12.2 Inpatient Review Prioritization: High-Risk & High-Volume
Key Takeaways
Define local review scope, capacity, and responsibilities rather than assuming a national bed ratio or universal impossibility of full coverage.
Prioritize meaningful unresolved issues and time sensitivity; high resource use or a low working DRG does not prove missing diagnoses.
A maximum severity tier does not complete accuracy review, and queries must not be driven solely by a search for backup CCs or MCCs.
Diagnosis removal requires evaluating the complete remaining grouping inputs rather than assuming every claim falls to the lowest tier.
Validate worklist tools with underlying records and appropriate unflagged samples; an algorithmic flag is not a diagnosis or proof of error.
Prioritizing Reviews by Actual Need and Available Capacity
A CDI worklist determines which records receive attention and when. Its design should support accurate documentation, clinically relevant clarification, timely handoffs, and the program’s authorized scope. Prioritization is necessary when work competes for limited time, but no universal staffing ratio proves that reviewing every admission is impossible or that a particular population is unworthy of review.
Define the program’s population, services, responsibilities, and available capacity. Some programs review broadly, some target particular areas, and some combine coverage with sampling. Describe that local design accurately. Do not turn an unsupported bed-to-reviewer ratio or a practitioner’s preferred caseload into a national professional requirement.
Identify a Review Need Rather Than a Desired Diagnosis
Useful signals can include a documented conflict, an important missing procedural detail, new information that affects a prior conclusion, a pending response near a local deadline, or an unresolved issue approaching discharge. A clinical change may make a record worth examining, but it does not establish a new complication. A working DRG that seems unexpected may reflect incomplete information, a mapping problem, or a legitimate case classification.
| Potential signal | What to verify | What not to assume |
|---|---|---|
| Transfer, rapid response, or new organ support | Actual event, available documentation, and any unresolved issue | The event automatically establishes shock, respiratory failure, or a complication |
| Single qualifying severity diagnosis | Support, correct code, exclusions, and the complete grouping result | Another diagnosis must be found as insurance against an audit |
| Resource-intensive stay with no qualifying CC/MCC | Documentation completeness, services, coding inputs, and classification | High resource use proves under-documentation |
| Imminent discharge with a meaningful open question | Clinical relevance, appropriate recipient, and local response process | Discharge should be delayed solely to reach a preferred DRG |
| Automated alert | Source data, reason for the alert, and actual record context | The algorithm has diagnosed the patient or proved a coding error |
A concern involving immediate clinical safety belongs with the appropriate treating or safety team. Administrative CDI prioritization must not substitute for clinical triage or delay necessary care.
Review All Severity Levels for Accuracy
A record already assigned to an MCC tier may still contain an unsupported diagnosis, wrong procedure, unresolved conflict, or incomplete discharge communication. A record without a CC/MCC may be entirely accurate. Do not equate the current payment tier with the completion or value of the review.
The effect of removing a diagnosis depends on the actual grouping logic and remaining inputs. In a hypothetical family with MCC, CC, and no-CC/MCC levels, removal of the only qualifying MCC could leave a qualifying CC and move the case to the CC level. It does not necessarily move every case to the lowest level, and it does not imply recoupment of the entire claim. Re-run the appropriate grouper with complete verified inputs when evaluating a real classification change.
The objective is complete and accurate reporting of supported conditions. Searching for “backup” CCs or MCCs solely to protect payment is an inappropriate premise for queries. A genuinely missing, reportable condition may warrant clarification, but the clinical and documentation need must stand on its own. Do not invent a minimum number of diagnoses for a defensible record.
Use Several Prioritization Dimensions
Consider time sensitivity, significance of the unresolved issue, availability of necessary information, responsible-provider access, and the effort needed to resolve it. Program scope and local operational deadlines also matter. A high projected financial effect can help describe an issue, but should not replace clinical and documentation considerations or justify different standards of truth across payers.
For an illustrative local morning queue, record A has a real contradictory diagnosis and anticipated discharge today. Record B already has a high severity tier but a newly identified unsupported diagnosis. Record C has complete reviewed information and no new event. A and B both have meaningful work; C does not become a higher priority merely because its diagnosis category has a high reimbursement weight. The local team still decides order based on actual deadlines, risk, information, and coverage.
Make deferred work visible. Assign an owner, reason, and next review trigger instead of letting a lower-ranked case disappear. A pending test, unavailable provider, or missing report may justify a planned return; it does not automatically establish that the case can be closed. Check the queue for aging or omitted cases as well as newly flagged ones.
Treat Technology as a Worklist Aid
Rules, natural language processing, and other models can suggest records for review using available data. A flag may reflect an ambiguous phrase, an actual mismatch, an old copied statement, or an extraction error. The reviewer must examine the underlying record and determine what question remains. Do not automatically add the suggested diagnosis or assume a model’s ranking is a clinical severity scale.
If the tool sends a patient-specific diagnostic clarification to a provider, it functions as a query and the 2026 brief’s query requirements apply. A human-reviewed worklist and an automatically delivered diagnostic question are different operations. Document which functions the system performs and who is responsible for review, correction, and monitoring.
Check performance in the population and workflow where the tool is used. Review false alerts and an appropriate sample of unflagged records; reviewing only alerts cannot show how much relevant work was missed. Examine changes after a template, interface, population, or rule update. Avoid claiming universal sensitivity, accuracy, or resource savings from a product description.
Worked Worklist Evaluation
In a fictional evaluation of 20 records, the tool flags seven. Human review finds that three of those seven contain the defined target documentation issue, while four do not. Among the 13 unflagged records, two also contain that issue. The reviewed sample therefore contains five target cases in total.
The proportion of flags that were correct is 3 / 7 = 42.9%, rounded. The tool found 3 / 5 = 60% of the target cases in this sample. These measures answer different questions. Neither is a diagnosis rate, a national benchmark, or proof that the tool works equally well in another service. The example shows why both flagged and unflagged records need attention in evaluation.
Use the results to refine the local rule or workflow and consider whether the time spent reviewing false alerts outweighs the benefit. A small sample supports investigation rather than a universal performance conclusion. Preserve an independent route for staff to identify meaningful issues the model did not flag.
Reassess the Prioritization Policy
Evaluate coverage, unresolved age, accuracy, response quality, and the consequences of deferred work alongside throughput. Review who is systematically missed and whether documentation needs in lower-ranked groups remain unaddressed. Changes in staffing or service mix may call for a different design.
A sound worklist supports thoughtful review rather than maximizing the number of severe diagnoses. The program should be able to explain why a record was selected, what the reviewer checked, what was resolved, and what remains pending. That explanation is more informative than a claim that every high-cost stay must contain an uncaptured complication.
Source anchors: 2026 ACDIS–AHIMA query practice brief, sections V–VI, IX, XI, and XIII; FY 2027 ICD-10-CM guidelines; final FY 2027 MS-DRG v44 definitions; ACDIS 2023 practitioner productivity examples.
In a hypothetical verified DRG family with MCC, CC, and no-CC/MCC levels, the only qualifying MCC is removed while one qualifying CC remains and no other input changes. What conclusion is appropriate?
The entire claim must be recouped because every diagnosis removal eliminates all payment.
The case necessarily moves to the no-CC/MCC level even though a qualifying CC remains.
The stated severity logic would leave a CC-level case; actual cases require complete-input regrouping rather than a universal lowest-tier assumption.
A second MCC must be created through a query to preserve the original classification.
A record is already at a high severity tier, but the reviewer identifies a significant unsupported diagnosis and an unresolved provider conflict. How should its review priority be considered?
Close it because high-tier records cannot contain material documentation errors.
Defer it permanently because accuracy matters only when a higher payment tier is available.
Search only for another MCC and leave the unsupported diagnosis unchanged.
Assess the actual unresolved issue, timing, and risk under the local process; the existing severity tier does not eliminate the need for accuracy review.
In a fictional 20-record evaluation, a tool flags seven records: three contain the defined documentation issue and four do not. Two of 13 unflagged records also contain the issue. What is the correct interpretation?
Three of seven flags were correct, about 42.9%; three of five target cases were found, 60%. These are different sample measures, not universal performance claims.
The tool found every issue because only the seven flagged records count in its evaluation.
Seven of 20 records must contain the issue because the algorithm flagged them.
The tool’s sensitivity is 100% because three flagged records were confirmed by a reviewer.
Sections you finish are checked off in the contents.