Data Analytics and Informatics: From Data Quality to Prediction
From data to knowledge
The DIKW hierarchy runs data (raw) to information (organized) to knowledge (interpreted) to wisdom (applied). Data quality dimensions are accuracy, completeness, consistency, timeliness, validity, and uniqueness. A data dictionary defines fields; a data model shows relationships; a data warehouse stores integrated, subject-oriented data; a registry is a focused collection for a defined population (for example, a cancer registry).
Analytics types
Descriptive analytics answers what happened; diagnostic answers why; predictive answers what will happen; prescriptive answers what should be done. Dashboards should match the decision being made and avoid clutter. Benchmarking compares performance to internal or external standards.
Statistics to know
Sensitivity = TP/(TP+FN); specificity = TN/(TN+FP); positive predictive value = TP/(TP+FP). Incidence counts new cases over a period; prevalence counts all existing cases at a point in time. Rates should be risk- or age-adjusted when comparing populations.
Interoperability and advanced tools
HL7 v2 messaging, CDA documents, and FHIR APIs are the main standards; HIE models are centralized, federated, and hybrid. Natural language processing extracts concepts from unstructured text. Secondary use of data and AI require governance: bias, validity, transparency, and privacy.
Trap and worked example
A common trap is confusing sensitivity with PPV, or a data warehouse with a registry. Worked example: 1,000 people are screened; 100 have the disease. The test is positive in 90 of the 100 and in 90 of the 900 without disease. Sensitivity = 90/100 = 90%. PPV = 90/180 = 50%. A high-sensitivity test in a low-prevalence population can still have poor PPV.