Diagnostics › Diagnostic accuracy

Interval likelihood ratios

Stratum-specific likelihood ratios report the LR for each bin of a predictor, with a 95% Wald CI on log(LR).

What is Interval likelihood ratios?

A single LR+ / LR− pair assumes the test is treated as binary at one cutoff. When the predictor is continuous and you bin it (e.g. quintiles), each bin has its own LR — the multi-level test framing of Sackett.

An LR profile that climbs cleanly from low to high bins is the ideal: the test gives graded information across the range, not just a single threshold-positive / negative.

When should I use Interval likelihood ratios?

  • Multi-level (not binary) diagnostic test reporting.
  • Showing that a test gives graded information across its range.

What data does it need?

Continuous predictor + binary outcome + optional bin breaks (default quintiles).

What does it report?

Per-bin LR + 95% CI on log(LR) + underlying positive / negative counts.

How do I interpret the result?

Convention: LR > 10 strongly raises post-test probability; LR < 0.1 strongly lowers it; LR ≈ 1 is uninformative.

See also

References

  • Sackett et al. (1985). Clinical Epidemiology, 2nd ed.