Key concepts and methods to establish the reliability32 and validity33 34 of SPADE
| Concept | Method | Validity and reliability |
| Symptom-disease pair | Test an association that is clinically plausible, linking a presenting symptom (chief complaint) and specific disease.28 | Face validity and biological plausibility of the target symptom-disease dyad |
| Bidirectional analysis | Use both look-back and look-forward methods to assess the same symptom-disease association.21 25 | Convergent construct validity of the symptom-disease dyad |
| Baseline comparisons (observed to expected) | Compare event frequency or rate of return with baseline or expected level (look back—OR25; look forward—HR21) or matched control population.23 | Strength of measured association relative to internal or external control |
| Temporal profiles | Plot temporal profile or trend (look back—time before index event25; look forward—time after index event21). | Temporality and biological plausibility/gradient based on disease natural history |
| Positive control comparisons | Test a similar association that is clinically plausible (look back—linked symptom19 25; look forward—linked disease35). | Coherence of the symptom-disease dyad or alternative form reliability |
| Negative control comparisons | Test an association that is not clinically plausible (look back—unlinked symptom25; look forward—unlinked disease21 22). | Discriminant construct validity (specificity) of the symptom-disease dyad |
| Subgroup analyses | Test for clinically plausible subgroup associations (eg, dizziness linked to missed ischaemic but not haemorrhagic stroke; headache linked to both25). | Face validity and biological plausibility/gradient of the measured associations |
| Associated diagnostic process failures | Correlate specific outcomes with known process failures (eg, missed stroke linked to improper use of CT rather than MRI51). | Coherence of the identified associations and construct validity |
| Triangulation of findings | Use alternative methods (eg, chart review, surveys, root cause analyses) to confirm the diagnostic error association.26 | Convergent construct validity, coherence of the measured associations |
| Impact analysis | Monitor the impact of interventions designed to reduce error or harms on the measure (‘flattening the hump’). | Predictive (criterion) validity and measure responsiveness |
| Reproducibility of analytical results | Repeat the analysis in multiple data sets21–23 or using resampling methods (eg, bootstrapping or split-halves). | Consistency of the measured associations or resampling73 reliability |
| Reproducibility of SPADE approach | Repeat the approach across other analogous symptom-disease dyads (eg, chest pain-myocardial infarction,24 fever-meningitis/sepsis36). | Analogy34 of the approach to related problems |
HR, hazard ratio; OR, odds ratio; SPADE, Symptom-Disease Pair Analysis of Diagnostic Error.