Data and Statistics
Number Needed to Treat and Harm Pairing
A Cardily data-literacy explainer on number Needed to Treat and Harm Pairing, focused on interpretation rather than invented precision.
What the numbers are trying to measure
Number needed to treat estimates how many people need an intervention for one additional benefit compared with control over the trial’s follow-up. It derives directly from absolute risk reduction, not relative risk reduction alone.
Number needed to harm is the reciprocal of absolute risk increase for a specified adverse outcome-myopathy, bleeding, hypotension-again tied to duration.
Confidence intervals for NNT are wide when event rates are low; point NNT alone overstates precision.
How to interpret them without being misled
A statin NNT of 50 over five years for MACE prevention differs from an SGLT2 NNT for HF hospitalization over two years-they are not interchangeable rankings of “better drug.”
When trials report composite endpoints, NNT prevents one composite event, which may not equal preventing one death.
Express paired NNT/NNH to the same population: trial entrants, not general population averages.
Common statistical traps
Calculating NNT from relative risk reduction without control event rate-invalid.
Using median follow-up when events accrue non-uniformly without person-time alignment.
Ignoring competing mortality that removes patients from harm denominators.
What the data imply for learning and policy literacy
Shared decision aids use NNT frames; policy models multiply prevented events by eligible population from NHANES prevalence estimates.
AHA summaries of therapeutic advances sometimes cite illustrative NNTs-trace back to source trials.
Equity: if only affluent patients receive therapy, realized population NNT is worse than trial NNT.
Responsible takeaways
Always state event type, horizon, and confidence interval when citing NNT/NNH.
Teach reciprocals as pedagogical tools, not mystical integers.
Not treatment recommendations.
Putting it together
NNT and NNH pairing is core Cardily statistics ethics: absolute, paired, time-stamped.
Links forward to aspirin, GLP-1, SGLT2, and ARNI articles with consistent framing.
Educational literacy for students and public readers.
Educational disclaimer
This article is part of Cardily’s educational Insights library. Cardily content is for educational and informational purposes only and does not constitute medical advice, diagnosis, or treatment. Always consult a qualified healthcare professional for personal medical concerns. In an emergency, call local emergency services.