Sepsis Mortality Metric Disagreement Between Two Dashboards
A 400-bed hospital's quality department reports a risk-adjusted sepsis mortality index of 0.82 for the fiscal year, based on an administrative claims-derived model using principal diagnosis and present-on-admission flags. The same year, the clinical analytics team builds a sepsis dashboard using a sepsis-3 definition applied to vital signs, lactate, and culture results from the EHR, and reports an observed mortality of 19% against an expected 14%, an index of 1.36. The chief medical officer sees both numbers in the same board packet and asks why the hospital is simultaneously better and worse than expected. Neither team can explain the other's method. The board wants a single number for the quality report.
1. Explain the methodological reasons the two indices diverge and why neither is simply wrong.
Read the worked answer
The two indices answer different questions with different denominators.
Case definition differs. The claims model identifies sepsis from coded diagnoses, which depend on documentation and coding practice. The EHR model applies sepsis-3 criteria (suspected infection plus organ dysfunction, e.g., SOFA change or lactate) to clinical data. The EHR definition captures patients who meet clinical criteria but were never coded as sepsis, and it excludes coded sepsis cases that do not meet the clinical criteria. The populations are overlapping but not identical.
Present-on-admission handling differs. The claims model relies on POA flags to separate community-acquired from hospital-onset sepsis. The EHR model must infer onset from timestamps, which introduces its own judgment.
Risk adjustment differs. The claims model adjusts using coded comorbidities, which are themselves sensitive to documentation depth. The EHR model may adjust on a narrower variable set, so it may under-adjust for case mix and inflate the expected-to-observed gap.
Timing differs. Claims data lag; the EHR dashboard may be near-real-time and include cases still open.
So the 0.82 reflects coded sepsis performance against a coded benchmark; the 1.36 reflects clinically defined sepsis against a clinical benchmark. Both are internally valid; they are not comparable.
2. The board demands one number. Recommend how the HIM and quality leaders should respond, including what to report and what governance decision is required.
Read the worked answer
Do not average the two or pick the more flattering one. Averaging two non-comparable indices produces a number with no defined meaning and no external benchmark.
Recommended response:
- Report the primary measure with its definition attached. Select the measure that matches the use: if the purpose is external benchmarking and regulatory comparison, use the claims-based risk-adjusted measure, because it has a national comparison group. If the purpose is internal clinical improvement, use the EHR-based measure, because it is actionable at the point of care.
- Report the secondary measure as context, labeled clearly - e.g., "clinically defined sepsis mortality, internal monitoring only, not benchmarked."
- Reconcile the case lists. Have both teams produce patient-level lists and quantify the overlap. Document the coded-sepsis cases that fail clinical criteria and the clinical-sepsis cases never coded. That reconciliation is the real deliverable; it usually reveals documentation gaps that explain much of the gap.
- Escalate the definition decision to the data governance or quality committee as a formal decision with a named owner, recorded in the data dictionary, and reviewed annually.
The governance point: when two valid measures conflict, the organization needs a documented decision rule, not a compromise number. The board should approve the rule and the rationale, so the question does not recur each year.