Web Content Display Web Content Display

Cochrane Clinical Answers: data extraction

Each Cochrane Clinical Answer is based on a single Cochrane Review. Extraction of data from Cochrane intervention reviews that use standard pairwise meta-analyses is semi-automated. The CCA Editor checks the automatically extracted data and inserts any additional relevant information from the Cochrane Review into the template. Since the PICOtron (data extraction software) only extracts data from the analysis section of the Cochrane Review, it cannot extract data from NMA, overviews of reviews, qualitative/narrative reviews, or reviews of DTA; these are extracted manually by a CCA Editor.

Automated data extraction

The table below shows the information extracted by the PICOtron from Cochrane intervention reviews with standard pairwise meta-analyses, and the source of that information.

Source of information from the Cochrane Review

Comparison title	Data and analysis > Comparison title
Narrative result	When there are meta-analyses, the narrative text is automatically generated by the PICOtron from several different locations in the Cochrane Review: The number of studies: Data and analyses > No. of studies The number of participants: Data and analyses > No. of participants Intervention and comparator names: graph labels on the forest plot Direction of effect: the python script calculates whether there is a statistically significant difference between groups, and if so, what the direction of effect is. It then selects a sentence structure depending on these calculations Outcome name: Data and analyses > Outcome or subgroup title If the data are not meta-analyzed, no information is included in this section or the quantitative results sections (below) and the fields are completed manually.
Quantitative result: relative effect or mean difference	Direction of effect: the python script calculates whether there is a statistically significant difference between groups, and if so, the direction of the effect. It then selects a sentence structure depending on these calculations Favoured intervention or comparator: graph label on the forest plot Relative effect/mean difference and associated confidence intervals: Data and analyses > Effect size
Quantitative result: absolute effect	The absolute risks (ARs) for dichotomous outcomes are estimated from relative risks (RRs) as follows and reported as rates per 1000 people. The comparator AR is estimated as the average (weighted median) AR from the trials in the source Cochrane Review meta-analysis. Intervention group AR is estimated by multiplying the comparator AR with the pooled RR from the meta-analysis. The confidence intervals (CIs) are calculated by multiplying the comparator AR with the lower and upper bounds of the 95% CIs for that RR. It is worth noting that there will sometimes be discrepancies in the absolute numbers reported in the CCA and those reported in the Summary of findings (SoF) tables in the Cochrane Review. This is due to the PICOtron calculating these using the median event rate and the SoF using the mean event rate; the median is less influenced by any outlying study that may have a particularly high or low comparator event rate.

Comparison title

Data and analysis > Comparison title

Narrative result

When there are meta-analyses, the narrative text is automatically generated by the PICOtron from several different locations in the Cochrane Review:

The number of studies: Data and analyses > No. of studies
The number of participants: Data and analyses > No. of participants
Intervention and comparator names: graph labels on the forest plot
Direction of effect: the python script calculates whether there is a statistically significant difference between groups, and if so, what the direction of effect is. It then selects a sentence structure depending on these calculations
Outcome name: Data and analyses > Outcome or subgroup title

If the data are not meta-analyzed, no information is included in this section or the quantitative results sections (below) and the fields are completed manually.

Quantitative result: relative effect or mean difference

Direction of effect: the python script calculates whether there is a statistically significant difference between groups, and if so, the direction of the effect. It then selects a sentence structure depending on these calculations
Favoured intervention or comparator: graph label on the forest plot
Relative effect/mean difference and associated confidence intervals: Data and analyses > Effect size

Quantitative result: absolute effect

The absolute risks (ARs) for dichotomous outcomes are estimated from relative risks (RRs) as follows and reported as rates per 1000 people.

The comparator AR is estimated as the average (weighted median) AR from the trials in the source Cochrane Review meta-analysis.

Intervention group AR is estimated by multiplying the comparator AR with the pooled RR from the meta-analysis. The confidence intervals (CIs) are calculated by multiplying the comparator AR with the lower and upper bounds of the 95% CIs for that RR.

It is worth noting that there will sometimes be discrepancies in the absolute numbers reported in the CCA and those reported in the Summary of findings (SoF) tables in the Cochrane Review. This is due to the PICOtron calculating these using the median event rate and the SoF using the mean event rate; the median is less influenced by any outlying study that may have a particularly high or low comparator event rate.

Manual checks and additional information added by the CCA Editor

The CCA Editor checks each of the auto-generated PICOtron fields to confirm accuracy. Other relevant information from the Cochrane Review is added to the template, as follows:

Population

The CCA Editor inserts details of the population(s) included in the studies that contribute to the analyses in the CCA; these may differ from the main inclusion criteria of the Cochrane Review.

Intervention/comparator

The CCA Editor inserts details of the intervention(s) evaluated in the studies that contribute to the analyses in the CCA. These may include regimen details (dose, route, and frequency of administration), concomitant therapy (unless administered to both groups, in which case this would be reported in the population field), duration of treatment, and/or components of complex interventions.