Article type
Year
Abstract
Background: Parents and carers of children with autism need information about the long-term health and well-being outcomes, or 'prognosis', of their children at the time of diagnosis. Clinicians and health service providers need valid and applicable information about the prognosis of autism to help parents make decisions about treatment options for their affected children and to develop best practice models of service delivery. However, there are many clinical and methodological differences in prognosis studies about autism that make meaningful data synthesis, and assessment of applicability of findings in a systematic review, problematic.
Objectives: To develop a method for classification of studies to assist quality assessment and to guide meta-analysis in a prognosis systematic review.
Methods: A comprehensive search strategy was used for published and unpublished studies about the prognosis of autism. Titles and abstracts were screened and full paper reviews completed to finalise inclusions based on predetermined criteria. Quality assessment tools for prognosis studies were reviewed. Three key issues related to potential 'risk of bias' were identified: the population type (clinical or population-based); the timing of assessment of diagnosis, prognostic factors and outcomes; and the proportion of children followed up. Classification criteria with subcategories were developed to address these issues. We classified studies on the basis of whether the diagnosis used in the study was made prior to or at the start of the study, whether the outcomes were assessed prospectively or retrospectively, and whether prognostic factors were measured prior to or at the same time as the outcomes. In addition, information was gathered about diagnostic definition and length of follow-up, as key clinical differences. Two reviewers categorised each paper according to each criteria. If there was inconsistency between reviewers a third reviewer classified the papers, blind to the first two reviewers' assessments.
Results: The initial search identified 8369 papers and after screening, using specific inclusion criteria, 88 papers were included. Consensus of at least two reviewers was reached for all classification criteria developed. There was considerable variation between studies according to the classification criteria identified. For example, within a subgroup of studies (12) that used similar definition for diagnosis and measured speech and language outcomes, all were derived from clinical populations and equal numbers used prospective and retrospective methods for diagnosis and outcome assignment. Duration of follow-up ranged from 1 to 15 years. Assessment of the effect of 'risk of bias' on reported outcomes using sensitivity analysis, and whether criteria to assess 'risk of bias' are independent factors, is ongoing.
Conclusions: For studies about prognosis of autism this newly developed classification system has provided a useful way of selecting important clinical subgroups for data synthesis and identifying 'risk of bias' to facilitate sensitivity analyses. It could be adapted to other health problems.
Objectives: To develop a method for classification of studies to assist quality assessment and to guide meta-analysis in a prognosis systematic review.
Methods: A comprehensive search strategy was used for published and unpublished studies about the prognosis of autism. Titles and abstracts were screened and full paper reviews completed to finalise inclusions based on predetermined criteria. Quality assessment tools for prognosis studies were reviewed. Three key issues related to potential 'risk of bias' were identified: the population type (clinical or population-based); the timing of assessment of diagnosis, prognostic factors and outcomes; and the proportion of children followed up. Classification criteria with subcategories were developed to address these issues. We classified studies on the basis of whether the diagnosis used in the study was made prior to or at the start of the study, whether the outcomes were assessed prospectively or retrospectively, and whether prognostic factors were measured prior to or at the same time as the outcomes. In addition, information was gathered about diagnostic definition and length of follow-up, as key clinical differences. Two reviewers categorised each paper according to each criteria. If there was inconsistency between reviewers a third reviewer classified the papers, blind to the first two reviewers' assessments.
Results: The initial search identified 8369 papers and after screening, using specific inclusion criteria, 88 papers were included. Consensus of at least two reviewers was reached for all classification criteria developed. There was considerable variation between studies according to the classification criteria identified. For example, within a subgroup of studies (12) that used similar definition for diagnosis and measured speech and language outcomes, all were derived from clinical populations and equal numbers used prospective and retrospective methods for diagnosis and outcome assignment. Duration of follow-up ranged from 1 to 15 years. Assessment of the effect of 'risk of bias' on reported outcomes using sensitivity analysis, and whether criteria to assess 'risk of bias' are independent factors, is ongoing.
Conclusions: For studies about prognosis of autism this newly developed classification system has provided a useful way of selecting important clinical subgroups for data synthesis and identifying 'risk of bias' to facilitate sensitivity analyses. It could be adapted to other health problems.