1/1
A Meta-Analysis of Diagnostic Accuracy (Dollaghan & Homer, 2011)
dataset
posted on 2011-08-01, 00:00 authored by Christine A. Dollaghan, Elizabeth A. HornerPurpose To describe quality indicators for appraising studies of diagnostic accuracy and to report a meta-analysis of measures for diagnosing language impairment (LI) in bilingual Spanish–English U.S. children.
Method The authors searched electronically and by hand to locate peer-reviewed English-language publications meeting inclusion criteria; the authors rated quality features, calculated accuracy metrics and confidence intervals, and generated forest plots.
Results Of 771 citations (86 unique) located initially, accuracy metrics could be calculated for 17 index measures studied in a total of 100 children with LI and 109 with typical language. Most studies lacked clear descriptions of reference standards, procedures, and controls for subjective bias, making it difficult to rate specific quality features with confidence. Positive likelihood ratios (LR+) for most measures were at least diagnostically suggestive (pooled LR+ = 4.12; 95% CI [2.94, 5.78]). Negative likelihood ratios (LR?) were also generally suggestive, but heterogeneity precluded averaging. For every measure, confidence intervals for LR+ and LR? included diagnostically uninformative values.
Conclusions The available evidence does not support strong claims concerning the diagnostic accuracy of these measures, but a number appear promising. Several steps are suggested for strengthening future investigations of diagnostic accuracy.
Method The authors searched electronically and by hand to locate peer-reviewed English-language publications meeting inclusion criteria; the authors rated quality features, calculated accuracy metrics and confidence intervals, and generated forest plots.
Results Of 771 citations (86 unique) located initially, accuracy metrics could be calculated for 17 index measures studied in a total of 100 children with LI and 109 with typical language. Most studies lacked clear descriptions of reference standards, procedures, and controls for subjective bias, making it difficult to rate specific quality features with confidence. Positive likelihood ratios (LR+) for most measures were at least diagnostically suggestive (pooled LR+ = 4.12; 95% CI [2.94, 5.78]). Negative likelihood ratios (LR?) were also generally suggestive, but heterogeneity precluded averaging. For every measure, confidence intervals for LR+ and LR? included diagnostically uninformative values.
Conclusions The available evidence does not support strong claims concerning the diagnostic accuracy of these measures, but a number appear promising. Several steps are suggested for strengthening future investigations of diagnostic accuracy.
