Data cleaning can have big impact on trials
Data cleaning can have big impact on trials
Watch for discrepancies
Clinical trial data increasingly is collected and communicated electronically. Sites often are dependent on how sponsors and clinical research monitoring organizations set up their data collection and editing systems, but there are issues they can prevent through better processes and preparation.
Experts offer this advice on how to improve several of these types of data management processes:
Edit checks: "Most electronic systems have edit checks built into the interactive environment, so if you enter something out of range it will automatically fire an edit check," says Kit Howard, MS, CCDM, CRCP, principal and owner of Kestrel Consultants of Ann Arbor, MI.
It's wise to make certain the data edit check conforms to expectations.
For example, if there is a field for listing an adverse event, there should be another field that lists the end date of the AE or a box to check if the AE is ongoing, she says.
An edit check should make certain that one of these boxes is filled in, and this should be caught at the point of data entry, she adds.
"This kind of issue doesn't require an external query, it pops up a question on the screen," Howard says.
Another data entry edit check involves ranges, including age, weight, lab results, and blood pressure measurements.
If someone types in a subject's weight as 300 pounds, there should be a screen that pops up, asking, "Do you really mean 300?" Howard explains. "If you click on 'yes' then the screen moves on."
These types of editing checks typically have to be programmed at the site level since vendors would not have these preset for clinical research, she notes.
Clinical research (CR) sites can save considerable time that otherwise will be spent on answering monitoring queries if they work with a data manager or information technology expert to program the system to incorporate these checks.
Most electronic data systems have the capability to set these parameters, but an individual at the site will need to set up the fields according to the study's patient population and disease or product being studied.
"Whoever is responsible for that next level of cleaning will take that information and run it through an external edit check," Howard says. "It's a series of programs that look at data and look for potential problems that are too complex to pick up at the time of data entry."
These include problems with electronic data related to adverse events, exacerbation of medical history conditions, or patients taking concomitant medications, she adds.
Such issues will need to be accompanied by an explanation.
"If a study patient is taking a new medication in the trial, then there has to be some reason for it," Howard says. "If the reason is an adverse event, then we'd expect to see an adverse event listed on the AE page that corresponds in terms of data and indication."
The edit check might be sophisticated enough to note whether the listed AE corresponds to the particular medication that was started, as well.
"You wouldn't expect to see that the person was started on a new medication before the AE occurred," Howard says.
"And you wouldn't expect a person to take ibuprofen for an AE of nausea," she adds. "You'd expect or think there was some kind of pain or inflammation."
External audit checks look across data for more subtle potential data problems.
"When the discrepancy is found, then the sponsor's data manager will write a query and send it to the site," Howard says.
Handle queries: When entered data do not match case report form documentation or when a data manager notices a potential issue, sites will receive a query from monitoring organizations.
For instance, if a site has sent data showing a subject's blood pressure to be in the normal range of 120/80 on visit one, then a second visit that indicates a blood pressure of 170/100 would be suspect, Howard says.
"That's a huge change, and you would think that either there's a data entry error, or the subject was having a panic attack or something was going on," she says. "The data manager would look to make sure there was a reasonable explanation, such as an adverse event."
If no explanation is apparent, then the discrepancy results in a query to the site.
"This could be an electronic or manual query," Howard says. "The manual query usually is typed into a form and sent via email or some electronic transmission method."
Monitoring organizations also might identify a discrepancy through a manual data check in which a data manager identifies a problem.
And sometimes the problem can be the result of sites listing data in slightly different ways that go unrecognized by the electronic system. For instance, one site coordinator might write "headache" under a list of AEs, while another coordinator writes "head ache" or the word "headache" with a period at the end, Howard says.
"The computer will see these as three different adverse events, and that's not helpful," she explains.
Sometimes these kinds of discrepancies will result in an automatic query, but they could also signal a programming change. The data manager needs to program the system to recognize the alternative versions of the word as identical to the preferred version of the word. And sites should create a list of data definitions to help prevent these kinds of glitches.
In another example, if a study coordinator uses an unrecognizable term to describe an adverse event, the monitor will send out a query that asks for a term that is acceptable in any medical or coding dictionary.
Occasionally, sites will receive large numbers of queries that appear to be irrelevant, Howard says.
"This usually is the result of a data manager whose data checks are overly zealous," she says. "They program every data check they can think of because they can."
For instance, the data check system might send a query asking for a date that was omitted in a medical history report of a past hernia. The point is that the hernia is entirely irrelevant to the study, and so the missing date is also irrelevant, Howard explains.
"The site will write back that the patient can't remember when the hernia happened, and that was not a useful query," she adds. "Most people would not want a query for something like that, so there has to be a measure of judgment applied here."
But if a site continues to receive an unreasonable number of queries, it could be because the site has been careless, she notes.
"Most sites do not have quality processes in place to make sure they do things accurately," Howard says.
CR sites can prevent some of these issues by meeting with the monitor and data manager to discuss all of the data points and what the sponsor's expectations are for the quality of data, she suggests.
They can ask:
What fields are okay to have blank?
What range checks should happen?
What kind of cross data point issues are likely to arise?
"They should not start with a blank slate and instead have a list of standard questions to ask," she says. "And then they go through these and make sure to add any specific questions that have to do with the data particular to that trial."
Clinical trial data increasingly is collected and communicated electronically. Sites often are dependent on how sponsors and clinical research monitoring organizations set up their data collection and editing systems, but there are issues they can prevent through better processes and preparation.Subscribe Now for Access
You have reached your article limit for the month. We hope you found our articles both enjoyable and insightful. For information on new subscriptions, product trials, alternative billing arrangements or group and site discounts please call 800-688-2421. We look forward to having you as a long-term member of the Relias Media community.