A patient arrives unconscious in the emergency department. The nurse pulls up the allergy list: it is blank. Does this patient have no allergies, or did nobody ever record them? That ambiguity, an empty field that could mean "none" or "unknown," is the kind of silent data defect that corrupts hospital dashboards and, occasionally, harms people.
This lesson shows you how to put numbers on that ambiguity. We will score four dimensions of clinical data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition → (completeness, timeliness, accuracy, and conformance) against real fields, and calculate the missing-data rates that leadership never sees on their polished screens.
Clinical data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition → is not one number. It is a scorecard. The four dimensions below are drawn from widely used frameworks, including the Kahn et al. harmonized data quality model that underpins many US research networks.
ICD-10 (International Classification of Diseases, 10th revision) is the standard diagnostic coding system used for billing and reporting in the US and across Europe. EHR means Electronic Health Record, the software system of record for a patient.
Allergy data has a special problem. An empty allergy field is ambiguous. Good EHR design forces a positive assertion: the clinician must click "No Known Allergies" (NKA) rather than leave the field blank.
So the completeness metric here is not "field populated." It is "field either lists an allergy OR carries an explicit NKA flag."
Worked calculation. Suppose your inpatient unit admitted 4,000 patients last month.
Completeness = 3,200 / 4,000 = 80%.
That 20% gap is not "20% of patients have no allergies." It is 20% of patients whose allergy status is *unknown to the system*, and therefore unknown to the next clinician and to any drug-interaction alert running in the background.
Set a threshold. Many quality programs target 95%+ documented allergy status. At 80%, this unit fails.
The discharge diagnosis drives billing, case-mix reporting, and public quality measures. Two things break here.
Conformance failure: a diagnosis stored as free text ("bad chest infection") instead of a valid ICD-10 code (J18.9, pneumonia, unspecified). Free text cannot be counted, trended, or risk-adjusted.
Completeness failure: the encounter is closed with no principal diagnosis coded at all, common when coding backlogs grow.
Here is a simple conformance check in SQLSQLSales Qualified Lead: a prospect the sales team has validated as ready for direct outreach and a proposal, having passed clear qualification criteria.View full definition →-style pseudocode you can hand to a data team:
SELECT
COUNT(*) AS total_discharges,
SUM(CASE WHEN dx_code REGEXP '^[A-TV-Z][0-9][0-9AB]'
THEN 1 ELSE 0 END) AS valid_icd10,
ROUND(100.0 * SUM(CASE WHEN dx_code REGEXP '^[A-TV-Z][0-9][0-9AB]'
THEN 1 ELSE 0 END) / COUNT(*), 1) AS conformance_pct
FROM discharges
WHERE discharge_month = '2026-01';If conformance_pct reads 91.4%, then nearly one in eleven discharges carries a diagnosis your dashboards silently drop or miscount.
Vitals (blood pressure, heart rate, respiratory rate, temperature, oxygen saturation) feed early-warning scores that flag deteriorating patients. Late data means late warnings.
Timeliness needs two timestamps: when the observation happened, and when it was entered. The gap is the charting lag.
Worked calculation. For 10,000 vitals sets on a ward:
Timeliness = 9,300 / 10,000 = 93%.
Now the accuracy layer. Among charted vitals, apply plausibility rules: reject a heart rate of 0, a temperature of 45 C, an oxygen saturation above 100%. If 120 of the 10,000 records fail plausibility, accuracy = (10,000, 120) / 10,000 = 98.8%.
Notice these dimensions multiply, not average. A vitals record is only trustworthy if it is present AND on time AND plausible. Chaining them: 93% timely of records that are 98.8% plausible gives a combined "usable" rate closer to 91.9%, below any single headline number.
Leadership wants one traffic light. Analysts need the components. Do both.
| Field | Dimension | Score | Threshold | Status |
|---|---|---|---|---|
| Allergy status | Completeness | 80% | 95% | Red |
| Discharge dx | Conformance | 91.4% | 98% | Red |
| Vitals | Timeliness | 93% | 90% | Green |
| Vitals | Accuracy | 98.8% | 99% | Amber |
The rule: never let a green timeliness score hide a red completeness score by blending them into a single average. Report by field and dimension, then flag the worst cell.
Here is the mechanism most executives miss. Dashboards usually compute rates over *recorded* data. If 20% of allergy statuses are blank, the "percentage of patients with penicillin allergy" is calculated over the 80% who have any record. The denominator is quietly wrong.
The same happens with mortality and readmission dashboards when discharge diagnoses are incomplete: unindexed encounters simply vanish from the numerator and denominator, biasing the rate in an unknown direction.
A blank field is not neutral. It is a data point with an unmeasured value, and treating it as absent (rather than unknown) is the most common analytical error in hospital reporting.
Knowledge check
1. Why does a blank allergy field represent a data quality problem rather than simply a piece of missing information?
2. A patient's discharge record lists a diagnosis code of 'Z99.XYZ,' which does not exist in the ICD-10 code system. Which data quality dimension does this most directly violate?
3. Vitals were charted, the value is clinically plausible, and the field is fully populated, but the entry was logged three hours after the reading was taken. Which dimension is compromised?
4. Select ALL correct answers about why forcing a clinician to click 'No Known Allergies' (NKA) improves data quality.
Select all the correct answers.
5. Select ALL correct answers about the four-dimension clinical data quality model.
Select all the correct answers.
There is no single global standard for "good" clinical data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition →, so treat all figures below as illustrative program targets, not regulatory mandates.
The practical lesson: thresholds are chosen, not discovered. Anchor them to clinical risk. A blank allergy field carries higher risk than a late-charted routine temperature, so it deserves a stricter threshold (95%) than vitals timeliness (90%).
1. Pick the field and its owner. Allergy status: pharmacy and nursing.
2. Define the "should be populated" population. All inpatient admissions, not outpatients where the field may not apply.
3. Choose one dimension per rule. Do not mix completeness and accuracy in one check.
4. Set a threshold tied to risk, and write down the as-of date.
5. Report the raw rate, the denominator, and the missing count, never just the percentage.
6. Trend monthly. A field drifting from 96% to 88% over a quarter signals a broken workflow, often a new form or a staffing change.
Attach a named owner to each cell in the scorecard. That is what turns measurement into improvement. A red cell with no owner is just a complaint.