Foundational Knowledge of Analytics in Healthcare
The analytics ladder
Healthcare analytics is usually taught as a maturity ladder. Descriptive analytics reports what happened (census counts, readmission rates, cost per case). Diagnostic analytics explains why (drill-downs, stratification, process mapping). Predictive analytics estimates what is likely to happen (risk scores, no-show models). Prescriptive analytics recommends an action (care-gap outreach lists, staffing optimization). Questions often swap predictive and prescriptive - if the output tells someone what to do, it is prescriptive.
Data in healthcare
Know the difference between structured data (claims fields, lab values, coded diagnoses) and unstructured data (notes, images). Know the common sources: EHR, claims and remittance, registries, health information exchanges, surveys, device and wearable feeds, and social-determinant datasets. Claims data covers a broad population but lacks clinical detail; EHR data is rich but incomplete and inconsistently coded.
Measurement basics
Be able to distinguish prevalence (existing cases at a point in time) from incidence (new cases over a period), and sensitivity (of true cases, the share detected) from positive predictive value (of positive results, the share truly diseased). PPV moves with prevalence; sensitivity and specificity do not. Quality measures are usually proportions with a carefully specified denominator (eligible population minus exclusions) and numerator.
Trap and worked example
A screening tool with 95% sensitivity and 95% specificity is used where prevalence is 1%. Of 10,000 people, 100 have disease; 95 are detected, and 5% of the 9,900 disease-free people (495) falsely test positive. PPV = 95/590 ≈ 16%. Rule of thumb: when prevalence is low, even a good test produces mostly false positives - always ask what the base rate is.