OBJECTIVE: For psychiatric diagnoses, solving the problem of false positives is thought to be a matter of tightening diagnostic criteria. But low prevalence illnesses by their nature have high false positive rates. A recent study of bipolar disorder found the predictive value of bipolar diagnoses to be <50%. Is it possible to achieve much higher diagnostic accuracy for psychiatric diagnoses? METHOD: We calculate predictive values while varying diagnostic sensitivity and holding specificity constant, and vice versa, for a given prevalence of illness. We then calculate predictive values while holding sensitivity and specificity constant, but varying prior probability (clinically feasible by assessing other factors associated with bipolar outcomes, such as family history and degree of recurrence). RESULTS: Assuming a sample in which the prevalence of illness is 10%, achieving positive predictive values (PPV) >50% requires diagnostic specificity of >95%. Holding specificity at a level already achieved clinically (86%), increasing prior probability yields predictive values as high as 83%. CONCLUSION: Systematic assessment of clinical factors that increase the prior probability of illness, before applying DSM/ICD criteria, could raise PPV substantially compared with targeting greater specificity via more stringent diagnostic criteria.
OBJECTIVE: For psychiatric diagnoses, solving the problem of false positives is thought to be a matter of tightening diagnostic criteria. But low prevalence illnesses by their nature have high false positive rates. A recent study of bipolar disorder found the predictive value of bipolar diagnoses to be <50%. Is it possible to achieve much higher diagnostic accuracy for psychiatric diagnoses? METHOD: We calculate predictive values while varying diagnostic sensitivity and holding specificity constant, and vice versa, for a given prevalence of illness. We then calculate predictive values while holding sensitivity and specificity constant, but varying prior probability (clinically feasible by assessing other factors associated with bipolar outcomes, such as family history and degree of recurrence). RESULTS: Assuming a sample in which the prevalence of illness is 10%, achieving positive predictive values (PPV) >50% requires diagnostic specificity of >95%. Holding specificity at a level already achieved clinically (86%), increasing prior probability yields predictive values as high as 83%. CONCLUSION: Systematic assessment of clinical factors that increase the prior probability of illness, before applying DSM/ICD criteria, could raise PPV substantially compared with targeting greater specificity via more stringent diagnostic criteria.
Authors: Lars Vedel Kessing; Klaus Munkholm; Maria Faurholt-Jepsen; Kamilla Woznica Miskowiak; Lars Bo Nielsen; Ruth Frikke-Schmidt; Claus Ekstrøm; Ole Winther; Bente Klarlund Pedersen; Henrik Enghusen Poulsen; Roger S McIntyre; Flavio Kapczinski; Wagner F Gattaz; Jakob Bardram; Mads Frost; Oscar Mayora; Gitte Moos Knudsen; Mary Phillips; Maj Vinberg Journal: BMJ Open Date: 2017-06-23 Impact factor: 2.692
Authors: Kevin Jenniskens; Joris A H de Groot; Johannes B Reitsma; Karel G M Moons; Lotty Hooft; Christiana A Naaktgeboren Journal: BMJ Open Date: 2017-12-27 Impact factor: 2.692