Risk stratification: scores, pretest probability, and disposition

A good score doesn’t make the decision for you — it turns scattered findings into one probability, and lets a threshold do the deciding.

The idea

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.

How it works

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.

When to use it

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.

Watch out for

• Applying a score outside its lane. HEART is for undifferentiated chest pain in the ED, not for a known STEMI or a post-op patient. Wells is for suspected PE/DVT, not chest pain. The right score depends on the exact question.
• Trusting the number over the patient. A low score in someone who looks genuinely ill, or with a dynamic ECG, doesn’t buy a discharge. Gestalt and serial data override a single snapshot.
• Forgetting the troponin is a trajectory. A single normal troponin very early can be falsely reassuring. Most pathways require serial or high-sensitivity protocols before the “low-risk” label sticks.
• Discharging without the safety net. “Low risk” is not “no risk.” The disposition is discharge with return precautions and timely follow-up — the safety-netting is part of the plan, not an afterthought.
• Double-counting or mis-scoring an element. Age, risk factors, and history each count once. Inflating “highly suspicious history” on a hunch quietly moves the needle without new evidence.

Worked example

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.