DOI: 10.1542/peds.2010-3222; originally published online May 2, 2011; 2011;127;e1436Pediatrics

Catherine R. Chittleborough, Debbie A. Lawlor and John W. LynchBirth Cohort

Young Maternal Age and Poor Child Development: Predictive Validity From a  

  http://pediatrics.aappublications.org/content/127/6/e1436.full.html

located on the World Wide Web at: The online version of this article, along with updated information and services, is

 

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

at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

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

WHAT’S KNOWN ON THIS SUBJECT: Teen-aged mothers and theirchildren are targeted by policies and programs aimed atimproving child development. Teenage motherhood may be a riskfactor for poor childhood development, but it may be aninaccurate criterion for predicting risk of developmentaloutcomes.

WHAT THIS STUDY ADDS: To reach the goal of improving childdevelopment outcomes across the population, factors such asmaternal education level, financial difficulties, smoking, anddepression during pregnancy should be considered in addition toyoung maternal age when recruiting women to preventive programs.

abstractOBJECTIVE: We aimed to examine the ability of mother’s age, and otherfactors measured during pregnancy (education, financial difficulties,partner status, smoking, and depression), to predict child develop-ment outcomes up to age 5 years.

METHODS: Data were obtained from the Avon Longitudinal Study ofParents and Children (ALSPAC). Poor child development was defined asscoring in the worst 10% of a parent-reported ALSPAC developmentalscale (ADS) at 18 months (n � 7546), the Strengths and DifficultiesQuestionnaire (SDQ) at 47 months (n � 8328), or teacher-reportedSchool Entry Assessment (SEA) scores at 4 to 5 years (n� 7345).

RESULTS: Only a small proportion of children with poor developmenthadmothers aged younger than 20 years at their birth (3.3%, 6.4%, and9.2%, for the ADS, SDQ, and SEA, respectively). A greater proportionwith each measure of poor development would be identified (48.9%,63.6%, and 74.4%, respectively) if all 6 predictors were used and awoman had at least 1 of these. Model discrimination was poor usingmaternal age only (area under the receiver operator characteristiccurve �0.5 for all 3 outcomes). This improved when all 6 predictorswere included in the model (ADS: 0.56; SDQ: 0.66; SEA: 0.67). Calibrationalso improved with the model including all 6 predictors.

CONCLUSIONS: Even if programs targeted at teen-aged mothers aresuccessful in improving child development, they will have little impacton population levels of poor child development if youngmaternal age isthe sole or main means of identifying eligibility for the program.Pediatrics 2011;127:e1436–e1444

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 andClinical Practice, University of Adelaide, Adelaide, Australia;cMRC Centre for Causal Analysis in Translational Epidemiology,University of Bristol, Bristol, United Kingdom; and dSansomInstitute for Health Research, Division of Health Sciences,University of South Australia, Adelaide, Australia

KEY WORDSALSPAC, child development, maternal age, maternal healthservices, predictive value of tests

ABBREVIATIONSALSPAC—Avon Longitudinal Study of Parents and ChildrenADS—ALSPAC developmental scaleSDQ—Strengths and Difficulties QuestionnaireSEA—School Entry AssessmentEPDS—Edinburgh Postnatal Depression ScalePPV—positive predictive valueAUROC—area under the receiver operator characteristic curve

All authors contributed to the conceptual development, analysisplan, and interpretation of results. Dr Chittleborough undertookthe analyses and wrote the first draft of the paper. Drs Lawlorand Lynch contributed later drafts. All authors approve the finalversion to be published and take responsibility for the analysesof 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 haveno financial relationships relevant to this article to disclose.

e1436 CHITTLEBOROUGH et al at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

The highest priority recommendationin the recent Marmot Review to reducehealth inequalities was to give everychild the best start in life,1 with actionsthat are universal “but with a scale andintensity that is proportionate to thelevel of disadvantage”1 so that moresupport goes to those with greaterneed. This progressive universalism ischallenging for early child developmentservices because an accuratemethod ofidentifying those most in need has to bebalanced against efficient use of limitedresources and the risk of stigmatizingmothers and families who may be la-beled as poor parents.2–7

Programs such as the Family NursePartnership in the United Kingdom,8

Family Home Visiting in Australia,9 andNurse Family Partnership in the UnitedStates10 offer extended services be-yond immediate postnatal contact to“vulnerable” families often identifiedby maternal age, with mothers aged�20 years eligible for the program.8,9

Although teen-age motherhood can bean important risk factor for poor child-hood development,11,12 it may be an in-accurate predictor of developmentalrisk in children.13 Other risk factors incombination might improve identifica-tion of those at greatest risk of poorerdevelopment. If such additional factorswere routinely obtainable in the ante-natal or early postnatal period, theywould increase the ability to effectivelytarget limited program resources tothose most likely to benefit. The pres-ent study examines the predictive va-lidity of young maternal age (�20years) and 5 other factors (maternaleducation, financial difficulties, partnerstatus, smoking during pregnancy, anddepression) in predicting poor develop-ment in childhood up to age 5 years.

METHODS

The Avon Longitudinal Study of Parentsand Children (ALSPAC) is a prospec-tive, geographically representative

study of children born to women res-ident in the Avon area of southwestEngland with an expected deliverydate between April 1, 1991, and De-cember 31, 1992. Details of the back-ground, methods, recruitment, and re-sponse rates have been reportedelsewhere (www.bristol.ac.uk/alspac/).14

The core ALSPAC sample consists of14 541 pregnancies (Fig 1). Ethical ap-proval was obtained from the ALSPACLaw and Ethics committee and local re-search ethics committees.

Child Outcomes

Child developmental abilities at 18months were assessed using theALSPAC developmental scale (ADS),created using items derived from theDenver Developmental Screening Testshown to be most predictive of devel-opmental abnormality.15 Becausemany of the Denver items were de-signed to be observed by trained exam-iners, the ADS was adapted for paren-tal report after focus group pilotingwith members of the ALSPAC cohort.

Parents reported whether their childcould do 56 activities within 4 develop-mental domains (grossmotor, finemo-tor, communication, and social skills).The number of passes, indicated by“yes, can do well” responses, wassummed in each of the 4 subscales,and the total development score wassummed across subscales. Age forcompletion of the ADS was restrictedto an 8-week window around 18months given the developmental age-specific nature of the questions.16

The parent version of the Strengthsand Difficulties Questionnaire (SDQ)17

was completed by the main caregiver(usually the mother) when the childwas 47 months old, using a scale from1 to 3 (does not apply, applies some-what, definitely applies). The scale con-sisted of 25 items in 5 subscales(prosocial behavior, hyperactivity,emotional symptoms, conduct prob-lems, and peer problems). A total diffi-culties score was created by summingthe scores from the last 4 subscales.

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

Eligible cohort N = 14 531

Excluded triplet and quadruplet births and women with missing data for age at last

menstrual period N = 10

No data provided at 47 mo N = 5109

No data provided at 18 mo N = 5949

Linked data unavailable N = 5195

Missing data on covariates N = 1036

Missing data on covariates N = 1094

Missing data on covariates N = 1991

ADS data, 18 mo N = 8582

SDQ data, 47 moN = 9422

SEA data, 4–5 y N = 9336

ADS data, 18 mo N = 7546

SDQ data, 47 mo N = 8328

SEA data, 4–5 y N = 7345

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

RESPONSE SAMPLE:

ANALYSIS SAMPLE:

IMPUTED SAMPLE:

Impute data on missing outcomes

and covariates

Impute data on missing outcomes

and covariates

Impute data on missing outcomes

and covariates

FIGURE 1Eligible cohort and numbers included for analyses.

ARTICLES

PEDIATRICS Volume 127, Number 6, June 2011 e1437 at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

The prosocial subscale was excludedbecause it measures positive aspectsof behavior. High scores on the SDQhave been shown to be predictive ofpsychiatric disorders among children.18

Children were rated by their teacher inthe School Entry Assessment (SEA) dur-ing the first half of their first term in re-ception class at ages 4 to 5 years.19 Thisassessment is undertaken in all recep-tion classes in England, and scores forALSPAC participants were obtainedthrough consented record linkage withdata provided by local education author-ities. Integer scores between 2 and 7 oneach of 4 required scales (language,reading, writing, and mathematics)were summed to provide a total SEAscore.

Potential Predictors of ChildhoodDevelopment Problems

Age of mother at last menstrual periodwas obtained for 14 531 women (Fig 1)and dichotomized at younger than 20years, the cutoff point commonly usedto identify mothers eligible for pro-grams.8,9 Highest education level wascategorized into “O level or higher”(where O level is ordinary level exami-nations most commonly taken at 16years of age, the legal minimum agefor leaving school in the United King-dom) and “less than O level” (Certifi-cate of Secondary Education com-monly taken at age 16 years byindividuals considered to be unable toobtain an O level in that subject, a vo-cational qualification, or no educa-tional qualifications). The financial dif-ficulties factor was assessed using 5questions asking how difficult themother found it to afford food, cloth-ing, heating, rent or mortgage, andthings she will need for the infant, witha score of 1 (very difficult) to 4 (notdifficult) for each response. The algo-rithm for calculating the overall finan-cial difficulties score was 20 minus thesum of the scores of each of the 5

items, resulting in an overall score inwhich 0 represented no financial diffi-culties and 15 the maximum financialdifficulties. Participants scoring �8were defined as experiencing financialdifficulties.20 Partner status at studyenrollment (married or cohabitatingversus no partner or not living withpartner) and whether women hadsmoked during the first 3 months oftheir pregnancy were assessed by us-ing a questionnaire.

Ten items that formed the depressionscale of the Edinburgh Postnatal De-pression Scale (EPDS)21,22 were admin-istered via questionnaire at 18 to 20weeks’ gestation. None of the 10 itemsis specific to the postnatal experience,and this scale has been validated foruse postnatally and during pregnan-cy.23–25 Each question had 4 responsecategories scored from 0 to 3 and re-ferred to the feelings of the mother inthe past week. A score�12 is used toindicate probable depressive disorder.21

Analysis

Distributions of continuous outcomevariables were skewed and so eachdistribution was dichotomized, withthe lower tail (or upper for SDQ) con-taining 10% of those with the poorestchild developmental outcome.16,17

We calculated the proportion of chil-dren with poor developmental out-comes whose mothers had each of theindividual binary predictive factors,and also whose mothers had at least 1and at least 2 of the 6 binary predic-tors. Specificity, positive predictivevalue (PPV) and likelihood ratio of eachbinary predictor were calculated (Sup-plemental Table 5). Univariable andmultivariable (with mutual adjustmentfor all other predictors) logistic re-gression examined associations ofpredictors with each child outcome.The predicted probability of poor childdevelopment was calculated fromthese regression models. In clinical

practice, the predictors would likely beused as binary variables, but becausecalibration statistics cannot be easily in-terpreted using a single binary predic-tor, maternal age, financial difficulties,and EPDS scorewere included as contin-uous variables in the prediction models.

The area under the receiver operatorcharacteristic curve (AUROC) wasused to assess the discriminatory ca-pability of the models, or how accu-rately each model separates mothersinto those with and without childrenwith poor outcomes. Model 1 con-tained only maternal age; model 2 in-cluded all 6 predictors. An AUROC of 1represents a model that perfectly dis-criminates the outcome; an AUROC of0.5 represents a prediction tool that isno better than chance at identifyingthose at risk. AUROCs were also calcu-lated using all binary predictor vari-ables, as would be more commonlyused in clinical practice, and thesewere lower but consistent with predic-tive models that included continuousvariables (data not shown).

Calibration of the 2 models, or theagreement between observed and pre-dicted outcomes, was assessed byranking mothers into deciles of theirpredicted risk and comparing the pre-dicted to observed proportion withineach decile. The Hosmer-Lemeshowgoodness-of-fit �2 statistic was used totest the accuracy of calibration.26 Thisstatistic tests the null hypothesis thatthe predicted proportion equals theobserved proportion within rankedgroupings (deciles) of predicted risk. Ahigh P value suggests good calibrationof predicted and observed risk.

The integrated discrimination im-provement27 for model 2 comparedwith model 1 was also calculated. Thisassesses discrimination without rely-ing on cutoff points and compares theaverage difference in predicted riskfor women whose children have poordevelopment with women whose chil-

e1438 CHITTLEBOROUGH et al at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

dren do not have poor development.The integrated discrimination im-provement is greater when the secondmodel correctly assigns individuals tohigher or lower probabilities of havingthe outcome in comparison to the firstmodel.

Missing Data

Sensitivity analyses were conductedon an imputed data set to examine theinfluence of missing data on the find-ings. Multiple imputation by chainedequation was used to impute missingdata on child outcomes and predictorsfor respondents who had data on atleast 1 child outcome (N� 12 570; Fig1) using the ice command in Stata(Stata Corp, College Station, TX).28 Theimputation model included all childoutcomes and predictors as well aspredictors of missingness (birthweight, parity, social class, ethnicity,

and reaction to pregnancy). We gener-ated 20 data sets and undertook 20 cy-cles of regression switching.28 Table 1shows the prevalence and amount ofdata available for each child outcomeand predictor.

RESULTS

Table 2 displays the proportion of chil-dren with poor development whose

mothers had each predictive factor. Asmall proportion of children with poordevelopment had mothers aged �20(3.3%, 6.4%, and 9.2% for ADS, SDQ, andSEA, respectively). High proportions ofpoor development could be identified ifinformation on all 6 predictors wasused and a woman had at least 1 ofthese predictors (48.9%, 63.6%, and74.4% for ADS, SDQ, and SEA). The pre-dictor that alone identified the highestproportion of each child outcome wasmother’s low education.

Table 3 shows univariable and multi-variable associations between thepotential predictors and child develop-mental outcomes. Associations be-tween the potential predictors andchild outcomes using the multiply im-puted data set (Supplemental Table 6)were consistent with analyses of com-plete cases.

Table 4 shows calibration and discrim-ination for both models. Discrimina-tion was poor using model 1 (maternalage only). This finding improved a littlefor ADS and more so for SDQ and SEAwhen all 6 predictors were used inmodel 2. AUROC values calculated us-ing the multiply imputed data set (Sup-plemental Table 7) were consistentwith complete case analyses. TheHosmer-Lemeshow goodness-of-fit testsindicated better calibration usingmodel 2 than model 1 for SDQ and SEA,whereas both models showed goodcalibration for ADS. Model 1 underesti-

TABLE 1 Prevalence and Amount of Data Available for Each Child Outcome and Potential PredictorMeasured During Pregnancy

Time of DataCollection

ResponseSamplea

Analysis Sample, %b ImputedSample,%c (N�12 570)

N % ADS(n� 7546)

SDQ(n� 8328)

SEA(n� 7345)

Child outcome

ADS 18 mo 8582 10.0 10.0 — — 10.2

SDQ 47 mo 9422 11.6 — 11.4 — 12.7

SEA 4–5 y 9336 10.1 — — 8.5 9.5

Predictor

Age�20 y Study enrollment 14 531 6.6 3.8 3.1 4.9 6.1

No partner ornotcohabitating

Study enrollment 13 485 8.8 6.4 6.0 7.5 8.7

Financialdifficulties

32 weeks’gestation

12 011 10.0 9.1 8.4 10.5 9.5

Depression 18–20 weeks’gestation

12 177 13.9 12.1 11.6 13.1 13.4

Smoking infirst 3 mo ofpregnancy

18–20 weeks’gestation

13 189 25.1 21.8 20.6 24.0 24.9

Education�Olevel

32 weeks’gestation

12 340 30.1 28.1 24.7 32.3 30.9

At least 1 of the6 predictors

10 955 51.2 50.0 46.5 54.3 53.9

At least 2 of the6 predictors

10 955 22.8 21.0 19.0 24.5 25.1

a Response sample is the number who responded to specific 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 1of the 3 child outcomes.

TABLE 2 Proportion of Child Outcome Cases That Would be Detected With Potential PredictorsMeasured During Pregnancy

Predictor Measured DuringPregnancy

ADS 18 Months(N� 7546;

n cases� 755), %

SDQ 47 Months(N� 8328;

n cases� 946), %

SEA 4–5 Years(N� 7345;

n cases� 621), %

Age�20 y 3.3 6.4 9.2No partner or not cohabitating 5.0 9.1 12.2Financial difficulties 9.1 16.0 17.7Depression 14.3 22.3 19.0Smoking 17.7 31.9 34.6Education�O level 29.0 34.5 55.4At least 1 of the 6 predictors 48.9 63.6 74.4At least 2 of the 6 predictors 19.5 35.1 44.1

ARTICLES

PEDIATRICS Volume 127, Number 6, June 2011 e1439 at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

mated the likelihood of poor develop-ment according to SDQ and SEA amongthose at highest risk (Table 4 and Fig2). Integrated discrimination improve-ments indicated that model 2 resultedin an improvement in calibration overmodel 1, particularly for SDQ and SEAwith �3% of the children being cor-

rectly reclassified by model 2 com-pared with model 1.

DISCUSSION

The main finding of our study is thatprograms, which have been shown tobe effective at improving child develop-ment, will have little impact on child

development outcomes at the popula-tion level if youngmaternal age is usedas the sole or main criterion for iden-tifying eligible mothers. This finding issound because, in general the propor-tion of births to women aged 15 to 19years is low, at 6.1% of live births inEngland and Wales in 2009,29 10.2% ofbirths in the United States in 200630

and 4.2% of births in Australia in2008.31 Therefore, only a small propor-tion of children with poor developmenthave teen-aged mothers. Maternal age�20 years identifies only 9% of thecases of poor development at 5 years,whereas 74% of these cases would beidentified among mothers with �1 ofthe 6 predictors, and 44% would beidentified among mothers with �2 ofthe 6 predictors. If the 23% of womenexperiencing�2 of these characteris-tics could be engaged in programsaimed at supporting the developmentof their children, 44% of cases of poorchild development at school entrycould be identified and potentiallyprevented. Furthermore, a model in-cluding all predictors provides bet-ter discrimination and calibrationfor predicting these child outcomesthan a model based solely on moth-er’s age. Mother’s low education isthe single characteristic that ac-counts for the highest proportion ofcases of poor child development,from almost 30% at 18 months to58% at 4 to 5 years. Low education ismore common than teen-age moth-erhood in the population, and par-ents with higher levels of educationare thought to positively influencetheir children’s academic achieve-ment through use of more varied andcomplex language and reading inter-actions, and exposing children to in-creased educational opportunities.32

The strengths of this study are thelarge sample size and longitudinal de-sign with inclusion of a large numberof relevant predictors measured dur-

TABLE 3 Univariable and Multivariable Associations of Potential Predictors With ChildDevelopmental Outcomes

Predictor MeasuredDuring Pregnancy

ADS 18 Months (N�7546; n cases� 755)

SDQ 47 Months (N�8328; n cases� 946)

SEA 4–5 Years (N�7345; n cases� 621)

OR(95% CI)

P OR(95% CI)

P OR(95% CI)

P

UnivariableAge group

�20 1 1 1�20 0.86 (0.57–1.31) .492 2.50 (1.86–3.36) �.001 2.13 (1.59–2.87) �.001Partner statusMarried/cohabitating 1 1 1No partner/notcohabitating

0.76 (0.54–1.06) .110 1.69 (1.32–2.15) �.001 1.82 (1.41–2.36) �.001

Financial difficultiesscore

�9 1 1 1�9 1.01 (0.78–1.31) .929 2.35 (1.93–2.85) �.001 1.98 (1.59–2.48) �.001EPDS score

�12 1 1 1�12 1.25 (1.00–1.55) .046 2.53 (2.13–3.00) �.001 1.63 (1.32–2.02) �.001Smoked in first 3 mo ofpregnancyNo 1 1 1Yes 0.76 (0.62–0.92) .005 1.99 (1.71–2.31) �.001 1.77 (1.48–2.11) �.001Highest education level

�O level 1 1 1�O level 1.05 (0.89–1.24) .568 1.72 (1.49–1.99) �.001 2.88 (2.44–3.40) �.001

Multivariablea

Age group�20 1 1 1�20 0.95 (0.61–1.48) .825 1.72 (1.25–2.37) .001 1.36 (0.98–1.88) .062Partner statusMarried/cohabitating 1 1 1No partner/notcohabitating

0.78 (0.55–0.12) .181 0.98 (0.75–1.28) .869 1.11 (0.84–1.48) .465

Financial difficultiesscore

�9 1 1 1�9 1.02 (0.78–1.33) 0.887 1.73 (1.41–2.12) �.001 1.54 (1.22–1.94) �.001EPDS score

�12 1 1 1�12 1.33 (1.06–1.66) 0.013 2.02 (1.69–2.42) �.001 1.27 (1.02–1.59) .036Smoked in first 3 mo ofpregnancyNo 1 1 1Yes 0.73 (0.60–0.90) 0.003 1.55 (1.32–1.81) �.001 1.29 (1.07–1.56) .007Highest education level

�O level 1 1 1�O level 1.10 (0.93–0.31) 0.246 1.44 (1.24–1.67) �.001 2.56 (2.15–3.04) �.001

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

e1440 CHITTLEBOROUGH et al at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

ing pregnancy. Self-reported smokingstatus may underestimate smokingprevalence among pregnant women,33

but self-reported smoking still contrib-uted to the prediction of poor child de-velopment and reflects the clinical sit-uation in which pregnant womenreport their smoking status at antena-tal consultations. Given that calibra-tion cannot be assessed with a single

binary predictor, we used the continu-ous age variable, which may underes-timate the poor calibration of mater-nal age with a cutoff �20 years, as isused in practice. Reduced power fromcohort attrition is not a major problemin a study of this size, and analyses us-ing multivariate multiple imputationproduced similar results to completecase analyses, suggesting little bias

due to missing data. Societal changes(eg, downward trends in the propor-tion of births to young mothers,29

smoking during pregnancy,34 and in-creased participation of young peoplein higher education35) mean that ourfindings do not necessarily generalizeto more contemporary populations orthose from other countries. However,our conclusion that young maternalage is likely to identify only a small pro-portion of childrenwith developmentalproblems is likely to hold across mosthigh-income countries where the prev-alence of young maternal age is low.Differences in prediction between theSDQ and SEA, despite being measuredat similar ages, reflect the fact thatchildren with behavioral difficulties(as reported by parents) were not nec-essarily the same group as those withpoor SEA scores (as assessed byteachers).

Association of mother’s age, and otherfactors, with child development out-comes have been shown previous-ly11,36–38 but this is the first study, to ourknowledge, to demonstrate discrimi-nation, calibration, and sensitivity ofthese maternal factors in predictingchild development. Such analyses maymore directly inform targeting of pro-grams to support parenting and childdevelopment. An analysis of the UK Mil-lennium Cohort Study found that ap-proximately one quarter of childrenwith poor development and behaviorwould be identified among the 10% ofchildren with the highest risk, as pre-dicted by a model using many predic-tors, including depression, smokingduring pregnancy, educational qualifi-cations, and socioeconomic position.38

Higher AUROC values (�0.80) for pre-dicting poor child development at age5 years than we have found were sub-sequently reported.39 These predictionmodels included, in addition to thecharacteristics we examined, birthweight, gender, and breastfeeding,

TABLE 4 Calibration and Discrimination of the 2 Models

Model 1 Model 2

Observed Predicted Ratio Observed Predicted Ratio

ADS 18 mo (n� 7546)Lowest 10th 7.7 8.3 1.07 7.0 6.8 0.982nd 8.8 9.0 1.02 8.6 8.1 0.943rd 8.1 9.3 1.15 7.4 8.8 1.194th 9.1 9.6 1.05 6.9 9.3 1.355th 10.4 9.9 0.95 11.5 9.7 0.856th 10.9 10.3 0.94 10.3 10.1 0.987th 12.3 10.5 0.85 12.1 10.6 0.888th 12.3 10.8 0.87 11.8 11.1 0.959th 10.8 11.2 1.03 11.1 11.9 1.07Highest 10th 10.9 12.3 1.13 13.4 13.6 1.01Hosmer-Lemeshow �2 testa 6.83, P� .555 12.35, P� .262AUROC (95% CI) 0.5395 (0.5183–0.5607) 0.5629 (0.5416–0.5842), P� .020b

IDI, % (95% CI) — 0.26 (0.15–0.37), P� .001c

SDQ 47 mo (n� 8328)Lowest 10th 10.0 7.5 0.75 5.9 5.1 0.862nd 8.7 8.9 1.02 5.1 6.2 1.213rd 9.6 9.8 1.02 6.5 7.0 1.094th 9.7 10.6 1.09 7.2 7.9 1.105th 9.8 11.1 1.14 9.5 8.9 0.946th 11.0 11.7 1.07 10.2 10.1 0.997th 11.1 12.3 1.11 12.0 11.6 0.978th 11.1 13.2 1.19 14.0 13.7 0.989th 15.0 14.5 0.97 18.0 17.0 0.94Highest 10th 21.9 16.9 0.77 25.1 26.0 1.03Hosmer-Lemeshow �2a 27.82, P� .001 5.28, P� .872AUROC (95% CI) 0.5691 (0.5488–0.5893) 0.6600 (0.6412–0.6787), P� .001b

IDI, % (95% CI) — 3.13 (2.76–3.51) , P� .001c

SEA 4–5 y (n� 7345)Lowest 10th 7.8 5.9 0.75 3.4 3.8 1.132nd 6.6 6.9 1.05 3.8 4.3 1.133rd 6.4 7.4 1.16 5.0 4.7 0.934th 6.4 7.9 1.22 5.0 5.1 1.035th 8.2 8.4 1.02 6.5 5.8 0.886th 7.7 8.8 1.15 7.2 6.8 0.947th 8.0 9.2 1.15 8.0 9.0 1.128th 10.9 9.8 0.89 11.6 11.3 0.979th 10.6 10.6 1.00 14.0 14.1 1.01Highest 10th 14.1 12.1 0.86 20.0 19.8 0.99Hosmer-Lemeshow �2a 16.52, P� .036 2.94, P� .983AUROC (95% CI) 0.5659 (0.5412–0.5907) 0.6732 (0.6509–0.6955), P� .001b

IDI, % (95% CI) 2.84 (2.46–3.22) , P� .001c

Model 1, maternal age; Model 2, maternal age, highest education level�O level, financial difficulties score, no partner or notcohabitating, smoked in first 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.

ARTICLES

PEDIATRICS Volume 127, Number 6, June 2011 e1441 at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

which we did not include becausethese factors would be unavailable forselecting women for programs duringpregnancy.

Although a broader range of risk fac-tors may more accurately identifymothers whose children are at highrisk of poor development, there are

many issues to consider. First, collec-tion of all of the characteristics wouldneed to be feasible in routine clinicalsettings and acceptable to pregnant

2 ledoM 1 ledoM

ALSPAC developmental scale 18m

Strengths and Difficulties Questionnaire 47m

School Entry Assessment 4-5y

.06

.08

.1

.12

.14

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

observedpredicted

.06

.08

.1

.12

.14

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

observedpredicted

.05

.1

.15

.2

.25

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

observedpredicted

.05

.1

.15

.2

.25

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

observedpredicted

.05

.1

.15

.2

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

observedpredicted

.05

.1

.15

.2

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

observedpredicted

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

e1442 CHITTLEBOROUGH et al at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

women. Forms completed by FamilyNurse Partnership nurses in thesecond-wave pilot sites in England hadmissing data for 8.8% of women onmarital status, 9.3% on employmentstatus, and 10.1% on education.40 Al-though responding to a research ques-tionnaire is different compared withanswering questions in a clinic setting,our study suggests thatmost pregnantwomen provide information on thecharacteristics we have examined. It ispossible that the 15% to 17% who didnot answer questions about education,depression, or financial difficulties arethose with children at particular highrisk of childhood development prob-lems. Second, a simple tool would beneeded for using the collected dataand generating a risk score for eachindividual. This could range from asimple checklist of predictors inwhich, for example, women with �2of the binary predictors are consid-ered for interventions, through tocomputer-based tools that make useof predictive risk algorithms contain-ing continuous variables. The former islikely to be feasible in most settings;the latter is becoming increasinglycommon, for example, in the predic-tion of cardiovascular risk. Third, al-though there is some randomized con-trolled trial evidence that theseinterventions improve outcomes forchildren of teen-aged mothers,41–43 forthe other predictors that we examinedthere is little such evidence, and it

would be important to determine theeffectiveness of programs amongwomen identified using a larger num-ber of predictive factors. Fourth, avail-able resources would be needed toprovide programs to all families iden-tified as at risk. One advantage of amultiple characteristic predictionscore with good calibration is that itallows policy makers to decide thenumber of people to whom they areable to provide the programs, and awell-calibrated prediction score willidentify those at most risk. For exam-ple, policy makers might decide to pro-vide programs to the top 10% of thoseat predicted risk.

CONCLUSIONS

Programs and services designed forteen-aged mothers may remain neces-sary to provide for the specific needsof this group.44,45 However, even if pro-grams for teen-aged mothers are suc-cessful in improving child outcomes,they will have little impact on improv-ing population levels of poor child de-velopment because maternal youngage is not an adequate singular predic-tor, and few children with poor devel-opmental outcomes have teen-agedmothers. If the goal of improving childdevelopment outcomes across thepopulation is to be reached, factorssuch as maternal education level, fi-nancial difficulties, smoking, and de-pression during pregnancy should beconsidered when recruiting women topreventive programs. Additional re-

search is needed to examine the feasi-bility of collecting these data in prac-tice, the effectiveness and cost-effectiveness of providing programs tothese targeted groups in real-worldsettings, and how this broader rangeof factors can be used in clinical deci-sions about which women should beoffered preventive programs.

ACKNOWLEDGMENTSThis research was funded by a grantfrom the UK Economic and Social Re-search Council (RES-060-23-0011). TheUK Medical Research Council (grant74882), the Wellcome Trust (grant076467), and the University of Bristolprovide core support for ALSPAC. ProfLawlor works in a center that receivessupport from the UK Medical ResearchCouncil (G0600705) and the Universityof Bristol. Prof Lynch is supported byan Australia Fellowship from the Na-tional Health and Medical ResearchCouncil of Australia. Dr Chittleboroughis also supported by funds from theAustralia Fellowship awarded to ProfLynch. The funding bodies had no rolein the decision to publish or the con-tent of this article.

We are extremely grateful to all thefamilies who took part in this study,the midwives for their help in recruit-ing them, and the whole ALSPAC team,which includes interviewers, com-puter and laboratory technicians, cler-ical workers, research scientists, vol-unteers, managers, receptionists, andnurses.

REFERENCES

1. The Marmot Review. Fair Society, HealthyLives. London, United Kingdom: UK Depart-ment of Health; 2010

2. Lynch JW, Law C, Brinkman S, Chittlebor-ough C, Sawyer M. Inequalities in childhealthy development: some challenges foreffective implementation. Soc Sci Med.2010;71(7):1244–1248

3. Hall D, Hall S. The “Family-Nurse Partner-ship”: Developing an Instrument for Identi-fication, Assessment and Recruitment ofClients. London, United Kingdom: Depart-

ment for Children, Schools and Families;2007

4. McDermott E, Graham H, Hamilton V. Expe-riences of Being a Teenage Mother in theUK: A Report of a Systematic Review of Qual-itative Studies. Lancaster, United Kingdom:Institute for Health Research, LancasterUniversity; 2004

5. Yardley E. Teenage mothers’ experiences ofstigma. J Youth Studies. 2008;11(6):671–684

6. Wilson H, Huntington A. Deviant (m)others:

The construction of teenage motherhood incontemporary discourse. J Soc Policy. 2005;35(1):59–76

7. Duncan S, Alexander C, Edwards R. What’sthe problem with teenage parents? In: Dun-can S, Edwards R, Alexander C, eds. TeenageParenthood: What’s the Problem? London,United Kingdom: The Turfell Press; 2010

8. Barnes J, Ball M, Meadows P, McLeish J,Belsky J, the FNP Implementation Re-search Team. Nurse-Family PartnershipProgramme: First Year Pilot Sites Imple-

ARTICLES

PEDIATRICS Volume 127, Number 6, June 2011 e1443 at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

mentation in England. London, UnitedKingdom: Institute for the Study of Children,Families and Social Issues, Birkbeck, Uni-versity of London; 2008

9. Children, Youth and Women’s Health Ser-vice. Family Home Visiting. Service Outline.Adelaide, Australia: Children, Youth andWomen’s Health Service; 2005

10. Olds D, Kitzman H, Cole R, Robinson J. Theo-retical foundations of a program of homevisitation for pregnant women and parentsof young children. J Community Psychol.1997;25(1):9–25

11. Shaw M, Lawlor DA, Najman JM. Teenagechildren of teenagemothers: psychological,behavioural and health outcomes from anAustralian prospective longitudinal study.Soc Sci Med. 2006;62(10):2526–2539

12. Paranjothy S, Broughton H, Adappa R, FoneD. Teenage pregnancy: who suffers? ArchDis Child. 2009;94(3):239–245

13. Pepe MS, Janes H, Longton G, Leisenring W,Newcomb P. Limitations of the odds ratio ingauging the performance of a diagnostic,prognostic, or screening marker. Am J Epi-demiol. 2004;159(9):882–890

14. Golding J, Pembrey M, Jones R, ALSPACStudy Team. ALSPAC—the Avon longitudinalStudy of Parents and Children. I. Studymethodology. Paediatr Perinat Epidemiol.2001;15(1):74–87

15. Frankenburg WK, Dodds J, Archer P, Sha-piro H, Bresnick B. The Denver II: a majorrevision and restandardization of the Den-ver Developmental Screening Test. Pediat-rics. 1992;89(1):91–97

16. Heron J, Golding J, the ALSPAC Study Team.Thimerosal exposure in infants and devel-opmental disorders: a prospective cohortstudy in the United Kingdom does not sup-port a causal association. Pediatrics. 2004;114:577–583

17. Goodman R. The Strengths and DifficultiesQuestionnaire: a research note. J Child Psy-chol Psychiatry. 1997;38(5):581–586

18. Goodman R, Ford T, Simmons H, Gatward R,Meltzer H. Using the Strengths and Difficul-ties Questionnaire (SDQ) to screen for childpsychiatric disorders in a community sam-ple. Br J Psychiatry. 2000;177:534–539

19. South Gloucestershire Professional andCurriculum Support Service. Entry Assess-ment in South Gloucestershire. Weston-super-mare, United Kingdom: WoodspringEducational Resource Centre; 1996

20. Bowen E, Heron J, Waylen A, Wolke D, ALSPACStudy Team. Domestic violence risk duringand after pregnancy: findings from a British

longitudinal study. BJOG. 2005;112(8):1083–1089

21. Cox JL, Holden JM, Sagovsky R. Detection ofpostnatal depression. Development of theEdinburgh Postnatal Depression Scale. Br JPsychiatr. 1987;150(6):782–786

22. Evans J, Heron J, Francomb H, Oke S, Gold-ing J. Cohort study of depressed mood dur-ing pregnancy and after childbirth. BMJ.2001;323(7307):257–260

23. Murray D, Cox JL. Screening for depressionduring pregnancy with the Edinburgh de-pression scale (EPDS). J Reprod Infant Psy-chol. 1990;8(2):99–107

24. Thorpe K. A study of the Edinburgh postnataldepression scale for use with parentgroups outside the postpartum period. JReprod Infant Psychol. 1993;11(2):119–125

25. Murray L, Carrothers AD. The validation ofthe Edinburgh Postnatal Depression Scaleon a community sample. Br J Psychiatr.1990;157(2):288–290

26. Steyerberg EW, Vickers AJ, Cook NR, et al.Assessing the performance of predictionmodels: a framework for traditional andnovel measures. Epidemiology. 2010;21(1):128–138

27. PencinaMJ, D’Agostino RB Sr, D’Agostino RBJr, Vasan RS. Evaluating the added predic-tive ability of a new marker: From area un-der the ROC curve to reclassification andbeyond. Stat Med. 2008;27(2):157–172; dis-cussion 207–212

28. Royston P. Multiple imputation of missingvalues. Stata J. 2004;4(3):227–241

29. Office for National Statistics. Births inEngland and Wales 2009. Available at: www.stat ist ics .gov .uk/statbase/Product .asp?vlnk�14408. Accessed October 1, 2010

30. Martin JA, Hamilton BE, Sutton PD, et al.Births: Final Data for 2006. National VitalStatistics Report, Vol 57, No. 7. Hyattsville,MD: National Center for Health Statistics;2009

31. Australian Bureau of Statistics. Births 2008.Catalogue No. 3301.0. Canberra: AustralianBureau of Statistics; 2009

32. Eccles JS, Davis-Kean PE. Influences of par-ents’ education on their children’s educa-tional attainments: the role of parent andchild perceptions. London Review of Educa-tion. 2005;3(3):191–204

33. Shipton D, Tappin DM, Vadiveloo T, CrossleyJA, Aitken DA, Chalmers J. Reliability of selfreported smoking status by pregnantwomen for estimating smoking prevalence:a retrospective, cross sectional study. BMJ.2009;339:b4347

34. The Information Centre for Health and So-

cial Care. Statistics on Smoking, England2006. London, United Kingdom: The Informa-tion Centre; 2006

35. Higher Education Funding Council for Eng-land. Trends in Young Participation inHigher Education: Core Results for England.London, United Kingdom: HEFCE; 2010

36. Turley RN. Are children of young mothersdisadvantaged because of their mother’sage or family background? Child Dev. 2003;74(2):465–474

37. Deave T. Associations between child devel-opment and women’s attitudes to preg-nancy andmotherhood. J Reprod Infant Psy-chol. 2005;23(1):63–75

38. Kiernan KE, Mensah FK.Maternal Indicatorsin Pregnancy and Children’s Infancy thatSignal Future Outcomes for Children’s De-velopment, Behaviour and Health: Evidencefrom the Millennium Cohort Study. York,United Kingdom: University of York; 2009

39. Hobcraft JN, Kiernan KE. Predictive Factorsfrom Age 3 and Infancy for Poor Child Out-comes at Age 5 Relating to Children’s Devel-opment, Behaviour and Health: Evidencefrom the Millennium Cohort Study. York,United Kingdom: University of York; 2010

40. Barnes J, Ball M, Meadows P, Belsky J.Nurse-Family Partnership Programme. Sec-ond Year Pilot Sites Implementation in Eng-land. The Infancy Period. London, UnitedKingdom: Institute for the Study of Children,Families and Social Issues, Birkbeck, Uni-versity of London; 2009

41. Olds DL, Henderson CR Jr, Tatelbaum R,Chamberlin R. Improving the delivery of pre-natal care and outcomes of pregnancy: arandomized trial of nurse home visitation.Pediatrics. 1986;77(1):16–28

42. Kitzman H, Olds DL, Henderson CR Jr, et al.Effect of prenatal and infancy home visita-tion by nurses on pregnancy outcomes,childhood injuries, and repeated childbear-ing. A randomized controlled trial. JAMA.1997;278(8):644–652

43. Olds D, Kitzman H, Cole R, Robinson J, SidoraK, Luckey DW, et al. Effects of nurse home-visiting on maternal life-course and childdevelopment: age six follow-up results of arandomized trial. Pediatrics. 2004;114(6):1550–1559

44. de Jonge A. Support for teenage mothers: aqualitative study into the views of womenabout the support they received as teenagemothers. J Adv Nurs. 2001;36(1):49–57

45. Department for Children Schools and Fami-lies. Teenage Parents: Who Cares? Notting-ham, United Kingdom: Department for Chil-dren, Schools and Families; 2008

e1444 CHITTLEBOROUGH et al at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from

DOI: 10.1542/peds.2010-3222; originally published online May 2, 2011; 2011;127;e1436Pediatrics

Catherine R. Chittleborough, Debbie A. Lawlor and John W. LynchBirth Cohort

Young Maternal Age and Poor Child Development: Predictive Validity From a  

ServicesUpdated Information &

htmlhttp://pediatrics.aappublications.org/content/127/6/e1436.full.including high resolution figures, can be found at:

Supplementary Material

/peds.2010-3222.DC1.htmlhttp://pediatrics.aappublications.org/content/suppl/2011/04/21Supplementary material can be found at:

References

html#ref-list-1http://pediatrics.aappublications.org/content/127/6/e1436.full.at:This article cites 29 articles, 11 of which can be accessed free

Citations

html#related-urlshttp://pediatrics.aappublications.org/content/127/6/e1436.full.This article has been cited by 4 HighWire-hosted articles:

Subspecialty Collections

y_health_services_subhttp://pediatrics.aappublications.org/cgi/collection/communitCommunity Health Services

y_pediatrics_subhttp://pediatrics.aappublications.org/cgi/collection/communitCommunity Pediatricsthe following collection(s):This article, along with others on similar topics, appears in

Permissions & Licensing

mlhttp://pediatrics.aappublications.org/site/misc/Permissions.xhttables) or in its entirety can be found online at: Information about reproducing this article in parts (figures,

Reprints http://pediatrics.aappublications.org/site/misc/reprints.xhtml

Information about ordering reprints can be found online:

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

at Indonesia:AAP Sponsored on August 23, 2014pediatrics.aappublications.orgDownloaded from