Incidental Findings in Imaging Research:  Evaluating Incidence, Benefit, and Burden FREE

An incidental finding (IF) in human subjects research is defined in a major consensus project as an observation “concerning an individual research participant that has potential clinical importance and is discovered in the course of conducting research, but is beyond the aims of the study.”^1(p219) Numerous reports have detailed how the detection of an IF can result in the early beneficial diagnosis of an unsuspected malignant neoplasm or aneurysm.^2- 5 However, others describe harm and excessive cost resulting from treatment of radiographically suspicious IFs.^6- 7 Moreover, clinical experience dictates that many IFs are of indeterminate clinical significance and generate uncertainty among both research participants and their physicians.⁶

At our institution alone there are approximately 4000 imaging examinations a year performed solely for research purposes. However, depending on the research study and institution, established mechanisms for handling imaging IFs may vary significantly or might not exist.⁸ For example, imaging data may or may not be evaluated in a timely manner or by a trained radiologist,^5,8 potentially resulting in the failure to offer a life-saving intervention early in a disease process.

Contributing to the wide variation in IF management is a lack of data available with which researchers, radiology departments, institutional review boards (IRBs), and institutions can estimate the expected frequency of IFs and consequences of different management strategies. Although the frequency of IFs has been well documented in a few specialized areas such as computed tomographic (CT) colonography and functional magnetic resonance (MR) imaging of the brain, other imaging modalities (eg, molecular imaging, ultrasonography, and plain film radiography) commonly used in clinical research are less studied. Even less is known regarding the subsequent clinical action and ultimate medical benefit or burden that may result from routinely identifying and disclosing imaging IFs.

The purposes of this study were to assess the frequency of IFs arising from multiple research imaging modalities at a large medical institution, to estimate the frequency with which clinical action is taken as a result of discovering an IF, and to attempt to examine the medical benefit or harm to human participants from identifying and working up IFs in imaging research.

Since 2003, it has been the policy of the Mayo Clinic Department of Radiology that imaging studies performed exclusively for research purposes are examined the day they are performed by a staff radiologist. This radiologist examines research images for IFs of potential clinical significance and dictates a report that appears in the patient's electronic medical record. The interpreting radiologist is expected to call the patient's primary care physician if an IF requiring immediate attention is identified. Further investigation and treatment is left to the discretion of the primary care physician and the patient.

This retrospective study was approved by the Mayo Clinic IRB. A database of research imaging examinations and their clinical reports was constructed from billing records of research imaging studies that listed IRB numbers. During January through March 2004 there were 1823 radiology examinations billed to research grants. Figure 1 details how 397 of these examinations were excluded, yielding a final study cohort of 1426 research imaging examinations.

An IF was defined as an observation noted in the dictated radiology report that was not directly related to the aims of the respective research study as listed in the protocol title. Comments about past surgical interventions, old injuries, nonpathologic anatomic variations, or normal line or pacemaker locations were not considered IFs.

Radiology reports were reviewed for each research examination, with all IFs, imaging modality (eg, CT or MR imaging), and body region being imaged (eg, head, chest, or abdomen) recorded to create 8 combinations of imaging modality and body region (Table 1). Medical records were available for all participants and were reviewed through February 2007 to obtain demographic data and information about any clinical action performed as a result of the IF. Specifically, actions that were recorded included further diagnostic imaging, referral to a subspecialist, diagnostic medical testing, invasive diagnostic procedures/biopsies, initiation of medical therapy, and surgical intervention. Descriptions of the clinical course of each IF were compiled for review by an expert panel.

An expert panel was assembled consisting of 6 physicians (including 4 radiologists [3 abdominal and 1 neurologic: J.G.F., J.D.P., E.G.M., and B.F.K.], 1 medical oncologist [H.C.P.], and 1 gastroenterologist [W.J.T.]) and 3 bioethics scholars (law, social science, and religious ethics [S.M.W., B.A.K., and M.E.R., respectively]). Physicians included a former IRB chair, the head of the IRB at our institution, and the chairs for Radiology Research and the Department of Radiology. The bioethics scholars all had significant experience with research ethics and included the director of the Mayo Center for Translational Science Activities ethics resource (B.A.K.) and the principal investigator of a National Human Genome Research Institute (NHGRI) study of IFs (NHGRI grant R01-HG003178). Because the NHGRI study investigator (S.M.W.) was not based at the Mayo Clinic, she did not participate in the review of any research participants' records or the ranking of individual cases.

The panel was charged with devising a ranking system for categorizing medical benefit/burden on a research participant imposed by clinical action resulting from an imaging IF. The panel spent considerable time creating a categorization scheme based on objective end points that could be deduced from the existing medical record. The panel determined that an objective marker to use was the initiation of medical or surgical treatment on the basis of an IF.^2,4 Medical burden/benefit rank was determined as follows: Clear medical benefit occurred when medical or surgical treatment was administered as a direct result of the discovery of an IF, with resolution or improvement in the disorder or disease. Clear medical burden occurred when medical or surgical treatment was administered as a direct result of the discovery of an IF with resulting mortality or morbidity or with lack of improvement in the disorder/disease. Potential medical benefit occurred when no treatment was initiated, but an improvement or resolution in a disorder could be realized in the future because of the knowledge of the IF. Similarly, potential medical burden occurred when no treatment was initiated but an adverse effect might occur if the IF were investigated clinically. Cases not fitting into these categories were rated as having “unclear” benefit/burden. Physicians on the panel were additionally asked to rate the medical gravity of an IF along a 5-point scale (from minimal/trivial to life-threatening) according to the original description in the radiology report (ie, not the final diagnosis). Psychological, social, and economic factors were not considered in creating this medical benefit/burden categorization given the difficulty of contacting research participants and assessing their subjective views regarding these factors.

Panel members independently rated medical benefit/burden. Rankings were considered in agreement when all members rated medical benefit/burden within 1 rank of each other, with the dissenting member given the opportunity to explain his or her perspective. Cases for which ratings were not in agreement were resolved by conference and consensus at a later meeting.

The number of IFs listed by staff radiologists in radiology reports was tabulated by imaging modality and body region. The number of examinations with at least 1 IF was also calculated. Multiple variable logistic regression was used to assess the association between the presence of an IF (the dependent variable in the model) and imaging modality and body region, adjusting for patient age and sex.

To assess the risk of an IF for any type of imaging examination relative to any other, odds ratios (ORs) were also reported for each pair of research imaging examinations. The significance level was set at .05 for statistical significance. The number of IFs generating further clinical action was also reported by imaging modality and body region.

A total of 1376 research participants underwent 1426 research imaging examinations from 91 different IRB-approved research protocols. Participants' mean age was 58 years (range, 3-97 years), with 690 male (48.4%) and 736 female (51.6%) participants.

Of the 1426 research participants, 567 had at least 1 IF (39.8%) for a total of 1055 IFs (284 examinations with multiple findings), with a subsequent IF to examination ratio of 1.86 (0.74 was the ratio for all 1426 examinations). Of research participants with IFs, mean age was 63 years (range, 3-97 years), with 251 male (44.3%) and 316 female (55.7%) participants.

The frequency of IFs reported varied widely depending on body region and research imaging modality (Table 1). Computed tomography of the abdomen/pelvis and thorax produced the highest percentages of imaging examinations with an IF (61% and 55%, respectively), providing an average of 1.29 and 1.16 IFs per research imaging examination. The most commonly encountered IFs stemming from CT of the abdomen/pelvis were aortoiliac calcifications (68 examinations), diverticulosis (25), and renal cysts (15). Ultrasonography and nuclear medicine scans infrequently produced an IF (9% and 4%, respectively).

Multiple variable logistic regression was used to assess associations between the odds of an IF and factors of interest (age [considered linearly], sex, and type of imaging study). Type of imaging study was significantly associated with an IF (P < .001). With ultrasonography used as the reference imaging study, each of the other examinations, with the exception of nuclear medicine examinations, was associated with a significantly higher odds of an IF (Table 2). The largest ORs were for CT of the abdomen/pelvis (OR, 18.9) and CT of the thorax (11.9). Table 3 reports the OR and 95% confidence interval (CI) for an IF for each pair of the 8 imaging studies of interest. Frequency of IFs was also compared between body regions for modalities infrequently resulting in IFs (ultrasonography, plain film radiography, and nuclear medicine). None of these comparisons was significant except for thoracic plain film radiography having increased odds of producing an IF relative to the extremity radiography (95% CI for OR, 1.2-5.6; P = .02).

Greater age was also significantly associated with increased odds of IF (OR, 1.5 per 10-year increase in age; 95% CI, 1.4-1.7). This increased risk translates into a 4.2% increase in the odds of having an IF per year of age (1.9% per year of age for MR imaging of the brain alone). There was no difference between male and female participants in the likelihood of IFs (male relative to female participants: OR, 1.04; 95% CI, 0.8-1.3]).

A second multiple variable model was also considered that included categorized age (<40, 40-64, and ≥65 years). Results again were that greater age was significantly associated with a greater odds of IF; age 40 to 64 years (relative to <40 years) had an OR of 4.1 (95% CI, 2.5-6.8) and age 65 or greater (relative to <40 years) had an OR of 9.7 (5.8-16.3).

Of the 1426 research imaging studies examined, 35 research participants (2.5%) (8 male and 27 female participants; mean age, 57 years; range, 31-87 years) received further clinical action based on an IF (Table 4).

Of these 35 research participants, 32 received follow-up imaging for a total of 71 additional studies (CT in 35, MR imaging in 12, ultrasonography in 23, and positron emission tomography in 1), and 16 underwent additional tests or procedures. Twenty-seven were referred by their primary care physician for subspecialty consultation. Five research participants underwent noninvasive diagnostic medical tests (serial cancer antigen 125 levels, dexamethasone and catecholamine levels, pulmonary function tests, fungal serologic studies, and coagulation tests), and 6 underwent invasive diagnostic procedures (bronchoscopy in 2, biopsy in 2, fine-needle aspiration in 1, and flexible nasopharyngoscopy in 1). Eight research participants underwent surgery for an IF, 2 underwent radiofrequency ablation (renal cell carcinoma in 1 and carcinoid liver metastasis in 1), and 2 received medical treatment (antifungal agents in 1 and antitussive drugs in 1).

Although many imaging modalities and body regions were found to generate large numbers of IFs, only chest CT, abdominopelvic CT, all other CT, and MR imaging of the head yielded IFs that received further investigation. No clinical investigation resulted from IFs from nonhead MR imaging examinations, ultrasonography, plain film radiography, or nuclear medicine. Abdominopelvic CT had the most IFs receiving action, with 19 acted on among 207 examinations (9.2%) (Table 1). The most frequent IFs receiving further action were ovarian/adnexal masses (n = 9) in the abdomen/pelvis and indeterminate lung nodules (n = 5) in the chest.

Medical burden/benefit and gravity of disease for each IF generating subsequent action are reported in Table 4. Six cases were found to be examples of clear medical benefit (rib osteomyelitis, renal cell carcinoma [Figure 2], small-bowel carcinoid tumor, sphenoid sinus Aspergillus colonization, ovarian mucinous cystadenoma, and grade 2 ependymoma [Figure 3]), with a mean gravity of disease score of 4.0. Twenty-four cases received an evaluation of unclear medical benefit/burden, whereas only 2 cases received the designation of potential medical burden. One-third of participants (n = 8) with unclear benefit/burden underwent serial cross-sectional imaging of their IF, and more than half (n = 13) underwent consultation with a specialist. Three cases were found to represent clear medical burden to the patient, with a mean gravity score of 2.3. These included suspicious mesenteric nodules that were found to be benign reactive lymph nodes at laparoscopy (Figure 4), an ovarian mass found to be a physiologic cyst at laparoscopy, and an adrenal mass found at surgery to be an adrenocortical adenoma. No deaths or postsurgical complications occurred.

In this study, 40% of research imaging examinations had at least 1 IF, with the risk of an IF increasing with age. Of these, a small minority of IFs (2.5%) eventually resulted in subsequent clinical action. Moreover, we found that some imaging modalities (eg, CT and MR imaging) and body regions (eg, thorax, abdomen/pelvis, and brain) generated significantly more IFs and were more likely to be acted on.

Despite the fact that clinical pursuit of most IFs was of unclear benefit/burden, 6 research participants (1.1% of all examinations with an IF) were believed to have clear medical benefit resulting in curative treatment of tumors and infections. In contrast, 3 participants (0.5% of all examinations with an IF) underwent laparoscopy or surgery for benign disease but experienced no long-term morbidity.

The body of literature concerning IFs in imaging research originates predominantly from work within CT colonography and functional MR imaging studies of the brain^2,4,8- 17 but includes studies of structural MR examinations of the head,^18- 19 body,²⁰ and others.²¹ In contrast, the present study estimates the frequency of IFs across all research imaging modalities at 1 institution, differentiating by imaging modality and body region scanned, and includes an attempt to assess resulting medical benefit or burden to the research participant.

Nonetheless, some of our observations can be compared with those of other studies that have examined individual imaging modalities. For example, at least half of asymptomatic participants have an extracolonic finding on CT colonography, but only 6% to 8% have extracolonic findings of potential medical significance,^{2,4- 5,9,15,22} paralleling our observation for all types of research CT of the abdomen/pelvis (Table 1). Overall, 0.6% (9 of 1426) of imaging examinations in our study were followed by surgery and/or radiofrequency ablation because of an IF compared with 0.2% to 1% of asymptomatic subjects participating in CT colonography studies^2,13,15,22 and higher rates in symptomatic patients or those with greater radiation dose and intravenous contrast.⁵

Previous studies examining IFs in brain imaging demonstrate a large range of IFs, from an incidence of about 20% for MR imaging performed on younger patients^16- 17 to 47% for MR imaging performed on older adults (mean age, 47 years).⁸ In the present study, among participants with a mean age of 63 years, 43% of all examinations had at least 1 IF. In these previous studies, 3% to 8% of brain MR images resulted in further clinical action compared with 2.2% in our study.^8,16- 17 In addition, we found an age-related increase in the likelihood of having an IF on brain MR imaging of 1.9% for each year of age, similar to others.¹⁸

This study demonstrates that research imaging IFs are common in certain types of imaging examinations, potentially offering an early opportunity to diagnose asymptomatic life-threatening disease, as well as a potential invitation to invasive, costly, and ultimately unnecessary interventions for benign processes. It should be noted that the clinical investigation of a suspicious IF that results in a benign diagnosis is not necessarily without clinical value or avoidable when presented with potentially life-threatening consequences (eg, a solid renal mass). Nevertheless, the majority of IFs seem to be of unclear significance. These instances represent a dilemma for researchers.^8,16- 17 As a result, clinical evaluation and serial imaging is often the course of action, with the participant, research grant, or medical insurance left to cover the cost. Even despite imaging surveillance, the final diagnosis often remains uncertain. Implications of uncertain IFs include emotional and financial distress in the research participant, clinical confusion for medical providers, and unplanned consequences for researchers.

Clinical researchers and IRBs may struggle with determining how to handle and plan for imaging IFs. This study attempts to inform research protocol design by supplying frequencies of IFs by imaging modality, body region, and age, but also the proportion of these findings that were considered worrisome enough to warrant further investigation. For example, 9.2% of all abdominopelvic CT examinations in this study prompted further investigation or treatment. In contrast, plain film radiography generated sizable numbers of IFs (39% had IFs), but none were followed by subsequent clinical action. It may be reasonable to devote fewer resources to potential IFs when the expected frequency of potential benefit is extremely low (eg, generating Sharp scores of rheumatoid arthritis from radiographs of the hand). Conversely, provisions should be developed for detection, disclosure, and follow-up of IFs when the expected frequency is high (ie, abdominopelvic CT and brain MR imaging).

Currently, management protocols for research IFs vary.²³ In some instances, this variation may raise ethical questions. For example, at some institutions, not all research images are evaluated for IFs by a trained radiologist. Functional MR images are often read by PhD or even non-PhD, non-MD researchers.²³ Even when a radiologist is included as a member of the investigative team, images may not be reviewed in a timely manner, thereby possibly losing opportunities to intervene early in a disease process. This study and recent recommendations set forth by the National Institutes of Health–supported Working Group on Managing Incidental Findings in Research should aid institutions and researchers in evaluating and drafting their IF policies.¹ When research imaging is performed within a department of radiology, a review of the images by a trained radiologist usually requires minimal time; however, significant practical hurdles may exist, such as those encountered with smaller institutions, stand-alone research imaging facilities, transmission or archiving of images, no standardized reporting system, and unclear responsibilities and communication mechanisms.

Finally, it should be recognized that research participants may overestimate the potential benefit that research imaging studies may offer. One study observed that, even when research participants knew that their images would not be reviewed by a physician, they still expected that if a brain abnormality existed it would be detected.²³ In contrast, experience demonstrates that few research participants appreciate the risk of anxiety, cost, and possible surgical morbidity or mortality for benign disease from radiographically suspicious IFs. Consequently, the research consent process should outline risks related to the identification of an IF when the expected frequency/potential benefit is high.

Our study has a number of limitations. (1) Variability exists among radiologists in selecting and classifying IFs. (2) We did not include an assessment of emotional/mental/anxiety-generating burden because this was thought to be highly subjective and impractical in a retrospective study. However, although not measured in this study, we recognize that a strong emotional impact does exist for research participants who have an IF (particularly if clinical consultation or serial imaging is recommended) and that this important factor affects the potential benefit/burden of research imaging. (3) The number of IFs that resulted in clinical investigation may be underestimated because some research participants may have pursued investigation outside our institution. However, our practice of dictating a clinical note detailing IFs the same day the research examination is performed and personally contacting the primary care provider in instances of life-threatening findings minimizes this likelihood. (4) Our attempt at assessing medical benefit/burden with an expert panel necessarily has the weaknesses of retrospective assessment with time-limited follow-up, categorization based on subjective benefit/burden definitions (particularly for the “potential” category), and clinical opinion. (5) We did not perform a cost analysis. (6) For cases classified as showing clear medical benefit, it is impossible to know with certainty how an earlier diagnosis precipitated by research imaging compares in final outcome with diagnosis at a later time precipitated by a natural onset of symptoms. (7) Our investigation occurred during a 3-month period and was limited to active investigations during that time interval.

Results of this study demonstrate that specific imaging modalities, body regions, and advanced age increase the likelihood of generating an IF during the course of imaging in clinical research. Research imaging studies at high risk for generating IFs can be identified. These data should inform researchers, radiology departments, and IRBs about the risk of an IF and subsequent clinical action and can be used in creating management plans for research imaging IFs.

Timely, routine evaluation of research images by radiologists can result in identification of IFs in a substantial number of cases that can result in significant medical benefit to a small number of patients.

Correspondence: Joel G. Fletcher, MD, Mayo Clinic, 200 First St SW, Rochester, MN 55905 (fletcher.joel@mayo.edu).

Accepted for Publication: February 26, 2010.

Author Contributions: Dr Fletcher had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Orme, Fletcher, Tremaine, McFarland, Koenig, and Wolf. Acquisition of data: Orme, Fletcher, Siddiki, Port, and King. Analysis and interpretation of data: Orme, Fletcher, Siddiki, Harmsen, O’Byrne, Port, Tremaine, Pitot, Robinson, and Wolf. Drafting of the manuscript: Orme, Fletcher, Port, and McFarland. Critical revision of the manuscript for important intellectual content: Orme, Fletcher, Harmsen, O’Byrne, Port, Tremaine, Pitot, Robinson, Koenig, King, and Wolf. Statistical analysis: Harmsen and O’Byrne. Obtained funding: Fletcher and Wolf. Administrative, technical, and material support: Orme, Fletcher, and McFarland. Study supervision: Fletcher and Port. Review of ethics content: Koenig.

Financial Disclosure: Dr Wolf reports owning stock in Open MRI of Amsterdam.

Funding/Support: The work of Drs Orme, Fletcher, and Koenig and Ms Robinson on this publication was made possible by grants SP 1 TL1 RR024152 (Drs Orme and Fletcher) and 1 UL1RR024150 (Dr Koenig and Ms Robinson) from the National Center for Research Resources (NCRR), a component of the National Institutes of Health (NIH), and the NIH Roadmap for Medical Research. Information on the NCRR is available at http://www.ncrr.nih.gov/. Information on Reengineering the Clinical Research Enterprise can be obtained from http://nihroadmap.nih.gov. Dr Wolf's work on this article was aided by NIH NHGRI grant R01-HG003178 on “Managing Incidental Findings in Human Subjects Research” (Dr Wolf, principal investigator). Dr Wolf did not have access to the Mayo Clinic case data analyzed herein.

Disclaimer: The contents of this article are solely the responsibility of the authors and do not necessarily represent the views of NCRR, NIH, or NHGRI.