Assessment of Diagnostic Tests When Disease Verification is Subject to Selection Bias

Begg, Colin B.; Greenes, Robert A.

doi:10.2307/2530820

Cited by 617 publications

(484 citation statements)

References 5 publications

Supporting

Mentioning

469

Contrasting

Unclassified

Order By: Relevance

“…46 In this way, verification bias is avoided that arises because only some women are referred for colposcopy. 47 The sensitivities and specificities of abnormal cytology are comparable to figures in review studies. 11,48,49 Besides basecase estimates, high-and low-sensitivity estimates are presented.…”

Section: Screening Modelsupporting

confidence: 54%

Natural history and screening model for high‐risk human papillomavirus infection, neoplasia and cervical cancer in the Netherlands

Berkhof

Bruijne

Zielinski

et al. 2005

Intl Journal of Cancer

View full text Add to dashboard Cite

A simulation model is presented that assumes that persistent infection with high-risk human papillomavirus (hrHPV) is a necessary cause of cervical cancer. For the estimation of the model parameters, data of recent Dutch follow-up studies were reanalyzed. The predicted incidences of cervical cancer, cervical intraepithelial neoplasia (CIN1, CIN2 and CIN3) and abnormal cytology were validated with nationwide figures and population-based screening results. The model predicted a lifetime risk for cervical cancer of 2.9% with a peak at age 48 years. The predicted lifetime risk dropped to 0.4% when attending cervical screening. For women who were not hrHPV infected at 30 years, the lifetime risk was 1.6%. Sensitivity analyses were performed to check natural history assumptions that were only weakly identified from available data sets. The incidence of CIN3 observed with screening appeared a useful clinical end point as the predicted incidence was robust against changes in the sensitivity of cervical cytology and the duration to CIN3. The model can be used to study the healtheconomic benefits that can be achieved in nationwide screening when including an hrHPV test. ' 2005 Wiley-Liss, Inc.Key words: cervical neoplasms; cytology; cost effectiveness; human papillomavirus; longitudinal studies; models; screening Population-based screening is regarded as an effective method for reducing the incidence of invasive cervical cancer. Epidemiologic studies report that in countries where cytologic screening has been implemented, the incidence of cervical cancer has shown a larger decrease than in countries without screening. 1,2 Although many countries subscribe to the necessity of screening, each country has its own screening program and different opinions about the optimal program remain.Because persistent infection with the high-risk human papillomavirus (hrHPV) is the key causative agent of cervical cancer, 3,4 many recent and ongoing longitudinal studies have examined whether a test for detection of hrHPV in addition to cytologic inspection increases the efficacy of screening. 5-9 Combined screening (hrHPV and cytology) leads to an earlier detection of women who are at risk for developing high-grade lesions 10 and has a higher sensitivity for the detection of high-grade lesions. 11 However, hrHPV testing also induces extra screening costs so that the determination of the optimal screening scenario involves an elaborate cost-effectiveness analysis.An attractive way to study the efficacy of different screening scenarios is simulation modeling. 12-16 Simulation enables us to combine the results of several studies, compare a large number of scenarios and provides an assessment of the costs and effectiveness in preventing cancer. A simulation model requires the specification of an underlying natural history model that describes health trajectories of women in an unscreened population. We present a natural history model for women in the Netherlands between 30 and 80 years of age and apply the model to predict the incidence of ne...

show abstract

Section: Screening Modelsupporting

confidence: 54%

Natural history and screening model for high‐risk human papillomavirus infection, neoplasia and cervical cancer in the Netherlands

Berkhof

Bruijne

Zielinski

et al. 2005

Intl Journal of Cancer

View full text Add to dashboard Cite

show abstract

“…Begg and Greenes (1983) proposed bias correction methods for estimating the sensitivity and specificity. Consider the data given in Table 1a with the sample of size n when we know that a sub-sample n 1 has verified disease status.…”

Section: Incomplete Data Methods For Verification Biasmentioning

confidence: 99%

“…While for the remaining n 2 = n − n 1 subjects, we do not know their true disease status. Existing methods for correcting for verification bias are as follows: Begg-Greenes (B&G) interval: Begg and Greenes (1983) proposed a method to derive sensitivity and specificity under the ignorability assumption for the verification process. If we follow the notation of Table 1a , it follows that the sensitivity estimate iŝ…”

Section: Incomplete Data Methods For Verification Biasmentioning

confidence: 99%

“…The results are summarized in Tables 2-3. When we tried to use the bootstrap-t method for the B&G (Begg and Greenes, 1983) and Logit B&G methods, there was a zero cell in many of the simulated data sets and the estimate of the pivotal quantity was close to zero as well. Hence, the use of this method resulted in very unstable results.…”

Section: Simulation Studymentioning

confidence: 99%

“…Under verification bias we can no longer estimate the sensitivity and specificity using two separated binomial distributions as before. The most widely used correction method for this method was developed by Begg and Greenes (1983) under the ignorable verification bias assumption, that assumes the reason for selecting a sample for verification depends only on observed data. Zhou (1993) extended that method using a maximum likelihood approach.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Multiple imputation for correcting verification bias

Harel

Zhou

2006

Statistics in Medicine

View full text Add to dashboard Cite

In the case in which all subjects are screened using a common test and only a subset of these subjects are tested using a golden standard test, it is well documented that there is a risk for bias, called verification bias. When the test has only two levels (e.g. positive and negative) and we are trying to estimate the sensitivity and specificity of the test, we are actually constructing a confidence interval for a binomial proportion. Since it is well documented that this estimation is not trivial even with complete data, we adopt multiple imputation framework for verification bias problem. We propose several imputation procedures for this problem and compare different methods of estimation. We show that our imputation methods are better than the existing methods with regard to nominal coverage and confidence interval length. Copyright © 2006 John Wiley & Sons, Ltd.

show abstract

Trends in the Population-Based Incidence of Squamous Cell Carcinoma of the Skin First Diagnosed Between 1984 and 1992

Gray

Suman²,

et al. 1997

Arch Dermatol

193

View full text Add to dashboard Cite

Assessment of Diagnostic Tests When Disease Verification is Subject to Selection Bias

Cited by 617 publications

References 5 publications

Natural history and screening model for high‐risk human papillomavirus infection, neoplasia and cervical cancer in the Netherlands

Natural history and screening model for high‐risk human papillomavirus infection, neoplasia and cervical cancer in the Netherlands

Multiple imputation for correcting verification bias

Trends in the Population-Based Incidence of Squamous Cell Carcinoma of the Skin First Diagnosed Between 1984 and 1992

Contact Info

Product

Resources

About