Data quality report

Completeness of each variable by site, outliers, entry delay and loss to follow-up, with the rules behind each number.

Before an interim analysis, an annual registry report or a publication you need to know how far the data can be trusted. The Data quality report gathers four checks on one page: completeness of each variable by site, outliers, entry delay and loss to follow-up. Everything is computed from the live data of the sites you see, without the patients who withdrew the use of their data.

Completeness

For each variable: how many values are filled among the pages where the field was shown. A field hidden by its show condition (say, “cigarettes a day” for a non-smoker) is not expected and does not spoil the percentage. Drafts count: an empty field of a draft is a missing value today too. A “no data” mark (ND, UNK, NA) is an answer: it is shown apart in the cell's tip and is not counted as a filled value.

• Green is 95 % and more, amber 80–95 %, red below 80 %.
• The Only variables below 95 % button keeps only the variables with gaps in the table.
• Required fields of completed forms are always filled; completeness shows first of all the optional variables, which are what registries lose.

Why there is no data. Once the site gives reasons for “no data” marks, a table under completeness counts the marks of each variable by reason, and “Reason not given” shows what the site left without one. Another table counts the unscheduled visits that took place by why the patient came, by site: when most visits are “worsening”, the visit count follows the patient’s state. Both tables are in the Excel file too.

Outliers

For number fields with 10 values or more, the modified z-score of Iglewicz and Hoaglin is computed: z = 0.6745 × (value − median) / MAD, where MAD is the median of distances from the median. A value with |z| above 3.5 goes to the list. The method is robust: the outlier itself hardly moves the median and the MAD. The list starts with the farthest values, and a link opens the patient's page. An outlier is not necessarily an error: it is a reason to look and, if needed, raise a query.

Entry delay

How many days pass from the visit (its actual date when the site recorded it, otherwise the planned one) to the first save of the page. Only pages typed at the site count: imports and ePRO are left out. The table gives the median, the 90th percentile and the share of pages entered later than 30 days, by site.

Loss to follow-up

For each site: how many patients were enrolled, how many are lost to follow-up or withdrawn, and their share. The At risk column shows active patients with an overdue visit or without new data for 180 days: call them before they are lost. Attempts to reach a patient are kept on the patient page.

From a number to the patients

Every number of the report with patients behind it is a link. It opens a list from which you go straight to the patient or to the page:

• a completeness cell: the pages at that site where the field was shown and not filled; the Study column, across all your sites;
• Completeness: every page with fields not filled, each with the fields it lacks;
• Outliers: jumps to the outlier table on the same page, with patient, visit, value and a link to the form;
• Entry delay and the “later than 30 days” percentage: the pages first saved more than 30 days after the visit, the latest first;
• Lost and withdrawn, and the lost, withdrawn and at-risk numbers of the follow-up table: those patients, with why they are at risk.

The list has exactly as many rows as the number stands for: the rules are the report's. Only members who see patients get the lists; a site monitor sees their own sites only, and a blinded member does not see the forms for the unblinded team. In a big study the list, like the report, is kept for a few minutes.

MH_OTHER is 48 % complete at site 02. Clicking the number shows the pages where the field is empty; a “no data” mark (ND) is in brackets.

Excel and permissions

The Excel button downloads the same tables and an “About the file” sheet with the rules; the download goes to the audit trail. The project administrator, the data manager, the monitor and the sponsor's viewer see the report. The list of outliers with patient codes is there only for those who see patients; the viewer sees how many outliers each variable has. A member with access to some sites sees only those sites.