Article type
Year
Abstract
Background: Systematic reviews (SR) with the statistical methods of meta-analysis represent the highest level on the evidence to make decision. Extracting data can be a problem when randomized clinical trials (RCT) publish figures as the only source of outcome data. Thus, none included RCTs in the SRs can be lost to perform meta-analysis. In this way, alternative methods to data extraction need to be developed.
Objectives: To analyze the precision of alternative methods (rule and software) to data extraction from figures to perform meta-analysis.
Methods: All data were produced in a statistical package and were generated 14 RCTs. Figures were generated from same RCTs; mean and 95% confidence interval (CI) were used as statistical parameters from figures to elevated the complexity of data extraction; standard deviations was obtained applying the equation: ‘Sqrt(N) × (upper limit of CI − lower limit of CI)/3.92’. Differences between researchers (VS and AJG) to data extraction were previously compared. Using other datas, one blind researcher (AJG) performed data extraction from figures with a simple rule (Fig. 1a) and another blind researcher (VS) performed the data extraction with the Digitizelt software (Fig. 1b). Meta-analyses were performed and compared with data extracted from a rule, software and original data.
Results: There are absence of differences (Chi2 = 0.01, df. = 1, P = 0.93; I2 = 0%) between researchers (VS and AJG) to data extraction. The meta-analyses results are presented through the Figure 2.
Conclusions: The alternative methods, such as rule and software, evaluated in this study to procedure data extraction from figures showed high precision and seem to be useful as a option to perform meta-analysis when RCTs publish figures as the only source of outcome data. The authors are planning further studies—using real data from a SR—to confirm these results.
Objectives: To analyze the precision of alternative methods (rule and software) to data extraction from figures to perform meta-analysis.
Methods: All data were produced in a statistical package and were generated 14 RCTs. Figures were generated from same RCTs; mean and 95% confidence interval (CI) were used as statistical parameters from figures to elevated the complexity of data extraction; standard deviations was obtained applying the equation: ‘Sqrt(N) × (upper limit of CI − lower limit of CI)/3.92’. Differences between researchers (VS and AJG) to data extraction were previously compared. Using other datas, one blind researcher (AJG) performed data extraction from figures with a simple rule (Fig. 1a) and another blind researcher (VS) performed the data extraction with the Digitizelt software (Fig. 1b). Meta-analyses were performed and compared with data extracted from a rule, software and original data.
Results: There are absence of differences (Chi2 = 0.01, df. = 1, P = 0.93; I2 = 0%) between researchers (VS and AJG) to data extraction. The meta-analyses results are presented through the Figure 2.
Conclusions: The alternative methods, such as rule and software, evaluated in this study to procedure data extraction from figures showed high precision and seem to be useful as a option to perform meta-analysis when RCTs publish figures as the only source of outcome data. The authors are planning further studies—using real data from a SR—to confirm these results.
