Double data entry vs confidence based review is a choice between keying every field twice and comparing, or letting software read the form and sending only its low-confidence fields to a person. Double entry catches typing slips everywhere but doubles the labor. For most fields, confidence based review is the better trade, provided its scores track real errors.
What double data entry catches and what it costs
Double data entry is a check where two people key the same form separately and software compares the two versions. Wherever they differ, someone looks at the paper and settles the value. It catches typing slips that one keyer makes and the other does not, and it costs a full second pass of labor on every field.
The method is described in Field Trials of Health Interventions by Smith, Morrow and Ross: two data entry clerks "independently enter the data without any knowledge of the other's work," a program compares the entries, and most often "a third person resolves the discrepancies, by referring to the original forms or questionnaires." The same chapter argues that a second entry can cost less than cleaning up the errors single entry leaves behind.
There is evidence that it works on typing errors. In a study of patient questionnaires, Paulsen, Overgaard and Lauritsen's comparison of single entry, double entry and automated forms processing found that double-key entry produced fewer errors than single-key entry on the checkbox questionnaires. That result comes from one study of hip replacement patients' forms; it is not a benchmark for intake forms or EOBs.
The cost is the catch. Every field is keyed twice, whether it was hard to read or perfectly clear.
How confidence based review compares on errors caught
Confidence based review catches the errors software is unsure about and misses the ones it is wrongly sure about. Each field gets a confidence score, and fields under a threshold go to a reviewer who checks them against the page. Faint ink, smudges and messy handwriting tend to lower scores, so those misreads reach a person.
The weak spot is a confident misread. If the software reads a clean-looking 7 as a 1 and scores it 98%, the field skips review. Double entry has its own blind spot: if both clerks misread the same illegible digit the same way, the entries agree and nothing is flagged. Neither method catches an error written on the form itself, such as a patient writing the wrong birth year.
Per 100 intake forms, side by side
This comparison uses example numbers for 100 fictional intake forms with 40 fields each, so 4,000 fields. Every figure is an illustration, not a measured rate.
| Measure (example) | Double data entry | Confidence based review |
|---|---|---|
| Fields keyed by people | 8,000 (4,000 twice) | none keyed; 240 checked (6% routed) |
| Keystrokes at 8 per field | about 64,000 | about 480 (60 corrections typed) |
| Staff minutes at 5 seconds a keyed field, 15 seconds a checked field | about 667, plus discrepancy time | about 60 |
| Errors caught | Slips one keyer made and the other did not | Misreads that scored under the threshold |
| Errors missed | Same misread by both keyers; errors on the form | Confident misreads; errors on the form |
In this example, confidence based review uses under a tenth of the staff time. That gap closes if the routing rate climbs, which it does on faxed or handwritten pages. The review queue those 240 fields land in has its own design questions, covered in the guide to designing a document review queue, and the people it needs are worked out in the post on review queue staffing.
Is double data entry still worth it for any field?
Double data entry is still worth it for a few fields where one wrong character causes serious harm and the score alone is not trusted. Candidates include a patient's date of birth and member ID, a consent checkbox, and any value that links a record to a person. Here a second independent look can justify the extra seconds.
You do not need two keyers to get that second look. A practical middle path is to route every value of a chosen high-stakes field to review regardless of score, while letting other fields pass on the threshold. That gives you a person on each date of birth without re-keying the patient's favorite pharmacy.
Double entry also still fits when no software reads the form at all, such as a one-off research questionnaire with a hand-drawn layout. The Paulsen study concluded that automated processing suits highly structured forms; free-form or unusual layouts are where manual methods keep their place. For Schema Care, the routing rule works per field: anything under the threshold you choose waits for a reviewer who sees it marked on the scanned intake form, and the list of documents Schema Care reads includes intake forms with demographics, insurance, history and consent checkboxes.
What do you lose when you drop double data entry?
When you drop double data entry, you lose a check that touched every field, including the ones software is confident about. Confidence based review never looks at a field that scored above the threshold, so a confident misread can pass. The safety net you need instead is a regular sample of accepted fields, checked by a person.
Three replacements cover most of what double entry did:
- Sampling. Pull a small random set of fields that skipped review each week and check them against the page. If errors show up, lower the threshold for that field type. The method is set out in the post on why to sample fields that already passed the confidence threshold.
- Always-review fields. Route every value of a short list of high-stakes fields to a person.
- Validation rules. Check formats and ranges, such as a birth date not in the future or a member ID with the right length, so obvious misreads are flagged even when scored high.
With those in place, you keep a check on confident errors while spending reviewer time where the software is unsure. More on intake paperwork, from form design to getting data into the EHR, is in the patient intake category.