A diagnostic accuracy study lives or dies on pairing: every participant needs the index test result, the reference standard result, the timing between them and a clear record of who read what. Capture keeps those pieces in one study with one audit trail, ready for the two-by-two table. Build it free in the sandbox.
Free sandbox · No credit card · 21 CFR Part 11 aligned
| Reference positive | Reference negative | Total | |
|---|---|---|---|
| Index test positive | 78 | 14 | 92 |
| Index test negative | 12 | 88 | 100 |
| Index test indeterminate | 3 | 5 | 8 |
| Total | 93 | 107 | 200 |
What a diagnostic accuracy study needs from its data capture
The study design
A diagnostic accuracy study asks how well an index test (a rapid assay, an imaging algorithm, a questionnaire, a wearable measurement) identifies a condition that is established by a reference standard. The main outputs are sensitivity, specificity and often predictive values and likelihood ratios, with confidence intervals. The reporting guideline most commonly followed is STARD 2015, which lists 30 essential items intended to help readers judge risk of bias and applicability. Several of those items are about data that has to be captured at the time, such as when tests were done and by whom, rather than reconstructed later.
That makes a diagnostic study structurally different from an interventional trial. There is usually no randomization and no treatment, but a high premium on the integrity of pairs. Every participant should have a recorded index result, a recorded reference result, the interval between them and, when more than one reader is involved, which reader read which case. Participants are often recruited consecutively or by a defined sampling strategy, so the screening log, the number approached and the reasons for exclusion are data too. The eligibility screening template and the screening and enrollment log template show how to structure that.
The practical risks are well known. If the reference standard is only done on people with a positive index test, the study has verification problems. If the reader of the reference standard sees the index result, the results are no longer independent. If indeterminate results are removed without being counted, accuracy looks better than it is. None of these are software problems, but a data system can make each of them either visible or invisible. The aim is to capture the information a methodologist needs to tell which of these happened.
Building the forms
Build the index test and the reference standard as separate eCRF forms, each with its own date, time, operator or reader, device or kit lot where relevant, and the result. Keep results as typed fields: a numeric value and a unit for a continuous measure, a coded category for a binary or ordinal one, and a distinct option for invalid, indeterminate or not performed with a reason field. Where the index test produces a continuous score, capture the raw value so that thresholds can be explored in the analysis, rather than only the categorized result.
Calculated fields can help with timing. Capture can calculate the minutes or hours between two date-times and show it read-only during entry, which makes the interval between index test and reference standard explicit and testable. An edit check can flag an interval beyond the protocol limit for query. Use the same pattern for age from date of birth, and for calculators relevant to the topic such as BMI or eGFR where they feed eligibility. The edit checks page describes how rules and auto-queries work, and the query management page shows the review workflow.
For studies with multiple readers, build one reading form that repeats per reader and carries the reader ID, the read date, the result and a confidence rating if the protocol uses one. Capture does not review images itself; the read is entered or imported by your team from whatever reading tool you use. Blinded roles never receive treatment-arm values, which is relevant to randomized designs; for independence between readers or between tests, use separate forms and role assignments and confirm in the sandbox what each role can see. If the index test is a device, EDC for medical device clinical trials covers device-specific fields, and for digital measures, digital biomarker validation study software is a close companion.
Index test
Test date and time
IXDTCOperator ID
IXOPERRaw value
IXVALResult
IXRESMinutes to reference standard
IXGAPCalculated from the two date-times
95 min CalculatedProtocol to build
| Need | What the data looks like | Where it lives in Capture |
|---|---|---|
| Participant selection | Approached, eligible, enrolled, reasons for exclusion | Screening form and enrollment log; eligibility edit checks |
| Index test | Date, time, operator, raw value, categorized result | Site eCRF form with typed fields and a distinct indeterminate option |
| Reference standard | Date, time, reader, method, final diagnosis | Separate eCRF form linked to the same participant |
| Timing between tests | Minutes or hours between date-times | Calculated field with a range edit check |
| Multiple readers | Reader ID, read order, result, confidence | Repeating reader form |
| Corrections and re-reads | Old result, new result, reason | Field-level audit trail |
| Analysis dataset | Wide file ready for 2x2 tables | CSV and Excel export with data dictionary |
Capture records the data. Accuracy statistics, thresholds and confidence intervals are computed in your analysis software.
Create both forms in the free sandbox, enter practice participants with an indeterminate result and a late reference test, and check how it exports. No credit card.
Planning and analysis
Sample size for a diagnostic accuracy study is driven by the target sensitivity or specificity, the precision wanted around it and the prevalence of the condition in the recruited population, which determines how many participants are needed to get enough cases. The power analysis calculator and the sample size calculator give a first estimate for comparisons, and your statistician will choose the diagnostic-specific method. If your recruitment strategy is likely to overstate prevalence, plan for it explicitly.
On the data side, the export is a wide CSV or Excel file with an automatic data dictionary, raw and decoded versions, and filtering by site and date range. Export EDC data to R, SAS and SPSS shows the usual path, and in R or SAS the two-by-two table and its confidence intervals are a short script on top of the index and reference columns. Because indeterminate and missing results are explicit values, your analysis can report them and apply whichever handling rule the protocol pre-specified.
For a multi-site or multi-reader study, the multi-site clinical trial management and audit trail pages explain by-site exports and the append-only trail. A small single-center study may only need EDC for single-site clinical trials. Consent for a diagnostic study is often simpler, and the eConsent page covers on-screen signature if you need it.
Before first participant
What it is, who reads it and the maximum interval after the index test.
Who sees which result and when, written before the first read.
How inconclusive, invalid and missing results are counted and reported.
Reader IDs, experience, read order and a re-read rule.
Consecutive or sampled, with reasons for exclusion recorded.
A mock two-by-two table built from practice participants.
Yes. Build them as separate site eCRF forms on the same participant record, each with its own date, time and operator or reader, so a single export holds both.
Give the result field explicit options for indeterminate, invalid and not done, with a reason field. They then export as values, and your analysis plan decides how to report them.
No. Capture captures and exports the data; you compute accuracy statistics in R, SAS, SPSS or similar from the exported file.
Yes, with a repeating reader form carrying reader ID, date and result. Image review itself happens in your own reading tool.
STARD 2015 is a reporting guideline for your publication, not a software feature. Capturing timing, readers, flow and indeterminate results as structured data makes it easier to complete the items that need them.
Keep exploring
Free sandbox with every feature. No credit card, and you pay only when you go live with real participants.