander C, Edwards R. Whats the problem with teenage parents? In: Duncan S, Edwards R, Alexander C, eds. Teenage Parenthood: Whats the Problem? London, United Kingdom: The Turfell Press; 2010 8. Barnes J, Ball M, Meadows P, McLeish J, Belsky J, the FNP Implementation Research Team. Nurse-Family Partnership Programme: First Year Pilot Sites Imple-

PEDIATRICS Volume 127, Number 6, June 2011

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

e1443

mentation in England . London, United Kingdom: Institute for the Study of Children, Families and Social Issues, Birkbeck, University of London; 2008 9. Children, Youth and Womens Health Service. Family Home Visiting. Service Outline. Adelaide, Australia: Children, Youth and Womens Health Service; 2005 10. Olds D, Kitzman H, Cole R, Robinson J. Theoretical foundations of a program of home visitation for pregnant women and parents of young children. J Community Psychol. 1997;25(1):9 25 11. Shaw M, Lawlor DA, Najman JM. Teenage children of teenage mothers: psychological, behavioural and health outcomes from an Australian prospective longitudinal study. Soc Sci Med. 2006;62(10):2526 2539 12. Paranjothy S, Broughton H, Adappa R, Fone D. Teenage pregnancy: who suffers? Arch Dis Child. 2009;94(3):239 245 13. Pepe MS, Janes H, Longton G, Leisenring W, Newcomb P. Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker. Am J Epidemiol. 2004;159(9):882 890 14. Golding J, Pembrey M, Jones R, ALSPAC Study Team. ALSPACthe Avon longitudinal Study of Parents and Children. I. Study methodology. Paediatr Perinat Epidemiol. 2001;15(1):74 87 15. Frankenburg WK, Dodds J, Archer P, Shapiro H, Bresnick B. The Denver II: a major revision and restandardization of the Denver Developmental Screening Test. Pediatrics. 1992;89(1):9197 16. Heron J, Golding J, the ALSPAC Study Team. Thimerosal exposure in infants and developmental disorders: a prospective cohort study in the United Kingdom does not support a causal association. Pediatrics. 2004; 114:577583 17. Goodman R. The Strengths and Difculties Questionnaire: a research note. J Child Psychol Psychiatry. 1997;38(5):581586 18. Goodman R, Ford T, Simmons H, Gatward R, Meltzer H. Using the Strengths and Difculties Questionnaire (SDQ) to screen for child psychiatric disorders in a community sample. Br J Psychiatry. 2000;177:534 539 19. South Gloucestershire Professional and Curriculum Support Service. Entry Assessment in South Gloucestershire. Westonsuper-mare, United Kingdom: Woodspring Educational Resource Centre; 1996 20. Bowen E, Heron J, Waylen A, Wolke D, ALSPAC Study Team. Domestic violence risk during and after pregnancy: ndings from a British

21.

22.

23.

24.

25.

26.

27.

28. 29.

30.

31.

32.

33.

34.

longitudinal study. BJOG . 2005;112(8): 10831089 Cox JL, Holden JM, Sagovsky R. Detection of postnatal depression. Development of the Edinburgh Postnatal Depression Scale. Br J Psychiatr. 1987;150(6):782786 Evans J, Heron J, Francomb H, Oke S, Golding J. Cohort study of depressed mood during pregnancy and after childbirth. BMJ. 2001;323(7307):257260 Murray D, Cox JL. Screening for depression during pregnancy with the Edinburgh depression scale (EPDS). J Reprod Infant Psychol. 1990;8(2):99 107 Thorpe K. A study of the Edinburgh postnatal depression scale for use with parent groups outside the postpartum period. J Reprod Infant Psychol. 1993;11(2):119 125 Murray L, Carrothers AD. The validation of the Edinburgh Postnatal Depression Scale on a community sample. Br J Psychiatr. 1990;157(2):288 290 Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21(1): 128 138 Pencina MJ, DAgostino RB Sr, DAgostino RB Jr, Vasan RS. Evaluating the added predictive ability of a new marker: From area under the ROC curve to reclassication and beyond. Stat Med. 2008;27(2):157172; discussion 207212 Royston P. Multiple imputation of missing values. Stata J. 2004;4(3):227241 Ofce for National Statistics. Births in England and Wales 2009. Available at: www. statistics.gov.uk/statbase/Product. asp?vlnk14408. Accessed October 1, 2010 Martin JA, Hamilton BE, Sutton PD, et al. Births: Final Data for 2006. National Vital Statistics Report, Vol 57, No. 7. Hyattsville, MD: National Center for Health Statistics; 2009 Australian Bureau of Statistics. Births 2008. Catalogue No. 3301.0. Canberra: Australian Bureau of Statistics; 2009 Eccles JS, Davis-Kean PE. Inuences of parents education on their childrens educational attainments: the role of parent and child perceptions. London Review of Education. 2005;3(3):191204 Shipton D, Tappin DM, Vadiveloo T, Crossley JA, Aitken DA, Chalmers J. Reliability of self reported smoking status by pregnant women for estimating smoking prevalence: a retrospective, cross sectional study. BMJ. 2009;339:b4347 The Information Centre for Health and So-

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

cial Care. Statistics on Smoking, England 2006. London, United Kingdom: The Information Centre; 2006 Higher Education Funding Council for England. Trends in Young Participation in Higher Education: Core Results for England. London, United Kingdom: HEFCE; 2010 Turley RN. Are children of young mothers disadvantaged because of their mothers age or family background? Child Dev. 2003; 74(2):465 474 Deave T. Associations between child development and womens attitudes to pregnancy and motherhood. J Reprod Infant Psychol. 2005;23(1):6375 Kiernan KE, Mensah FK. Maternal Indicators in Pregnancy and Childrens Infancy that Signal Future Outcomes for Childrens Development, Behaviour and Health: Evidence from the Millennium Cohort Study. York, United Kingdom: University of York; 2009 Hobcraft JN, Kiernan KE. Predictive Factors from Age 3 and Infancy for Poor Child Outcomes at Age 5 Relating to Childrens Development, Behaviour and Health: Evidence from the Millennium Cohort Study. York, United Kingdom: University of York; 2010 Barnes J, Ball M, Meadows P, Belsky J. Nurse-Family Partnership Programme. Second Year Pilot Sites Implementation in England. The Infancy Period. London, United Kingdom: Institute for the Study of Children, Families and Social Issues, Birkbeck, University of London; 2009 Olds DL, Henderson CR Jr, Tatelbaum R, Chamberlin R. Improving the delivery of prenatal care and outcomes of pregnancy: a randomized trial of nurse home visitation. Pediatrics. 1986;77(1):16 28 Kitzman H, Olds DL, Henderson CR Jr, et al. Effect of prenatal and infancy home visitation by nurses on pregnancy outcomes, childhood injuries, and repeated childbearing. A randomized controlled trial. JAMA. 1997;278(8):644 652 Olds D, Kitzman H, Cole R, Robinson J, Sidora K, Luckey DW, et al. Effects of nurse homevisiting on maternal life-course and child development: age six follow-up results of a randomized trial. Pediatrics. 2004;114(6): 1550 1559 de Jonge A. Support for teenage mothers: a qualitative study into the views of women about the support they received as teenage mothers. J Adv Nurs. 2001;36(1):49 57 Department for Children Schools and Families. Teenage Parents: Who Cares? Nottingham, United Kingdom: Department for Children, Schools and Families; 2008

e1444

CHITTLEBOROUGH et al

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013

Young Maternal Age and Poor Child Development: Predictive Validity From a Birth Cohort Catherine R. Chittleborough, Debbie A. Lawlor and John W. Lynch Pediatrics 2011;127;e1436; originally published online May 2, 2011; DOI: 10.1542/peds.2010-3222
Updated Information & Services Supplementary Material including high resolution figures, can be found at: http://pediatrics.aappublications.org/content/127/6/e1436.full. html Supplementary material can be found at: http://pediatrics.aappublications.org/content/suppl/2011/04/21 /peds.2010-3222.DC1.html This article cites 29 articles, 11 of which can be accessed free at: http://pediatrics.aappublications.org/content/127/6/e1436.full. html#ref-list-1 This article has been cited by 1 HighWire-hosted articles: http://pediatrics.aappublications.org/content/127/6/e1436.full. html#related-urls This article, along with others on similar topics, appears in the following collection(s): Office Practice http://pediatrics.aappublications.org/cgi/collection/office_pra ctice Information about reproducing this article in parts (figures, tables) or in its entirety can be found online at: http://pediatrics.aappublications.org/site/misc/Permissions.xht ml Information about ordering reprints can be found online: http://pediatrics.aappublications.org/site/misc/reprints.xhtml

References

Citations

Subspecialty Collections

Permissions & Licensing

Reprints

PEDIATRICS is the official journal of the American Academy of Pediatrics. A monthly publication, it has been published continuously since 1948. PEDIATRICS is owned, published, and trademarked by the American Academy of Pediatrics, 141 Northwest Point Boulevard, Elk Grove Village, Illinois, 60007. Copyright 2011 by the American Academy of Pediatrics. All rights reserved. Print ISSN: 0031-4005. Online ISSN: 1098-4275.

Downloaded from pediatrics.aappublications.org at Indonesia:AAP Sponsored on February 27, 2013