Do we need a Bayes Collaboration? Proposal for a Diagnostic Database

Pewsner D., Bleuer JP., Jüni P., Battaglia M., Bucher H., Egger M. Swiss Academy of Medical Sciences DOKDI, Bern, Switzerland

The aim of this project is to develop a comprehensive database ("Bayes Library") to provide rapid access to data on frequently used diagnostic and screening tests, including signs and symptoms, laboratory tests, x-rays etc. Relevant data on pretest probabilities, test characteristics (sensitivity, specificity, likelihood ratios) and study designs will be identified in comprehensive literature searches, critically appraised, and presented in a standardised format to inform the choice of tests and interpretation of test results. Thomas Bayes (1702-1761, see picture) showed that the post-test probability of a single or a series of tests can be calculated using pretest probability and likelihood ratios, allowing clinicians and consumers to weigh complex information according to its diagnostic importance. The Box presents an example for suspected appendicitis. 1. Pre-test probability: In ambulatory care 0.7-1.6% of adults with abdominal pain have appendicitis. 2. Selection of studies and critical appraisal: Characteristics of selected studies 1 2 are presented in Table1 Study Study Type Size Prev SB Description of patients Description of test RB VB Wagner et al1 MA of CS 5275 - considerable given: heterogeneous given unclear small Andersson et al2 CS 502 39% considerable given: homogeneous given unclear small Table 1. 1Wagner et al (JAMA 1996;276:1589) and 2Andersson et al (World J Surg 1999;23:133). MA: Meta-analysis; CS: Cohort study; Size: Number of patients included; Prev: Prevalence; SB: Spectrum bias; RB: Review bias; VB: Verification bias. 3. Characteristics of diagnostic tests: Likelihood ratios for diagnostic tests are provided in Tables 2 and 3 Signs and symptoms LR+ [ 95% CI] LR- [95% CI] Right lower quadrant pain* 7.31-8.46 0-0.28 Rigidity 3.76 [2.96-4.78] 0.82 [0.79-0.85] Migration of pain to right lower quadrant 3.18 [2.41-4.21] 0.50 [0.42-0.59] Pain before vomiting 2.76 [1.94-4.94] - Psoas sign 2.38 [1.21-4.67] 0.90 [0.83-0.98] Rebound tenderness test* 1.10-6.30 0-0.86 Guarding* 1.65-1.78 0-0.54 No similar pain previously 1.50 [1.36-1.66] 0.32 [0.25-0.42] Rectal tenderness* 0.83-5.34 0.36-1.15 Nausea* 0.69-1.20 0.70-0.84 Vomiting 0.92 [0.82-1.04] 1.12 [0.95-1.33] Table 2. Data from 1Wagner et al (JAMA 1996;276:1589). *Heterogeneous studies. LR+: Positive likelihood ratio; LR-: Negative likelihood ratio. Leukocytes (x10-9/L) LR [95% CI] < 8 0.16 [0.10-0.26] 8 to = 10 0.83 [0.53-1.28] 10 to = 12 1.12 [0.75-1.65] 12 to = 15 2.44 [1.63-3.65] > 15 7.03 [4.11-12.1] Rectal Temp (°C) LR [95% CI] < 37.7 0.44 [0.34-0.55] 37.7 - 37.9 1.61 [1.01-2.57] 38.0 - 38.4 1.77 [1.24-2.51] = 38.5 3.01 [1.99-4.58] Table 3. Data from 2Andersson et al (World J Surg 1999;23:133). 4. Example: Thomas, an 18-year old student suffers from moderate (1) right lower abdominal pain with (2) no similar previous episodes. The pain was initially generalised, but has now (3) migrated to the right lower quadrant. Examination shows (4) moderate guarding, and (5) the leukocyte count is 13 x 10-9/L. The combined power of this sequence of tests (1-5) is calculated by multiplying their LRs: 7.3 x 1.5 x 3.2 x 1.6 x 2.4 = 135. Assuming a pretest probability of 1%, a post-test probability of 60% is obtained from a Fagan nomogram. Given potential biases, 40%-50% is a more realistic estimate of the post-test probability of appendicitis Preparing, maintaining and promoting the accessibility of critically appraised diagnostic information will require an international collaborative effort, similar to the Cochrane Collaboration. The proposed Bayes Collaboration and Bayes Library has considerable potential in promoting evidence-based health care.

Cape Town O37