Computer‐aided diagnosis with potential application to rapid detection of disease outbreaks

Abstract

Our objectives are to quickly interpret symptoms of emergency patients to identify likely syndromes and to improve population-wide disease outbreak detection. We constructed a database of 248 syndromes, each syndrome having an estimated probability of producing any of 85 symptoms, with some two-way, three-way, and five-way probabilities reflecting correlations among symptoms. Using these multi-way probabilities in conjunction with an iterative proportional fitting algorithm allows estimation of full conditional probabilities. Combining these conditional probabilities with misdiagnosis error rates and incidence rates via Bayes theorem, the probability of each syndrome is estimated. We tested a prototype of computer-aided differential diagnosis (CADDY) on simulated data and on more than 100 real cases, including West Nile Virus, Q fever, SARS, anthrax, plague, tularaemia and toxic shock cases. We conclude that: (1) it is important to determine whether the unrecorded positive status of a symptom means that the status is negative or that the status is unknown; (2) inclusion of misdiagnosis error rates produces more realistic results; (3) the naive Bayes classifier, which assumes all symptoms behave independently, is slightly outperformed by CADDY, which includes available multi-symptom information on correlations; as more information regarding symptom correlations becomes available, the advantage of CADDY over the naive Bayes classifier should increase; (4) overlooking low-probability, high-consequence events is less likely if the standard output summary is augmented with a list of rare syndromes that are consistent with observed symptoms, and (5) accumulating patient-level probabilities across a larger population can aid in biosurveillance for disease outbreaks.

Inspector interfaces to facilitate subjective reasoning about quantities in trends

“…Anecdotal evidence suggests the need to reason quantitatively.To a certain extent the approach is similar to what one might envision, for example, in recognizing unnatural disease outbreaks or in modern quality control applications. In monitoring for unnatural disease outbreaks (Burr et al 2007) one must use observed patient symptoms that are measured with error, account for the background incidence of possible diseases (the "prior" probability of each candidate disease), and somehow aggregate estimated probabilities of candidate diseases by patient over space and time. Burr et al (2007) allowed the user to vary the prior disease probability for the same reason our safeguards sliders are provided.…”

“…In monitoring for unnatural disease outbreaks (Burr et al 2007) one must use observed patient symptoms that are measured with error, account for the background incidence of possible diseases (the "prior" probability of each candidate disease), and somehow aggregate estimated probabilities of candidate diseases by patient over space and time. Burr et al (2007) allowed the user to vary the prior disease probability for the same reason our safeguards sliders are provided. In modern quality control, described for example in (Hines et al, 2007), there is a need to monitor both for product quality drift and for instrument calibration drift, so the task of learning normal behaviour is challenging for reasons that are similar to the reasons it is challenging to learn normal behaviour in our described safeguards application.…”

“…Marking the most appropriate points can often be a challenge . Further details on this representation can be found in Howell &Scothern, 2000 andBurr, 2007. Further details on trend marking can be found in Miller and Howell (1999) …”

Data processing and analysis in real‐world traditional Chinese medicine clinical data: challenges and approaches

“…Therefore, the problem of missing values challenges reliable data analysis. It is important to determine whether the unrecorded positive status of a symptom means that the status was negative or that the status was unknown . Because the missing patterns of clinical data are highly relevant to different TCM physicians, this indicates that physician name or code should be included as one of the essential fields in TCM clinical data, which would be valuable for data analysis.…”