Effect of different sampling techniques on odds ratio estimates using hospital-based cases and controls

Area of Science:

• Veterinary epidemiology
• Biostatistics
• Retrospective study design

Background:

• Hospital admission records are frequently used in veterinary research, but potential biases from sampling methods are often overlooked.
• Retrospective case-control studies rely heavily on accurate selection of control groups from a defined study base population.

Purpose of the Study:

• To evaluate the impact of different control selection techniques on odds ratio (OR) estimates in a veterinary retrospective case-control study.
• To identify sampling methods that minimize bias in disease risk factor analysis using hospital records.

Main Methods:

• Utilized University of California, Davis Veterinary Medical Teaching Hospital (VMTH) patient records for horses with Corynebacterium pseudotuberculosis abscesses.
• Derived a 'gold standard' OR and calculated sampling ratios (SRs) for various factors (admission type, age, breed, sex) under different sampling techniques (random, matched, diagnosis-based).
• Analyzed SR distributions and OR estimate biases using analysis of variance and covariance.

Main Results:

• Sampling ratios (SRs) deviating from 1.0 introduced bias into OR estimates compared to the 'gold standard'.
• Matched and diagnosis-based sampling ('colic', 'fractures') resulted in wider SR ranges and larger OR errors.
• Simple random and systematic sampling yielded SR distributions closest to 1.0 with smaller standard deviations, resulting in least biased OR estimates.

Conclusions:

• The choice of sampling technique significantly influences the accuracy of OR estimates in retrospective veterinary studies.
• Simple random and systematic sampling are recommended for control selection to minimize bias when using hospital admission records.
• Proper definition of the study base population and appropriate sampling design are critical for reliable epidemiological findings.