Selection of variables using ‘independence bayes’ in computer‐aided diagnosis of upper gastrointestinal bleeding

Ohmann, Ch.; Künneke, M.; Zaczyk, R.; Thon, K.; Lorenz, W.

doi:10.1002/sim.4780050515

Cited by 18 publications

(9 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An alternative generalisation, which remains within the mould of independence Bayes, has been explored by Ohmann et al (1986), Langley & Sage (1994, Singh & Provan (1996), Kontkanen er al. (1999), and others, sometimes called 'selective Bayesian classifiers'.…”

Section: Modifications To the Basic Independence Modelmentioning

confidence: 99%

Idiot's Bayes—Not So Stupid After All?

Hand¹,

2001

Int Statistical Rev

495

196

View full text Add to dashboard Cite

Folklore has it that a very simple supervised classification rule, based on the typically false assumption that the predictor variables are independent, can be highly effective, and often more effective than sophisticated rules. We examine the evidence For this, both empirical, as observed in real data applications, and theoretical, summarising explanations for why this simple rule might be effective.

show abstract

Section: Modifications To the Basic Independence Modelmentioning

confidence: 99%

Idiot's Bayes—Not So Stupid After All?

Hand¹,

2001

Int Statistical Rev

495

196

View full text Add to dashboard Cite

show abstract

“…Crichton et aL [65,66] reported a small increase in accuracy using five or six selected variables rather than all 41 on a database of acute abdominal pain, although the improvement does not appear to be statistically significant. However, Ohmann et aL [67] warned of the danger of bias in selecting variables; in their study of .. patients with gastrointestinal bleeding, ~ they found that the estimated true diagnostic accuracy generally improved as variables were included. Similarly, in a simulation study of cases of vaginal discharge, Chard and Rubenstein [68] found that optimal diagnostic accuracy was achieved using all variables.…”

Section: Discussionmentioning

confidence: 99%

The Relative Accuracy of a Variety of Medical Diagnostic Programs

Todd

Stamper²

1994

Methods Inf Med

View full text Add to dashboard Cite

Abstract:Acute abdominal pain is one of the most widely studied applications of computer-aided diagnosis. The usual approach is to apply Bayes’ theorem with the assumption of conditional independence (“independence Bayes”). We compared various approaches to designing diagnostic programs for abdominal pain of suspected gynaecological origin. The methods range from statistical to knowledge-based. All programs were evaluated using a database of 1,270 cases collected retrospectively. Our results suggest that in this application no significant improvement in accuracy can be made by taking interactions into account, either by statistical or by knowledge-based means; independence Bayes is near-optimal. As far as accuracy is concerned, there appears to be little point in pursuing knowledge-based approaches. However, the “nearest neighbours” method using a new metric appears to be at least as accurate as independence Bayes. We argue that the nearest neighbours method is more suitable than independence Bayes for clinical use because of greater accountability.

show abstract

“…[6][7][8][9][10][11][12][13][14][15][16] Bayes' theorem can work well when independence is assumed, 6,7,12,15,16 but violations of the independence assumption can degrade diagnostic accuracy. [7][8][9][10] Proposed solutions include rankordering variables according to degree of intercorrelation, 8 selecting variables based on whether they increase diagnostic accuracy, 13,14 or using discriminant function analysis. 10 However, clinicians are urged to use LRs published in medical literature, particularly with regard to the clinical examination, to calculate disease probabilities at the bedside.…”

mentioning

confidence: 99%

Quantitative Assessments from the Clinical Examination: How Should Clinicians Integrate the Numerous Results?

Holleman

Simel

1997

J GEN INTERN MED

View full text Add to dashboard Cite

OBJECTIVE:To describe strategies for using multiple clinical examination items to estimate disease probabilities; and to evaluate the diagnostic accuracy of each strategy. DESIGN: Prospective observational study.SETTING: Medical preoperative evaluation clinic at a university-affiliated Veterans Affairs Medical Center. PATIENTS:Previously reported consecutive series of patients referred for outpatient medical preoperative risk assessment. MEASUREMENTS AND MAIN RESULTS:Pulmonary clinical examination and spirometry were the measurements. A strategy of using likelihood ratios (LRs) from seven clinical examination items was least accurate ( p Ͻ .0001). Three alternative strategies were equivalent in diagnostic accuracy ( p Ն .2): (1) using the single best clinical examination item and its LR, (2) using the LRs from three clinical examination items chosen by logistic regression, and (3) using the adjusted LRs chosen in strategy 2. When compared with using LRs from all seven items, the strategies of using three LRs chosen by logistic regression or using adjusted likelihood ratios better discriminated patients with airflow limitation from those without ( he clinical examination comprises the medical history and the physical examination. Each component of the clinical examination can be considered a separate diagnostic test. As such, clinicians must choose how to incorporate the numerous results obtained from eliciting the patient's history and performing a physical examination into their clinical decision making. Some of these results will be highly relevant, whereas others ought to have little impact on clinical decisions. This process of determining the relevance of clinical examination results is twofold: first, is the result statistically associated with the target disorder; and second, how much does the result contribute to the diagnosis above other clinical examination results? To make these decisions about relevance, clinicians need a summary measure of the impact of a given clinical examination result on the likelihood of the target disorder.Likelihood ratios (LRs) are convenient summary measures of diagnostic test performance. They describe the prevalence of a diagnostic test result in patients with a condition of interest versus the prevalence of the same test result among people without the condition. 1 Whether the test yields dichotomous, ordinal, or continuous results, LRs convey a similar type of diagnostic information. Clinicians can use LRs to modify their estimated prior odds of a target disorder to arrive at the posterior odds of the disorder. When there is a single diagnostic test, posterior odds is simply the product of prior odds and the LR for the test result.

show abstract

Selection of variables using ‘independence bayes’ in computer‐aided diagnosis of upper gastrointestinal bleeding

Cited by 18 publications

References 15 publications

Idiot's Bayes—Not So Stupid After All?

Idiot's Bayes—Not So Stupid After All?

The Relative Accuracy of a Variety of Medical Diagnostic Programs

Quantitative Assessments from the Clinical Examination: How Should Clinicians Integrate the Numerous Results?

Contact Info

Product

Resources

About