The goal of this website is to hold the numbers that a brain just can't. These numbers are important: the strength to rule-in or rule-out disease of any "thing" that is done upon a patient should be known. Useful tests should be pursued and less useful investigations should not. This both saves time and may avoid unnecessary and potentially harmful tests. This database should reinforce that no test is perfect and whether a test is "good" always depends on how serious the disease and the pre-test probability.
Likelihood ratios will be continuously added to this site, but, to save time, focus is placed on investigations that are commonly used in clinical practice. Emphasis will also be placed on tests that are not often used in clinical practice but perhaps should be (see TST skin test for acute tuberculosis) and tests that are often used but maybe shouldn't be (see the absence of cardiac risk factors to rule-out MI).
To get started simply follow the links on the front-page to find the sought after test quickly. Once the investigation is found, select whether it is negative or positive, guess a pre-test probability and see what happens to your patient's probability. Suggestions for pre-test probabilities are also provided from the primary literature for selected investigations. If this is the case, the pre-test probability will become a link to this data. Information on the specific study or review from which the likelihood ratios are derived is also shown. Data is from an adult population unless otherwise noted. I recently wrote a tutorial with examples at the CanadiEM blog.
Details on Data Collection
The data here is biased. The entire literature was not culled, although meta-analyses, when found, were preferred over individual studies. Attempts were made to find the most recent studies and to include data from those that were strong. If a study did not report likelihood ratios specifically, these were calculated using the excel spreadsheet found here. If specificity or sensitivity was 100% in these studies (meaning one subgroup of the 2x2 table had zero patients), all cells had a value of 0.5 added to prevent a zero or infinite LR without a confidence interval. Confidence intervals are reported to reinforce the fact that an average value only represents a range of values that, in some cases, can vary over a very large range.
Please send suggestions for new LRs or comments on website design to firstname.lastname@example.org