the story of how a population study of developmental language disorder came to be….

Professor Courtenay NorburyPsychology and Language Sciences, UCL30 March 2017

…a story in three acts

• conceiving SCALES – challenges, set-backs, and what helped get it off the ground

• a research baby is born! – the tears and the glory

• looking into the future – possibilities and impact

Developmental Language Disorder

• child’s language abilities are below chronological age expectations

• language deficits are not explained by other developmental concerns such as sensory impairment, autism, extreme deprivation, head injury, intellectual impairment

• language impairments interfere with everyday life at home or at school

• formally ‘specific language impairment’

• prevalence estimate 7.4%– non-verbal ability > 86

– language cut-off -1.25SD on 2/5 composite scores

– identifies many false positives

– largely ignores co-morbidity

DSM-5 (APA, 2013); Tomblin et al. (1997)

conceiving SCALES – challenges, set-backs, and what helped get it off the

ground

Discuss need for population study with G. Baird at conference dinner

2007

You need an expert team!

2008

Submit 1st

grant application – test screening measures

2009

Start discussions with key players in Surrey about a screening study to identify cohort of children with DLD

2009

Submit 2nd

application to Wellcome

2009

DESPAIR!!

2009

Clarify thinking!Write it again… and again… and again… and again…

2010

now the fun begins…• planned a Jan 2011 start

• The plan

– Get teachers to screen ~4000 children in reception

– Relate language screen to standard academic measures

– Assess 500 in detail in Year 1 – over select low language

– Re-assess in Year 3

• HOW??

– School liaison colleagues hugely helpful

• Money in grant to buy teachers out for a day to do screening

• Set up meetings with Headteacher/SENCO forums

• Assisted with ethics (opt-out consent for screen)

• Advised use of unique pupil numbers to track children through school

• Tie screen into Early Years Foundation Stage Profile reporting period

– Visited many, many schools to get buy in…

Debbie Gooch starts MARCH 2011

Screening starts May

2011

a research baby is born! – the tears and the glory

Stage 1: Screening for language/communication ‘risk’

8340 children in sample

( 176 schools; 67%)

42 schools opted out

45 schools no reply( 4,058 children)

7267 children screened (87%)

(161 schools, ~264 teachers)

• Background information:• Child ethnicity• Home language• SEN status• Existing diagnoses• Teacher concerns

• Children’s Communication Checklist-Short (Bishop, 2003)

• Strengths and Difficulties Questionnaire (Goodman, 1997)

• New Early Years Foundation Stage Profile

12,398 children in target population

(263 schools)

20 parental opt-outs

Screens incomplete(1053 children)

• Population: 7267 children starting mainstream school in 2011 (59% of total)

• Gender: 51% boys and 49% girls• Ethnicity: 5959 children (82%) of white

British ethnic origin (83% England; 83% Surrey)

• English as additional language: 797 (11%) were rated as having English as an additional language (17% UK total; 10% Surrey)

• Socio-economic status: Income Deprivation Affecting Children Index (IDACI)

SCALES: a population study

How many children start school with language disorder?

What other developmental concerns are present?

How do these co-occurring deficits change over time?

teacher report of language

Children’s Communication Checklist –Short

males

females14% of total population

1% No Phrase Speech

• younger children rated as having more language difficulties, behaviour problems and poorer academic progress• mismatch between language abilities of 4-year-olds and curriculum

demands

• fewer than 5% of those with teacher-rated language difficulties achieved “Good Level of Development” on the EYFSP

• language best predictor of academic attainment

Should we…

Have different cut-off scores for girls and boys?

Include NPS children?

Include EAL?

Use language cut-off of -1.25SD or -1.5SD?

Include children with other diagnoses?

636 children selected for stage 2

7267 children screened (Stage 1)

529 children seen for Y1 assessment (83%)

(150 schools)

107 children did not participate in stage 2

48 NPS5499 LR 912 HREAL pilot study

61/80 EAL children seen for

assessment (76%)

777 EAL children

31 children attending special school

Exclusions

Remaining 6411 stratified by

gender and SOB(bottom 14% =

HR)

588 children randomly selected for Stage 2

sampling fraction LR<HR

Oversampled girls

499 children seen for Y3 assessment (85%)

(180 schools)

SCALES: diagnostic framework

Non-verbal ability assessed using WPSSI block design and matrix reasoning

-1.5SD on 2/5 composite scores

NVIQ composite >70

Vocabulary

composite

Grammar

composite

Narrative

composite

Expression

composite

Comprehension

composite

ROWPVT

Receptive

vocabulary

EOWPVT

Expressive

vocabulary

TROG

receptive

grammar

SASIT-E32

sentence recall

ACE Narrative

comprehension

ACE Narrative

Recall

(info units)

Stage 2: additional assessmentsDevelopmental area Assessment

Hearing Pure tone audiometer screen

Non-verbal reasoning WPPSI Block Design

WPPSI Matrix reasoning

Clinical language

‘markers’

Past tense task (Conti-Ramsden et al. 2011)

Non-word repetition

Speech DEAP articulation screen

Diadokokinetic rate

Literacy YARC Letter Sound Knowledge

YARC Phoneme Deletion

CastlesColtheart-2: regular, irregular, non-word

reading

Teacher questionnaire

Speed of processing WISC Coding

Visual search

Rapid automatic naming

Stage 2: additional assessmentsMotor skills Go task (reaction time)

Coin posting

Bead threading

Developmental Co-ordination Disorder-

Questionnaire

Attention/executive

control

Go no-go task

Visual search

Self-ordered pointing tasks (Cragg & Nation, 2009)

SWAN Questionnaire

Social understanding,

interaction and

communication

Theory of Mind experimental battery (Wellman &

Liu, 2004)

Social Responsiveness Scale

Children’s Communication Checklist-2

Learning tasks Serial Reaction Time task

Paired associate learning task

definitely out of my comfort zone…• use probability weights so

findings are representative of the population

• standardise all core language /IQ tests

• learn(ing) both Stata and Mplus

• pushing limits of statistical knowledge every day…

Expr

essi

ve V

ocab

.

Age

0

50

100

150

50 57 64 71 78 85 92 99 106 113 120

Prevalence and profilePrevalence Year 1 % of

population

Language Disorder (causeunknown)

7.58%

higher NVIQ 4.80%

lower NVIQ 2.78%

Gender (boys:girls) 1.22:1

Language Disorder (known cause and/or intellectualimpairment)

2.34%

Language Disorder (cause unknown):Only 11% achieve ‘good level of development’9.68% clinical levels of social, emotional and behavioural deficit

Language Disorder (known diagnosis):More severe deficits 50% clinical levels of social, emotional and behavioural deficit

just for fun… (reviewer 3)Weighted prevalence Achieving

Good Level of Development

SCALES: ‘specific’ LILanguage -1.5SD, NVIQ > 70

7.58% 11%

Tomblin: SLILanguage -1.25SD, NVIQ > 85

7.74% 27.60%

Tomblin: DSM-5Language -1.25SD, NVIQ > 70

11.11%

ICD-11 (proposed): SLILanguage -2SD, NVIQ > 85

1.07% 0

use Stata for data

analysis –annotate and save

all do-files

One do-file for each

paper

stability in language function to YEAR 3

-4-2

02

-4 -2 0 2 4

3 z_total_comp yTotal Language Year 1 (z-score)

Tota

l Lan

guag

e Ye

ar 3

(z-

sco

re)

Better than

Year 1

Worse than

Year 1

ICC = .94

growth in language (raw scores)

Theoretically interesting

-same rate of growth, why are the intercepts so different?

Practically important

-same rate of growth, possible to narrow the gap?

some data analytic techniques

Only two time points for core language data…

Linear mixed effects models

Differences in slopes relative to TD?

What predicts slopes?

not NVIQnot SESnot behaviour

All action is

in the intercept

some data analytic techniques

SOP

Year 1

Coding

Visual search

RAN

Simple RT (mean)

.69

.38

.52

-.42SOP

Year 3

Coding

Visual search

RAN

Simple RT (mean)

.67

.35

.61

-.46

LANG

Year 3

ROWPVT

TROG

EOWPVT

Narrative Comp

.73

.75

.81

.63

Narrative Recall

SASIT.76

.57

.95 [.88, 1.02]

(.95 [.89, 1.02])

.52

[.3

1, .

72

]

(.4

6 [.

22

, .7

0])

.51

[-.

55

, 1.5

7]

(.3

8 [-

.59

, 1.3

5])

.16

ADHD

Year 1

ADHD

Year 3

.50 [.36, .64]

(.50 [.36, .65])

.55

[-.

21

, 1.3

0]

(.5

4 [-

.17

, 1.2

5])

.56

[.3

5, .

76

]

(.5

4 [.

31

, .7

8])

LANG

Year 1

ROWPVT

TROG

EOWPVT

Narrative Comp

.73

.72

.80

.73

Narrative Recall

SASIT .78

.73.35

.97 [.72, 1.21]

(.95 [.69, 1.21])

.49

[.3

3, .

64

]

(.4

3 [.

26

, .6

0])

.18

[-.

10

, .4

6]

(.2

1 [-

.08

, .5

0])

linking in with other big data

• Because we have unique pupil numbers…

– We can link our screening data to national pupil data

• Phonics screen, SATS, receipt of pupil premium grant, receipt of EHC plan

– Also possible to link to health records

• How many children access speech-language therapy?

• Does our screen (at reception) improve prediction of referral to CAMHS in adolescence?

looking into the future –possibilities and impact

the TO DO list…

• write, write, write, write…– make sure you build time into the grant for writing– prioritise key papers

• applied for follow-up funding – fingers crossed• data archiving!!

– how to ensure anonymity– what of all this vast data is useful to others– where to put it so people can find it– sharing analysis scripts

the TO DO list…

• Impact

– Blog

– Write for the Conversation

– Use the Press Office

– Use policy organisations

• Decide your key message

– Keep it simple

– Offer the solution

Me discussing SCALES with Lord Ramsbothamin House of Commons, November 2016

the TO DO list…

• Legacy

– Creating an on-line assessment calculator

– Encourage use of standard testing protocol

– Depositing narrative samples with CHILDES

– Clinicians/educators can access up-to-date population norms for key tests

– Researchers in other countries can access same data and compare with own norms

– Different academic disciplines explore data in ways I can’t imagine!

challenges for the field

• growing statistical expertise• ethical issues around collecting/ sharing big (anonymised)

data• pre-registration – really difficult with this kind of project!• fragmented education services• incentives for health/education services to collaborate in

research– Get them on board early– Figure out what works for them– Keep them in the loop!

the SCALES team

Andrew PicklesGillian Baird Tony Charman

Emily Simonoff

PhD students: Charlotte Wray & Katie Whiteside; Grad RAs: Harriet Maydew & Claire Sears; Student RAs: Caroline, Natalie, Hayley, Charlotte, Naomi & Tanya; Becca Lucus

Debbie Gooch George Vamvakas

thanks to…

• Wellcome Trust for funding us• Surrey County Council especially:

Virginia Martin and Jennifer Saunders

• Children and their families for taking part• Teachers, SENCOs and all school staff for

their enthusiastic support – we could not do it without them!

thank you for listening!find out more about

language impairment and the impact of language

impairment on children and young people!

https://www.youtube.com/RALLIcampaign

c.norbury@ucl.ac.uk