By Gary Burd, SVP, Global Medical Director
Data accuracy is one of the basics tenets of scientific rigor and is fundamental to the quality-control process. In the fast-paced world of medical communications, where data are being made available from stats teams sometimes on a daily/hourly basis and multiple content changes can be required throughout the delivery of a project, a single data check at one stage in the process has become obsolete. Keeping track of inserted data, new data, checked data, unchecked data, latest data sets, and expired data sets has become essential to meet expectations. Furthermore, source data itself can sometimes contain errors, and so medical writers have to be extra vigilant to spot oddities and have the confidence to speak up.
Here are some tips for how you can track data quality to minimize the risk of errors.
- Annotate documents with data sources, providing the document name, page number, and helpful hints on how to locate it if appropriate. In Word, you can do this with comments boxes. You can also use the annotations to show calculations performed or if the source is not validated. This helps the data checker by making it easy for him/her to find the data source.
- Keep all the data sources for a document in one location, making sure they are the latest versions. Move old versions into a separate location so that there are not multiple documents purporting to be the right data set. Having well-organized archives really helps when, 6 months after the deliverable is done, someone needs to find something.
- Data checks should be performed at the first draft stage and following the introduction of new data. A last check by a fresh pair of eyes wouldn’t go amiss either.
- If someone changes data during the review process, be sure to look for evidence that the change is appropriate; ask them for their rationale and source documents if you need to.
- I’m old school and like to work on printouts; I tick every piece of checked data and write queries. I then scan this in and upload it as a record of the data check having being performed. Electronically, you could highlight every piece of checked data.
I think it’s important to remember that the person doing the data check is a human being, not a machine. Any process can minimize the risk but it can’t negate it completely. Perhaps it won’t be long before machines are doing data checks and getting everything right, right?