| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Perspective |
1 Department of Radiology-Breast Imaging, University of California at San
Francisco, P.O. Box 1667, San Francisco, CA 94143-1667.
2 University of Texas Southwestern Medical Center, Dallas, TX 75390.
3 Buckingham, Doolittle, and Burroughs, LLP, Cleveland, OH 44114.
Received February 8, 2005; accepted after revision May 9, 2005.
Address correspondence to R. J. Brenner.
OBJECTIVE. The use of computer-aided detection (CAD) in radiology has been studied for different organ systems. As with any new technology, its impact on determinations of standards of clinical practice is an evolving one that is often defined by its acceptability not only in medical forums but also as defined by courts of law.
CONCLUSION. We discuss the first known appellate legal decision regarding the acceptability of CAD as it relates to the clinical practice of mammography.
Keywords: breast • mammography • radiology practice
Since the approval by the U.S. Food and Drug Administration (FDA) in June 1998 for the first commercial use of a computer-assisted detection (CAD) system, this technology has become increasingly available in the United States and worldwide. Sales of new equipment continue to grow, and currently four different systems are available in the market-place. Research and development of neural network applications and improved algorithms designed to increase both sensitivity and specificity of significant lesion detection promise to advance both the technology and efficacy of this interpretation aid.
One consequence of the incorporation of new technology into clinical practice is its implications on the determination of standards of care. Sufficient longitudinal study of appropriateness and impact on outcome and general acceptance by the medical community are two benchmarks used both in the medical and legal communities for assessing the validity of standards of care technologies [1]. Advocates for new approaches may advance the notion that such developments have become a standard of care before a general consensus. In the legal arena, judges are responsible for determining the admissibility of such arguments, usually attempting to permit reasonable opinions that are subject to vigorous cross examination to establish the truth or lack thereof of any stated position [2].
According to the Physician Insurers Association of America [3], delay in the diagnosis of breast cancer is the leading reason that physicians are sued for malpractice and radiologists are the most common defendants. Nondetection of mammographic signs that eventually are shown to represent cancer remains an important cause for litigation. It is perhaps not coincidental that early CAD research was directed toward improving mammographic accuracy [4, 5].
The efficacy of CAD for mammography has been studied with different results in different clinical settings [6, 7]. In this context, the use of CAD has raised several questions regarding potential legal liability. A recent study reconciling CAD detection of subthreshold abnormalities that were eventually associated with malignancies indicated that a majority of experts would not have selected such mammographic findings for recall or biopsy [8]. Indeed, expert testimony is required to establish the standard of care for reasonable detection and diagnosis of breast cancer. The authors of that study concluded that the simple presence of a CAD mark at a particular anatomic site does not of itself indicate significance, especially given the large number of false-positive marks that current CAD technology assigns. In other words, the failure to respond to a CAD mark was not considered, in that report, to constitute a breach in the standard of care.
Such studies help establish the relative impact of CAD in clinical practice. However, legal cases in which technologies such as CAD are applied provide another objective, external perspective as to the validity of their proposed use. More important, when trial case verdicts are subjected to appellate review—an uncommon phenomenon in most malpractice cases— appellate court decisions provide a legal precedent for future reference and establish a practical basis for assessing the impact of technology on a standard of medical care.
An exhaustive search of medical and legal databases revealed no prior appellate decisions regarding the use of CAD in mammography or any other imaging field. We report a case of alleged delayed diagnosis of breast cancer in which CAD results were obtained and used during litigation presented at trial, and the defense verdict was affirmed by the appellate court. We discuss the potential legal implications of CAD technology in the medical practice setting.
Report of the Case
In June 1998, an asymptomatic woman in her early 60s underwent a screening mammography examination. The study was compared with studies in 1996 and 1997. It was interpreted as showing no features of malignancy, and neither recall nor recommendation for biopsy was included in the interpretative report. A second screening mammographic study 1 year later was also interpreted as normal. Six months later, the woman developed a clinical lump and was referred for a diagnostic imaging study, revealing a mass with microcalcifications in the lower inner quadrant of the right breast. Biopsy showed infiltrating ductal carcinoma, and surgery confirmed a 4.8-cm infiltrating ductal carcinoma with a spread to 17 of 19 examined axillary lymph nodes. Both estrogen and progesterone receptors were negative and the HER2/neu oncogene was overexpressed (stage IIB)— poor prognostic indicators.
The patient initiated a malpractice lawsuit alleging a breach in the standard of care by the radiologists interpreting both the 1998 and 1999 screening mammograms, each study interpreted by different radiologists. The plaintiff dismissed the defendant radiologist who interpreted the 1999 mammography study after a confidential settlement. As a result, the only issue at trial regarded the remaining defendant radiologist's interpretation of the 1998 study.
All four of the retained standard-of-care experts (two for the plaintiff, two for the defendant) agreed that some calcifications were present at the anatomic sight focus in 1998 and that by 2000 the mammography study showed a suspicious mass with clustered microcalcifications. Defense experts testified that there were calcifications at the site in 1998, but they were similar to other background scattered microcalcifications and therefore neither distinguishable nor suspicious. One expert retained by the plaintiff suggested there were six calcifications in the area, not commenting on distribution, and another expert called by the plaintiff indicated there were five tightly clustered calcific particles that she would have referred for additional evaluation. All experts agreed there were no calcifications in 1996, and none of the experts could be sure if there were calcifications at this site in 1997.
The primary plaintiff's expert indicated that the group of calcifications on the 1998 study should have prompted a recall and that a recommendation for biopsy probably would have been the result of an additional study. A second expert, who initially had been named as an expert by the dismissed codefendant, was subpoenaed to testify at the trial on behalf of the plaintiff. When this expert reviewed the 1998 films retrospectively, she thought the calcifications in question were suspicious for malignancy. However, on cross-examination she admitted that she had not detected the calcifications during a prospective review of the same films.
The defense experts characterized the calcifications as fewer than five, smooth and punctate, without an identifiable ductal distribution, and radiographically faint. They concluded that such findings did not meet generally recognized criteria for recall and that recommending routine screening was reasonable.
Recognizing a potential conflict of opinion regarding the appropriate management of the mammographic findings, before the trial, the defense attorney engaged one of the defense experts who had used CAD in his own practice to digitize and analyze the 1998 films using CAD. The films had not undergone such analysis at the time of interpretation. The plaintiff objected to the testimony regarding CAD results as hearsay, indicating that they should be excluded from information for the jury to consider.
CAD (version 3.2, R2 Technology), using a threshold of 3 calcium particles within a 1-cm diameter (Roehrig J, personal communication), did not indicate a mark at the site of the calcifications that were eventually associated with malignancy or at any other area of either breast. The films, in the interest of validity and reproducibility, were subjected to the CAD system two more times with the same results.
The defense expert who had CAD experience in clinical practice testified about the sensitivity of CAD for identifying microcalcifications and introduced the results of the analysis of the mammograms in question. He testified that CAD was not being advocated as a standard of care but rather as an analytic tool for resolving the controversy regarding reasonableness of nonrecall. The defense attorney argued that CAD's failure to mark any suspicious calcifications supported the first defense expert's position and the proposition that the initial interpreting radiologist comported with the standard of care.
A jury rendered a unanimous verdict for the defendant. After the trial, the plaintiff's attorney interviewed all 12 jurors to understand their reasoning during the trial, a procedure commonly used by both plaintiff and defense counsel. The jurors who agreed to comment indicated that they believed the credentials and opinions of the plaintiff's expert were stronger than those of the first defense expert. However, the jurors held the second defense expert's opinions in similar high regard. This expert's testimony was directed exclusively toward the introduction of the CAD results. As a result, the jurors believed that the plaintiff's expert established too high a standard of care for the defendant.
Because CAD retrospective results were introduced as evidence during the trial over the objection of the plaintiff's attorney, the plaintiff filed an appeal for review of the trial court's ruling. The plaintiff at trial argued that the use of the CAD results was an improper out-of-court statement (hearsay) and that the CAD results were not sufficiently probative or accurate to be considered relevant for consideration by a jury. The trial court determined that the introduction of the CAD results was sufficiently reliable and relevant to be considered by the jury and the information was properly admitted into evidence through the defense expert's testimony. The appellate court upheld the trial judge's admission of the evidence and affirmed the jury verdict in the case.
Discussion
Standard of care is generally defined as what a reasonable and prudent physician would do under similar circumstances [9]. When technology is introduced that purports to enhance the ability of the radiologist to interpret studies, clinical trials and accumulated experience help to define the extent to which such technology fulfils its promise. Various outcomes can occur. Sometimes enthusiastic initial reports are tempered by subsequent study, and other times subsequent studies suggest even greater potential than initial experience indicated [6, 7, 10, 11]. Part of this changing perspective, of course, may be due to evolving improvements in technology, as is likely the case for CAD.
The use of CAD has been studied from different perspectives. Investigations that formed the basis for initial FDA approval and early reports subjected mammographic studies that were interpreted as showing no features of malignancy on a previous year to CAD reevaluation. CAD in such studies identified a large number of calcifications that were not recalled by the initial interpreting radiologist and that were subsequently associated with malignancy. However, CAD did not perform as well with masses [10]. What could not be determined from these reports was whether the initial radiologist identified the calcifications and elected not to recall them or if the calcifications were overlooked. In the former case, CAD marks would not affect outcome, whereas in the latter case, they would.
Subsequent studies of CAD under similar conditions concluded that a majority of subtle cases that CAD identified would not have been recalled even by experts, suggesting that not all CAD marks are tantamount to standard-of-care recall [8, 12]. This is supported by the incidence of breast cancer (variable but about 5/1,000) and false-positive marks (about 4) per four-view study, suggesting that CAD provides a useful mark in approximately 1/2,000 times. Nonetheless, as a safety measure, such aid may be important because various factors affect visual detection and recognition of abnormalities, and a mark at an area that might have been otherwise overlooked will provide a potential reconsideration of a given finding, as was shown in a prospective study using CAD in clinical practice [13, 14].
Most medical-legal considerations in the literature have focused on the potential impact of CAD in marking an area that is eventually shown to be cancer as a way of determining the reasonableness of conduct of a radiologist in not recalling such a patient for further evaluation. This case, by contrast, shows how CAD was used in litigation to help determine the reasonableness of a radiologist in not recalling a patient for a subtle, subthreshold, or nonspecific finding similar to those previously reported [8].
The case is important for two primary reasons. It illustrates that under current circumstances, at least one appellate court has affirmed a trial court's admission of CAD results as reliable supportive evidence—not proof— of what may constitute a reasonable standard of care. It also shows how the use of negative CAD results may assist the defense in a case in which a lesion was not recalled. CAD was not advocated as a standard of care but rather as a technology sufficient to support the reliability and probative value of a defense expert's testimony that referred to its results. The appellate court saw the CAD results as "...a scientific analysis conducted by computer, which performed a series of complex mathematic calculations based on detailed information it drew from an X-ray. Therefore, we do not find the computer analysis to be hearsay" [15].
The trial judge is the gatekeeper for all evidence admitted at the trial. As the gatekeeper, the judge has the discretion to exclude evidence if it is not reliable according to the Federal Rules of Evidence. Evidence rule 702(c) [16] provides that if an expert is testifying about the results of a test, the testimony is reliable only if the theory behind the test is objectively verifiable or is validly derived from widely accepted knowledge, facts, or principles; the design of the test reliably implements the theory; and the particular test was conducted in a way that will yield an accurate result.
The U.S. Supreme Court in Daubert v Merrill Dow Pharmaceuticals [2] stated that in evaluating reliability, the following factors may be considered: whether the theory or technique has been tested, whether it has been subject to peer review, whether there is a known or potential rate of error, and whether the methodology has gained general acceptance.
In this case, the plaintiff did not raise a Daubert objection specifically at the trial, but raised it on appeal. The appellate court was not required to perform a formal Daubert analysis because technically the appellant (plaintiff) had not properly preserved the issue at the trial by raising an objection at that time. However, the appellate court stated that its ruling should not be construed as indicating that if a rule 702(c) challenge had been properly presented, it would have been successful.
In commenting on the fairness of the trial, the appellate court extended its comments to indicate that it is not implying ("a conclusion we do not reach") that a Daubert challenge would have been successful. The trial court did not exercise its discretion to exclude the evidence. The appellate court took judicial notice of the issue and affirmed the trial court's admission of the evidence.
It remains unclear whether CAD may assist in the plaintiff's position that a lesion not subject to recall should have been recalled if the CAD mark were placed, especially in two views, at the anatomic site. The conclusions reached by previous authors [8, 12], in the context of multiple false-positive marks by CAD, suggest that the liability for nonrecall of a given finding would be dependent on expert opinion.
By contrast, the absence of a CAD mark, given the sensitivity of CAD for calcium, would likely buttress a defense argument that nonrecall was not below a reasonable standard of care. Challenges, however, might still be directed toward specific technology and reproducibility [11]. Given the exit interviews of the jury in this case, it appears that CAD results may be influential in other defense cases. The probative value of CAD results in a plaintiff's context are untested.
Several caveats need be considered. The impact of the credibility of defendants, expert witnesses, and the attorneys often influences jury decisions. Whether intended to support either a plaintiff's or defendant's position, CAD was neither advanced by testimony nor considered tantamount to the determination of the standard of care. Rather, as this case illustrates, its impact on that determination may nonetheless carry influence in supporting a defense expert's position that nonrecall in the finding of some calcifications is reasonable. What is unclear is whether a jury, partial to a given expert position, would change that perspective based primarily on CAD results.
CAD results were determined by a trial court and a court of appeals to provide relevant, probative, and reliable evidence for a jury to consider when determining the standard of care applicable to interpretation in the context of not recalling a mammographic finding. The version and particular CAD program used in this case may be different from other programs or newer versions. Improvements in CAD technology will struggle to balance sensitivity and specificity. Programs that emphasize one over the other may influence the results of retrospective review. Rather than the year in which they were developed, the emphasis on sensitivity or specificity of one algorithm versus another may invite competing results that could, if admitted into evidence, compete in establishing the credibility of expert testimony. This notion cannot be resolved now, but reinforces the subordinate manner in which CAD results should be seen with respect to specific expert testimony under current circumstances.
CAD's introduction for the plaintiff may be viewed differently. The negative predictive value of the absence of a CAD mark for calcifications may be more powerful than the positive predictive value for the reasons discussed regarding false-positive marks. In fact, given the disproportionately high number of false-positive marks, unless CAD is used at the time of the initial interpretation, subsequent analysis producing positive results are arguably not reliable or probative by the previously stated criteria.
This issue, however, has yet to be addressed in the legal setting. It should be noted that this argument applies only to calcifications—not masses, architectural distortions, or neodensities—for which CAD results have not been as successful [17]. Finally, appellate courts, while deferring to one another, sometimes reach different conclusions and the admissibility of CAD results as a universally accepted science remains subject to future decisions.
The patient's estate declined to release the actual films for publication. We sought formal legal opinion to see if we could use them without direct consent and were counseled we could not (the public domain aspect of the mammography pertains only to the trial). However, the description of the case is both accurate and a matter of public record [17], and the issues for consideration by the practicing radiologist do not require a review of the specific images. Indeed, if CAD fails to mark a spiculated mass that an expert witness persuades a jury is suspicious for malignancy, it is unlikely that absence of a CAD mark will have much impact. In like manner, if CAD marks an area that is subtle and nonspecific, then prior reports support the notion that the failure to recall such a case does not constitute a breach in the standard of care [8, 12]. Each litigation is specific to the facts of the case.
A majority of cases that reach trial are decided in favor of the defendant [3]. This is in part due to the plaintiff having the burden of proving by a preponderance of evidence that a breach of a standard of care has been committed. Assuming plaintiffs provide sufficient evidence to meet their burden, for cases involving nonspecific mammographic findings such as this one, the use of CAD alone may not provide sufficient evidence to rebut a plaintiff's allegations, but it does provide corroborative information that the defendant met the applicable standard of care.
In summary, a trial court admitted a retrospective analysis by CAD. A trial judge has the responsibility to admit only evidence that is reliable and probative. Over objection of plaintiff's counsel, the trial court in this case made the determination that CAD was reliable and probative under the circumstances in which the evidence was presented. Unpublished trial court rulings usually have no precedent-setting value. However, this ruling was affirmed by the Ohio Court of Appeals consisting of three appellate judges and therefore has authoritative value in that jurisdiction [15] and is considered persuasive authority in all other Ohio jurisdictions. Other appellate courts throughout the United States often look to the reasoning of other state appellate courts when cases are presented to them. It is only a matter of time before CAD's admissibility is tested in other state courts.
References
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  J. G. Elmore and R. J. Brenner The More Eyes, the Better to See? From Double to Quadruple Reading of Screening Mammograms J Natl Cancer Inst, August 1, 2007; 99(15): 1141 - 1143. [Full Text] [PDF]
|
|
|
|
 |
|
 B. F. Branstetter IV Basics of Imaging Informatics: Part 1 Radiology, June 1, 2007; 243(3): 656 - 667. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. J. Fenton, S. H. Taplin, P. A. Carney, L. Abraham, E. A. Sickles, C. D'Orsi, E. A. Berns, G. Cutter, R. E. Hendrick, W. E. Barlow, et al. Influence of Computer-Aided Detection on Performance of Screening Mammography N. Engl. J. Med., April 5, 2007; 356(14): 1399 - 1409. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. S. Greenberg An Appellate Court Ruling and Potential Implications for CAD Technology in the Courtroom Am. J. Roentgenol., January 1, 2006; 186(1): 52 - 53. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
