6.2 Documentation Improvement Project Identification & Scoping

Select a project by establishing the problem

A promising CDI project begins with a question that matters to patient care, record integrity, or an important workflow. A dashboard can show where to investigate; a record review establishes what is happening. Separate three decisions: whether the signal is reliable, what explains it, and which change the organization can test. This prevents a revenue or severity metric from becoming a diagnosis quota before anyone has examined the underlying cases.

For Medicare-participating hospitals, 42 CFR 482.21 requires a hospital-wide, data-driven quality assessment and performance improvement program. Its priorities include high-risk, high-volume, or problem-prone areas, with attention to incidence, prevalence, severity, and effects on outcomes, safety, and quality. Hospitals must document their projects, reasons, and measurable progress. This supports connecting CDI work to the hospital's quality priorities. It does not prescribe one CDI charter template or a national minimum query-agreement percentage.

Interpret signals as competing hypotheses

Consider the following examples before choosing a diagnosis-specific initiative:

CMI is a summary of assigned DRG relative weights. A difference does not by itself distinguish a documentation problem from a change in services or patient complexity. Similarly, a high query-agreement rate can coexist with leading questions or unsupported diagnoses. A defensible project identifies a specific defect through an audit rather than treating an indicator as its own explanation.

Read mortality and length-of-stay comparisons carefully

For a measure defined as observed deaths divided by expected deaths, an O/E value of 1.48 means observed deaths are 1.48 times the model's expected count for the stated cohort and period. Interpretation also depends on the model, sample size, uncertainty, and data quality. It does not prove that documentation caused the result or establish a universal national threshold for every mortality program.

Risk adjustment is measure-specific. Expected outcomes may incorporate demographics, clinical factors, timing, and other variables specified by the model. They are not universally calculated entirely from secondary diagnoses marked POA Y. Adding a diagnosis need not change a given model's expectation. Clinical review may identify either missing supported risk information or genuine care concerns; both deserve an honest account. Do not target additional comorbidities simply to make an unfavorable result appear better.

Length-of-stay comparisons also require matching definitions. A service's arithmetic average LOS and a DRG's geometric mean LOS are different statistics. A difference can suggest cases to review, but it does not establish missing complications or an individual patient's appropriate discharge date. Separate medical complexity, clinical delays, placement barriers, operational delays, and data errors. Payment and transfer rules require their own application, as discussed in the grouping chapter.

Validate the data before auditing the clinical hypothesis

Write a short measure dictionary before comparing results. Specify the population, setting, dates, unit of analysis, numerator, denominator, exclusions, data source, and code/model version. State whether a denial count represents claims, diagnosis lines, unique encounters, or all appeal events. Deduplicate the same case when the intended unit is a unique encounter.

Check whether the denominator has had enough follow-up for the outcome. A recent discharge cohort may have few denials simply because payers have not completed review. Compare cohorts with a consistent opportunity for an outcome to appear, and keep pending decisions distinct from upheld or overturned denials. Record extraction changes, missing data, and changes in payer mix. Otherwise, a reporting change may look like a clinical improvement.

Choose a sample that answers the question

A targeted review is useful for investigating an unusual subgroup, such as repeatedly denied records. A probability sample can estimate how often a defect occurs in a defined population, subject to sampling uncertainty and sound execution. These designs answer different questions. A set containing only denied claims cannot establish the defect rate among all discharges. A convenience sample from the most cooperative providers may miss problems elsewhere.

Build an audit instrument with explicit criteria and sources. Separate a missing provider explanation, an unsupported diagnosis, a coding error, and a compliant record. Reviewers should apply the same definitions and reconcile disagreements about the instrument before interpreting trends. Include cases without the suspected defect so that the team can test alternative explanations. A project is better scoped when the team can say what would disprove its initial assumption.

Turn findings into a charter

Suppose a hospital suspects that nutrition assessments are not being reconciled with provider documentation. In an illustrative local audit, reviewers examine 100 eligible adult inpatient records selected from a defined discharge period. They find 20 with an unresolved discrepancy after applying the same review criteria. This is a 20% observed defect rate in that audit. The finding concerns the sampled population and stated method; it does not establish that all 20 patients had malnutrition or that a particular severity code was omitted.

A useful charter could contain:

• Problem: unresolved nutrition-assessment/provider-documentation discrepancies in 20 of 100 audited eligible records.
• Aim: reduce the observed unresolved-discrepancy rate to 10% or less in the defined follow-up audit by the end of the next quarter, using the same eligibility and adjudication rules.
• Scope: the selected adult medical unit and the handoff from dietitian assessment to provider review; other units remain outside the initial test.
• Team: an accountable sponsor, project lead, dietitian, diagnosing-provider representative, CDI, coding, quality, and an analyst; involve other affected staff as needed.
• Proposed change: test a reliable notification and review process that makes the assessment visible and supports an appropriate clarification when needed.
• Measures: unresolved-discrepancy rate, completion of the notification step, provider workload, and audit of the resulting documentation and queries.
• Decision: review the test findings with the sponsor and decide what to revise before broader implementation.

These numbers and roles are a worked local design, not AHIMA requirements. A follow-up audit of 100 eligible records with 10 unresolved discrepancies would be 10%, an absolute decrease of 10 percentage points and a relative decrease of 50% from 20%. That arithmetic describes the samples; a causal or statistical conclusion requires an appropriate analysis of variation, comparability, and other changes.

The aim should permit all clinically appropriate resolutions: a supported diagnosis and degree, a different explanation, no additional condition, or an inability to determine. “Obtain severe malnutrition in 90% of queried records” would distort that purpose. Financial effects may be estimated in program planning, with explicit assumptions and costs. Including financial information in a charter is not itself a False Claims Act violation. The compliance concern is whether actual documentation, queries, coding, and claims are truthful and supported, not whether the project has a budget.

Prioritize and authorize the test

Compare candidate projects on clinical risk, verified defect frequency, affected volume, equity concerns, feasibility, staff burden, and available resources. A low-volume issue with serious safety consequences may deserve attention before a higher-volume administrative inconvenience. Conversely, a large apparent financial opportunity with an unverified clinical premise needs investigation before implementation.

Present the sponsor with the evidence, uncertainty, scope, proposed measures, owner, and resources. Obtain the applicable local clinical, compliance, technology, and operational approvals for the change being tested. A charter makes the work accountable; it does not authorize a template to insert a diagnosis or bypass provider judgment. The next section explains how to learn from the test and decide whether to adapt, implement, or stop it.

Primary references

• 42 CFR 482.21, hospital QAPI requirements; project example above is locally designed.
• CMS Measures Management System risk-adjustment overview.
• IHI Model for Improvement, aims and measures.
• 2026 ACDIS/AHIMA compliant query practice brief.
• AHIMA Standards of Ethical Coding, 2016 version, truthful, supported reporting.
• DOJ False Claims Act overview, liability for false claims and related conduct.