The effects of disease verification and referral on the relationship between symptoms and diseases

Reference symptoms of individual diseases, which are usually gleaned from hospital-based observations, may not be pertinent to primary health care. Observed relationships between symptoms and diseases can be biased by consultation, disease verification, and referral patterns. These biases are shown to affect sensitivity, specificity, predictive values, the likelihood ratio, and the odds ratio. On the assumption that consultation and disease verification rates are positively influenced by symptoms, but not directly by the yet unknown disease status of the patient, the general practitioner will observe a higher sensitivity and a lower specificity than in an unselected population. A positive correlation between symptom and disease will decrease the likelihood ratio, while the predictive values and the odds ratio are not changed after consultation. Referral is influenced by both symptom status and disease verification status. In general, the disease verification and referral patterns of general practitioners have not been quantified. While the influence of referral cannot easily be corrected for, the likely magnitude and direction of selection bias can be evaluated using simple formulas. A possible way to make unbiased estimations from populations preselected by referral is to compute the predictive values and the odds ratio from the category of patients referred without pre-existing knowledge of the disease status.