[Skip to Navigation]
Views 427
Citations 0
Translational Science With Clinical Promise
July 2013

Personalized Medicine: Bayesian Inference as Applied to the Measurement of Glaucomatous Visual Field Loss

Author Affiliations
  • 1New Jersey Medical School, Institute of Ophthalmology and Visual Science, Newark
JAMA Ophthalmol. 2013;131(7):837-838. doi:10.1001/jamaophthalmol.2013.5145

If presented with visual fields from a patient with central islands of vision, most of us would have difficulty in determining, on the basis of this information alone, the correct diagnosis. If we were given additional information that the intraocular pressure was 35 mm Hg OU, we might be more inclined to suggest that the visual field results signify the presence of advanced glaucoma. However, if we were then presented with fundus photographs illustrating diffuse waxy pallor of the optic nerves with no cupping, attenuated retinal vascular caliber at the optic nerve heads, and bone-spicule hyperplasia of the retinal pigment epithelium in a bilaterally symmetric pattern, we would probably conclude that the cause of the visual field findings is retinitis pigmentosa (RP). If we then obtained a family history and learned that the patient’s father and paternal grandfather had been diagnosed as having RP, we would likely conclude that this is a case of autosomal dominant RP. If we obtained genetic testing and learned that there is a mutation in this patient’s rhodopsin gene (eg, P23H), we would be confident in the diagnosis of RP. This patient illustrates how our assessment of the probability of a diagnosis is influenced by both the prior probability and the addition of new information. The probability of a diagnosis being correct increases as additional information that is consistent with that diagnosis is acquired, and it decreases as information that is not consistent with that diagnosis is acquired. Beyond a certain point, additional information only changes our level of certainty minimally. We also use this sort of reasoning to reject the validity of clinical data. If, for example, as part of a preoperative assessment, a patient’s testing reveals an abnormally high prothrombin time (PT)/partial thromboplastin time (PTT) but the patient has no history of abnormal bleeding, does not use an anticoagulant, and does not have a history of rheumatological disease, we are likely to conclude that the result is spurious and are likely to either ignore the test or repeat it rather than initiate treatment with fresh frozen plasma before major surgery on the basis of the PT/PTT result alone. This approach is sound because the likelihood of the abnormal PT/PTT being a valid result, in view of all the other information available to us, is low.

Add or change institution