YouEDC Biostat: the study's statistics by itself

Ten charts and the tables of efficacy and safety from the live data, by arm when the reader may see the arms, and the PDF of the report.

YouEDC Biostat is the seventh and last section of the portal: the statistics of every YouEDC Capture study count themselves from the data as it is now: who takes part, how the key variables change over the visits, which adverse events there are. It is a quick look at the study between the analyses and a simple report for studies of dietary supplements, observational studies and registries. The report downloads as a PDF with one button.

The manager, data manager, statistician and the sponsor's viewer open the section; the investigator and the monitor do not see it. A member limited to their sites sees only those sites' patients. The report shows no patient codes, only summary numbers.

Four steps from the data to the PDF. Nothing is copied: it is counted again every time.

Where to find it

The YouEDC Biostat tab is the last one in the left menu, below YouEDC Supply. The section's first page lists the studies whose report you may read: how many patients and sites, how many randomised, whether the numbers are by arm or for the whole study together (blinded), whether the variables of the report were chosen. The Report button opens the study's report, the PDF button downloads it at once. Inside a study the same report opens from YouEDC Biostat in the study menu, after Data quality.

To see the section without real data, create a demo study: in both demos (type 2 diabetes and arterial hypertension) every chart and the PDF are ready right after creation.

The YouEDC Biostat section: the studies whose report you may read, with the report and PDF buttons.

What the report holds

At the top are tiles: patients, randomised, completed, withdrawn and lost, with adverse events. Then three parts.

• Patients. Enrolment by month, cumulative; how the patients take part (active, completed, withdrawn, lost, died); age and sex; Table 1 with the baseline values of the key variables.
• Efficacy. For each key variable a chart over the visits: the mean with its 95 % confidence interval for numbers, or the shares of the answers for a single choice. For the first numeric variable, the change from baseline: each patient as a dot, the quartiles, the median and the mean. Below the charts, a table by visit: n, mean (SD), median [Q1; Q3], minimum and maximum, the change from baseline with its own 95 % CI.
• Safety. Patients by their worst adverse event, the most frequent terms, a summary table (serious AEs, relationship to the product) and charts of the safety variables, such as ALT or creatinine.

YouEDC Biostat in the demo study: the tiles, enrolment by month and how the patients take part, by arm.

Arms and blinding

If the study has randomisation, the report shows the numbers by arm, but only to readers who may see the arms. In an open study every reader of the report sees them. In a blinded study only members marked unblinded do; the others see the whole study together, with a note saying so. Forms for the unblinded team stay out of the report for readers they are hidden from.

In a demo project everyone sees the arms, because the data is invented: this shows what the report by arm looks like. In the demo the active arm differs from placebo in efficacy (HbA1c and weight in diabetes, blood pressure in hypertension) and in side effects (nausea, for example, is more frequent).

Variables of the report

Key variables of efficacy: up to four numbers or single choices filled at two visits or more (outside demographics, history, adverse events and therapy). Safety variables: up to three numbers. Until someone chooses them, the report takes the first numbers, critical ones first, and offers to choose the variables of your protocol.

The manager, data manager or statistician chooses: the Variables button leads to the list at the bottom of the page. Every change is recorded in the audit trail with the old and the new list.

Efficacy in the demo: means with 95 % CI over the visits and the shares of the patient diary's answers.

Hy’s law plot (eDISH)

When the study collects ALT and total bilirubin (fields or lab table rows coded ALT and BILI) and keeps its own lab normal ranges, the Safety part shows the Hy’s law plot. Each patient is a point: their highest ALT and their highest total bilirubin, each divided by the upper limit of normal (ULN) of the range for their site, sex, age and visit date. Both axes are logarithmic; the dashed lines stand at 3 × ULN for ALT and 2 × ULN for bilirubin, after the FDA guidance on drug-induced liver injury (2009).

• Upper right: potential Hy’s law. A patient is listed when their bilirubin reached 2 × ULN on the day of an ALT of 3 × ULN or within 30 days after it.
• Lower right: Temple’s corollary, ALT raised without bilirubin.
• Upper left: bilirubin raised without ALT (hyperbilirubinaemia, cholestasis).

Readers who may open patients (manager, data manager) see the patient code on each point, and the point opens the patient; a statistician or the sponsor's viewer sees the points without codes, and the PDF never carries codes. Without the study's ranges the plot is not drawn: the portal never guesses the upper limit of normal. The plot is a signal for the safety physician, not a diagnosis: alkaline phosphatase and other causes are not checked, and every case is reviewed by a physician.

In both human demos one patient has a drug-induced liver injury caught at a chemistry visit, so the plot has a point in the upper right corner.

Time to event and exposure

The “Time to event and exposure” part shows Kaplan–Meier curves by arm for three events: the first adverse event, the first serious adverse event and early discontinuation. The curve is the share of patients who have not had the event yet; the light band is its 95 % confidence interval, and a vertical tick is a patient followed that long without it. Under the axis, the patients still at risk. The table gives for each arm the patients with the event, the median time to it with its 95 % CI and the rate per 100 patient-years (EAIR) with its exact Poisson 95 % CI, and the total follow-up.

Time to event in the demo: Kaplan–Meier curves by arm and adverse events per 100 patient-years.

• Day 0 is the patient's enrolment date. An event before it is not counted.
• Follow-up ends on the date of completion, withdrawal or death. For a patient still taking part, on their last contact: the latest actual visit date, the day a form was saved (imported ones aside) or an adverse event onset, never after today.
• The onset of an adverse event is the field AESTDAT of the form with AETERM. An event without a date leaves its patient out of that endpoint, and the table says how many.
• Rate per 100 patient-years: patients with a first event divided by their time at risk (to the first event, else to the end of follow-up). A year is 365.25 days. It is fairer than a plain share when the arms were followed for different times.

Shift tables of lab results

The “Lab results against the normal range” part shows, for the lab tests with the study's normal ranges, a shift table (ICH E3, section 14.3.4): rows are the patient's first result (low, normal, high), columns their worst later result. The “normal → out of range” cells are highlighted. The tests with the most such shifts come first, eight are shown. The worst result is the one furthest outside the range (as a share of its width). A result without a range is left out: the portal never guesses the range.

Checked against references

The YouEDC Biostat section page has a Checked against references button. There the portal's methods run on published data sets, and each result stands next to the reference: the NIST StRD accuracy sets NumAcc1–4 and the QUARTILE.INC example (mean, SD, quartiles), the example of R's survival package on the aml data (Kaplan–Meier curve, standard error, CI limits, median), Student's t quantiles and exact Poisson limits by SciPy. For each number you see how many significant digits agree. The page is counted again each time it opens, on your installation, and PDF for the validation file saves a record with the build, the author and a line for the reviewer's signature; every download is in the audit trail.

It tests the calculations against an independent reference, as GAMP 5 asks of software made in-house. It does not replace the validation of a study's own analysis, and before the numbers are used for decisions a biostatistician reviews them.

The PDF report

The PDF report button downloads the whole report on A4 pages: a cover with the study, sponsor, sites and arms, then the same charts and tables as on the page, two charts to a row. The PDF holds the file biostat.json with the same numbers, and its SHA-256 is printed at the bottom of every page: it shows that the numbers were not changed. Every download is recorded in the audit trail with the number of patients and the start of the hash.

The top of every page says “Descriptive statistics, not for advertising”. The report does not prove efficacy and does not replace the analysis by the statistical analysis plan. For dietary supplements this matters: claims about their properties in advertising are restricted by law (in Russia, article 25 of the Law on Advertising), and the descriptive numbers of a study are no such evidence.

How it is counted

• Patients of the sites you see, without those who withdrew the use of their data.
• A visit is a scheduled or recurring visit of the plan, in its order. The baseline of a variable is its first visit with data.
• Change from baseline is the patient's value at the visit (or at their latest visit) minus their own baseline; only patients with both values count.
• Mean with its 95 % confidence interval by Student's t, median and quartiles, SD is the sample standard deviation. There are no p-values and no tests between arms.
• A hollow point with a dashed line: fewer than 5 patients at that visit; its interval is cut at the edge of the chart and given in full in the table.
• Adverse events come from the form with the CDASH variable AETERM; terms are grouped as written (case and spaces aside), not coded. A patient counts once per term and by their worst severity (AESEV); seriousness is AESER, relationship AEREL.
• Hy’s law plot: each ALT and bilirubin result is divided by the upper limit of the study’s range for the patient's site, sex, age and visit date (else the day the form was first saved); results without a range are left out, and their patients are counted under the table.
• When there are many patients, the report is kept for 10 minutes; the “Count again” link next to the time of counting counts it anew.