6.3 Executing, Piloting & Measuring CDI Project Outcomes

Execute a CDI project as a testable change

A charter states the problem and intended result. Execution turns the proposed solution into a process that people can try, measure, and improve. Keep the clinical purpose visible: more accurate, supported documentation with a workable process. A larger query count or a favorable payment shift cannot substitute for that result.

Choose a method appropriate to the project and the organization's improvement program. PDSA and DMAIC are useful structures, not competing national CDI requirements. A small test can reveal a broken notification step quickly; a complex recurring process problem may require more extensive measurement and analysis. Neither a fixed 90-day pilot nor a particular baseline duration guarantees a valid conclusion.

Apply PDSA to a specific uncertainty

IHI's Model for Improvement connects the aim, the evidence of improvement, and the proposed change with iterative Plan-Do-Study-Act testing. Each cycle should answer a practical question. Early tests can be small, followed by revisions and tests under different conditions before implementation.

Continue the illustrative nutrition-assessment project from section 6.2. The first uncertainty is whether a new notification reaches the responsible provider without creating duplicate requests.

Meeting one prediction does not require immediate hospital-wide rollout. Conversely, an unsuccessful test still provides useful information. Record what was learned and what the next cycle will test. Repeated testing under varied conditions helps reveal dependencies that a receptive volunteer group may not encounter.

Use DMAIC for a recurring process problem

ASQ describes DMAIC as Define, Measure, Analyze, Improve, and Control. In a CDI workflow, Define establishes the scope and users' needs; Measure establishes reliable baseline observations; Analyze investigates causes; Improve tests and refines changes; and Control establishes ownership, monitoring, and a response plan to sustain the improved process.

For repeated misrouted queries, map the route from reviewer to responsible provider, including coverage changes and handoffs. A cause-and-effect diagram can organize possible contributors such as unclear assignment rules, missing roster updates, and interface failures. These are hypotheses to investigate. Interview the people performing the steps and check actual routing records before attributing every delay to provider resistance.

A Pareto chart ranks observed categories by frequency or another stated quantity. For example, 30 reviewed routing failures might include 15 outdated rosters, nine ambiguous service assignments, and six interface errors. The first category represents 50% of the failures, not an assumed 80%. A frequently discussed “80/20” pattern is not a result to impose on every dataset. Define the categories, handle multiple causes consistently, and use the actual counts when prioritizing a test.

Match the intervention to the cause

Use education when the verified gap concerns knowledge or interpretation. Use a workflow change when the problem is that an assessment cannot be found or a question reaches the wrong person. A new educational presentation will not repair an interface that drops messages. An EHR prompt should make relevant information and the clinical question easier to address; it should not preselect a diagnosis or copy an unsupported condition into each encounter.

Prepare the users for the test. Identify who owns the updated procedure, how covering clinicians receive questions, how staff report defects, and who can pause the change. Clinical, coding, compliance, and technology reviewers should examine the parts within their responsibilities. Query wording still needs the individual record's relevant facts and reasonable response options. A diagnosis-specific template cannot make every patient meet that diagnosis.

Measure the result, the work, and unintended effects

IHI distinguishes outcome, process, and balancing measures. Apply all three to the aim rather than equating a busy program with an effective one:

For query response, define whether the denominator is all issued queries or those whose response window has elapsed, and how closed, withdrawn, and duplicate queries are handled. Report unresolved cases rather than silently excluding them. Agreement, response speed, and coverage can be useful process information; high values do not independently establish nonleading wording or clinical support.

Some teams call an early operational measure a leading indicator and a later result a lagging indicator. Availability depends on the data and workflow. A documentation audit may be performed during the pilot; a payer denial outcome may require extended follow-up. Do not postpone all quality assessment until discharge or assume that every outcome arrives months later. Use timely review to identify unsupported diagnoses or misleading templates while the test is still running.

Distinguish a run chart from a control chart

A run chart places observations in time order. Label the measure and interval, keep definitions consistent, and annotate intervention dates and relevant external events. IHI's run-chart instructions include a median line with at least ten observations. A median provides a reference for interpreting patterns; it is not an upper or lower control limit.

A statistical control chart adds limits calculated using a method appropriate to the data and chart type. NIST describes three-sigma limits as common, with interpretation depending on distribution and estimation assumptions. A count, proportion with changing denominators, and continuous elapsed time do not all call for the same calculation. Work with qualified quality or analytic staff to select the method and interpretation rules before using it for decisions.

A point or pattern signaling unusual variation warrants investigation. It does not prove that the CDI intervention caused the change. Changes in case mix, audit practice, staffing, or data extraction may also matter. A pair of before-and-after averages can conceal substantial variation; show the underlying sequence and denominators. Avoid labeling every favorable fluctuation “statistically significant” without an appropriate analysis.

Decide whether to adapt, implement, or stop

Suppose the pilot's query-agreement rate is 94%, but an audit finds a template that steers clinicians toward one diagnosis despite weak or contradictory evidence. Pause or restrict the defective process as appropriate, examine affected records under the organization's correction procedures, revise the template and training, and retest. Do not preserve the template because its agreement rate exceeds a target. A provider's agreement does not cure a leading query, and a denial increase alone does not prove which defect caused it.

For a promising pilot, compare the outcome, process, and balancing findings with the aim and predictions. Check unresolved cases, staff feedback, and whether clinical and coding support remain sound. A small, uncontrolled pilot can establish feasibility and guide a next test; it does not automatically establish causation or predict results in every service line. Test the workflow with different coverage patterns, users, and conditions before broader implementation.

Sustainment needs an accountable owner, an approved procedure, maintained training and technology, a defined monitoring schedule, and an action when performance deteriorates. Continue checking documentation accuracy, not just whether a preferred phrase appears without a query. New staff, guidance, code versions, or service changes may require revision and another test. Where the project forms part of hospital QAPI, preserve the reasons, measures, progress, and follow-up evidence required by the program.

Primary references

• IHI Model for Improvement, testing changes, and establishing measures.
• ASQ DMAIC.
• IHI Run Chart instructions and NIST control-chart overview.
• 42 CFR 482.21, QAPI monitoring and project documentation.
• 2026 ACDIS/AHIMA compliant query practice brief.