Relationships between child, family and neighbourhood characteristics and childhood injury: A cohort study

ARTICLE IN PRESS

0277-9536/$ - se

doi:10.1016/j.so

�Correspondfax: 0115 84669

E-mail addr

(D. Kendrick).

Social Science & Medicine 61 (2005) 1905–1915

www.elsevier.com/locate/socscimed

Relationships between child, family and neighbourhoodcharacteristics and childhood injury: A cohort study

Denise Kendricka,�, Caroline Mulvaneya, Paul Burtonb, Michael Watsonc

aDivision of Primary Care, Floor 13, Tower Building, University Park, Nottingham NG7 2RD, UKbDepartment of Health Sciences, University of Leicester, 22-28 Princess Road West, Leicester LE1 6TP, UK

cUniversity of Nottingham, Queen’s Medical Centre, Nottingham NG7 2HA, UK

Available online 31 May 2005

Abstract

There has been little research into the role of neighbourhood effects in childhood injury. We report results from a

cohort study, comprising 1717 families (2357 children aged 0–7 years) registered at 47 general practices in Nottingham,

UK. Multi-level Poisson regression examined relationships between electoral ward (neighbourhood), family and child

characteristics and medically attended injury rates. Primary care attendance rates were higher for children in rented

accommodation and those aged 2–3 years. An n-shaped relationship was found between geographical access to services

and the primary care attendance rate. Accident and Emergency (A&E) department attendance rates were higher

amongst boys, children in rented accommodation, with a teenage mother, aged 2–5 years and living in wards with a

higher number of parks and play areas. They were lower for children whose families had a smoke alarm. Hospital

admission rates were higher amongst children living in more deprived wards and wards with higher violent crime rates.

They were lower in children whose families had smoke alarms, stair gates and stored sharp objects safely. Primary care

and A&E attendance rates varied significantly between families. Variation between wards in the A&E attendance rate

was explained by family characteristics. We conclude that characteristics of wards, families and children are associated

with medically attended childhood injury rates. This study did not find a neighbourhood effect for A&E attendances

that could not be explained by family level characteristics. Studies with greater power and a measure of injury severity

independent of health service utilisation are needed to explore the relationship between neighbourhood effects and more

severe injuries. The greater variation in injury rates vary between families than between neighbourhoods suggests

reducing inequalities in injury rates may be achieved more effectively by focussing prevention at families rather than

neighbourhoods, but in practice interventions at both levels are likely to be necessary.

r 2005 Elsevier Ltd. All rights reserved.

Keywords: Childhood accidents; Multi-level modelling; Deprivation; UK

Introduction

Injuries are the most common cause of death in

childhood (Roberts, DiGuiseppi, & Ward, 1998) and the

e front matter r 2005 Elsevier Ltd. All rights reserve

cscimed.2005.04.003

ing author. Tel.: 0115 8466914;

04.

ess: [email protected]

social gradient for deaths from injury is steeper than that

for any other cause of death in childhood.(Botting,

1995) Social gradients are particularly steep for deaths

caused by fire and flames, pedestrian injury, cyclist

injury, falls and poisoning (Roberts, 1997) and there is

evidence that the difference in death rates between the

more and less advantaged is increasing. There are also

steep social gradients for admissions to hospital in

childhood for pedestrian injury, burns and scalds and

d.

www.elsevier.com/locate/socscimed

ARTICLE IN PRESSD. Kendrick et al. / Social Science & Medicine 61 (2005) 1905–19151906

poisoning (Hippisley-Cox et al., 2002) and little evidence

that gradients in hospital admissions following severe

traffic injury are reducing (Coupland et al., 2003).

In recent years there has been increasing interest in

examining the effect of where people live as well as their

individual characteristics on health behaviours and

health outcomes (Diez Roux, 2001; Duncan, Jones, &

Moon, 1998; Pickett & Pearl, 2001; Reijneveld, 2002)

The role of neighbourhood effects in childhood injury

would seem particularly relevant as characteristics of the

neighbourhood such as traffic volume, the quality of

housing, the availability of safe play areas or off street

parking may be causally related to childhood injury

(Cubbin, LeClere, & Smith, 2000; O’Campo, Rao,

Gielen, Royalty, & Wilson, 2000; Reading, Langford,

Haynes, & Lovett, 1999, Soubhi, 2004) Furthermore, if

there is evidence of a neighbourhood effect this would

provide support for the use of neighbourhood level

interventions as well as interventions directed at the level

of children and families to prevent injury.

There has been little work in this area related to

childhood injury. Two recent UK studies investigated

the relationship between the Townsend score of social

areas, individual level characteristics and secondary care

attendances for injury. (Haynes, Reading, & Gale, 2003;

Reading et al., 1999) An increasing level of deprivation

of the social area was associated with an increasing odds

of injury. There was a small but significant amount of

unexplained variation in the injury rates for children

aged 0–4 years between social areas but most of the

unexplained variation was at child level. They suggested

that the area level effects may be explained by road

safety measures, housing conditions, access to amenities

or cultural attitudes to child safety and supervision.

One study in the USA (O’Campo et al., 2000)

examined the relationship between neighbourhood

characteristics and risk of events with injury-producing

potential amongst children aged 0–4 years. Higher rates

of events with injury producing potential were asso-

ciated with younger parental age and with poorer

quality housing. The authors postulated that the area

effect may be explained by children living in structurally

dangerous homes or playing in or around dilapidated

housing. A second study from Canada (Soubhi, 2004)

using self-reported data from the National Longitudinal

Survey of Children and Youth found that neighbour-

hood problems and parental perceptions of having a

difficult child were associated with a higher odds of

injury amongst children less than 2 years of age, whilst

positive and consistent parenting was associated with a

lower odds of injury for children above 2 years of age.

Neighbourhood cohesion was associated with a lower

odds of injury only amongst children who were

perceived as ‘‘difficult’’ and neighbourhood disadvan-

tage was associated with a higher odds of injury only

amongst children with aggressive behaviour. The author

concludes that intervening at the level of the neighbour-

hood would be insufficient to reduce injury rates if

parenting and child behaviour were not also addressed.

It is possible that unmeasured child or family level

characteristics known to be associated with childhood

injury, (Erens, Primatesta, & Gillian Prior, 2001; Tobin,

Milligan, Shukla, Crump, & Burton, 2002) individual

level measures of socio-economic disadvantage (Agran,

Winn, Anderson, & Del Valle, 1998; Alwash &

McCarthy, 1988; Faelker, Pickett, & Brison, 2000;

Pomerantz, Dowd, & Buncher, 2001) or safety practices

(Azizi, Zulkifli, & Kassim, 1994; DiGuiseppi, Roberts, &

Li, 1998; Elkington, Blogg, Kelly, & Carey, 1999;

Marshall et al., 1998; Petridou et al., 1998; Runyan,

Bangdiwala, Linzer, Sacks, & Butts, 1992; van Rijn,

Bouter, Kester, Knipschild, & Meertens, 1991) could

explain the area level effects found in these studies. In

addition, area characteristics could provide direct

explanations for variations in injury rates. This study

was therefore undertaken to examine the relationships

between a range of area, family and child characteristics

and medically attended unintentional injury in child-

hood. An examination of the utility of safety practices in

predicting childhood injury will be presented elsewhere.

Methods

These analyses are based on data from a cohort study

nested within the control arm of a randomised

controlled trial of childhood injury prevention in

primary care (Watson, Woods, & Kendrick, 2002).

The trial evaluated the effectiveness of health visitor

advice plus access to free or low-cost safety equipment,

fitted in the homes of families with children under 5. All

families (n ¼ 9909) with children aged under 5 years on

the caseloads of 62 health visitors attached to 47 general

practices in deprived areas (Townsend score 40) of

Nottingham, UK, were invited to participate in the trial.

A total of 3428 families (35%) agreed to participate and

were randomly allocated to the intervention and control

arms. The control arm comprised 1717 families with

2357 children. Data on socio-demographic character-

istics and safety practices were collected by postal

questionnaire at recruitment. Data on child age and

gender were obtained from the health visitor caseloads.

Questions on safety practices were validated by a home

visit (Watson, Kendrick, & Coupland, 2003) to a sample

of families to compare observed and self-reported

practices.

The neighbourhood characteristics that families con-

sidered important in determining the risk of childhood

injury were ascertained from a survey of families plus

interviews with a sample of families. All 1717 control

arm families were sent a postal questionnaire 21 months

after recruitment which contained open-ended questions

ARTICLE IN PRESSD. Kendrick et al. / Social Science & Medicine 61 (2005) 1905–1915 1907

asking respondents to identify neighbourhood charac-

teristics that they thought were important in determin-

ing the risk of injury. They were also asked if they would

agree to be interviewed to discuss their views of their

neighbourhood (defined as the area within 10 minutes

walking distance of their home). This included the

families’ concerns for their child’s safety inside and

outside the home and suggestions for ways of improving

child safety. Ten of these families were interviewed,

chosen at random from a sampling frame stratified by

Townsend score. Data from interviews were categorised

into themes and where possible, routinely collected ward

level data covering these themes were obtained from

Local Authorities in Nottingham. Ward level informa-

tion on populations, the percentage of housing by

Council Tax Band and the Index of Multiple Depriva-

tion (IMD 2000) were obtained (Department of the

Environment Transport and Regions, 2000). As the

domains of the IMD 2000 were highly correlated with

each other, we used the child poverty index (the

percentage of families with children under 16 receiving

means tested benefits) as the ward level measure of

deprivation because we considered it most relevant to

childhood injury. For primary care attendances, the

geographical access to services domain was also

considered for inclusion in the modelling as it included

access to a general practitioner, as well as to other

community facilities. Distance from the centre of the

ward of residence to the only Accident and Emergency

department in the District was calculated in kilometres.

The outcome measures were the primary care and

A&E attendance rates and the hospital admission rate

for unintentional injury, during the 2 year follow-up

period. These data were collected from primary and

secondary care records by members of the research

team. Children withdrawing from the study, those

placed on the child protection register during the course

of the study and those who moved outside Nottingham

for whom the date of moving was unknown were

excluded from the analysis. Those with a known date of

moving were included for the period during which they

resided in Nottingham. The number of years at risk of

injury was calculated for each child, taking account of

date of entry to the study and length of follow-up.

Data analysis

Continuous data were described using means and

standard deviations (SD) where they were normally

distributed and medians and interquartile ranges (IQR)

where they were non-normally distributed. Categorical

data were described using frequencies and percentages.

Poisson regression was used to examine the univariate

and multi-variable relationships between ward, family

and child level characteristics and each of the injury

outcomes. Where there was significant variation in

injury rates between wards, a 3 level random intercepts

model was used, with child at level 1, family at level 2

and ward at level 3 and child years at risk of injury as the

offset term. Where there was significant variation

between families a 2 level model was used, and where

there was no significant variation in injury rates between

wards or families a single level model was used. Random

slopes were added for covariates where it was considered

plausible that the relationship between the covariate and

the outcome variable might differ between wards or

families. They were retained in the models if they were

significant at the 5% level. Wald tests were used to assess

significance for fixed parameter estimates and the Self

and Liang w2 test for estimates of the variances of

random effects (Self & Liang, 1987). Covariates were

categorised where there was evidence of non-linearity.

Models were built by adding variables (fixed effects)

with a p-value of p0:1 in the univariate analysis into the

model, in order of their significance on univariate

analysis. They were retained in the model if they were

significant at the 5% level. At the end of the modeling,

excluded variables were reassessed for inclusion. Two-

way interactions between covariates which seemed

theoretically plausible were examined. Correlations

between coefficient estimates of explanatory variables

were assessed using the covariate correlation matrix.

Models were checked by examining plots of residual

values, leverage and influence. Analyses were under-

taken using Stata version 7 (Stata Corporation, 2001)

and MLwiN version 1.1 (Rasbash, Browne, Healy,

Cameron, & C., 2000)

Results

Interviews to determine neighbourhood characteristics

associated with risk of injury

Sixty-six per cent of families (957) returned the

neighbourhood characteristics questionnaire (excluding

276 who had changed address and 1 who had died) and

44% (418) of these agreed to be interviewed. The

neighbourhood characteristics most frequently identified

by families as being associated with a risk of injury were

speeding traffic and dangerous roads (377, 39.9%), lack

of parental supervision (231, 24.4%), lack of safe play

areas and leisure facilities (184, 19.5%) and a lack of

street and play area cleaning and maintenance (127,

13.4%). Four main themes emerged from the interviews.

These were housing (including steep stairs, exposed

central heating pipes, shared access to the back of

houses, no front garden and small back yards), traffic

(including traffic volume, on street parking, lack of

traffic calming measures), crime (vandalism) and facil-

ities for children (lack of safe play areas and affordable


leisure facilities). Thus we obtained routinely collected

data on housing value, road safety measures, crime and

local amenities including council-owned parks and

leisure centres. We also obtained data on the number

of nursery and childminder places in each ward as we

hypothesised that the provision of out of home child

care may also be important in determining injury risk.

Cohort study

Baseline data on socio-demographic characteristics,

family characteristics and safety practices were available

for 1642 families (95.6%). Data on ward characteristics

were available for 1630 families (94.9%) residing in 70

wards in Nottingham. Outcome data were available on

primary care injury attendances for 2066 children

(87.7%) and on A&E attendances and hospital admis-

sions for 2273 children (96.4%). The primary care injury

attendance rate was 44.2 per 1000 child years (95% CI

37.88–51.37), the A&E attendance rate was 174.1 per

1000 child years (95% CI 161.8–187.1) and the hospital

admission rate was 13.6 per 1000 child years (95% CI

1.8–17.6).

The socio-demographic and family characteristics and

safety practices of study families are shown in Table 1.

Table 2 describes ward level characteristics. The number

of children per ward ranged from 1 to 135 (median 18,

IQR 3–56.5).

Table 1

Socio-demographic, family characteristics and safety practices of stud

Characteristics

Child level characteristics

Mean age at end of study

Male

Family level characteristics

Overcrowding [27]

Teenage motherhood [34] (82 questionnaires were completed by fathe

X4 children under 16 years of age at baseline [5]

Non white ethnic minority group [39]

Single parent [28]

Rented accommodation [13]

No employed parents in the family [27]

No car [47]

Receives means tested benefits [66]

Median child poverty index [12]

Safety practices

Fitted fire guards [47]

Fitted stair gates [17]

Fitted and working smoke alarms [79]

Fitted window locks on upstairs windows [104]aSafe storage of medicines in kitchen [36]aSafe storage of cleaning products in kitchen [17]aSafe storage of sharp objects in kitchen [17]

aSafe defined as stored at adult eye level or above or in locked cup

The univariate relationships between ward character-

istics and injury outcomes are shown in Table 3. Two

ward characteristics were significantly associated with

the primary care attendance rate, three with the A&E

attendance rate and nine with the hospital admission

rate. Associations between ward characteristics and

injury outcomes tended to be stronger for hospital

admissions than for A&E attendances.

The univariate relationships between socio-demo-

graphic and family characteristics and safety practices

and injury outcomes are shown in Table 4. Ethnic

group, overcrowding, possession and use of a fireguard

and storage of medicines and cleaning products in the

kitchen were not significantly associated with any of the

injury outcomes (results not shown). Children living in

rented accommodation and those in families with two

unemployed parents had significantly higher primary

care attendance rates. All family characteristics in Table

4 were associated with significantly higher A&E

attendance rates. Children living in rented accommoda-

tion and those in single parent families had significantly

higher hospital admission rates. Families with a fitted

and working smoke alarm had a significantly lower rate

of A&E attendances and hospital admissions than those

without. Families with a fitted smoke alarm, a fitted stair

gate and with safe storage of sharp objects had a

significantly lower rate of hospital admission than those

without.

y families (n ¼ 1642) [missing data points]

Number (%) unless otherwise specified

4.18 (SD 1.80)

1232 (52.3)

191 (11.8)

rs) 354 (23.2)

134 (8.2)

243 (15.2)

459 (28.4)

747 (45.9)

552 (34.2)

499 (31.3)

789 (50.1)

51.6 (IQR 44.9, 62.0)

745 (46.7)

738 (45.4)

1180 (75.5)

707 (46.0)

1418 (88.3)

842 (51.8)

628 (38.7)

boards or drawers

ARTICLE IN PRESS

Table 2

Characteristics of wards in the study

Ward characteristics Median number (IQR)

Facilities (number per 1000 children under 5)

Parks and play areasc 8.01 (4.66, 11.83)

Nursery placesd 91.34 (0, 193.86)

Child minder placesd 71.25 (43.82, 109.90)

Leisure centresc 0 (0, 0.21)

Road safety measures (number per 1000 children under 5)

School crossing patrolse 3.86 (2.12, 6.59)

Zebra/pelican crossingse 1.78 (0, 3.81)

Pedestrian controlled lightse 3.37 (0, 7.06)

Small traffic calming schemesa,e 0 (0, 1.28)

Large traffic calming schemesb,e 0 (0, 0.84)

Crime reported to police Median percent (IQR)

Percentage of all dwellings reporting domestic burglariesf,g 2.45 (1.40, 4.13)

Percentage of population reporting vehicle crimef,g 2.30 (1.40, 3.40)

Percentage of population reporting violent crimef,g 1.20 (0.70, 1.75)

Child Poverty Index (percentage of families with children under 16 reliant on means tested benefits)h 33.73 (18.96, 50.62)

Housing value

Percentage of dwellings in Council Tax band A (housing of lowest value)i 44.75 (16.69, 65.16)

aSchemes with p5 adjacent roads traffic calmed.bSchemes with 45 adjacent roads traffic calmedcData obtained for year 2002.dData obtained for year 2003.eData obtained for year 2002/2003.fData obtained for Nottingham City wards for year 1999/2000.gData obtained for Nottingham County wards for 2000/2001.hIndicators for Child Poverty Index based on data for 1998/1999.iData obtained for 2001.

D. Kendrick et al. / Social Science & Medicine 61 (2005) 1905–1915 1909

The results of the multi-variable analyses are shown in

Table 5. Primary care attendance rates were 46% higher

for children living in rented accommodation and 3 times

higher for those aged 2–3, compared to those aged 0–1

years. There was an n-shaped relationship between

geographical access to services and the primary care

injury attendance rate. There was significant variation

between families in the primary care injury rate.

Model building resulted in two possible models for

A&E attendances; one containing rented accommoda-

tion and the other containing smoke alarms. This

occurred because each confounded the relationship

between the other and the A&E attendance rate. In

the rented accommodation model A&E attendance rates

were 25% higher for children in rented accommodation,

26% higher for those with a teenage mother, twice as

high for those aged 2–3 and 1.5 times higher for those

aged 4–5 than those aged 0–1 years and 32% higher for

boys. An increase of 1 in the number of parks and play

areas per 1000 children under 5 was associated with a

2% increase in the A&E attendance rate. In the smoke

alarm model those with a teenage mother had a 40%

higher and those with a smoke alarm had a 20% lower

A&E attendance rate. The effects of age and gender were

similar to those in the rented accommodation model.

There was a small but significant variation

(variance ¼ 0.05, SE (0.03); Self and Liang w2ð1Þ ¼ 2:78,

p ¼ 0:048) between wards in the A&E attendance rate in

both models, but this reduced and became non-

significant when rented accommodation was added to

the model, and reduced further when teenage mother-

hood was added. In the smoke alarm model the between

ward variation reduced and became non-significant

when smoke alarms were added to the model.

There was significant unexplained variation between

families in the A&E attendance rate in both models. The

difference in the injury rates between children aged 6–7

and those aged 0–1 varied significantly between families

in the rented accommodation model (1.34 (SE 0.74); Self

and Liang w2ð1Þ ¼ 3:27, p ¼ 0:04). In the smoke alarm

model the relationship between gender and the injury

rate differed significantly between families (1.49 (SE

0.60); Self and Liang w2ð1Þ ¼ 6:25, p ¼ 0:01) and there was

a significant negative covariance (�0.77 (SE 0.33); Self

and Liang w2ð1Þ ¼ 5:39, p ¼ 0:01) between the intercept

and slope, suggesting there was a smaller difference in

the injury rates between the sexes in families with high

injury rates than in those with lower injury rates.

In the analysis of hospital admission rates, several of

the ward level characteristics significant on univariate

ARTICLE IN PRESS

Table 3

Incidence rate ratios for univariate relationships between ward characteristics and primary and secondary care attendances and

hospital admissions for injury

Ward characteristics Primary care attendances A&E attendances Hospital admissions

IRR (95% CI)

Child poverty index

Least deprived fifth 1.00 1.00 1.00a

2nd 1.29 (0.78, 2.13) 1.15 (0.83, 1.60) 6.33 (1.88, 21.31)a

3rd 0.79 (0.44, 1.40) 1.26 (0.90, 1.78) 3.82 (1.06, 13.68)a

4th 1.11 (0.66, 1.87) 1.32 (0.94, 1.87) 5.84 (1.73, 19.72)a

Most deprived fifth 0.85 (0.48,1.50) 1.47 (1.06, 2.04)

Geographical access to services

Least deprived fifth 1.00

2nd 1.26 (0.67, 2.36)3rd 2.55 (1.41, 4.61)4th 1.96 (1.08, 3.55)Most deprived fifth 1.79 (0.98, 3.29)

Distance from hospital

Nearest quarter (0–3.3 km) 1.00 1.00 1.00

2nd (3.4–4.7 km) 1.01 (0.63, 1.62) 1.11 (0.88, 1.40) 0.80 (0.41, 1.55)

3rd (4.8–6.8 km) 0.79 (0.48, 1.31) 0.83 (0.65, 1.06) 0.47 (0.21, 1.04)

Farthest quarter (46.8 km) 0.81 (0.49, 1.33) 0.78 (0.61, 1.00) 0.45 (0.20, 1.00)

Crime reported to police (percentage per year)

Domestic dwellings experiencing a burglary 1.01 (0.94, 1.08) 1.03 (0.99, 1.07) 1.13 (1.03, 1.23)

Population experiencing vehicle crime 0.99 (0.89, 1.10) 1.02 (0.96, 1.08) 1.20 (1.08, 1.32)

Population experiencing violent crime 1.05 (0.94, 1.17) 1.02 (0.95, 1.09) 1.17 (1.07, 1.28)

Housing in Council Tax band A 1.00 (0.99, 1.01) 1.01 (1.00, 1.01) 1.01 (1.00, 1.03)

Facilities (number per 1000 children under 5)

Nursery places 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)

Child minder places 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 0.99 (0.99, 1.00)

Parks and play areas 1.02 (0.97, 1.06) 1.03 (1.00, 1.05) 1.02 (0.95, 1.09)

Leisure centres 1.07 (0.88, 1.29) 1.07 (0.96,1.20) 1.36 (1.06, 1.76)

Road safety measures (number per 1000 children under 5)

School crossing patrols 1.05 (0.99, 1.10) 0.98 (0.95,1.02) 1.01 (0.92, 1.11)

Zebra crossings 1.01 (0.99, 1.03) 1.00 (0.99, 1.01) 1.03 (1.01, 1.05)

Pedestrian controlled lights 1.02 (1.00, 1.03) 1.00 (0.99, 1.01) 1.03 (1.01, 1.04)

Small areas of traffic calmingb 0.98 (0.87, 1.10) 1.01 (0.95, 1.08) 1.08 (0.91, 1.29)

Larger areas of traffic calmingc 0.99 (0.85, 1.16) 1.08 (0.99,1.18) 0.83 (0.62, 1.12)

aThe child poverty index was categorised into quarters due to the small numbers of admissions in the most affluent category when

divided into fifths.bSchemes with p5 adjacent roads traffic calmed.cSchemes with45 adjacent roads traffic calmed.

D. Kendrick et al. / Social Science & Medicine 61 (2005) 1905–19151910

analysis were highly correlated with each other. For this

reason we only included the percentage of the popula-

tion experiencing violent crime (chosen based on

information from interviews) and the number of zebra

crossings. As the number of zebra crossings may result

from, rather than explain high injury rates in an area, we

built models with and without this variable. As there

was evidence of possible reverse causation it has been

excluded from the analyses presented in Table 5.

Model building resulted in 2 models, one containing

smoke alarms and one containing stair gates, as each

confounded the relationship between the other and the

hospital admission rate. There was no between ward or

between family variation in either model. In both models

families living in wards in the 2nd, 3rd and most

deprived quarters of all wards had hospital admission

rates between 3 and 7 times higher than those in the least

deprived quarter of wards. In the smoke alarm model,

families with a fitted and working smoke alarm had a

45% lower admission and those storing sharp objects

safely had a 56% lower admission rate than those

without these practices. In the stair gate model families

ARTICLE IN PRESS

Table 4

Incidence rate ratios for univariate relationships between socio-demographic characteristics and safety practices and primary and

secondary care attendances and hospital admissions for injury

Characteristics Primary care attendances A&E attendances Hospital admissions

IRR (95% CI)

Teenage mother 1.08 (0.71, 1.62) 1.37 (1.14, 1.66) 1.73 (0.96, 3.12)

4 or more children in family 1.25 (0.72, 2.16) 1.32 (1.01, 1.72) 1.21 (0.52, 2.84)

Single parent family 1.15 (0.78, 1.68) 1.21 (1.01, 1.45) 1.82 (1.03, 3.19)

Rented accommodation 1.48 (1.05, 2.09) 1.32 (1.12, 1.55) 1.92 (1.10, 3.36)

Number of unemployed parents

None 1.00 1.00 1.00One 1.41 (0.89, 2.24) 1.01 (0.82, 1.25) 0.78 (0.36, 1.69)Two 1.78 (1.12, 2.81) 1.33 (1.08, 1.64) 1.70 (0.87, 3.32)

No car 1.37 (0.96, 1.97) 1.21 (1.02, 1.44) 1.58 (0.89, 2.80)

Receives means tested benefits 1.42 (1.00, 2.01) 1.25 (1.05, 1.48) 1.26 (0.72, 2.22)

Male 1.18 (0.86, 1.62) 1.29 (1.10, 1.51) 1.21 (0.72, 2.04)

Age

0–1 1.00 1.00 1.002–3 2.15 (0.97, 4.73) 1.88 (1.32, 2.69) 3.30 (0.80, 13.74)

4–5 1.63 (0.73, 3.62) 1.40 (0.97, 2.00) 1.50 (0.49, 6.52)

6–7 1.52 (0.66, 3.53) 1.07 (0.73, 1.57) 0.97 (0.20, 4.81)

Safety practices

Fitted stair gates 1.09 (0.77, 1.53) 0.97 (0.82, 1.15) 0.46 (0.26, 0.83)

Fitted and working smoke alarms 0.88 (0.59, 1.30) 0.79 (0.65, 0.95) 0.51 (0.30, 0.89)aSafe storage of sharp objects in kitchen 0.89 (0.62, 1.28) 0.84 (0.70, 1.00) 0.50 (0.27, 0.93)

aSafe defined as stored at adult eye level or above or in locked cupboards or drawers.

D. Kendrick et al. / Social Science & Medicine 61 (2005) 1905–1915 1911

with a fitted stair gate had a 46% lower hospital

admission rate and those storing sharp objects safely

had a 56% lower admission rate than those without

these practices. A one-unit increase in the percentage of

the population experiencing violent crime was associated

with a 14% increase in the hospital admission rate for

injury.

Discussion

Principal findings

Two ward characteristics were associated with the

primary care attendance rate, three with the A&E

attendance rate and nine with the hospital admission

rate. The associations between ward characteristics and

injury outcomes tended to be stronger for hospital

admissions than for A&E attendances. Higher hospital

admission rates were found in more deprived wards and

those with a higher percentage of the population

experiencing violent crime. The A&E attendance rate

was higher in wards with a higher number of parks and

play areas per 1000 children under 5. There was an

n-shaped relationship between geographical access to

services and the primary care attendance rate. There was

significant variation in the injury rates between wards

only for the A&E attendance rate and this was explained

by family characteristics, including teenage motherhood,

rented accommodation and smoke alarm ownership.

Strengths and weaknesses of the study

This is the first prospective UK multi-level study of

childhood injuries to collect data on a wide range of

socio-demographic characteristics and safety practices

from families using validated questions on safety

practices (Watson et al., 2003) and the first to collect

data on ward characteristics defined by families living

within those wards as being important in determining

the risk of childhood injury. Hence as well as having the

potential to demonstrate an effect of neighbourhood

above that of the families living within the neighbour-

hood, these characteristics also had the potential to

provide direct explanations for the neighbourhood

effect.

The study had sufficient power to demonstrate

relationships between ward, socio-demographic charac-

teristics, safety practices and injury rates. This was

achieved as a result of high follow-up rates, especially

for A&E attendances and hospital admissions. However,

the small number of children with primary care

ARTIC

LEIN

PRES

STable 5

Incidence rate ratios for multivariable relationships between ward, family and child characteristics and primary and secondary care attendances and hospital admission rates for

injury

Characteristic Primary care attendance A&E attendance Hospital admission

IRR (95% CI)

Ward characteristics Rented accommodation model Smoke alarm model Stair gate model Smoke alarm model

Geographical access 1.00

Least deprived fifth

2nd 1.22 (0.65, 2.29)

3rd 2.41 (1.34, 4.34)

4th 1.92 (1.04, 3.52)

Most deprived fifth 1.65 (0.90, 3.03)

Child poverty index

Least deprived quarter 1.00 1.00

2nd 5.21 (1.52, 17.90) 7.04 (2.07, 23.94)

3rd 3.43 (0.95, 12.35) 4.23 (1.16, 15.40)

Most deprived quarter 4.50 (1.32, 15.40) 4.13 (1.17, 14.64)

Parks and play areas (number per 1000

childreno5 years old)

1.02 (1.00, 1.04)

% of population experiencing violent crime 1.14 (1.03, 1.27)

Distance from hospital

Nearest quarter 1.00

2nd 0.65 (0.32, 1.32)

3rd 0.36 (0.15, 0.86)

Farthest quarter 0.41 (0.17, 0.95)

Family characteristics

Rented accommodation 1.46 (1.03, 2.08) 1.25 (1.02, 1.52)

Teenage mother 1.26 (1.01, 1.56) 1.40 (1.15, 1.71)

Smoke alarms 0.80 (0.65, 0.99) 0.55 (0.31, 0.96)

Fitted stair gate 0.54 (0.30, 0.98)aSafe storage of sharp objects in kitchen 0.44 (0.23, 0.85) 0.44 (0.23, 0.84)

Child characteristics

Age

0–1 1.00 1.00 1.00

2–3 2.80 (1.11, 7.04) 2.18 (1.45, 3.29) 2.03 (1.32, 3.13)

4–5 2.18 (0.90, 5.27) 1.62 (1.07, 2.44) 1.55 (1.01, 2.39)

6–7 2.05 (0.77, 5.47) 1.31 (0.87, 1.98) 1.31 (0.83, 2.06)

Male 1.32 (1.11, 1.58) 1.31 (1.10, 1.56)

Random effects Variance (SE)

Variation in injury rates between wards 0.00 (0.00) 0.02 (0.02) 0.02 (0.02) 0.00 (0.00) 0.00 (0.00)

w2(1) ¼ 0.75, p ¼ 0.19 w2(1) ¼ 0.68, p ¼ 0.20

Variation in injury rates between families 1.44 (0.38) 0.56 (0.11) 0.76 (0.19) 0.00 (0.00) 0.00 (0.00)

w2ð1Þ ¼ 14:34, po0:001 w2ð1Þ ¼ 24:82, po0:001 w2ð1Þ ¼ 16:67, po0:001

aSafe defined as stored at adult eye level or above or in locked cupboards or drawers.

D.

Ken

drick

eta

l./

So

cial

Scien

ce&

Med

icine

61

(2

00

5)

19

05

–1

91

51912


attendances and hospital admissions in each ward meant

there was insufficient power to detect variations in injury

rates between wards.

The ward level data used in this study were those

routinely collected by Local Authorities or national

organisations. Many of these were crude measures of

exposure to risk, such as the number of parks or play

areas per 1000 children aged 0–4 years or housing value

as proxied by Council Tax banding. Measures of

exposure to risk such as time spent playing in parks

and play areas, measures of housing quality, traffic

volume, distance walked or cycled or time spent playing

in the street would have been more useful (Roberts &

Crombie, 1995; Roberts & Norton, 1994; Towner,

Jarvis, Walsh, & Aynsley-Green, 1994). These are not

routinely collected and it was not feasible to collect them

as part of this study. Many of the ward level variables

were inevitably highly correlated with each other,

excluding many from the analysis.

The geographical access domain of the IMD 2000

measures physical distance to shops, post offices,

primary schools and general practitioners. It has been

criticised because it ignores factors such as mobility,

access to transport and language barriers to accessing

services (Association of London Government and

Greater London Enterprise, 2003). It may therefore be

of limited use as a measure of access to primary care

services.

Explanation of study findings

The n-shaped relationship between access to services

and the primary care attendance rate for injury suggests

that areas with poor access to primary care services have

lower primary care attendance rates for injury than

those with better access. As areas with poor access were

those with a high degree of deprivation on the other

domains of the IMD 2000, it seems unlikely that these

areas will have lower injury rates. It is more plausible

that injured children are not presenting in primary care,

and are either presenting in secondary care, using

alternative services such as walk-in centres or the NHS

Direct telephone service or not receiving medical

treatment for more minor injuries. However this finding

requires confirmation due to the limitations of using the

geographical access domain of the IMD 2000 as a

measure of access to primary care services (Association

of London Government and Greater London Enter-

prise, 2003).

The finding of higher A&E attendance rates in wards

with a higher number of parks and play areas may

reflect greater exposure to hazards within parks and

playgrounds or on the journey to parks and play-

grounds. Our finding that the significant variation

between wards in the A&E attendance rate was

explained by family level factors, suggests characteristics

of families are more important in determining the risk of

A&E attendance for injury than the ward characteristics

we measured in this study. There was significant

unexplained variation in primary care and A&E

attendance rates between families and the effect of age

and gender on the A&E attendance rate also varied

significantly between families. Possible explanations for

this include differential exposure to hazards, perceptions

of risk, supervisory practices or safety rules.

There was some evidence that ward characteristics

were more strongly associated with hospital admissions

than with A&E or primary care attendances. Clinically

this is plausible as most ‘‘severe’’ injuries in this age

group such as fractures occur outside the home (e.g.

leisure or road traffic accidents) (Hippisley-Cox et al.,

2002; Office for National Statistics, 2001; Roberts

et al., 1998) and are likely to be influenced by

neighbourhood characteristics such as parks and play

areas, leisure centres and roads. However, as injury

severity was not measured in this study, care must be

taken in assuming that hospital admission necessarily

represents more severe injuries than A&E or primary

care attendance.

The higher rate of hospital admissions in wards with a

higher percentage of the population experiencing violent

crime is interesting. One possible explanation for this is

that the crime rate of a neighbourhood may reflect the

‘‘control’’ a community exerts over its members; in

which case children may be allowed to undertake more

risky activities in high crime rate areas. Alternatively the

crime rate may act as a proxy measure of the quality of

the environment of the neighbourhood which may

include factors such as unsafe housing, dilapidated

buildings, lack of safe and well-maintained play or

leisure areas or less safe roads.

Comparisons with previous work

Our findings of higher hospital admission rates, but

not A&E attendance rates amongst children living in

deprived wards confirm the findings of other large

ecological UK studies (Hippisley-Cox et al., 2002; Lyons

et al., 2000; Lyons, Lo, Heaven, & Littlepage, 1995;

Stark, Bennet, Stone, & Christi, 2002; Walsh, Jarvis,

Towner, & Aynsley-Green, 1996). As these ecological

studies were not able to adjust for individual socio-

demographic factors, it is unsurprising that some found

a stronger relationship between ward level deprivation

and injury (Hippisley-Cox et al., 2002) than found in this

study; as this may be partly explained by individual

measures of deprivation.

Two previous multi-level studies examining individual

and neighbourhood factors and their association with

childhood injury found that ‘‘social area’’ Townsend

score was significantly associated with injuries in

children aged 0–4 and 5–14 years after adjusting for


individual level variables (Haynes et al., 2003; Reading

et al., 1999). Deprivation of the area completely

explained the variation in injury rates between wards

for older children and partly explained that for younger

children. This contrasts with our findings that family

level characteristics explained the variation between

wards in A&E attendance rates. Possible explanations

for the conflicting findings include greater power to

detect between area variations in injury rates; the use of

‘‘social areas’’ which may have produced areas more

homogenous in terms of injury risk than in the previous

studies; use of children in different age groups and

previous studies have not included data on housing

tenure or safety practices, both of which we found to be

important in explaining the variation between wards in

the A&E attendance rates.

The significant unexplained variation in injury rates

between families in the primary care model and in both

A&E attendance models requires further exploration

and recent work suggests that factors such as parenting

practices, child temperament and child behaviour may

be important (Soubhi, 2004).

Implications for practice and further research

Although there may be an additional effect of the

neighbourhood, over and above the effect of family level

disadvantage on the risk of injury, we have been unable

to demonstrate such an effect, despite using neighbour-

hood characteristics considered by families with young

children to be important in terms of injury risk. The

finding in this and previous studies that most of the

variation in injury rates occurs at the level of children

and families rather than at the level of the neighbour-

hood suggests that there is greater potential to reduce

inequalities in injury rates by focussing prevention at the

level of children and families than at the level of the

neighbourhood. However, research evidence from injury

prevention and the broader field of public health

suggests a range of interventions appropriate to both

levels is likely to be necessary to reduce inequalities in

childhood injury (Haynes et al., 2003; Reading et al.,

1999; Towner, Dowswell, Mackereth, & Jarvis, 2001;

Dowswell & Towner, 2002; Tones & Tilford, 2001;

WHO, 1986; WHO, 1997).

Larger studies are required to examine whether there

is any variation between neighbourhoods in hospital

admission rates and severe injury rates measured using

an injury severity scale and the neighbourhood char-

acteristics which are associated with more severe injury.

These studies need to adjust for a range of family level

factors including living in rented accommodation, teen-

age motherhood, child age and gender, safety practices

and characteristics such as parenting behaviour and

child temperament and behaviour.

References

Agran, P. F., Winn, D. G., Anderson, C. L., & Del Valle, C.

(1998). Family, social, and cultural factors in pedestrian

injuries among Hispanic children. Injury Prevention, 4(3),

188–786.

Alwash, R., & McCarthy, M. (1988). Accidents in the home

among children under 5: Ethnic differences or social

disadvantage? British Medical Journal, 296, 1450–1453.

Association of London Government and Greater London

Enterprise. (2003). The review of the Index of Multiple

Deprivation 2000. London: Association of London Govern-

ment and Greater London Enterprise.

Azizi, B. H., Zulkifli, H. I., & Kassim, M. S. (1994).

Circumstances surrounding accidental poisoning in chil-

dren. Medical Journal of Malaysia, 49(2), 132–137.

Botting, B. E. (Ed.). (1995). The health of our children. London:

The Stationery Office Books.

Coupland, C. A. C., Hippisley-Cox, J., Kendrick, D., Groom,

L., Cross, E., & Savelyich, B. (2003). Severe traffic injuries

to children, Trent, 1992–7: time trend analysis. British

Medical Journal, 327, 593–594.

Cubbin, C., LeClere, F. B., & Smith, G. S. (2000). Socio-

economic status and injury mortality: Individual and

neighbourhood determinants. Journal of Epidemiology &

Community Health, 54(7), 517–524.

Department of the Environment Transport and Regions.

(2000). Indices of Deprivation 2000. London: Department

of the Environment, Transport and Regions.

Diez Roux, A. V. (2001). Investigating neighborhood and area

effects on health. American Journal of Public Health, 91(11),

1783–1789.

DiGuiseppi, C., Roberts, I., & Li, L. (1998). Smoke alarm

ownership and house fire death rates in children. Journal of

Epidemiology & Community Health, 52(11), 760–761.

Dowswell, T., & Towner, E. (2002). Social deprivation and the

prevention of unintentional injury in childhood: A systema-

tic review. Health Education Research, 17(2), 221–786.

Duncan, C., Jones, K., & Moon, G. (1998). Context,

composition and heterogeneity: Using multilevel models in

health research. Social Science & Medicine, 46(1), 97–117.

Elkington, J., Blogg, S., Kelly, J., & Carey, V. (1999). Head

injuries in infants: A closer look at baby walkers, stairs and

nursery furniture. NSW Public Health Bulletin, 10(7), 82–83.

Erens, B., Primatesta, P., & Gillian Prior (Eds.). (2001). Health

Survey for England—the health of minority ethnic groups ‘99.

London: The Stationary Office.

Faelker, T., Pickett, W., & Brison, R. J. (2000). Socioeconomic

differences in childhood injury: A population based

epidemiologic study in Ontario, Canada. Injury Prevention,

6(3), 203–786.

Haynes, R., Reading, R., & Gale, S. (2003). Household and

neighbourhood risks for injury to 5–14 year old children.

Social Science & Medicine., 57(4), 625–636.

Hippisley-Cox, J., Groom, L., Kendrick, D., Coupland, C.,

Webber, E., & Savelyich, B. (2002). Cross sectional survey

of socioeconomic variations in severity and mechanism of

childhood injuries in Trent 1992–7. British Medical Journal,

324, 1132–1786.

Lyons, R. A., Delahunty, A. M., Heaven, M., McCabe, M.,

Allen, H., & Nash, P. (2000). Incidence of childhood


fractures in affluent and deprived areas: Population based

study. British Medical Journal, 320, 149.

Lyons, R. A., Lo, S. V., Heaven, M., & Littlepage, B. N. (1995).

Injury surveillance in children—usefulness of a centralised

database of accident and emergency attendances. Injury

Prevention, 1(3), 173–786.

Marshall, S. W., Runyan, C. W., Bangdiwala, S. I., Linzer, M.

A., Sacks, J. J., & Butts, J. D. (1998). Fatal residential fires:

Who dies and who survives? Journal of the American

Medical Association, 279, 1633–1637.

O’Campo, P., Rao, R. P., Gielen, A. C., Royalty, W., &

Wilson, M. (2000). Injury-producing events among children

in low-income communities: The role of community

characteristics. Journal of Urban Health, 77(1), 34–49.

Office for National Statistics. (2001). Mortality statistics: Injury

and poisoning—DH4. London: Office for National Statistics.

Petridou, E., Trichopoulos, D., Mera, E., Papadatos, Y.,

Papazoglou, K., Marantos, A., & Skondras, C. (1998).

Risk factors for childhood burn injuries: A case-control

study from Greece. Burns, 24(2), 123–128.

Pickett, K. E., & Pearl, M. (2001). Multilevel analyses of

neighbourhood socioeconomic context and health out-

comes: A critical review. Journal of Epidemiology &


Pomerantz, W. J., Dowd, M. D., & Buncher, C. R. (2001).

Relationship between socioeconomic factors and severe

childhood injuries. Journal of Urban Health, 78(1), 141–151.

Rasbash, J., Browne, W., Healy, M., Cameron, B., & C., C.

(2000). MLwiN (Version 1.10.006). London: Institute of

Education, University of London.

Reading, R., Langford, I., Haynes, R., & Lovett, A. (1999).

Accidents to preschool children: Comparing family and

neighbourhood risk factors. Social Science & Medicine, 48,

321–330.

Reijneveld, S. A. (2002). Neighbourhood socioeconomic con-

text and self reported health and smoking: A secondary

analysis of data on seven cities. Journal of Epidemiology &


Roberts, I. (1997). Cause specific social class mortality

differentials for child injury and poisoning in England and

Wales. Journal of Epidemiology & Community Health, 51(3),

334–335.

Roberts, I., & Crombie, I. (1995). Child pedestrian deaths:

Sensitivity to traffic volume—evidence from the USA.

Journal of Epidemiology and Community Health, 49,

186–188.

Roberts, I., DiGuiseppi, C., & Ward, H. (1998). Childhood

injuries: Extent of the problem, epidemiological trends, and

costs. Injury Prevention, 4(4 (Suppl)), S10–S16.

Roberts, I., & Norton, R. (1994). Auckland children’s exposure

to risk as pedestrians. New Zealand Medical Journal,

107(984), 331–333.

Runyan, C. W., Bangdiwala, S. I., Linzer, M. A., Sacks, J. J., &

Butts, J. (1992). Risk factors for fatal residential fires. New

England Journal of Medicine, 327(12), 859–863.

Self, G. S., & Liang, K. Y. (1987). Asymptotic properties of

maximum likelihood estimators and likelihood ratio tests

under nonstandard conditions. Journal of the American

Statistical Association, 82, 605–610.

Soubhi, H. (2004). The social context of childhood injury in

Canada: Integration of the NLSCY findings. American

Journal of Health Behavior, 28(Suppl 1), S38–S50.

Stark, A. D., Bennet, G. C., Stone, D. H., & Christi, P. (2002).

Association between childhood fractures and poverty:

Population based study. British Medical Journal, 324, 457.

Stata Corporation. (2001). Intercooled Stata 7.0 for Windows.

College Station, TX: Stata Corporation.

Tobin, M. D., Milligan, J., Shukla, R., Crump, B., & Burton, P.

R. (2002). South Asian ethnicity and risk of childhood

accidents: An ecological study at enumeration district level in

Leicester. Journal of Public Health Medicine, 24(4), 313–466.

Tones, K., & Tilford, S. (2001). Health promotion. Effectiveness,

efficiency and equity. Cheltenham: Nelson Thornes.

Towner, E., Dowswell, T., Mackereth, C., & Jarvis, S. (2001).

What works in preventing unintentional injuries in children

and young adolescents? An updated systematic review.

London: Health Development Agency.

Towner, E. M., Jarvis, S. N., Walsh, S. S., & Aynsley-Green, A.

(1994). Measuring exposure to injury risk in schoolchildren

aged 11–14. British Medical Journal, 308(6926), 449–452.

van Rijn, O. J., Bouter, L. M., Kester, A. D., Knipschild, P. G.,

& Meertens, R. M. (1991). Aetiology of burn injuries among

children aged 0–4 years: Results of a case-control study.

Burns, 17(3), 213–219.

Walsh, S. S., Jarvis, S. N., Towner, E. M., & Aynsley-Green, A.

(1996). Annual incidence of unintentional injury among

54,000 children. Injury Prevention, 2(1), 16–786.

Watson, M., Kendrick, D., & Coupland, C. (2003). Validation

of a home safety questionnaire used in a randomised

controlled trial. Injury Prevention, 9(2), 180–183.

Watson, M., Woods, A., & Kendrick, D. (2002). Injury

prevention: Working together on an RCT. Community

Practitioner, 75(5), 172–175.

WHO. (1986). Ottawa charter for health promotion. Copenha-

gen: WHO.

WHO. (1997). The Jakarta declaration on leading health

promotion into the 21st century. Health Promotion Inter-

national, 12, 261–264.

Documents

Relationships between child, family and neighbourhood characteristics and childhood injury: A cohort study