Creatinine Units Lost in a New Interface: 340 Results Posted Unconverted
Northside Health Partners runs one EHR with a central laboratory interface. The core lab sends ORU^R01 results in mg/dL. In January, a newly acquired outpatient draw station was connected through a middleware engine; its analyzer reports creatinine in µmol/L. The mapping table was copied from the core lab build, and no unit value is populated because the receiving system treats OBX-6 as optional. Over nine days, 340 creatinine results from that station posted unconverted. A primary care physician treated a displayed value of 6.8 as mg/dL (it was 6.8 µmol/L, roughly 0.08 mg/dL) and referred the patient to nephrology; a second patient's 540 µmol/L appeared as 540 mg/dL and triggered a spurious dialysis consult. The lab director wants the station disconnected immediately. Operations objects because 60 patients per day depend on it.
1. Reconstruct the failure and identify which controls failed, distinguishing the technical defect from the process defect.
Read the worked answer
The technical defect is narrow: the middleware mapping was cloned from a build where the source analyzer used mg/dL, the destination does not require OBX-6, and no conversion or plausibility rule sat between the interface and the chart. The result is a unitless number displayed as if it were mg/dL.
The process failures are what let it persist for nine days:
- No interface-specific validation. A new source system was treated as a copy of an existing one rather than a new integration requiring its own test plan. Testing with the analyzer's default units would have caught it in minutes.
- Unit field treated as optional. Optional in the specification is not optional in practice; the build should have made OBX-6 mandatory or applied a source-specific conversion table.
- No result plausibility rules. A creatinine of 540 mg/dL is physiologically implausible and should have been flagged or held. The system had no range check on the receiving side.
- No post-go-live reconciliation. Nine days and 340 results passed without a lab or clinical reviewer comparing a sample of posted values to the source.
- Alerting silence. The spurious dialysis consult and the nephrology referral were both signals that did not route back to the interface owner.
So the defect is one missing conversion; the process defect is that a new source could go live without a source-specific unit test, plausibility checks, or reconciliation.
2. What are your first 72-hour actions, including whether and how to communicate with affected patients?
Read the worked answer
Sequence by immediacy of harm.
Hour 0-4: Stop the bleeding. Either disconnect the station or, preferably, put its results into a manual-review hold so the 60 patients per day are still collected and reported by the lab. Disconnecting entirely pushes 60 patients a day into repeat draws or delays, so holding results is the better operational tradeoff if the lab can report by phone or fax. Freeze the mapping table and preserve logs.
Hour 4-24: Identify all 340 affected results by source and time window, convert them correctly, and correct the charted values with an audit trail. Do not silently overwrite; use the EHR's result-correction workflow so clinicians see that a change occurred. Build a list of patients whose result crossed a clinical decision threshold after conversion, not just everyone with a result.
Day 1-3: Contact the ordering clinicians directly by phone for the patients with corrected values that change management, including the two known cases. For the remainder, send a documented communication and make corrected results visible.
Disclosure: This is a patient safety event, not a privacy breach. Follow the organization's safety-event and disclosure policy; for the two patients whose care was altered, an open disclosure conversation is appropriate. Report internally as a sentinel-type event and preserve the timeline for root-cause analysis. Notify risk management and, if state reporting thresholds are met, the appropriate agency.
3. Design the governance change that prevents the next new source system from repeating this.
Read the worked answer
The durable fix is to make unit handling a required, tested property of every interface rather than a build detail.
- Make OBX-6 mandatory at the receiving system and reject or hold messages without it. Configuring the field as optional transferred the risk to the clinician reading the screen.
- Maintain a source-system unit registry. Each analyzer and reference lab gets an entry with its native units; the middleware applies an explicit conversion table keyed to that source rather than inheriting another build's assumptions.
- Add receiving-side plausibility rules for high-risk analytes (creatinine, potassium, glucose, INR, hemoglobin) with hold-and-review thresholds, not just warnings.
- Require a source-specific interface test plan with test cases in non-default units, including a deliberately unitless message that must fail closed.
- Mandate two weeks of 100% reconciliation for any new source, then sampled reconciliation thereafter, with a named owner in lab and informatics.
- Route anomalies back automatically. The spurious dialysis consult should have generated a ticket to the interface owner; build a low-effort path for clinicians to flag implausible results.
Finally, add new-source interfaces to the change advisory board with a go-live checklist that cannot be waived by operations alone.