Use of a New Method in Reaching Consensus on the Cause of Cytologic-Histologic Correlation Discrepancy

Transcription

1 Anatomic Pathology / CONSENSUS ON DISCREPANCY CAUSES Use of a New Method in Reaching Consensus on the Cause of Cytologic-Histologic Correlation Discrepancy Stephen S. Raab, MD, 1 Chad H. Stone, MD, 2 Eva M. Wojcik, MD, 3 Kim R. Geisinger, MD, 4 Laila Dahmoush, MD, 5 Fernando U. Garcia, MD, 6 Dana Marie Grzybicki, MD, PhD, 1 Janine E. Janosky, PhD, 7 Frederick A. Meier, MD, 2 and Richard J. Zarbo, MD, DMD 2 Key Words: Diagnostic error; Pathology; Patient safety; Cancer; Interobserver agreement; Consensus DOI: /1790JN2YWCG833VU Abstract Pathologists exhibit very poor agreement in adjudicating the cause of cytologic-histologic correlation discrepancies, which contributes to problems in designing interventions to reduce discrepancy frequency. In this observational study, we developed a visual method of adjudicating discrepancy cause, termed the No-Blame Box method, which consisted of initially assessing interpretability by separately evaluating quality and the presence of tumor. Five pathologists blindly adjudicated the cause of discrepancy in pulmonary s from 40 patients. The κ statistic of all pathologist pairs in adjudicating discrepancy cause using the No-Blame Box method ranged from to 0.796, indicating acceptable to excellent agreement. Pathologists ranged in their assessment of interpretability from 13% to 20%, and in no case did all 5 pathologists concur that a was interpretable. Most discrepancies resulted from pathologists diagnosing noninterpretable samples. Pathologists who used the No-Blame Box showed significant agreement in the adjudication of discrepancy cause. Health care practitioners long have recognized the limitations in cancer diagnostic testing, and test performance metrics (eg, sensitivity, specificity, and likelihood ratios) express testing failures in a probabilistic manner. 1 The impact of the Institute of Medicine report To Err Is Human 2 on diagnostic testing was to shift the focus from measuring test performance to error reporting, measuring the impact of error on patients and society, root cause analysis, and error reduction. This shift has highlighted the current inability of health care practitioners to determine the reasons why diagnostic errors occur. 3 Many clinicians and pathologists begrudgingly accept current test performance metrics and lack the ability to develop and implement nontechnological initiatives to reduce diagnostic testing errors. Testing for cancer involves the performance of a number of steps, which lie separately within the clinical and laboratory domains, and errors may occur anywhere along this complex pathway. 4 Many current methods of diagnostic testing root cause analysis are not process-oriented and crudely classify errors as sampling when there is a presumed failure in tissue procurement or as interpretation when there is a presumed mistake in diagnosis. 5-7 Pathologists differ widely in their assessment of error or discrepancy cause. 3,5 We believe that a component of the currently used process is blame-centered and identifies failures as either laboratory- or clinician-related. A multihospital consortium reported that institutional interpretive errors composed 5% to 42% of all diagnostic errors, with the remainder being sampling. 3 When pathologists from these institutions performed root cause analysis of the same set of cases using the traditional dichotomous classification system, the pairwise κ statistic ranged from 0.02 to 0.74 with 60% of pairwise κ statistics less than 0.4, generally indicating poor agreement Am J Clin Pathol 2006;126: DOI: /1790JN2YWCG833VU

2 Anatomic Pathology / ORIGINAL ARTICLE This initiative grew out of 3 years of failure to achieve consensus in adjudicating diagnostic discrepancy root cause in an Agency for Healthcare Research and Quality multi-institutional patient safety project. 3 At a consensus meeting, representatives from 6 institutions designed a new method to perform diagnostic discrepancy root cause analysis that focused on evaluating all components of the diagnostic testing pathway rather than on individuals responsible for error. Interobserver agreement using this method is presented. We believe that this method, which we term the No-Blame Box method, allows for more reliable evaluation of diagnostic discrepancies. Materials and Methods Diagnostic Error Detection The accurate anatomic pathology diagnosis of cancer depends on the successful completion of many processes, including tissue procurement, transport, processing, pathologist interpretation, and diagnostic reporting. 4 Each of these processes includes a number of subprocesses, and errors may occur at any or multiple points. Errors in cancer diagnosis generally are detected after a report has been issued, and detection usually involves case review. 7,8 Cytologic-histologic correlation is one method of error detection that involves reviewing the diagnoses of cytologic and histologic s that are obtained from the same body site. 3,5 For example, suppose a patient has a lung mass and undergoes 2 tests to reach a diagnosis of cancer. In this scenario, the patient first may be evaluated with transthoracic fine-needle aspiration. If the result is negative and the clinical suspicion is sufficiently high, a bronchoscopic examination and histologic biopsy may be performed and the cytologic and histologic diagnoses correlated. The performance of cytologic-histologic correlation is congressionally mandated through the Clinical Laboratory Improvement Amendments of The federal regulations do not mandate how cytologic-histologic correlation is to be done, and, consequently, the performance is not nationally standardized. 10 Cytologic and histologic diagnostic schema are different, and we developed a scaled hierarchy of categories to determine whether a discrepancy occurred. 3 For simplification, we assumed that the cytologic and histologic diagnoses were binary (ie, benign or malignant). We defined a discrepancy as a difference between the cytologic and histologic diagnoses that indicated the presence or absence of a pathologic entity or a definite difference in the degree to which the pathologic condition was judged to be present. 3 In the most commonly used method of cytologic-histologic correlation, a pathologist reexamines all microscopic slides and determines whether the cytologic, histologic, neither, or both diagnoses are erroneous. 10 In assigning the root cause of error, the pathologist uses a binary system: interpretation or sampling. 5,6 An interpretation error is an error in diagnostic categorization. A sampling error is an error in which diagnostic material is not present on the discrepant slide and often is attributed to a failure in tissue procurement. Consensus Conference Rationale and Goals Since the start of this Agency for Healthcare Research and Quality project in September 2002, pathologists at the participant laboratories have attempted to reach agreement on the cause of error in cases detected by the cytologic-histologic correlation process. Our goal has been to achieve pairwise error assessment agreement levels measured by the κ statistic of 0.40 or more. 3 During the past 3 years, the principal investigators from the original 4 project sites evaluated 3 study sets of 40 cytologic-histologic case pairs, and approximately 60% of pairwise κ statistics were less than Attempts to reach agreement through conference calls involving individual case discussion, telepathology conferences examining cases realtime, and face-to-face meetings failed. On December 3 and 4, 2005, the principal investigators from all 6 currently participating sites, consultants, and coinvestigators had a consensus meeting in Pittsburgh, PA. The goals of this conference were to explore the reasons underlying the lack of consensus and to test modifications to the correlation process that might improve agreement. We conjectured that the prior failure to reach consensus was because of variability in assessing the presence of interpretable tumor and, more important, the inability to agree on the appropriate cutoff for the separation of interpretation and sampling errors. We developed a new method, termed the No-Blame Box method, to assess the root cause of error in cases detected by cytologic-histologic correlation. A No-Blame Box is depicted in Figure 1, and the method involves a probabilistic estimation of interpretability. We assumed that interpretability for a patient who has cancer consists of 2 components: (1) the presence of recognizable tumor, and (2) the overall quality of the. In the No-Blame Box, the amount of tumor is depicted vertically, increasing from no tumor at the top of the box to abundant tumor at the bottom. For cytologic s, nondefinitive diagnoses (eg, atypical or suspicious) are probabilistically depicted along the vertical axis. Thus, a that contains many cancer cells would be graded on the lower portion of the vertical axis and a that contains only a few possible cancer cells would be graded higher on the vertical axis. The overall quality is depicted horizontally, increasing from a poor quality at the left to an excellent quality at the right. Aspects of excellent quality include absence of obscuring factors, absence of technical limitations Am J Clin Pathol 2006;126: DOI: /1790JN2YWCG833VU 837

3 Raab et al / CONSENSUS ON DISCREPANCY CAUSES Amount of tumor Quality of A. B. C. Poor quality Poor quality (eg, crush artifact or air-drying), and overall cellularity with elements expected from that anatomic site (eg, lung parenchymal elements from a bronchial washing ). Specimen quality relates to preanalytic ( procurement and transport) and analytic (laboratory preparation) processes. The 4 squares of the No-Blame Box divide s into combinations of good quality poor quality and cancer no cancer. The perfect diagnostic test of a patient who has cancer would be scored in the right lower corner of the box. Completing the No-Blame Box generally shows that the error cause is multifactorial. The No-Blame Box is a visual tool and does not involve the evaluation of individual practitioners. A separate No- Blame Box is completed for the cytologic and histologic s. After examining the slides, the review pathologist places a mark in the box at the location that best describes the amount of tumor and the quality of the. Pathologists adjudicate error differently if they see little or abundant tumor and if the is of excellent or poor quality. By using the No-Blame Box, the combination of factors that lead to error becomes apparent, and adjudication becomes less oriented toward blame and more toward interpreting limitations. Some pathologists require an abundant amount of tumor and a of high quality as necessary to adjudicate an error as interpretation. Other pathologists tend to ignore quality and make a review diagnosis of malignancy if only rare malignant cells D. Figure 1 No-Blame Box. The No-Blame Box is divided into 4 quadrants depicting s of different interpretability ranging from excellent to poor quality and abundant to no tumor. are present; these pathologists have a lower threshold of interpretability in adjudicating an interpretation error. Pathologists could examine a group of cases and assign each case to the identical region in the No-Blame Box but still have marked differences in their adjudication of an interpretation or a sampling error. Patient s that are classified in square D are viewed as interpretable for tumor because they are of sufficient quality and contain a sufficient amount of tumor for diagnosis. Patient s that are classified in squares A, B, and C are considered essentially noninterpretable for tumor because they are limited in quality and/or sufficient amount of tumor. This classification exercise is performed for cases reviewed following an identification of an error. In the initial case evaluation, we recognize that some pathologists would interpret patient s in square C as malignant, although on review, we assume that the quality of the limits interpretation. Testing of the No-Blame Box We tested the use of the No-Blame Box in a study set of 40 cases. We measured the interobserver agreement of error cause adjudication in pulmonary s because the lung is a common anatomic location associated with cytologic-histologic correlation error. 3 The s were bronchial washing or brushing and histologic s. The 6 project sites each selected 10 discrepant cases for review, and 1 of the coinvestigators (D.M.G.) selected 40 of these cases for blinded review. The 6 project sites selected an equal mix of cases originally adjudicated as a false-negative interpretation or sampling error (37 cytologic and 3 histologic). The cytologichistologic correlation process detects a higher frequency of cytologic error compared with histologic error because cytologic testing often is the front-line test, followed by histologic testing if a diagnosis is negative or nondefinitive. Before the evaluation of the 40 test cases, 15 cases were reviewed jointly by using the No-Blame Box method. The case slides and original reports were deidentified (institution, patient, and provider identifiers removed). The original diagnoses, clinical history, gross and microscopic findings, and diagnostic comments were not removed because this information is routinely available to pathologists performing correlation. 10 Five principal investigators performed cytologic-histologic correlation root cause analysis. The pathologists examined the cases independently; cases were circulated in groups of 8. Each pathologist was provided a No-Blame Box and a standardized data collection form that recorded the cytologic and histologic diagnoses and an assessment of cytologic quality (eg, obscuring inflammation, blood, and necrosis, air-drying artifact, and thickness of the ) graded on a 5-point Likert scale. For each case, the pathologists placed a mark in the appropriate square. Finally, 838 Am J Clin Pathol 2006;126: DOI: /1790JN2YWCG833VU

4 Anatomic Pathology / ORIGINAL ARTICLE each pathologist assessed whether the discrepancy was due to the sampling process or the interpretation process based on the location of the mark. Project Sites The participating institutions are located in the Mid-Atlantic region or the Midwestern region of the United States. Institutional review board approval was granted at each institution. Statistical Analysis All statistical analyses were performed using the statistical software package SPSS (version 12.0, SPSS, Chicago, IL) by 1 of the coauthors (J.E.J.). Pairwise κ statistics with 95% confidence intervals were used to measure interobserver agreement. We did not perform hypothesis testing because our purpose in measuring agreement was to assess its level and not to compare levels between pathologist pairs. Pairwise κ statistics were used to measure the agreement for the adjudication of error cause (ie, interpretation or sampling process) following the completion of the No-Blame Box. In addition, we measured agreement for the adjudication of cytologic interpretability using the No-Blame Box (eg, categories A and B vs C and D). We interpreted the pairwise κ statistic as follows: 0.80 or more to 1.0, excellent agreement; 0.60 or more to 0.80, good agreement; 0.40 or more to 0.60, fair agreement; and less than 0.40, poor agreement. Associations between the individual pathologist and No Blame-Box category for cytologic cases were examined using a χ 2 test. Statistical significance was assumed at a P value of.05 or less. Results The pairwise κ statistics for adjudication of discrepancy cause following pathologist use of the No-Blame Box are shown in Table 1. All the pairwise κ statistics were or more, indicating acceptable to excellent agreement. Use of the No-Blame Box for the interpretability of cytologic s is illustrated in Figure 2A and Figure 2B, showing the combined 40 responses of pathologists A and E, respectively. These two pathologists used the No-Blame Box quite differently on a case-to-case basis. Pathologist A classified more cases as having sufficient tumor (in excellent or poor quality s), and pathologist E classified more cases as being of poor quality and having no tumor. The categorization of cytologic cases by quality and presence of tumor are shown in Table 2 and Table 3. With square D representing interpretable and squares A, B, and C representing noninterpretable s, pathologists ranged in their assessment of interpretable s from 10% to 20%. A statistically significant association existed between pathologist and assignment of interpretability (P =.044), indicating relatively high disparity in how pathologists classified individual cases. Pathologists disagreed on the amount of tumor present, which previously led to poor agreement when simply adjudicating whether a misinterpretation was made; this is the reason why traditional adjudication methods failed to show consensus. Table 3 shows the number of cases classified in the 4 squares combining quality and amount of tumor. For example, the category of poor quality with tumor identified showed unanimous agreement in 7 cases. No other category showed unanimous agreement, and in no case did all 5 pathologists concur that a was interpretable (square D). In 27 cases, the pathologists chose only 1 of 2 categories, varying in their assessments of quality or tumor amount. The pathologists unanimously agreed that no tumor was present in 9 cases, and these cases could be attributed to pure sampling error. By using the No-Blame Box, the pairwise κ statistic for pathologists in assessing the presence or absence of tumor (categories A and B vs C and D) ranged from to Discussion We believe that the significance of our findings is 2-fold. First, we were able to get pathologists to agree on the cause of diagnostic discrepancy, and this has not been previously achieved. Second, we developed a method to perform root cause analysis that allows for a much richer understanding of the cause of diagnostic discrepancy. Table 1 Pairwise κ Statistics for Assigning Cause of Error After Using the No-Blame Box Method of Correlation * Pathologist Pathologist B C D E A ( ) ( ) ( ) ( ) B ( ) ( ) ( ) C ( ) ( ) D ( ) * Data are given as κ statistics (95% confidence intervals). Am J Clin Pathol 2006;126: DOI: /1790JN2YWCG833VU 839

5 Raab et al / CONSENSUS ON DISCREPANCY CAUSES A Quality of B Quality of Amount of tumor A. B. Poor quality Amount of tumor A. B. Poor quality C. Poor quality The first steps of improving patient safety involve error detection, reporting, and determination of cause. 2,11-13 Previous work showed that congressionally mandated cytologic-histologic correlation, 10 one of the main diagnostic error detection methods, is severely limited by institutional variability in reporting and, more important, by the inability to establish consensus on root cause of error. 3 The interinstitutional comparison of root cause data is essentially incoherent because pathologists cannot agree on whether discrepancies in cancer diagnosis are secondary to pathologists making mistakes or clinicians obtaining uninterpretable or limited samples. 3 In the first 3 years of study, our consortium has failed to reach consensus on discrepancy cause. The problem lies in the D. C. Poor quality method of root cause analysis that assigns fault to 1 of 2 components of a very complex pathway. This method grew out of the notion that anatomic pathologists are the final arbitrators of the pathologic detection of disease and sometimes assume an individual-centered rather than a process-oriented view. 14 A challenge for practitioners who perform root cause analysis is that they often lack understanding of the clinical or pathologic domain, and there is a tendency to attribute discrepancies to the other domain without appropriate communication. Cancer diagnostic discrepancies are secondary to system failures and not to pathologist or clinician incompetence. We consider our level of pairwise pathologist agreement on the cause of diagnostic discrepancy using the No-Blame D. Figure 2 No-Blame Box classification of 40 cytology cases by pathologists A (A) and E (B), respectively. Each circle represents 1 of 40 cytology cases. Classification depends on a number of factors, including experience, risk aversion, and ability. Table 2 Pathologist Categorization of Error by No-Blame Box Square * No-Blame Box Square Pathologist A B C D A B C D E * Squares were designated as follows: A, poor quality and no tumor identified; B, excellent quality and no tumor identified; C, poor quality and tumor identified; and D, excellent quality and tumor identified. Table 3 Number of Cases in Which the Following No-Blame Box Squares Were Used * No-Blame Box Square Combination No. of Cases C only 7 C or D only 10 A or B only 9 B or C only 6 A or C only 2 A, B, or C 3 B, C, or D 2 A, C, or D 1 * Squares were designated as follows: A, poor quality and no tumor identified; B, excellent quality and no tumor identified; C, poor quality and tumor identified; and D, excellent quality and tumor identified. 840 Am J Clin Pathol 2006;126: DOI: /1790JN2YWCG833VU

6 Anatomic Pathology / ORIGINAL ARTICLE Box method as a breakthrough. Limitations in this study are the relatively few participants, case selection bias, and pathologist bias resulting from investment in reaching consensus. Additional interobserver variability studies are needed with a focus on the study of mechanisms by which pathologists evaluate and adjudicate s. However, the underlying notion, regardless of the method adopted, is that correlation root cause analysis must shift from assigning individual blame to one investigating process. 2,15,16 National adoption of such a method will not be straightforward or easy. We hypothesize that when pathologists assessed interpretability, their focus shifted to process, and better agreement occurred, even when they shifted back to adjudicate errors as sampling process vs interpretation process. The No-Blame Box method of assessing interpretability is probabilistic and follows in the tradition of Schwartz et al, 17 who described the value of probabilistic reporting of pathology diagnoses. In contrast with the probabilistic assessments by Schwartz et al, 17 the No-Blame Box primarily is visual and nondiagnostic and allows for the expression of uncertainty, which intuitively makes sense when examining diagnostic errors that occurred sometime in the past. Further study is needed to establish whether the metrics of tumor amount and quality are the main components of interpretability, whether other metrics necessitate measuring, the degree of independence of these variables, the reproducibility of assessment, and association between low metric scores and error. In addition, further study is needed to determine whether probabilistic quality assessments could be used with the original interpretation and not simply during discrepancy adjudication. Currently, pathologists evaluate the adequacy of some nongynecologic types, but these evaluations rarely have been linked to outcome. Nodit et al 18 performed root cause analysis of pulmonary cytology errors and showed that poor quality, related to clinical and/or pathology processes, contributed to 97% of errors. The majority of false-negative diagnoses are not simply a result of pathologist misinterpretation. 18 The No- Blame Box illustrates that the overwhelming majority of false-negative errors occur because of 3 reasons: (1) Pathologists make interpretative errors when they evaluate poor quality s. (2) Uninterpretable s are not diagnosed as unsatisfactory. (3) Some interpretable s simply do not contain tumor. Each of these error types requires a different focus. Error reduction requires a unified clinical and pathology effort because errors result from combined failures along the clinical and pathology portions of the diagnostic pathway. A challenge in cancer care is in simultaneously improving testing sensitivity and specificity. Efforts to improve diagnostic testing are not new in the clinical or pathology domain and have focused on individual components of the pathway. For example, pathologists reported that increasing the number of cervical histologic sections correlated with fewer missed diagnoses of cervical intraepithelial lesion. 19,20 Improvement efforts generally occur at a local level, and national standardization has not occurred. Cancer diagnostic testing process improvement is not widespread, and many practitioners believe that error reduction should be technological rather than process-oriented. This belief has been secondary to the lack of understanding or the difficulty in addressing the complexity of the cause of error. We believe that the use of the No-Blame Box method of root cause analysis results in pathologists exhibiting better agreement in discrepancy adjudication. The evaluation of interpretability rather than metrics of sampling and misinterpretation allows for practitioners to evaluate underlying system causes of error. In reality, sampling error is secondary to a number of factors such as the nature of the lesion, sampling device, type of preparation, experience of the clinician, and other factors. Interpretation error is secondary to factors such as inadequate cellularity and obscuring elements, as well as pathologist experience and vigilance. Although pathologists may disagree on how the individual quality and amount of tumor enter into diagnostic discrepancy, pathologists generally agree that interpretation of poor quality s is the underlying cause of most discrepancy. Consequently, diagnostic testing error reduction necessitates involving process redesign in laboratory and clinical services to target interpretability components. From the Departments of 1 Pathology and 7 Family Medicine and Clinical Epidemiology, University of Pittsburgh School of Medicine, Pittsburgh, PA; 2 Pathology, Henry Ford Health System, Detroit, MI; 3 Pathology, Loyola University Medical Center, Maywood, IL; 4 Pathology, Wake Forest School of Medicine, Winston-Salem, NC; 5 Pathology, University of Iowa Healthcare, Iowa City, IA; and 6 Pathology, Drexel University College of Medicine, Philadelphia, PA. Supported by grant HS from the Agency for Healthcare Research and Quality, Rockville, MD. Address reprint requests to Dr Raab: Dept of Pathology, University of Pittsburgh School of Medicine, 5230 Centre Ave, Pittsburgh, PA References 1. Sacket DL, Haynes RB, Guyatt GH, et al. Clinical Epidemiology: A Basic Science for Clinical Medicine. 2nd ed. Boston, MA: Little Brown; 1991: Kohn LT, Corrigan JM, Donaldson MS. To Err Is Human: Building a Safer Health System. Washington, DC: National Academy Press; Raab SS, Grzybicki DM, Janosky JE, et al. Clinical impact and frequency of anatomic pathology errors in cancer diagnosis. Cancer. 2005;104: Am J Clin Pathol 2006;126: DOI: /1790JN2YWCG833VU 841

7 Raab et al / CONSENSUS ON DISCREPANCY CAUSES 4. Condel JL, Jukic DM, Sharbaugh DT, et al. Histology errors: use of real-time root cause analysis to improve practice. Pathol Case Rev. 2005;10: Clary JM, Silverman JF, Liu Y, et al. Cytohistologic discrepancies: a means to improve pathology practice and patient outcomes. Am J Clin Pathol. 2002;117: Jones BA, Novis DA. Cervical biopsy-cytology correlation: a College of American Pathologists Q-Probes study of 22,439 correlations in 348 laboratories. Arch Pathol Lab Med. 1996;120: Renshaw AA. Measuring and reporting errors in surgical pathology: lessons from gynecologic cytology. Am J Clin Pathol. 2001;115: Raab SS, Nakhleh RE, Ruby SG. Patient safety in anatomic pathology: measuring discrepancy frequencies and causes. Arch Pathol Lab Med. 2005;129: Department of Health and Human Services, Health Care Financing Administration. Clinical Laboratory Improvement Amendments of 1988 (CLIA), 57 Federal Register 7146 (1992). 10. Vrbin CM, Grzybicki DM, Zaleski MS, et al. Variability in cytologic-histologic correlation practices and implications on patient safety. Arch Pathol Lab Med. 2005;129: Battles JB, Kaplan HS, Van der Schaaf TW, et al. The attributes of medical event reporting systems: experience with a prototype medical event reporting system for transfusion medicine. Arch Pathol Lab Med. 1998;122: Fernald DH, Pace WD, Harris DM, et al. Event reporting to a primary care patient safety reporting system: a report from the ASIPS collaborative. Ann Fam Med. 2004;2: Rubin G, George A, Chinn DJ, et al. Errors in general practice: development of an error classification and pilot study of a method for detecting errors. Qual Saf Health Care. 2003;12: Vermoch KL. Should medical laboratories admit mistakes? Clin Leadersh Manag Rev. 2000;14: Waring JJ. Beyond blame: cultural barriers to medical incident reporting. Soc Sci Med. 2005;60: Odwazny R, Hasler S, Abrams R, et al. Organizational and cultural changes for providing safe patient care. Qual Manag Health Care. 2005;14: Schwartz WB, Wolfe JH, Pauker SG. Pathology and probabilities: a new approach to interpreting and reporting biopsies. N Engl J Med. 1981;305: Nodit L, Balassanian R, Sudilovsky D, et al. Improving the quality of cytology diagnosis: root cause analysis for errors in bronchial washing and brushing s. Am J Clin Pathol. 2005;124: Golbang P, Scurry J, de Jong S, et al. Investigation of 100 consecutive negative cone biopsies. Br J Obstet Gynaecol. 1997;104: Joste NE, Wolz M, Pai RK, et al. Noncorrelating Pap tests and cervical biopsies: histological predictors of subsequent correlation. Diagn Cytopathol. 2005;32: Am J Clin Pathol 2006;126: DOI: /1790JN2YWCG833VU