Study design and participants
This case–control study enrolled poisoning cases and injured and sick controls through the Sydney Children’s Hospital Emergency Department (SCHED) from 22 February 2005 to 14 January 2007. The Sydney Children’s Hospital is located in Randwick, NSW Australia and is one of the state’s three children’s tertiary care hospitals. A third control group, healthy controls, was enrolled into the study from 18 September 2005 to 31 October 2006 from the local community. Mothers were enrolled with their children into all four groups.
Poisoning cases were defined as children aged 1–3 years who presented to the SCHED for treatment of a poisoning after accessing a substance themselves (i.e., not given to them by a caregiver or other person). Injury and sick controls were defined as children aged 1–3 years who presented to the SCHED for treatment of an unintentional injury (other than poisoning) or an illness, respectively. Healthy controls were defined as children aged 1–3 years who attended a playgroup or a child care centre in the geographic areas served by the SCHED.
The sample size required for each of the four groups was 36 children aged 1–3 years. This was based on a required sample size of 35 for multiple regression with six predictors, an effect size of 0.5 for differences between groups in child temperament scores and parenting stress scores with 80% power and a level of significance of 5%. The sample size was increased to 36 to allow for equal numbers (6) in each single year of age and sex group (e.g., 6 males aged 1 year, 6 females aged 1 year).
The cases and three controls were to be matched by age (within 3 months) and sex to control for development-related aspects that may contribute to poisoning risk (e.g., physical growth) [14]. The South-eastern Sydney Area Health Service Research Ethics Committee approved the study.
Procedures
Enrolment of emergency department cases and injury and sick controls
Children who met the poisoning case and injury and sick control group definitions were identified from all SCHED presentations on a weekly basis (excluding presentations ending in death). Letters explaining the study and inviting participation were sent to the residential address of mothers of a stratified sample (by child’s age and sex). Interested mothers were asked to contact the first author (MS). Letters were sent to 102 poisoning cases, 674 injury controls and 1014 sick controls.
Mother-child dyads were excluded if the mother reported any known developmental delay in her child or her child had any health conditions requiring long stays in hospital. Dyads were also excluded where the mother was not the primary caregiver, was unable to complete questionnaires in English, did not reply to the letter within 4 weeks or was not available to attend a 1.5 hour interview at the University of New South Wales which is located close to the SCHED.
Enrolment for each age-sex combination closed once six dyads had completed the interview and returned all study materials.
Enrolment of healthy community controls
Playgroups and child care centres in the geographic areas of interest were identified from lists of playgroups and centres registered with the NSW Playgroups Association and Australian Child Care Access Hotline, respectively. Playgroup coordinators and child care centre directors were asked to distribute study letters to mothers of eligible children (aged 1–3 years). A total of 2738 letters were sent to 67 child care centre directors for distribution. Playgroups NSW distributed approximately 1450–1740 study invitation letters to 29 playgroup coordinators (2 rounds of 25–30 letters/playgroup).
Interested mothers from both playgroups and child care centres contacted the first author. Healthy mother-child dyads underwent the same screening and group enrolment procedures used with the poisoning cases and injured and sick controls.
Interview
Mother-child dyads attended a three-part 1.5 hour digitally recorded appointment. First, the mother read and signed consent forms for herself and her child regarding participation in the study. Next, the child’s ability to perform a number of developmentally-related tasks was measured by the first author using the Denver Developmental Screening Test (DDST) [41]. Then the mother was asked to interact with her child during two 10 minute activities- a structured puzzle task (puzzle) and an unstructured free play session with toys in the room (free play). The first author was present in the room during the first task, but not the second. Finally, each mother was instructed to ask her child to pack away the toys (clean-up). The clean-up task was limited to three minutes. At the end of each appointment, each mother was given an envelope of questionnaires to complete and return.
Measures
The measures involved questionnaire, performance and observational methods and have been organised into the general domains of child, mother and environment (Figure 1).
Questionnaire – child domain
Demographics
Mothers completed the child section of the socio-demographic questionnaire (SDQ) which captured information on the following child variables: age (from date of birth), sex, number of hours in child care per week in own home and/or outside own home and total number of hours in child care per week.
Developmental level
The first author administered the Denver II (DDST) [41] to assess the current level of development of the child. This screening tool is a standardised instrument that measures the ability of children aged 0–6 years to perform developmentally-related tasks in four areas: personal-social (25 items), gross motor (39 items), fine motor-adaptive (29 items) and language (32 items). The child’s age at the time of the test (exact age) was calculated and the test was administered according to the test protocol.
Parental report was permitted for some items (as per test instructions). Each item attempted was scored as ‘pass’, ‘not pass (fail)’, ‘refused’ or ‘no opportunity’. For each item, the DDST documentation provided a value for the age at which 25, 50, 75 and 90 percent of the test group of children passed the item. The 90th percentile age value for the highest item passed was compared to the child’s age at the time of the test. The level of developmental ability was scored for each scale as ‘below expected’, ‘at expected’ or ‘above expected’.
Temperament
Mothers reported their child’s temperament using the Short Temperament Scale for Toddlers (STST). This questionnaire was developed through the Australian Temperament Project [42] and measures six factors reflecting temperament dimensions- approach, cooperation–manageability, persistence, rhythmicity, distractibility and reactivity. Mothers rated each item from 1–6, where 1 = ‘almost never’ and 6 = ‘almost always’. Six subscale scores and a composite easy/difficult score (average of approach, cooperation and reactivity scores) were calculated according to test instructions. Higher scores indicated the mother perceived her child as withdrawing (less approaching), uncooperative/unmanageable, not persistent, arrhythmic, non-distractible/non-soothable, highly reactive/irritable and difficult.
Compliance
Mothers reported their child’s ability to comply with maternal ‘do’ and ‘don’t do’ requests using a child compliance checklist (Ccomp). This checklist was adapted from a compliance checklist developed by Gralinski and Kopp [43] and measures the frequency of child compliance with parental requests. It comprises 31 items, with 15 items assessing the frequency of compliance with ‘do’ requests by parents (e.g., I have asked my child to put/pack his/her toys away) and 16 items assessing compliance with ‘don’t do’ requests (e.g., I have asked my child not to climb on furniture). Mothers rated their child’s frequency of compliance for each of the 31 items as ‘never complies’, ‘rarely complies’, ‘sometimes complies’, ‘often complies’, ‘always complies’ or ‘not applicable’ (i.e., I have never asked my child to do this). Maternal ratings were scored with the following numerical values: 1 = ‘never complies’, 2 = ‘rarely complies’, 3 = ‘sometimes complies’, 4 = ‘often complies’, 5 = ‘always complies’ and 9 = ‘non-applicable’. Ratings for the ‘do’ and ‘don’t do’ items were summed separately and in total. Scores for ‘do’ and ‘don’t do’ items were adjusted for ‘not applicable’ or missing answers. Higher total scores indicated the child complies more frequently with parental requests.
Questionnaire – mother domain
Demographics
Mothers completed the mother section of the SDQ which captured information on nine variables: age, marital status, highest level of education attained, country of birth, language spoken at home, employment status, number of working hours/week (if working outside the home), cigarette and alcohol use. Marital status was categorised as married/living with partner or single (i.e., never married, divorced). Highest level of education was categorised as a university level education or less attained or post university education.
Life events
Mothers completed a life events questionnaire adapted from the Holmes and Rahe Social Readjustment Rating Scale [44]. Mothers identified whether any of 30 listed life events occurred in their life in the year preceding study participation. The stress value weight(s) associated with each life event indicated were summed to identify a total life stress value.
Health status
Mothers rated their level of health on the General Health Questionnaire 28 (GHQ-28) [45]. The four GHQ-28 subscales were scored- somatic symptoms (A), anxiety (B), social dysfunction (C) and depression (D). The original numeric scores (1–4) were summed for each subscale using GHQ scoring. Answers of ‘3’ or ‘4’ (two-right hand columns) were scored ‘1’, whereas answers of ‘1’ or ‘2’ (other two columns) were scored ‘0’. These scores were summed for each variable and for a total test score. Higher scores indicated more of the attribute measured. A total sum of 5 or more indicated psychiatric ‘caseness’.
Parental stress
Mothers reported their level of parenting stress on the Parenting Stress Index- Short Form (PSI-SF) [46]. This questionnaire contains 36 items which correspond to three scales- Parental Distress (P_D) (12 items), Parent/Child Dysfunctional Interaction (PC_DI) (12 items) and Difficult Child (D_C) (12 items). Each item was scored from 1–5, where 1 = ‘strongly disagree’ and 5 = ‘strongly agree’. Variables scores and a total score were calculated according to test instructions. Higher scores indicate more of the attribute measured.
Supervision attributes
Mothers indicated their supervision attributes on the Parenting Supervision Attributes Profile Questionnaire (PSAPQ) [47]. The questionnaire comprised two parts. Part I included the following five scales: protectiveness (11 items), vigilance/proximity (6 items), worry (3 items), confidence (6 items) and values risk taking (4 items). Each item in Part 1 was scored from 1–5, where 1 = ‘strongly disagree’ and 5 = ‘strongly agree’. Part II consisted of three items: supervision during play activities (10 items), self-care (8 items) and risk activities (3 items). Each item in Part II was scored from 1–5 where 1 = ‘I’m often in another room and I go to my child when he/she calls me’ to 5 = ‘I’m often in the same room as my child and within arms reach’. When the statement was not applicable to the home or child, ‘NA’ was recorded. Scales in Part I and II were coded and summed according to test instructions. Higher scores indicate more of the attribute measured (Part I) and closer supervision during activities (Part II).
Questionnaire – environment domain
Socio-demographic data
Mothers indicated their post code, type of residence, number of bedrooms, ownership status and number and age of occupants on the environment section of the SDQ. The level of socio-economic disadvantage was derived using the Australian Socio-economic Index for Areas (SEIFA) [48] by postal code.
Poisons safety
Mothers indicated their poison safety practices for medicinal and non-medicinal (household) substances on the poison safety section of the SDQ. Poisoning storage questions assessed the height of usual storage of medicinal and household substances in different rooms and presence of child safety or other locks on usual places of storage. The number of accessible locations of medicinal and household substances in various rooms was derived (i.e. number of usual storage locations minus number of locations stored >= 1.4 metres or locked). The percent of total storage locations that were accessible was calculated for each room. Aspects of temporary storage for medicinal and household substances were also measured. Caregivers indicated if substances were intentionally stored in a temporary location and how often the substance was left out after use.
Observed mother-child interaction factors
Mother-child interaction factors were measured by applying the Parent–Child Interaction System (PARCHISY) [49, 50] to the digital recordings of the puzzle and free play tasks. This observational coding system for mother-child interactions comprises 18 items; however, only five caregiver items (i.e., positive and negative control, positive and negative affect, responsiveness), five child items (positive and negative affect, responsiveness, independence, noncompliance) and three dyadic items (conflict, cooperation, reciprocity) were used. Each item was scored from 1 to 7 where 1 = ‘none of attribute shown’ to the 7 = ‘constant or exclusive use of the attribute measured’. The coding system was adapted to suit children aged 1–3 years and smaller coding intervals (i.e., 8 minutes total in 1 minute intervals). Medians of the eight 1-minute data points were used for the 13 PARCHISY variables scored for each task.
Observed child compliance
Children’s observed compliance with a maternal directive was measured by applying a compliance rating system employed by Kochanska et al. [51] to the digital recordings of the clean-up task. This compliance coding system used the following terms to describe child compliance: ‘committed compliance’, ‘situational compliance’, ‘passive noncompliance’, ‘overt resistance’, ‘defiance’ and ‘other’.
Reliability of coding
The authors, MS and AW, trained a person independent of the study to apply the PARCHISY and child compliance coding systems to the two mother-child interaction tasks and the child cleanup task, respectively. The coder was blinded to both the objectives of the study and study group membership. The observational data from the three tasks were copied to a DVD for each study dyad. Each DVD was labelled with a unique identifier and given to the coder in random order. The independent person coded all the DVDs initially and then re-coded a random selection of 10 percent of the DVDs. Intra-rater reliability for all three observation tasks was measured using Spearman’s rank correlations.
Intra-rater reliability for the puzzle task was 0.909 and 0.903 for the free play task. Across the two tasks, five individual items showed intra-rater reliability less than 0.6 and were excluded from further analysis (i.e., maternal positive control (free play), child positive affect (both tasks), child independence (puzzle), dyad cooperation (puzzle)). Intra-rater reliability for the observed compliance measure showed that all 10 DVDs matched on the rating assigned by the coder.
Circumstances of poisoning event
The socio-demographic questionnaire (SDQ) completed by mothers in the poisoning group included an ‘Event’ section. In this section, mothers provided a narrative regarding the circumstances of their child’s poisoning event and provided responses to a series of questions pertaining to the poisoning event. These questions assessed maternal perception of their child’s activities in accessing the substance as well as the substance type, use and storage. In addition, the questions assessed caregiver use of the NSW Poisons Information Centre, poisoning symptoms and actions taken upon presentation to hospital. The information collected was used for descriptive purposes and was not included in the analysis of factors predicting a poisoning event.
Data analysis
This study was designed as a case–control study with one poisoning case matched to three separate controls (injury, sick and healthy) on age (within three months) and sex. However, some age-sex combinations could not be recruited into the study for the poisoning case group (i.e., females aged 1 and 3 years); therefore, cases and controls were not matched as planned. Instead, an unmatched analysis was done and the effects of age and sex were controlled in the analysis phase.
Data were analysed in case–control pairs - poisoning-injury (PI), poisoning-sick (PS) and poisoning-healthy (PH). Descriptive analyses were performed for variables by case–control pair. Univariate logistic regression analyses were conducted to assess the association between poisoning (outcome) and each independent variable (IV). A Likelihood ratio chi-square p-value of <0.20 was used to select IVs for the multivariate analyses.
Two IVs eligible for the multivariate models contained imputed values due to limited missing data. One variable in the PI multivariate model, PSI total score, contained an imputed value for one injury control. Data for one healthy control in the PH multivariate model contained an imputed value for two variables, PSI difficult child and PSI total score. Mean values of the PSI difficult child and PSI total score variables were imputed for the groups missing these values.
Multivariate logistic regression analyses were used to identify the predictors associated with a poisoning. Models were built interactively for each case–control pair with the child’s exact age (using date of birth at the time of the interview) and sex forced into each model. Forward selection was used in conjunction with two criteria to select IVs for the final model for each case–control pair - the lowest Akaike’s Information Criterion (AIC) with a Wald p-value of 0.1 or less. Correlations and multi-collinearity between IVs were checked before each new IV was added to the model. Where a new IV was highly correlated (>=0.5) or showed evidence of multi-collinearity (variance inflation factor >=2.5) with one or more IVs already in a model, the new IV was not added. Adjusted odds ratios (OR’s) and profile likelihood 95% CIs were estimated for all explanatory variables in the final model for each case–control pair. Model fits were assessed using the C-statistic. Analysis was performed using SAS V9.1.3 SP3.