Probabilistic Bayesian reasoning can help identifying potentially wrong immunoassays results in clinical practice: even when they appear ‘not-unreasonable’

Abstract

Immunoassays are susceptible to analytical interferences including from endogenous immunoglobulin antibodies at a rate of ∼0.4% to 4%. Hundreds of millions of immunoassay tests (>10 millions in the UK alone) are performed yearly worldwide for measurements of an array of large and small moieties such as proteins, hormones, tumour markers, rheumatoid factor, troponin, small peptides, steroids and drugs. Interference in these tests can lead to false results which when suspected, or surmised, can be analytically confirmed in most cases. Suspecting false laboratory data in the first place is not difficult when results are gross and without clinical correlates. However, when false results are subtle and/or plausible, it can be difficult to suspect with adverse clinical sequelae. This problem can be ameliorated by using a probabilistic Bayesian reasoning to flag up potentially suspect results even when laboratory data appear "not-unreasonable". Essentially, in disorders with low prevalence, the majority of positive results caused by analytical interference are likely to be false positives. On the other hand, when the disease prevalence is high, false negative results increase and become more significant. To illustrate the scope and utility of this approach, six different examples covering wide range of analytes are given, each highlighting specific aspect/nature of interference and suggested options to reduce it. Bayesian reasoning would allow laboratorians and/or clinicians to extract information about potentially false results, thus seeking follow-up confirmatory tests prior to the initiation of more expensive/invasive procedures or concluding a potentially wrong diagnosis.