The point of departure for the doctor’s estimate of the pretest probability of disease is its prevalence in the population. This probability is then readjusted using relevant data such as the setting of the encounter, the patient’s age and gender, and the presence of risk indicators for disease , of symptoms , and of the findings. For example, the LR positive of ischemic heart disease (gold standard: coronary angiography) is 5.8 for typical angina, 1.2 for atypical angina, and 0.1 for non-anginal chest pain; 3.8 for prior myocardial infarction; and 2.3 for diabetes [10].

Doctors vary in their subjective assessments of the pretest probability of disease [11], and appear to need guidance in probability estimation [12]. To provide such guidance and thereby reduce the variability of doctor assessments of disease probability, an effort has been made to develop “clinical prediction rules” that combine findings derived from the patient’s history, physical examination, and other easily available tests to provide an assessment of the probability of disease and the need for further testing.

The objective of clinical prediction rules is to help physicians identify patients who require either further diagnostic testing or treatment. These rules are derived from systematic clinical observations. Some clinical prediction rules assign relative weights to specific clinical data found to be associated with specific outcomes. Other prediction rules consist of algorithms or more complex mathematical models [13]. However, the derivation of a prediction rule is just the first step [14]. Similar to any diagnostic test, prediction rules require prospective validation that the predictions are indeed consistent with observed outcomes; that they are applicable to patients in specific clinical settings; and that their use improves a doctor’s decisions [15]. Similar to other diagnostic tests , the value of prediction rules is affected by differences in disease prevalence in different settings, and the failure to recognize and adjust for these differences may either thwart decision-making or lead to the premature dismissal of valid rules [16].

The number of published studies on prediction rules suggests that their popularity has been increasing since the 1990s [14]. For example, clinical prediction rules have provided three-level scores that distinguished between patients with a low, intermediate, and high probability of pulmonary emboli [17] and bacteremia [18]. Other clinical prediction rules have attempted to estimate the probability of streptococcal tonsillitis in adults with a sore throat. Its prevalence at a primary-care hospital clinic in New York was 24 %; it increased to 83 % if associated with all of the following: tonsillar exudates, cervical lymphadenitis, exposure to streptococcal contact, fever above 36.1 °C, and no history of cough [19]. Therefore, although a 2012 systematic review of the literature concluded that symptoms and signs, either individually or combined into prediction rules, cannot be used to definitively diagnose or rule out streptococcal pharyngitis [20], they do have a value in assessing its probability.