AHRQ Quality Indicators: What Your PSI Report Actually Tells You

A PSI report can set a hospital in motion before the team knows whether the signal is real, reliable, or actionable. A postoperative sepsis rate moves in the wrong direction. The result appears in the monthly quality dashboard. It is repeated in an internal benchmarking report. Leaders need to know whether this is a real patient safety signal before they commit staff time, explain the trend, or begin corrective action.

That pressure is understandable, and it is exactly where PSI interpretation tends to go wrong. Patient Safety Indicators (PSIs) are a set of measures from the Agency for Healthcare Research and Quality (AHRQ) that flag potentially preventable adverse events in hospital records, one module within the broader family of AHRQ Quality Indicators, and the stakes reach well beyond a hospital’s quality department. PSI 90, the Patient Safety and Adverse Events Composite, is among the hospital quality measures with a direct connection to Medicare payment methodology, feeding the Centers for Medicare & Medicaid Services (CMS) Hospital-Acquired Condition (HAC) Reduction Program. Hospitals in the worst-performing quartile receive a 1 percent Medicare payment reduction. PSI data also feeds CMS’s Hospital Inpatient Quality Reporting Program, covered in more detail by American Data Network (ADN)’s Hospital IQR Program guide, and shapes external ratings such as Healthgrades’ Patient Safety Excellence Award and Leapfrog’s Hospital Safety Grade. Despite those stakes, PSI results were never meant to be definitive. The original AHRQ PSI work described PSIs as screening tools based on routinely collected administrative data rather than definitive proof of preventable harm.

Key Takeaways

• PSI 90 carries a real financial stake. Hospitals in the worst-performing quartile lose 1 percent of Medicare payments, and the same data shapes Healthgrades and Leapfrog ratings too.
• A PSI rate is not a verdict. The same result can reflect a genuine patient care gap, a risk adjustment effect, a data quality issue, or small-denominator instability.
• Most PSI misreads trace back to one of three causes: an unadjusted risk profile, flawed abstraction data, or a denominator too small to support a real conclusion.
• Reviewing the cases behind a rate, not just the number itself, is what separates a real safety signal from statistical noise.

Is This a Real Signal or Statistical Noise?

A PSI result can spike on a dashboard and still be statistical noise rather than a real signal. This is especially common when the denominator is small.

A hospital may see a sharp increase in a PSI rate after only two or three events. That change can look dramatic on a dashboard when it moves the hospital away from a peer benchmark. But a rate based on a small number of eligible cases does not carry the same meaning as one based on a larger denominator.

Confidence intervals help clarify this point. When the interval is wide, the estimate is less precise. When confidence intervals overlap across periods or peer groups, the apparent difference may not represent a reliable performance gap. AHRQ does not publish a single minimum case count that applies across all PSIs. Instead, reliability is assessed statistically, and lower-frequency indicators in particular can produce hospital-level rates too unstable to interpret with confidence at typical annual volumes. That is part of why a single-period spike deserves scrutiny before it becomes a corrective action plan.

For example, picture two hospitals that both post the same observed rate for a given PSI. Hospital A’s rate is built on 40,000 eligible discharges, so its confidence interval is narrow, tight enough to say with confidence that its performance sits above or below the benchmark. Hospital B’s identical rate is built on 3,000 eligible discharges, producing a much wider interval, one wide enough to overlap the benchmark entirely. The dashboard shows two hospitals performing the same. The underlying statistics say only one of them has a result precise enough to act on.

Consider a hospital that sees a quarterly increase in postoperative respiratory failure. The response is to schedule respiratory therapy education and prepare a summary for leadership. Before the intervention begins, the analyst reviews the cases behind the rate. The denominator is small. Two events drove most of the movement. One case involved a transfer patient with a high expected risk. Another had documentation that did not clearly support whether the respiratory condition was present on admission.

The hospital does not dismiss the finding. Instead, the team validates the cases, confirms denominator logic, and monitors the next reporting period before launching a corrective action plan. The case review may still identify an opportunity to improve documentation or escalation. But the hospital avoids treating an unstable measurement signal as confirmed evidence of a systemwide care failure.

A PSI rate becomes more credible when it aligns with related signals. A postoperative respiratory failure rate that rises alongside lower HCAHPS (the Hospital Consumer Assessment of Healthcare Providers and Systems) communication scores and a climbing 30-day readmission rate for respiratory diagnoses points to something real. A postoperative respiratory failure rate that rises in isolation, with no corresponding movement in related measures, is more likely to be noise.

ADN’s Data Analytics Services can support this kind of review by helping hospitals identify patterns, trends, and priorities across clinical, quality, financial, and patient safety data. One unstable PSI result may be noise, whereas the same pattern across multiple periods or related data sources may indicate a real patient care gap.

How Can PSI Risk Adjustment Change the Story?

Risk adjustment is meant to make PSI results more comparable across hospitals, but results can still be misread when the model is treated as a black box. A higher PSI rate does not automatically mean a hospital has a worse safety problem. A lower rate does not automatically mean performance is strong.

The AHRQ PSI risk adjustment model accounts for age, sex, Major Diagnostic Categories, MS-DRG (Medicare Severity Diagnosis Related Group) categories, Elixhauser comorbidities present on admission, and patient origin, including whether the patient was transferred from another facility. Those factors shape the expected rate against which observed performance is compared.

That matters operationally. A hospital that receives more high-acuity transfers may have a different expected rate than a hospital with fewer complex cases. A hospital with consistently captured comorbidities may also look different from one where patient complexity is underdocumented. Before acting on the rate, quality teams should ask whether transfer volume changed, whether case mix shifted, and whether comorbidities were captured consistently. One practical check: pull the hospital’s case mix index (CMI) trend for the same reporting period. If CMI rose alongside the PSI rate, some or all of the increase may reflect a genuine shift toward more complex patients rather than a decline in care quality. If CMI held steady while the PSI rate moved, risk adjustment is less likely to explain the change, and the signal deserves closer attention.

Comparative context matters here. ADN’s Clinical Benchmarking System can help hospitals compare PSI performance against peer and national benchmarks rather than relying on the rate in isolation.

Is the Right Data Entering the PSI Calculation?

Even when risk adjustment is understood, the PSI result can still be misleading if the data entering the calculation is incomplete, inconsistent, or based on an outdated specification. PSI reports rely on administrative and coded data. The reliability of the rate depends on how accurately diagnoses, exclusions, present-on-admission status, and comorbidities were captured before the report was generated.

A missed comorbidity can make a patient population appear less complex than it actually is. That can alter expected performance and make outcomes look worse than the underlying care supports. An incorrect present-on-admission (POA) indicator can also shift interpretation. If POA status is wrong, a case may be counted as a PSI event when it should not be, or excluded when it should have been reviewed. A practical way to catch this: each quarter, pull the five to ten highest-impact cases behind a PSI component and trace each diagnosis and POA flag back to the original chart documentation. This targeted chart-to-code reconciliation catches the errors most likely to move a rate, without requiring a full audit of every discharge.

The International Classification of Diseases, 10th Revision (ICD-10) diagnosis codes also matter. They help determine which cases are included, excluded, or counted as PSI events. Because AHRQ updates PSI specifications each year, teams need to make sure they are using the current version. The scope of these updates is not trivial. AHRQ’s v2025 PSI update included multiple revisions to code lists, software documentation, technical specifications, and calculation logic. A report built on last year’s specification may be based on rules that no longer match the latest methodology.

This is where PSI interpretation connects to the quality of abstraction. A PSI report may look like a downstream analytics output, but its reliability is shaped when clinical information is interpreted from the record and converted into reportable data.

ADN’s Clinical Data Abstraction Services apply Inter-Rater Reliability (IRR) methodology to sustain a 98.4% abstraction accuracy rate before data ever reaches a PSI calculation, backed by ADN’s standing as a CMS-approved quality measures vendor and an AHRQ-listed Patient Safety Organization since 2009. ADN’s Medical Chart Abstraction guide covers what disciplined abstraction requires in more detail.

How Should Quality Teams Review PSI Results Before Acting?

A PSI result should be reviewed before it is treated as a confirmed performance problem. The first step is to look behind the rate and confirm that the cases were counted correctly. That means checking whether the numerator cases truly meet the current specification, whether the right cases were included in the denominator, and whether exclusions were applied appropriately. In practice, this means pulling the discharge-level detail behind the rate, not just the summary number, and checking a sample of flagged cases against the current year’s specification rather than assuming last year’s logic still applies.

The next step is to review the data that informed the result. POA status should be supported by the record. Comorbidities should be captured consistently. Diagnosis coding should align with the current AHRQ specification. If those inputs are uncertain, the PSI rate may indicate a data quality issue rather than a clinical care gap.

Only after that review should the team decide what response is justified. A one-period change with a small denominator may warrant validation and monitoring. A stable pattern that repeats over time and is supported by case review may justify escalation to a focused performance improvement effort. The goal is to match the response to the strength of the signal.

Using AHRQ Patient Safety Indicators to Find Reliable Safety Signals

AHRQ Patient Safety Indicators can help hospitals identify potential patient care gaps, but a PSI report does not speak for itself. When teams validate cases, review inputs, compare results in context, and assess reliability before acting, PSI data becomes more useful for deciding which signals need action and which need further review.

Reliable PSI interpretation starts with the data underneath it. ADN’s Clinical Data Abstraction Services help ensure that diagnosis coding, present-on-admission status, and comorbidity capture reflect the current AHRQ specification before a PSI rate is ever calculated. From there, ADN’s Clinical Benchmarking System puts that rate in peer and national context, and ADN’s Data Analytics Services help quality teams track whether a signal repeats across periods or data sources before it becomes a corrective action plan.