Compare direct-to-bill and coder-in-the-loop workflows, including human review, exception handling and how to choose an automation tier.
Direct-to-bill coding sends AI-coded charts to billing without a coder reviewing them. Coder-in-the-loop sends every chart to a coder, with the AI's codes already drafted. In practice most deployments that use direct-to-bill are hybrids, because charts below a confidence threshold still go to a coder, and that threshold is something someone has to own. That threshold is a release policy, and it belongs to your HIM and compliance leaders, not to a vendor's default settings.
MEDIQCODE™ keeps a coder's approval on every chart, so we aren't neutral on the answer. This page sets out the trade-offs anyway, including the case for direct-to-bill, because a policy you can defend has to survive the strongest argument against it.
Direct-to-bill coding and coder-in-the-loop, defined
Direct-to-bill coding (also called straight-to-bill or touchless coding) is a release path in which software assigns the codes for a chart and sends it to billing with no human review before the claim is created. Charts below a confidence threshold go to a coder instead. Coder-in-the-loop means every chart reaches a coder, who approves or changes the AI's draft before release.
| Coder-in-the-loop | Hybrid | Direct-to-bill | |
|---|---|---|---|
| Who reads the chart before the claim | A coder, every chart | A coder, below-threshold charts only | Nobody, for released charts |
| Headline metric | Coder change rate, throughput | Automation rate plus sampled accuracy | Automation rate |
| How errors are found | At review, before billing | At review or in post-release sampling | Post-release sampling, denials, audits |
| Who should own the threshold | Not applicable | The hospital; confirm it in the contract | The hospital; confirm it in the contract |
These are release policies, not product categories, and the same engine can run under any of them.
What changes when no coder reads the chart
The claim still carries your certification
The software doesn't sign the claim. The institutional claim form carries a certification that the billing information is "true, accurate and complete" and that the submitter did not knowingly or recklessly disregard material facts (UB-04 / CMS-1450, maintained by the National Uniform Billing Committee). Under the False Claims Act, "knowingly" includes reckless disregard (31 U.S.C. 3729(b)(1)). A release rule that nobody monitors is the practical risk that definition points at.
No regulator draws the review line for you
Software intended for administrative support of a health care facility, including processing claims and billing information, is excluded from the medical device definition by section 520(o)(1)(A) of the Federal Food, Drug, and Cosmetic Act (FDA Digital Health Policy Navigator). In practice, no US agency tells you which coding output must be reviewed. Your compliance program sets the line, and your compliance program answers for it.
Errors move from before the bill to after it
Under coder-in-the-loop, a wrong code is usually caught before the claim drops. Under direct-to-bill, the same error surfaces through a payer denial, an external audit or your own sample, after the money has moved. If that error produced a Medicare overpayment, the 60-day report-and-return clock applies.
The sampling math behind direct-to-bill
Direct-to-bill replaces 100 percent review with a sample. The question is what a clean sample actually proves. When a sample of n charts contains zero errors, the upper bound of the 95 percent confidence interval for the true error rate is roughly 3 divided by n. This is the "rule of three" (Hanley and Lippman-Hand, JAMA, 1983).
Worked example: 5,000 released charts a month
Suppose a hospital releases 5,000 charts a month through direct-to-bill and its QA team finds no errors in its monthly sample. The volume is illustrative; the arithmetic is not.
| Clean QA sample (charts) | 95% upper bound on error rate | Charts that could still be wrong each month |
|---|---|---|
| 30 | about 10% | up to 500 |
| 100 | about 3% | up to 150 |
| 300 | about 1% | up to 50 |
| 1,000 | about 0.3% | up to 15 |
A 30-chart spot check with no findings is compatible with one released chart in ten being wrong. It takes roughly 300 clean charts a month to bound the rate near 1 percent, and even then that bound applies to the population sampled. Errors cluster by chart type, so a sample drawn across all released charts can miss a problem concentrated in one service line. Stratify by chart type, and treat every code-set update as a reason to resample.
Coder-in-the-loop produces a different signal. Because every chart is seen, you measure how often coders change the AI's codes, on every chart, instead of estimating an error rate from a sample.
Choosing the tier by chart type, not by vendor
The release decision works best chart type by chart type. A single hospital can reasonably keep inpatient coding under full review while it debates release policy for a narrow outpatient stream.
| Factor | Points toward considering release without review | Points toward coder-in-the-loop |
|---|---|---|
| Documentation | Short, templated, low variation | Long, free-text, multiple authors |
| Code-set breadth | A narrow set of codes covers most charts | Many valid combinations per chart |
| Payment impact per chart | Low; one code rarely changes payment much | High; one secondary diagnosis can change the MS-DRG |
| Guideline judgment | Little interpretation needed | Principal diagnosis selection, sequencing, POA |
| Query dependency | Rarely needs a physician query | Frequently waits on a clinical documentation integrity (CDI) query |
| Audit exposure | Low payer and OIG attention | Named in audit programs or payer policies |
Inpatient charts under Medicare Severity Diagnosis Related Groups (MS-DRG) sit in the right-hand column on almost every row, because a single complication or comorbidity (CC) or major CC (MCC) can move the payment. Narrow outpatient streams sit further left. The table tells you where the debate is worth having. It doesn't tell you where to put the threshold.
Revisit the scoring at every code-set update and whenever a payer adds a chart type to a review program.
The case for direct-to-bill, stated fairly
The strongest argument for direct-to-bill is that human coding isn't reviewed at 100 percent either. Most departments audit a sample of human-coded charts, so a well-sampled release policy holds the AI to the same standard applied to coders. Proponents add that for narrow, high-volume chart types, review time adds cost and discharged-not-final-billed (DNFB) days without catching much.
Both points are fair. The difference is who carries the judgment. A coder who assigns a code has read the chart and can explain it. Under direct-to-bill, nobody has read the released chart, and the explanation lives in the vendor's model and threshold settings. That matters most when an auditor asks why a code was reported.
Why MEDIQCODE keeps a human decision point
MEDIQCODE's most automated level has the AI code the whole chart before a coder opens it. The coder still approves or adjusts every code before anything is released, and no level of the product releases a chart without that approval. Three design choices make review fast enough to keep:
- Every proposed code carries written reasoning and a link to the source text in the record, so the coder checks evidence rather than re-reading the chart from scratch.
- Output is constrained to a closed codebook, with confidence thresholds that shape what the coder is prompted to check. No threshold releases a chart.
- The audit trail records who coded, who audited and the reasoning for each code, so the answer to an auditor's question exists before it is asked.
The trade-off is real. We give up the throughput of charts no one touches. What we keep is a named, accountable person on every claim, and a measurement (how often coders change the AI's codes) taken on every chart instead of inferred from a sample.
Questions to ask any vendor offering direct-to-bill
| Question | Why it matters |
|---|---|
| Who sets the release threshold, and can we set it per chart type? | The threshold is your compliance decision, not a vendor default |
| What happens to the threshold on October 1 and January 1? | Code-set changes can invalidate a threshold tuned on last year's codes |
| Who performs post-release sampling, and at what size? | Vendor-run QA grades its own work |
| Can we see the reasoning for any released chart, on demand? | You'll need it for audits and appeals |
| How quickly can we pause release for one chart type? | Rollback speed limits the size of any error |
| How does the contract allocate responsibility for released-chart errors? | The certification on the claim stays with you either way |
Frequently asked questions
Is direct-to-bill coding compliant?
It can be, if it is governed. Medicare's claim certification puts responsibility on the provider rather than prescribing who, or what, assigns the codes, and software intended for claims and billing support is generally outside FDA device regulation. Compliance therefore depends on your sampling, your monitoring and your ability to explain every released code.
What automation rate should we expect?
It depends entirely on the denominator and the chart types in scope, which is why what counts as a good automation rate starts with the definition rather than a number.
Who should own the release threshold?
The hospital. The HIM director or coding manager should propose it per chart type, the compliance officer should approve it, and the contract should say the vendor can't change it without written sign-off. A threshold the vendor tunes on its own schedule is a release policy you no longer control.
Can we start with coder-in-the-loop and move to direct-to-bill later?
Yes. Many hospitals start with full review and consider release chart type by chart type once they have months of coder change data. MEDIQCODE doesn't offer a release-without-review option, so that step would be your decision with a different tool.
Your next step: classify your chart types
List your chart types, score each against the factor table, and decide the release policy for each with your compliance officer before any vendor demo. Direct-to-bill coding is a choice you can make one chart type at a time, and you can decide never to make it.
