Respiratory disease detection using clinical category information

Page 75 of 201 populations at different frequencies. For example, it may be possible that respiratory disease occurrence might be different in one specific group of animals compared to another. If the prevalence of a disease is different in different populations then applying the same diagnostic test to those different populations may produce different predictive values because predictive value estimates are influenced by the prevalence. For respiratory disease, gross necropsy has a positive predictive value PPV of 0.9 and a negative predictive value NPV of 0.85. This means that if a gross necropsy is performed during a voyage and the gross necropsy diagnosis is respiratory disease, then that animal has a 90 probability of truly having respiratory disease as the cause of death positive predictive value. Conversely if a gross necropsy is performed and the gross necropsy diagnosis is not respiratory disease, then that animal has an 85 probability of having died from a cause other than respiratory disease negative predictive value. Having information on diagnostic test performance and gold standard test outcomes also provides a more detailed understanding of prevalence estimates. The prevalence of respiratory disease as a cause of death is the probability that a mortality case died of respiratory disease. The diagnostic test result gross necropsy classified animals as test positive or test negative and using these numbers we can generate a prevalence estimate of respiratory disease: 100215 = 47. The term apparent prevalence is used to refer to the prevalence estimate derived from the diagnostic test gross necropsy because it is really a prevalence of a positive test result and not the prevalence of true disease since the diagnostic test is not perfect. If a gold standard test result is available then these results provide an estimate of the true prevalence, based on the final cause of death results.

7.10.2 Respiratory disease detection using clinical category information

Table 16: Summary of classification of 215 cases of mortality by gold standard test final cause of death and the clinical category information outcome, using respiratory disease as the outcome of interest. Respiratory disease Gold standard test Final cause of death Disease + Disease - Clinical category Test + 9 6 15 Test - 98 102 200 107 108 215 Page 76 of 201 Table 17: Statistical measures of diagnostic performance for clinical category information as a test for detecting respiratory disease as a cause of death in cattle. Based on data in Table 16. Se= sensitivity; Sp=specificity; PPV = positive predictive value; NPV = negative predictive value; App Prev = apparent prevalence; True Prev = true prevalence; CI = Confidence Interval. Parameter Estimate 95 CI Lower Upper Se 0.08 0.04 0.15 Sp 0.94 0.88 0.97 PPV 0.6 0.36 0.8 NPV 0.51 0.44 0.58 App Prev 0.07 0.04 0.11 True Prev 0.5 0.43 0.56 The results clearly show that clinical category information alone is not useful for gaining any reasonable understanding of the extent of respiratory disease as a cause of death on export voyages. The sensitivity in particular is very low indicating that clinical category information is particularly poor at detecting those animals that truly have died of respiratory disease. The predictive values appear to be little better than tossing a coin to determine the probability of a disease outcome given the findings of the clinical category information. The apparent prevalence estimate proportion of all deaths that are due to respiratory disease when based on clinical category information is seriously inaccurate and an under- representation of the true prevalence. The results confirm that clinical category information is insufficient to provide a reasonable understanding of the contribution of respiratory disease to mortality on export voyages and that gross necropsy diagnosis is essential to achieve a good understanding of the contribution of respiratory disease to voyage mortalities.

7.10.3 Musculoskeletal injury detection using gross necropsy