A systematic review of the prevalence of parental concerns measured by the Parents’ Evaluation of Developmental Status (PEDS)indicating developmental risk

Search strategy

A protocol was developed and registered with the University of York Centre for Reviews and Dissemination (PROSPERO) on 6/11/2012 and updated on the 13/02/2014, registration number CRD42012003215(http://www.crd.york.ac.uk/PROSPERO/index.asp).

A systematic search of the literature was undertaken using the following inclusion criteria: primary observational studies (cohort study, cross-sectional studies) in geographically defined population or a community sample (including samples from primary health care services) of children aged under 8 years using the PEDS[15] with available prevalence data (Additional file 1). Studies using the modified “Survey PEDS” were also included in this review [14]. Electronic databases searched were Web of Science and Google Scholar, PubMed (Nov 2012), EMBASE (Nov 2012), Medline (Nov 2012), Psychinfo (Nov 2012), Global Health (Nov2012) CINAHL (Nov 2012), the Cochrane Library (Nov 2012), LILACS (Nov 2012), ERIC (Nov 2012), and Proquest (Nov 2012). Secondary searches of citations in review articles, requests to experts in the field and additional searches of the USA based PEDStest and RCH PEDS website for key studies were undertaken. Advice from the Cochrane Child Development, Psychosocial and Learning Groups was sought regarding search terms which were specific for early child development, developmental risk and the PEDS. There were no language limitations. Studies using specific clinical samples, for example, neonatal intensive care graduates or with participants who had a known developmental disorder were excluded.

The study titles, abstracts and full papers of “potentially relevant articles” were reviewed independently by two authors (SW&VE). Disagreements about inclusion were resolved through consensus and discussion with a third author (KW). Study characteristics, prevalence, and risk factors, were extracted independently by SW and VE on a data extraction form that was piloted and modified prior to use. Where insufficient data were reported, study authors were contacted. If no reply was forthcoming or full data not made available, data were included in analysis where possible. Methodological quality was assessed independently by SW and VE based on a validity of the study methods (design, sampling frame, sample size, outcome measures, measurement and response rate), interpretation of the results and applicability of the findings [19], a score of 6 or greater was rated by the reviewers as high quality.

Statistical analysis

Risk factor analysis

We conducted a meta-analysis for risk factors for having parental concerns on the PEDS indicating high versus low/no developmental risk. We extracted odds ratios and 95% confidence intervals from each study. If odds ratio (OR) with a 95% confidence interval was not provided, we calculated the odds ratio and 95% confidence interval. We extracted adjusted odds ratios when possible, but we were unable to calculate these for studies in which they were not provided. We obtained pooled estimates of unadjusted odds ratios (uOR) using meta-analysis with random effects. Where studies presented adjusted odds ratios (aOR) for similar child and family variables these were combined in a separate meta-analysis.

Investigation of heterogeneity

For all meta‒analyses and meta‒regressions of prevalence, we modelled within-study variability using the binomial distribution [21]. We then examined heterogeneity through meta-regression models, as described in previous systematic reviews of prevalence [22]. We quantified the reduction in the between study variance from the inclusion of the study characteristics compared to the ‘base’ model (i.e., the model of prevalence without any covariates). This provides an estimate of the proportion of heterogeneity that is explained by that characteristic. For our meta-analysis of risk factors, the degree of heterogeneity was investigated by estimating the I² statistic (which describes variation in the summary effect due to genuine variation rather than a sampling error as a percentage, a low I² indicates low heterogeneity and high I² indicates significant between study variability) and visual inspection of forest plots [22].

Studies identified

The search strategy identified 17,272 titles (excluding duplicates). Seventy-eight articles underwent a text screen and 41 of these were excluded (Figure 1) [23].

Included studies

Fifteen studies used the PEDS as a research tool to measure prevalence of developmental risk of which 12 were population surveys in high income countries and three were community samples. The remaining studies used the PEDS as a developmental surveillance tool in primary health care and early childhood education/early primary school settings [14, 24–28, 31–33, 35, 38, 41–46, 51, 53, 58, 59]. Eight of the studies were conducted in low and middle income countries [24, 42, 44–46, 51, 53, 59] and two studies were in socioeconomically disadvantaged communities in the USA [33].

Study sample sizes ranged from 20 to 54602 (median = 467). There were 210,242 subjects in total. Ages ranged from less than 1 month to 7 years and 11 months consistent with the age range for administration of the PEDS. Twenty seven of the studies used translated versions of the PEDS for at least part of their sample.

Study quality

Prevalence of developmental risk

The pooled estimate of the prevalence of parental concern on the PEDS indicating high developmental risk was 13.8% (95% CI 10.9-16.8%), meaning that almost 14% of parents raised concerns associated with a high risk for developmental problems (Figure 2). The pooled estimate of for moderate developmental risk was 19.8% (95% CI 16.7-22.9%). The pooled estimate for high or moderate developmental risk was 31.5%(95% CI 27.0-36.0%), meaning that more than 31% raised concerns associated with either high or moderate risk of developmental problems.

Pooled estimates for biological and psychosocial risk factors

Narrative summary of single studies, cumulative risk and life course analysis

A wide range of additional child, family, and service level factors were noted in single studies [36, 37, 39, 40, 56]. Child level factors were ear infections prior to age 2 (p < 0.001) [40], history of hospital admissions aOR 1.80 (95% CI 1.35–2.40) [37] and being underweight aOR 2.66 (95% CI 1.68–4.24) [37]. Family level factors were low scores on contentment/relaxation during pregnancy aOR 2.5 (95% CI 1.4 -4.2) [39], poor parenting morale when the child was 3 years old aOR 3.9 (95% CI 2.1- 7.3) [39], maternal history of domestic violence at pregnancy aOR 2.2 (95% CI 1.3- 3.7) [39, 40], household food-insecurity (aOR 1.76 (95% CI 1.26 - 2.46) [37], severe energy insecurity aOR 1.82 (95% CI 1.38 -2.39) [36], geographic site differences in the USA (p = 0.003) [37] and poor overall social support (p = 0.003) [39]. Service level factors were referral to early intervention (p < 0.001), speech pathology (p < 0.001) or audiology (p < 0.001) [40], lack of care coordination aOR 0.33 (95% CI 0.24–0.46), referrals aOR 0.40 (95% CI 0.25–0.65), family-centred care aOR 0.47 (95% CI 0.36–0.62) [30] and parental difficulty understanding the doctor uOR 3.35 (95% CI 2.1-5.4) [48].

Two studies reported a dose–response relationship between the number of risk factors and the increased likelihood of parental concerns on the PEDS indicating high developmental risk [7, 39, 40]. In one study having one risk factor was associated with an aOR 1.7 (95% CI: 1.20–2.38); two risk factors aOR 3.28, (95% CI: 2.27–4.73), three risk factors aOR 4.69 (CI: 2.84–7.73), and four risk factors aOR 14.58 (95% CI: 4.98–42.64) compared to a child with zero risk factors [7]. In addition, the greater the number of risk factors experienced by the child the more likely the child was to not receive comprehensive well child care [7].

The only longitudinal cohort in the review reported that at the second follow up when a child was 5 years of age male gender aOR 2.3 (1.3, 4.1), maternal history of abuse at pregnancy aOR 2.4 (1.3, 4.4) and poor parenting morale when the child was 3 years old aOR3.9 (2.1, 7.3) were predictors of parental concerns on the PEDS indicating high developmental risk [39].

Prevalence and associated risk factors for parental concerns on the PEDSindicating developmental risk

This systematic review provides synthesised critically appraised evidence of the substantial global prevalence of parental concerns on the PEDS that indicate high and moderate developmental risk, which increases with biological and psychosocial adversity. This information is useful for researchers, service providers and clinicians to quantify the level of parental concern and to estimate the risk of children having developmental problems in the general population and to identify vulnerable subpopulations. Gender, low birth weight, poor maternal mental health, low family SES, minority ethnicity, speaking a language other than English and a lack of stimulation, such as a child not being read to, are all associated with adverse impacts on development in the literature and this was supported by the synthesised evidence [1, 60–63]. The increasing parental concerns with age of a child regardless of SES demonstrated in this review reflect the increasing developmental demands with age. The impact of child’s poor general health on developmental risk may reflect a true increase as some chronic illnesses and syndromes are associated with adverse developmental outcomes. However concerns about their child’s health may increase parental concerns generally [64, 65].

This review demonstrated that lack of access to usual and comprehensive health care in the USA and Canada was associated with an increased prevalence of parental concerns on the PEDS indicating high developmental risk. Interestingly the evidence for access to services such as early childhood education which has been found to particularly benefit children from disadvantaged backgrounds was not demonstrated [66–68].

Two studies demonstrated that parental concerns on the PEDS indicating high developmental risk increased with the number of risk factors a child was exposed to, consistent with our understanding of the burden of multiple risk factors on early childhood development [7, 39, 40, 62]. In addition, the “inverse care law” applied in one USA study, with the greater the number of risk factors, the less access to comprehensive health care [7, 69].

Comparison with other measures of developmental risk

The confidence intervals around the pooled prevalence estimates of high and moderate developmental risk using the PEDS (27.0-36.0%) is similar to rates using the Denver Developmental Screening Test (DDST)[70–72] but higher than those using the Australian Early Development Index (AEDI)[1], and Ages and Stages Questionnaire (ASQ)[38, 43, 52]. While the PEDS gives an estimate of high and moderate developmental risk based on parental concerns this is not synonymous with a comprehensive developmental assessment. The PEDS specificities of 73-86% for parental concerns indicating high and/or moderate developmental risk means that some children identified by parental concern will be false positives [14, 17]. If parental concerns indicating only high developmental risk are examined the specificity of the PEDS improves to 89%, reducing the number of false positives but the sensitivity drops substantially to less than 50% giving an unacceptable level of false negatives [38, 43]. Thus, the true prevalence of actual developmental problems indicated by parental concerns is likely to lie somewhere between the values indicating high and moderate developmental risk [38, 73]. This is reflected in how the PEDS is used in clinical practice with those children identified as at high developmental risk on parental concerns referred on for a comprehensive developmental assessments and those at moderate risk undergoing a secondary developmental screen with a tool such as the ASQ and if they fail that then being referred on for a comprehensive developmental assessment [38, 43, 52]. Systematic reviews of the diagnostic test accuracy (DTA) of the tools that measure developmental risk such as the PEDS in relation to the reference-standard diagnostic batteries in nationally representative samples with an inclusive analysis of vulnerable subpopulations would be useful in understanding how useful developmental risk is as a way to estimate the burden of developmental problems in a population. This systematic review only included studies which had used the PEDS. Prevalence and DTA systematic reviews of other tools such as the ASQ and AEDI would also be useful for further comparison.

Limitations

There was considerable variation in quality of the individual studies included in this systematic review. The major sources of bias were an inadequate sampling method, sampling frame, sample size and response rate and a lack of information to aid interpretation and applicability of the results including reporting of confidence intervals and subgroups. It is suggested that future research focus on designing studies that address these issues. Where community samples were used, parents most concerned about their children may be over-represented and this could lead to an overestimation of prevalence. However meta-regression using quality of the study as a variable did not find significant differences in pooled prevalence estimates. There were significant differences in pooled prevalence estimates of developmental risk between studies when the subgroups of publication type and country income were examined. We did not have individual patient data to undertake our own multivariate analysis. Although covariates were similar between studies, how these were measured and the breadth of variables measured varied. This highlights the need for agreed tools on measurement of psychosocial risk in research.

The cross sectional nature of the majority of papers in the review means that only associations of prevalence with risk factors can be examined not causality. In an attempt to address the issue of misclassification bias due to false positives we only examined the relationship between risk factors and high versus low/no developmental risk. Thus any significant relationship is likely to be an underestimate of the true strength of association [17].