Accepted Manuscript Child Injury: Does Home Matter? Jodie M. Osborne, Tamzyn M. Davey, Anneliese B. Spinks, Roderick J. McClure, Neil Sipe, Cate M. Cameron PII:
S0277-9536(16)30072-7
DOI:
10.1016/j.socscimed.2016.02.017
Reference:
SSM 10518
To appear in:
Social Science & Medicine
Received Date: 10 August 2015 Revised Date:
9 February 2016
Accepted Date: 11 February 2016
Please cite this article as: Osborne, J.M., Davey, T.M., Spinks, A.B., McClure, R.J., Sipe, N., Cameron, C.M., Child Injury: Does Home Matter?, Social Science & Medicine (2016), doi: 10.1016/ j.socscimed.2016.02.017. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT Title: Child Injury: Does Home Matter? Authors: Jodie M. Osbornea, Tamzyn M. Daveyb, Anneliese B. Spinksa, c, Roderick J. McClured, Neil Sipee, Cate M. Cameronf
a
RI PT
Affiliation:
School of Medicine, Griffith University, Meadowbrook, Australia, [email protected], [email protected]
The University of Queensland, School of Public Health, Herston, Australia, [email protected]
c
Commonwealth Scientific and Industrial Research Organisation (CSIRO), Dutton Park, Australia,
SC
b
d
Harvard Injury Control Research Center, Harvard School of Public Health, Boston, USA, [email protected]
e
School of Geography, Planning and Environmental Management, The University of Queensland, St Lucia, Australia, [email protected]
CONROD Injury Research Centre, Menzies Health Institute Queensland, Griffith University,
TE D
f
M AN U
[email protected]
Meadowbrook, Australia, [email protected]
Osborne,
School
AC C
Jodie
EP
Corresponding Author:
of
Medicine,
Griffith
University,
[email protected], phone: +61733821377, fax: +61 7 3382 1338
Meadowbrook,
Australia,
ACCEPTED MANUSCRIPT CHILD INJURY: DOES HOME MATTER? ABSTRACT This study examined the relationship between home risk and hospital treated injury in Australian children up to five years old. Women with children between two and four years of age enrolled in the
RI PT
[REMOVED] were invited to complete a Home Injury Prevention Survey from March 2013 to June 2014. A total home risk score (HRS) was calculated and linked to the child’s injury related state-wide hospital emergency and admissions data and [REMOVED] baseline demographic surveys.
SC
Data from 562 households relating to 566 child participants were included. We found an inverse relationship between home risk and child injury, with children living in homes with the least injury
M AN U
risk (based on the absence of hazardous structural features of the home and safe practices reported) having 1.90 times the injury rate of children living in high risk homes (95% CI 1.15-3.14). Whilst this appears counter-intuitive, families in the lowest risk homes were more likely to be socio-economically disadvantaged than families in the highest risk homes (more sole parents, lower maternal education
TE D
levels, younger maternal age and lower income). After adjusting for demographic and socio-economic factors, the relationship between home risk and injury was no longer significant (p>0.05). Our findings suggest that children in socio-economically deprived families have higher rates of injury,
EP
despite living in a physical environment that contains substantially fewer injury risks than their less deprived counterparts. Although measures to reduce child injury risk through the modification of the
AC C
physical environment remain an important part of the injury prevention approach, our study findings support continued efforts to implement societal-wide, long term policy and practice changes to address the socioeconomic differentials in child health outcomes.
KEY WORDS Childhood injury, socio-economic factors, cohort studies, home safety, home hazards, data linkage, Australia
1
ACCEPTED MANUSCRIPT INTRODUCTION Within Australia, as well as in other developing countries, the majority of injuries to children under five years of age occur in the home or neighbouring environment (Flavin, Dostaler, Simpson, Brison, & Pickett, 2006; Gulliver, Dow, & Simpson, 2005; Harris & Pointer, 2012; Phelan, Khoury,
RI PT
Kalkwarf, & Lanphear, 2005). A recent Australian Institute of Health and Welfare (AIHW) report on hospitalised injury in New South Wales found that 89% of children under 1 year and 78% of children 1-4 years were injured in the home (Harris & Pointer, 2012). Young children are particularly vulnerable to injuries as their ability to explore their physical surroundings increases before they
SC
develop the skills they need to identify and respond to potential risks (Garzon, 2005)
M AN U
Both parental supervision and environmental modification strategies are important tools for reducing home injuries in young children (Morrongiello, Ondejko, & Littlejohn, 2004). Environmental modification strategies limit a child’s injury risk by either eliminating household hazards or by using safety equipment and practices to restrict access to these hazards. Evidence linking home safety practices/equipment to a reduction in injury rates has primarily come from studies examining the
TE D
effect of specific safety devices on their related area of injury. These include for example, working smoke alarms on burn injuries (DiGuiseppi, Roberts, & Li, 1998; Marshall et al., 1998; Runyan, Bangdiwala, Linzer, Sacks, & Butts, 1992), pool fencing on drowning rates (Thompson & Rivara,
EP
2000) child resistant closures on poisoning rates (Rodgers, 1996) and stair gates on fall injuries
AC C
(Elkington, Blogg, Kelly, & Carey, 1999). The broader relationship between measures of families’ home safety practices and childhood injury has been the focus of more recent research. Of particular interest has been whether home safety practices are a useful predictor of injury and whether there are differences in the home safety practices of injured and uninjured children. The findings of these studies have been mixed with some showing a link between specific types of home safety practices or hazards and childhood injury (Kamal, 2013; Kendrick, Watson, Mulvaney, & Burton, 2005b; LeBlanc et al., 2006; Ramsay et al., 2003) and others finding no significant association between safety practices and injury occurrence (Hapgood, Kendrick, & Marsh, 2001). One study by Pearce et al. (2012) found few indicators of the child’s household 2
ACCEPTED MANUSCRIPT environment were significantly associated with home injuries, but interestingly found that children living in homes with no safety equipment were less likely to have been injured than children with all four items of safety equipment examined. Whilst the safety of a child’s physical home environment is only one of many complex individual,
RI PT
family, neighbourhood and environmental risk factors for injury (Garzon, 2005), understanding its relationship to a child’s injury risk is important in informing the development of targeted injury prevention programs. The purpose of this study was to explore the relationship between home risk
SC
(based on home safety practices and physical features of the home) and injuries requiring hospital
MATERIALS AND METHODS Study design
M AN U
admission or emergency department (ED) treatment in Australian children less than 5 years of age.
This study used a nested cohort design that examined the association between parents’ responses to a self-report home safety questionnaire and child injury occurrences that resulted in attendance at either
TE D
an ED or admission to hospital in [REMOVED].
This research was conducted within the framework of an Australian longitudinal birth cohort study,
EP
[REMOVED] [Anonymous, 2012b] [Anonymous, 2012a]. [REMOVED] participants are the children of women who had registered to give birth at the main public maternity hospitals within a peri-urban
AC C
region over a six year period. Trained midwives approached all women waiting for third trimester antenatal clinic appointments, provided them with a comprehensive project description and invited them to take part in the study. [REMOVED] is listed on the Australian and New Zealand Clinical Trials Registry [REMOVED]. A detailed description of the methodology has been published previously [Anonymous, 2012a]. The [REMOVED] study population was mostly representative of births in the area, however there was a greater than the national average representation of families with lower incomes, younger maternal age, more overseas born parents and high proportions of maternal smoking in pregnancy, consistent with the public hospital setting [Anonymous, 2012b].
3
ACCEPTED MANUSCRIPT Participants Participants were families with children enrolled in the [REMOVED] study between two and four years of age. These families were recruited in 2009, 2010 and 2011 from the [REMOVED] Hospitals. A total of 1,249 families were active participants at the time of this research. Data was collected on
RI PT
the enrolled child only, not siblings or other children residing in the home.
Ethics Approval
Ethics approval was provided by the Human Research Ethics Committees of [REMOVED] and the
SC
hospitals participating in the [REMOVED] project. For the [REMOVED] study, each participant gave written informed consent for completion of a maternal baseline survey, follow-up contact, the
child’s ED and hospital admission records.
Data Sources and Collection Home Injury Prevention Survey
M AN U
release of hospital perinatal data related to the birth of their child, and state-wide linkage of their
TE D
Data on home safety was collected by a postal or online Home Injury Prevention Survey (HIP) from March 2013 to June 2014. Invitation letters and the survey were mailed to all 1,249 eligible families at their last recorded place of address. Surveys were addressed to the child’s mother, but may have been
EP
completed by either parent or other caregiver residing in the home. Email invitations and online survey links were sent to families with an email address recorded. Reminder letters were sent three
AC C
and five weeks after the first letter, an email/SMS reminder was sent at seven weeks, and follow up phone calls were made after 9 weeks. All participants who completed the survey were placed in a prize draw to win one of two $50 gift vouchers per cohort group. Return of the questionnaire was considered to indicate consent for participation in the study. The HIP survey was developed from three home safety questionnaires (Kids Health The Children's Hospital at Westmead, 2010; Lyons et al., 2008; Watson, Kendrick, & Coupland, 2003). A general section sought information on the participant’s home specifications, including the number of stories (levels), year of construction, details of renovations, whether it was owned or rented and the number
4
ACCEPTED MANUSCRIPT of rooms. The safety measures/hazards section included questions about the overall environment (electrical safety switches, safety plugs, hot water temperature, bath mats, safety guards around heaters/fires, window guards and access to garage/garden shed), smoke alarms, stairs, decks, medicines, chemicals and cleaning products and outdoor safety (pools/fishponds, play equipment and
RI PT
driveways). These questions reflect Australian based standards and building codes and best practice recommendations (Kids Health The Children's Hospital at Westmead, 2010; Queensland Government; Standards Australia). Photographs showing the safety mechanisms were included to facilitate
SC
participant responses.
A cross-sectional validation study, comparing parent’s self-administered responses to the safety
M AN U
measures/hazards component of the HIP survey with home observations undertaken by trained researchers was conducted in 2012 [Anonymous, 2015]. The majority of items had high levels of absolute agreement and 75% of items had sensitivities/specificities that were 70% or above where they could be calculated. Over-reporting of safe practice was found in half the items and underreporting in a third. Modifications were made to the HIP survey as a result of the validation study.
TE D
This included removing items with low specificity and sensitivity (blind and curtain cords) and deleting uncommon items (e.g. stove and oven guards, room locks). The poison questions were also
Home Risk Score
EP
rewritten to improve the accuracy of participant reporting.
AC C
A Home Risk Score (HRS) was calculated for each participating family based on their responses to 31 questions from the safety measures/hazards component of the HIP survey. These items included safety practices and house features that are established in the literature as being relevant to injury (DiGuiseppi et al., 1998; Elkington et al., 1999; Thompson & Rivara, 2000). Safety items that were applicable to all homes (e.g. safety switches, safety plugs, smoke alarms) were scored as either “1” item was present or “2” item was absent. For safety items which relied on the presence of a specific physical structure (e.g. stairs, balconies, pool, play equipment) multiple questions were combined and were numerically scored as being either a low risk “1” (physical structure was not on property), moderate risk “2” (physical structure was present and all safety measures were in place) or high risk 5
ACCEPTED MANUSCRIPT “3” (physical structure was present and one or more safety measures were not in place). These 18 items were clustered according to five injury domains: burns/electrocution, falls, poisoning, drowning and driveway accidents (Table 1), reflecting leading causes of injury in young children in Australia
Insert Table 1 here
RI PT
(Pointer, 2014).
A score for each injury domain was calculated by summing the scores for included items, with a possible range from 18-46. Weighting was then applied to ensure equal contribution to the final HRS
SC
with each domain contributing a total 20%. For example, the burns/electrocution domain obtained a maximum total score of 11 out of 46 (24%) and thus was weighted down to 20% by the following
M AN U
formula: Domain Raw Score x (0.2/0.24). The final HRS was calculated by summing the adjusted scores for the five injury domains. Total HRSs were divided into quartiles for analysis.
Demographic, Maternal and Household Variables
Information on child, maternal and household variables of interest was drawn from the [REMOVED]
TE D
baseline survey, completed by women during the second half of their pregnancy, and hospital perinatal records. Child characteristics were gender and age. Maternal variables were maternal level of education, maternal age, smoking in pregnancy and high-risk alcohol and drug use during
EP
pregnancy. Baseline household characteristics were marital status, other children living in the household, partner’s employment status, household income quintiles [Anonymous, 2012b] and home
AC C
ownership.
Outcome data
Injury presentations were extracted from the [REMOVED] Emergency Department Information Systems (EDIS) and injury related hospital admissions were extracted from the [REMOVED] Hospital Admitted Patients Data Collection [REMOVED] from the child’s date of birth until 31 December 2013 inclusive. ED records were only available for public hospitals, while state-wide inpatient records were drawn from all private and public hospitals. The [REMOVED] Department of Health linked demographic, child and maternal details through linkage software using 6
ACCEPTED MANUSCRIPT deterministic and probabilistic methodologies. Manual clerical reviews were undertaken where necessary.
Variables Injury classification
RI PT
Classification of injury presentations/admissions was based on ICD-10-AM, with injury cases being identified from Chapter 19 Injury and Poisonings (S00-T98) and Chapter 20 External Causes (U50Y98), excluding late effects from injury. All free text descriptions of ED records were manually
secondary diagnoses of injuries were included.
M AN U
Outcome measures
SC
checked to identify any injuries that were not recorded as injuries in the single diagnosis field. No
The outcome measure of interest was the total number of injury “episodes of care” derived from ED presentations and hospital admissions. Dates of admissions and ED presentations, transfer codes and diagnostic information were used to collapse multiple admissions, nested admissions and related ED
Calculation of person-years
TE D
presentations to single episodes of care that related to the one injury event.
EP
Children in this study were born between 1 July and 30 November from 2009 to 2011, with injury outcomes data available to the end of 2013. Therefore, the length of follow-up differed considerably.
AC C
Using dates of birth and coverage dates, individual person-years (PYs) were calculated for each child with the total time he or she was residing in the state of [REMOVED], alive and eligible for health care, from birth to 31 December 2013.
Analysis
Data cleaning and analyses were undertaken using SAS 9.4 software (SAS). The statistical significance of differences between groups was assessed by Pearson’s chi-squared test for categorical data, the Kruskall-Wallis test and Mann-Whitney test for ordinal data. All tests were two-sided with a
7
ACCEPTED MANUSCRIPT 5% level of significance. Rates of injury related episodes of care were calculated for each HRS quartile, taking into account PYs exposure data. Cases with missing data across more than 20% of Home Risk Score items were excluded from all analyses. Multiple imputation methods were used to impute incomplete multivariate exposure factors
RI PT
by Fully Conditional Specification (FCS) (Liu & De, 2015; Spratt et al., 2010) using 20 iterations. Factors associated with ‘missingness’ as well as the outcome variable were chosen for the imputation model. For binary variables, logistic regression; for ordered categorical variables, ordinal logistic
SC
regression; and continuous variables, regression; were specified in the imputation phase (UCLA : Statistical Consulting Group). For the analysis phase, zero-inflated poisson regression was used to
M AN U
estimate crude and adjusted rate ratios (RRs) between exposure (HRS quartiles) and outcome (count of injury related episodes of care) for child participants, accounting for individual PYs exposure time. A likelihood ratio test was used to compare poisson regression and negative binomial regression model fit. A Vuong test SAS macro (SAS; Vuong, 1989) was used to check whether a zero-inflated regression model was required to account for the presence of “excess zeroes” in the outcomes data.
TE D
Based on the likelihood ratio test, the Vuong test, and the goodness of fit criteria, the zero-inflated poisson regression model was the best model for the hospital count data (Gardner, Mulvey, & Shaw,
EP
1995; McCullagh & Nelder, 1989).
Checks for multicollinearity were conducted during the initial model preparation phase and all factors
AC C
had a correlation below 0.3 and thus not considered a concern. Factors shown to be associated with both the exposure (HRS) and the outcome (child injury episode rate) in univariate analysis where p0.05). For all items included in the HRS, the number of missing ranged from 0%-2.3%, with 83.9% of the items having
S0277-9536(16)30072-7
DOI:
10.1016/j.socscimed.2016.02.017
Reference:
SSM 10518
To appear in:
Social Science & Medicine
Received Date: 10 August 2015 Revised Date:
9 February 2016
Accepted Date: 11 February 2016
Please cite this article as: Osborne, J.M., Davey, T.M., Spinks, A.B., McClure, R.J., Sipe, N., Cameron, C.M., Child Injury: Does Home Matter?, Social Science & Medicine (2016), doi: 10.1016/ j.socscimed.2016.02.017. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT Title: Child Injury: Does Home Matter? Authors: Jodie M. Osbornea, Tamzyn M. Daveyb, Anneliese B. Spinksa, c, Roderick J. McClured, Neil Sipee, Cate M. Cameronf
a
RI PT
Affiliation:
School of Medicine, Griffith University, Meadowbrook, Australia, [email protected], [email protected]
The University of Queensland, School of Public Health, Herston, Australia, [email protected]
c
Commonwealth Scientific and Industrial Research Organisation (CSIRO), Dutton Park, Australia,
SC
b
d
Harvard Injury Control Research Center, Harvard School of Public Health, Boston, USA, [email protected]
e
School of Geography, Planning and Environmental Management, The University of Queensland, St Lucia, Australia, [email protected]
CONROD Injury Research Centre, Menzies Health Institute Queensland, Griffith University,
TE D
f
M AN U
[email protected]
Meadowbrook, Australia, [email protected]
Osborne,
School
AC C
Jodie
EP
Corresponding Author:
of
Medicine,
Griffith
University,
[email protected], phone: +61733821377, fax: +61 7 3382 1338
Meadowbrook,
Australia,
ACCEPTED MANUSCRIPT CHILD INJURY: DOES HOME MATTER? ABSTRACT This study examined the relationship between home risk and hospital treated injury in Australian children up to five years old. Women with children between two and four years of age enrolled in the
RI PT
[REMOVED] were invited to complete a Home Injury Prevention Survey from March 2013 to June 2014. A total home risk score (HRS) was calculated and linked to the child’s injury related state-wide hospital emergency and admissions data and [REMOVED] baseline demographic surveys.
SC
Data from 562 households relating to 566 child participants were included. We found an inverse relationship between home risk and child injury, with children living in homes with the least injury
M AN U
risk (based on the absence of hazardous structural features of the home and safe practices reported) having 1.90 times the injury rate of children living in high risk homes (95% CI 1.15-3.14). Whilst this appears counter-intuitive, families in the lowest risk homes were more likely to be socio-economically disadvantaged than families in the highest risk homes (more sole parents, lower maternal education
TE D
levels, younger maternal age and lower income). After adjusting for demographic and socio-economic factors, the relationship between home risk and injury was no longer significant (p>0.05). Our findings suggest that children in socio-economically deprived families have higher rates of injury,
EP
despite living in a physical environment that contains substantially fewer injury risks than their less deprived counterparts. Although measures to reduce child injury risk through the modification of the
AC C
physical environment remain an important part of the injury prevention approach, our study findings support continued efforts to implement societal-wide, long term policy and practice changes to address the socioeconomic differentials in child health outcomes.
KEY WORDS Childhood injury, socio-economic factors, cohort studies, home safety, home hazards, data linkage, Australia
1
ACCEPTED MANUSCRIPT INTRODUCTION Within Australia, as well as in other developing countries, the majority of injuries to children under five years of age occur in the home or neighbouring environment (Flavin, Dostaler, Simpson, Brison, & Pickett, 2006; Gulliver, Dow, & Simpson, 2005; Harris & Pointer, 2012; Phelan, Khoury,
RI PT
Kalkwarf, & Lanphear, 2005). A recent Australian Institute of Health and Welfare (AIHW) report on hospitalised injury in New South Wales found that 89% of children under 1 year and 78% of children 1-4 years were injured in the home (Harris & Pointer, 2012). Young children are particularly vulnerable to injuries as their ability to explore their physical surroundings increases before they
SC
develop the skills they need to identify and respond to potential risks (Garzon, 2005)
M AN U
Both parental supervision and environmental modification strategies are important tools for reducing home injuries in young children (Morrongiello, Ondejko, & Littlejohn, 2004). Environmental modification strategies limit a child’s injury risk by either eliminating household hazards or by using safety equipment and practices to restrict access to these hazards. Evidence linking home safety practices/equipment to a reduction in injury rates has primarily come from studies examining the
TE D
effect of specific safety devices on their related area of injury. These include for example, working smoke alarms on burn injuries (DiGuiseppi, Roberts, & Li, 1998; Marshall et al., 1998; Runyan, Bangdiwala, Linzer, Sacks, & Butts, 1992), pool fencing on drowning rates (Thompson & Rivara,
EP
2000) child resistant closures on poisoning rates (Rodgers, 1996) and stair gates on fall injuries
AC C
(Elkington, Blogg, Kelly, & Carey, 1999). The broader relationship between measures of families’ home safety practices and childhood injury has been the focus of more recent research. Of particular interest has been whether home safety practices are a useful predictor of injury and whether there are differences in the home safety practices of injured and uninjured children. The findings of these studies have been mixed with some showing a link between specific types of home safety practices or hazards and childhood injury (Kamal, 2013; Kendrick, Watson, Mulvaney, & Burton, 2005b; LeBlanc et al., 2006; Ramsay et al., 2003) and others finding no significant association between safety practices and injury occurrence (Hapgood, Kendrick, & Marsh, 2001). One study by Pearce et al. (2012) found few indicators of the child’s household 2
ACCEPTED MANUSCRIPT environment were significantly associated with home injuries, but interestingly found that children living in homes with no safety equipment were less likely to have been injured than children with all four items of safety equipment examined. Whilst the safety of a child’s physical home environment is only one of many complex individual,
RI PT
family, neighbourhood and environmental risk factors for injury (Garzon, 2005), understanding its relationship to a child’s injury risk is important in informing the development of targeted injury prevention programs. The purpose of this study was to explore the relationship between home risk
SC
(based on home safety practices and physical features of the home) and injuries requiring hospital
MATERIALS AND METHODS Study design
M AN U
admission or emergency department (ED) treatment in Australian children less than 5 years of age.
This study used a nested cohort design that examined the association between parents’ responses to a self-report home safety questionnaire and child injury occurrences that resulted in attendance at either
TE D
an ED or admission to hospital in [REMOVED].
This research was conducted within the framework of an Australian longitudinal birth cohort study,
EP
[REMOVED] [Anonymous, 2012b] [Anonymous, 2012a]. [REMOVED] participants are the children of women who had registered to give birth at the main public maternity hospitals within a peri-urban
AC C
region over a six year period. Trained midwives approached all women waiting for third trimester antenatal clinic appointments, provided them with a comprehensive project description and invited them to take part in the study. [REMOVED] is listed on the Australian and New Zealand Clinical Trials Registry [REMOVED]. A detailed description of the methodology has been published previously [Anonymous, 2012a]. The [REMOVED] study population was mostly representative of births in the area, however there was a greater than the national average representation of families with lower incomes, younger maternal age, more overseas born parents and high proportions of maternal smoking in pregnancy, consistent with the public hospital setting [Anonymous, 2012b].
3
ACCEPTED MANUSCRIPT Participants Participants were families with children enrolled in the [REMOVED] study between two and four years of age. These families were recruited in 2009, 2010 and 2011 from the [REMOVED] Hospitals. A total of 1,249 families were active participants at the time of this research. Data was collected on
RI PT
the enrolled child only, not siblings or other children residing in the home.
Ethics Approval
Ethics approval was provided by the Human Research Ethics Committees of [REMOVED] and the
SC
hospitals participating in the [REMOVED] project. For the [REMOVED] study, each participant gave written informed consent for completion of a maternal baseline survey, follow-up contact, the
child’s ED and hospital admission records.
Data Sources and Collection Home Injury Prevention Survey
M AN U
release of hospital perinatal data related to the birth of their child, and state-wide linkage of their
TE D
Data on home safety was collected by a postal or online Home Injury Prevention Survey (HIP) from March 2013 to June 2014. Invitation letters and the survey were mailed to all 1,249 eligible families at their last recorded place of address. Surveys were addressed to the child’s mother, but may have been
EP
completed by either parent or other caregiver residing in the home. Email invitations and online survey links were sent to families with an email address recorded. Reminder letters were sent three
AC C
and five weeks after the first letter, an email/SMS reminder was sent at seven weeks, and follow up phone calls were made after 9 weeks. All participants who completed the survey were placed in a prize draw to win one of two $50 gift vouchers per cohort group. Return of the questionnaire was considered to indicate consent for participation in the study. The HIP survey was developed from three home safety questionnaires (Kids Health The Children's Hospital at Westmead, 2010; Lyons et al., 2008; Watson, Kendrick, & Coupland, 2003). A general section sought information on the participant’s home specifications, including the number of stories (levels), year of construction, details of renovations, whether it was owned or rented and the number
4
ACCEPTED MANUSCRIPT of rooms. The safety measures/hazards section included questions about the overall environment (electrical safety switches, safety plugs, hot water temperature, bath mats, safety guards around heaters/fires, window guards and access to garage/garden shed), smoke alarms, stairs, decks, medicines, chemicals and cleaning products and outdoor safety (pools/fishponds, play equipment and
RI PT
driveways). These questions reflect Australian based standards and building codes and best practice recommendations (Kids Health The Children's Hospital at Westmead, 2010; Queensland Government; Standards Australia). Photographs showing the safety mechanisms were included to facilitate
SC
participant responses.
A cross-sectional validation study, comparing parent’s self-administered responses to the safety
M AN U
measures/hazards component of the HIP survey with home observations undertaken by trained researchers was conducted in 2012 [Anonymous, 2015]. The majority of items had high levels of absolute agreement and 75% of items had sensitivities/specificities that were 70% or above where they could be calculated. Over-reporting of safe practice was found in half the items and underreporting in a third. Modifications were made to the HIP survey as a result of the validation study.
TE D
This included removing items with low specificity and sensitivity (blind and curtain cords) and deleting uncommon items (e.g. stove and oven guards, room locks). The poison questions were also
Home Risk Score
EP
rewritten to improve the accuracy of participant reporting.
AC C
A Home Risk Score (HRS) was calculated for each participating family based on their responses to 31 questions from the safety measures/hazards component of the HIP survey. These items included safety practices and house features that are established in the literature as being relevant to injury (DiGuiseppi et al., 1998; Elkington et al., 1999; Thompson & Rivara, 2000). Safety items that were applicable to all homes (e.g. safety switches, safety plugs, smoke alarms) were scored as either “1” item was present or “2” item was absent. For safety items which relied on the presence of a specific physical structure (e.g. stairs, balconies, pool, play equipment) multiple questions were combined and were numerically scored as being either a low risk “1” (physical structure was not on property), moderate risk “2” (physical structure was present and all safety measures were in place) or high risk 5
ACCEPTED MANUSCRIPT “3” (physical structure was present and one or more safety measures were not in place). These 18 items were clustered according to five injury domains: burns/electrocution, falls, poisoning, drowning and driveway accidents (Table 1), reflecting leading causes of injury in young children in Australia
Insert Table 1 here
RI PT
(Pointer, 2014).
A score for each injury domain was calculated by summing the scores for included items, with a possible range from 18-46. Weighting was then applied to ensure equal contribution to the final HRS
SC
with each domain contributing a total 20%. For example, the burns/electrocution domain obtained a maximum total score of 11 out of 46 (24%) and thus was weighted down to 20% by the following
M AN U
formula: Domain Raw Score x (0.2/0.24). The final HRS was calculated by summing the adjusted scores for the five injury domains. Total HRSs were divided into quartiles for analysis.
Demographic, Maternal and Household Variables
Information on child, maternal and household variables of interest was drawn from the [REMOVED]
TE D
baseline survey, completed by women during the second half of their pregnancy, and hospital perinatal records. Child characteristics were gender and age. Maternal variables were maternal level of education, maternal age, smoking in pregnancy and high-risk alcohol and drug use during
EP
pregnancy. Baseline household characteristics were marital status, other children living in the household, partner’s employment status, household income quintiles [Anonymous, 2012b] and home
AC C
ownership.
Outcome data
Injury presentations were extracted from the [REMOVED] Emergency Department Information Systems (EDIS) and injury related hospital admissions were extracted from the [REMOVED] Hospital Admitted Patients Data Collection [REMOVED] from the child’s date of birth until 31 December 2013 inclusive. ED records were only available for public hospitals, while state-wide inpatient records were drawn from all private and public hospitals. The [REMOVED] Department of Health linked demographic, child and maternal details through linkage software using 6
ACCEPTED MANUSCRIPT deterministic and probabilistic methodologies. Manual clerical reviews were undertaken where necessary.
Variables Injury classification
RI PT
Classification of injury presentations/admissions was based on ICD-10-AM, with injury cases being identified from Chapter 19 Injury and Poisonings (S00-T98) and Chapter 20 External Causes (U50Y98), excluding late effects from injury. All free text descriptions of ED records were manually
secondary diagnoses of injuries were included.
M AN U
Outcome measures
SC
checked to identify any injuries that were not recorded as injuries in the single diagnosis field. No
The outcome measure of interest was the total number of injury “episodes of care” derived from ED presentations and hospital admissions. Dates of admissions and ED presentations, transfer codes and diagnostic information were used to collapse multiple admissions, nested admissions and related ED
Calculation of person-years
TE D
presentations to single episodes of care that related to the one injury event.
EP
Children in this study were born between 1 July and 30 November from 2009 to 2011, with injury outcomes data available to the end of 2013. Therefore, the length of follow-up differed considerably.
AC C
Using dates of birth and coverage dates, individual person-years (PYs) were calculated for each child with the total time he or she was residing in the state of [REMOVED], alive and eligible for health care, from birth to 31 December 2013.
Analysis
Data cleaning and analyses were undertaken using SAS 9.4 software (SAS). The statistical significance of differences between groups was assessed by Pearson’s chi-squared test for categorical data, the Kruskall-Wallis test and Mann-Whitney test for ordinal data. All tests were two-sided with a
7
ACCEPTED MANUSCRIPT 5% level of significance. Rates of injury related episodes of care were calculated for each HRS quartile, taking into account PYs exposure data. Cases with missing data across more than 20% of Home Risk Score items were excluded from all analyses. Multiple imputation methods were used to impute incomplete multivariate exposure factors
RI PT
by Fully Conditional Specification (FCS) (Liu & De, 2015; Spratt et al., 2010) using 20 iterations. Factors associated with ‘missingness’ as well as the outcome variable were chosen for the imputation model. For binary variables, logistic regression; for ordered categorical variables, ordinal logistic
SC
regression; and continuous variables, regression; were specified in the imputation phase (UCLA : Statistical Consulting Group). For the analysis phase, zero-inflated poisson regression was used to
M AN U
estimate crude and adjusted rate ratios (RRs) between exposure (HRS quartiles) and outcome (count of injury related episodes of care) for child participants, accounting for individual PYs exposure time. A likelihood ratio test was used to compare poisson regression and negative binomial regression model fit. A Vuong test SAS macro (SAS; Vuong, 1989) was used to check whether a zero-inflated regression model was required to account for the presence of “excess zeroes” in the outcomes data.
TE D
Based on the likelihood ratio test, the Vuong test, and the goodness of fit criteria, the zero-inflated poisson regression model was the best model for the hospital count data (Gardner, Mulvey, & Shaw,
EP
1995; McCullagh & Nelder, 1989).
Checks for multicollinearity were conducted during the initial model preparation phase and all factors
AC C
had a correlation below 0.3 and thus not considered a concern. Factors shown to be associated with both the exposure (HRS) and the outcome (child injury episode rate) in univariate analysis where p0.05). For all items included in the HRS, the number of missing ranged from 0%-2.3%, with 83.9% of the items having
