A good score doesn’t make the decision for you — it turns scattered findings into one probability, and lets a threshold do the deciding.
Validated scores like HEART (chest pain), Wells (PE/DVT), and CHA₂DS₂-VASc (stroke risk in atrial fibrillation) all do the same thing: they collapse a handful of findings into a single pretest probability of a bad outcome.
That probability isn’t the endpoint. You compare it to two thresholds — a test threshold and a treatment threshold — and where it lands decides disposition: discharge with safety-netting, keep and test, or admit. The score builder below uses HEART. Watch how one finding — an ischemic ECG, a rising troponin — can push the needle across a line and change the whole plan.
HEART score builder
Teaching model. MACE = major adverse cardiac event; band percentages are 6-week rates from HEART validation and are illustrative. Scores support, never replace, clinical judgment, serial troponins, local pathways, and shared decision-making.
HEART scores five elements 0–2 each, sums them, and reads the total against two thresholds. The total is a proxy for pretest probability; the thresholds convert probability into a plan.
History slightly / moderately / highly suspicious 0 / 1 / 2
ECG normal / non-specific / ST deviation 0 / 1 / 2
Age <45 / 45-64 / ≥65 0 / 1 / 2
Risk factors none / 1-2 / ≥3 or known disease 0 / 1 / 2
Troponin normal / 1-3x / >3x upper limit 0 / 1 / 2
total = 0 to 10
0-3 low 6-wk MACE ~1.7% -> discharge, safety-net, follow-up
4-6 moderate 6-wk MACE ~16.6% -> observe, serial troponin, test
7-10 high 6-wk MACE ~50% -> admit, early invasive strategy
Example: History 1 + ECG 0 + Age 1 + RF 1 + Trop 0 = 3 (low)
Add one finding -- ST deviation (ECG 0 -> 2): = 5 (moderate)
The disposition just flipped on a single element.
The reason the discharge threshold sits at such a low probability (~2%, not 50%) is that the cost of a miss is catastrophic. When missing the diagnosis can kill, the test threshold drops close to zero — you accept testing many people who turn out fine to avoid sending home the one who isn’t.
| A validated score fits when… | The trade-off / limit |
|---|---|
| The question is a specific, common outcome the score was built and validated for. | Using it outside its derivation population (wrong complaint, wrong setting) breaks the calibration. |
| You want a shared, defensible language for pretest probability across the team. | A number can create false precision — it’s a floor for judgment, not a ceiling. |
| Disposition genuinely turns on crossing a threshold. | Scores don’t capture the gestalt “this patient looks unwell” — that can override a reassuring number. |
An interviewer says: “A 58-year-old with atypical chest pain, one normal troponin, an unremarkable ECG, well-controlled hypertension. What’s your disposition?” A strong answer builds the score out loud: moderately suspicious history (1), normal ECG (0), age 45–64 (1), one or two risk factors (1), normal troponin (0) — total 3, low risk. “That maps to discharge with a serial-troponin pathway per our protocol, clear return precautions, and outpatient follow-up.” Then the sharp move: “But if the repeat troponin rises above the limit, that single element takes the score to 4–5 — moderate — and I’d switch to observation and further testing.” Naming the threshold, and the one finding that would cross it, is what shows real reasoning.
Check yourself
A patient scores 2 (low) but describes crushing pain radiating to the jaw and looks diaphoretic and unwell. What’s the disciplined move?
Why does the discharge threshold sit near ~2% probability rather than, say, 20%?