Women's Imaging
Original Research
Digital Mammography: Its Impact on Recall Rates and Cancer Detection Rates in a Small Community-Based Radiology Practice
Audio Available | 
Share
OBJECTIVE. The purpose of our study was to retrospectively evaluate the impact on recall rates and cancer detection when converting from film-screen to digital mammography in a small community-based radiology practice.
MATERIALS AND METHODS. Audits of our institution's records were compiled during a 4-year period: the year before conversion to digital mammography, and the first, second, and third years after conversion. We found that 4,838 film-screenings were performed during the period of audit 1 without the use of the digital system; 6,875 screenings were performed using the digital system during audit 2; 7,379 screenings during audit 3; and 7,294 during audit 4. Cancer detection and recall rates for each of the audits were determined retrospectively. Results were compared between audits using the chi-square test.
RESULTS. Cancer detection rates increased from 4.1/1,000 during film-screenings to 7.9/1,000 (p = 0.01) the year after the introduction of digital mammography. Recall rates also increased the year after digital mammography was introduced, from 6.1% to 10.2% (p < 0.001). Audit 3 showed no statistically significant difference in cancer detection rates from those in audit 1, decreasing to 5.1/1,000 (p = 0.42). However, audit 4 noted an increase in the cancer detection rate to 6.9/1,000.
CONCLUSION. In this community-based mammography practice, an increase in the cancer detection rate occurred initially during the conversion from film-screen to digital mammography, which subsequently decreased but remained higher than before digital conversion. This study suggests that the new technology alone is responsible for the increased number of cancers detected in patients with dense breasts that were not previously found using film-screening.
Keywords: breast cancer, cancer detection rates, digital mammography, recall rates
| Introduction |   |
|---|
Data obtained through the U.S. Food and Drug Administration indicate that an estimated 32% of mammography sites are digital, increased from 10% 2-3 years ago. Digital mammography offers considerable advantages over film-screen mammography [1-3]. Despite advantages, it has been slow to be adopted. This reluctance is due to many factors, including the high initial capital expenditure and the question of whether the added expense results in a better mammography “product” [4-7]. The reluctance to upgrade to digital mammography is especially true of small community-based imaging centers, where capital is less prevalent and patient volumes are lower than in larger metropolitan locations.
Despite these concerns, our desire to become digital and eliminate film completely led to our purchase of a digital mammography system in May 2005. At that time, we did not know whether the upgrade to digital mammography could affect the cancer detection or recall rates. As a result of our upgrade, we were required to perform a mid-year accreditation. In performing this accreditation, we noticed that our internal audit showed the cancer detection rate with our new digital equipment to be higher than it had been with our film-based equipment. To determine whether we were detecting more cancers, we decided to perform a more in-depth review of our cancer detection rates, analyzing cancer detection rates for individual radiologists and the facility as a whole.
To our knowledge, no prior study has compared cancer detection and recall rates at a single center before and after the installation of a digital mammography system, keeping the interpreting radiologists constant. Such a study would limit the number of uncontrolled variables, allowing potential outcomes to be mediated only by the introduction of the technology and the variability in the women undergoing screening.
In this article, we study digital mammography and its impact on recall and cancer detection rates for screening mammography in a small, community-based radiology practice.
| Materials and Methods | |
|---|
From July 1, 2004, through June 30, 2005, 100% of the mammograms at our center were performed on a DMR mammography film-screen system (General Electric Medical Systems). In May 2005, we began installation of a digital Senographe DS system (GE Healthcare). From July 1, 2005, through August 31, 2005, patients were imaged on both the DMR and the Senographe DS systems. By September 1, 2005, patients were imaged exclusively on the digital Senographe DS system. As a result, we performed a mammography audit that included the year of July 1, 2004, through June 30, 2005 (audit 1), our last year of film-screen mammography. The first year of digital mammography, when 100% of the screening mammograms were performed on a digital system, was from September 1, 2005, through August 31, 2006 (audit 2). To see whether there was an improved cancer detection rate using the digital system versus having found latent cancers that were lying undetected in women undergoing screening in a nondigital environment, we then performed a third audit from September 1, 2006, through August 31, 2007 (audit 3). Audit 4 included data from September 1, 2007, through August 31, 2008. We compared the third audit period with the second audit period. During audit 1, two of the radiologists who also performed screenings in audits 2 and 3 interpreted 92% of the screening mammograms. Each of these two radiologists had more than 20 years of experience in interpreting screening mammography. For audits 2, 3, and 4, a third radiologist, recently graduated and board-certified with only residency training in mammography, was added to our staff. Except for training in digital mammography as required under the Mammography Quality Standards Act, none of these physicians had had any prior experience with the interpretation of digital mammography. During audits 2, 3, and 4, these three radiologists performed 100% of the screening mammograms. All patients who presented with symptoms, along with the data that resulted from their diagnostic mammograms, were excluded from this study. This HIPAA-compliant study was performed with informed consent waived. The study was exempt from institutional review board review because it was a chart review of deidentified preexisting records.
During audit 1, we performed 4,838 screenings on the film-screen mammography system. During the second audit, 6,875 screenings were performed on the digital system. During the third audit period, 7,379 screenings were performed. Audit 4 included 7,294 screening examinations. All participants were women, and all imaging studies followed ACR (American College of Radiology) and ACS (American Cancer Society) guidelines for screening mammography.
The range of performance outcomes in screening mammography by practicing radiologists in the United States is well documented [8]. These performance benchmarks were used as a source for comparison with performance outcomes by mammographers in our study both before and after the implementation of digital mammography.
Recall rates and cancer detection rates between audits were compared using the chi-square test (2 × 2 contingency table) calculated using Minitab software, version 15 (Minitab, Inc.). A p value of less than 0.05 indicated a statistically significant difference.
| Results | |
|---|
Overall, 26,386 mammograms were obtained. We analyzed the cancer detection rate and recall rate during all four periods as both a facility total and for each individual radiologist. These results are summarized in Tables 1, 2, 3 and 4.
Overall, the recall rate was 5.9% (range, 3.3-8.9%) during audit 1 (before digital). The implementation of digital mammography resulted in a significant increase in the recall rate (p < 0.001). During audit 2, the overall recall rate increased to 10.2% (range, 4.0-13.7%). The recall rate for audit 3, 7.5% (range, 6.3-10.6%) was lower than that of the previous year but still significantly higher than that of audit 1 (p < 0.001). In audit 4, the facility recall rate was 9.0%. These values compare favorably with recall rates previously reported [8].
Overall, cancer detection during audit 1 was 4.1 cancers per 1,000 women screened, with a range of 4.0-4.9/1,000 screened. Surprisingly, and what has led to our desire to share these data, is that during audit 2, cancer detection unexpectedly increased to 7.9/1,000, with a range of 5.5-10.1/1,000 screened. This increase was highly significant when compared with the rate found in audit 1 (p = 0.012). Audit 3 showed a small overall increase in cancer detection rate compared with that of audit 1. However, the cancer detection rate in audit 3 (5.1/1,000) was not as high as that in audit 2 (7.9/1,000). When we compared this statistic from audit 3 with the detection rate found in audit 1, we discovered no significant difference between rates (p = 0.42). In other words, after the initial increase in detection rate following the introduction of digital mammography, the detection rate returned to a level that was higher than film-screen, but not statistically significantly higher. Interestingly, audit 4 shows a detection rate between the rates of audits 2 and 3 and elevated with respect to audit 1, although not quite to the point of statistical significance (p = 0.052). However, it is statistically significant when compared with national norms (p = 0.007). Further analysis of cancer detection rates in this coming audit period may shed light on the significance, or lack thereof, of this elevated detection rate.
As we stated earlier, our unique, small clinic environment allowed us to perform a statistical analysis of not just our facility as a whole, but also to analyze the analog-to-digital conversion, controlling for each radiologist. Looking at our cancer detection rate when controlling for radiologists 1 and 2, we have some very interesting and, we believe, expected results. When comparing the first postdigital year with the predigital statistics (audit 2 vs audit 1), we find that the combined cancer detection rate went from 4.5/1,000 to 7.1/1,000 (p = 0.10), and the recall rate went from 6.1% to 10.2% (p < 0.001). In audit 3, the second postdigital year, the cancer detection rate went back to 4.6/1,000 and the recall rate went back down to 6.4%. In comparison with audit 1, there was no significant difference in the cancer detection rates or in the recall rates in audit 3 (p = 0.92 and p = 0.60, respectively). Although there is an elevation in cancer detection rates between audit 4 and audit 1, this value is not statistically significant (p = 0.36). We did find, however, that between audit 4 and our predigital rates, a highly significant elevation occurred in recall rates, both for our facility (p < 0.001) and when controlled for radiologists (p < 0.001). These numbers suggest that once the latent cancers are detected in the population at risk, the cancer detection rate decreases toward a rate comparable with the incidence of disease. However, consistent with previous suggestions in the literature, a significant elevation in recall rates is maintained [6].
Radiologist 3 is clearly the outlier, showing a high cancer detection rate (10.1/1,000) and a high recall rate (12.1%) in audit 2. During audit 3, the cancer detection rate of radiologist 3 decreased to 6.7/1,000 and the recall rate dropped to 10.6%. However, in audit 4, radiologist 3 had an increase in cancer detection rate to 9.6/1,000 with a concomitant decrease in recall rate to 8.1%. In an attempt to improve their own diagnostic skills, radiologists 1 and 2 have begun to review the screening mammograms interpreted by radiologist 3 as positive that have resulted in a cancer diagnosis.
Our facility cancer detection rate did not entirely return to “normal” in audit 3 or audit 4. As a matter of fact, it increased to 6.9/1,000 in audit 4. It is possible that a slightly higher cancer detection rate is caused by a high percentage of patients with dense breasts, 40- to 50-year-old women, or premenopausal women. However, we do not have the statistics to support this premise.
| Discussion | |
|---|
Research regarding the comparison of digital and analog mammography has been performed primarily with large multicenter trials [2, 8]. Furthermore, to our knowledge, there have been no studies conducted that have looked prospectively or retrospectively at the impact of converting from analog mammography to digital mammography in a small community setting controlled for readers. This study shows how digital mammography, when implemented in a small community-based imaging center and controlled for radiologists, can result in a clear, statistically significant increase in cancer detection rates and an initial increase in recall rates.
When we opened our digital mammography service in May 2005, no other digital systems had yet been installed in our county. The nearest digital systems were more than 100 miles away. We implemented an aggressive marketing campaign, and, as a result, our screenings increased 42%, from 4,838 screenings in 2004-2005 to 6,875 screenings in 2005-2006. Most of our new patients had undergone their screening examinations at other centers in the area that used film-screen mammography. We had no new patients during the period of audit 2 who had previously been screened using digital technology.
Cancer detection rates for the first year of digital mammography (audit 2) showed significant increases in detection rates when compared with the year before implementation of digital mammography (audit 1). This finding is consistent with our premise that digital mammography increased the number of breast cancers detected by radiologists during the year after its implementation, because the expected incident cancers were discovered along with a number of breast cancers not found with film-screen technology but detected using superior digital technology. Audit 3 showed no significant difference in cancer detection when compared with the national norms (p = 0.572) or with the predigital period, suggesting that the increase in detection the year before was a result of the introduction of the new technology [8] and having a screening population with predominantly dense breasts. It is possible that a slightly higher cancer detection rate is due to a high percentage of patients with dense breasts or an influx of patients with dense breasts or an influx of other high-risk patients (e.g., 40- to 50-year-old women or premenopausal women). However, we do not have the statistics to support this premise. Interestingly, this slightly higher cancer detection rate continued to persist during audit 4.
The percentage of callback increased from audit 1 to audit 2 (after the installation of digital mammography). This finding is not surprising and is concordant with the experience of most others who convert to digital mammography [1]. The new technology allows greater visibility through dense breast tissue, allowing radiologists to visualize more lesions, and as a result, more screening patients are recalled. Contributing to the increase in callback rate is the fact that reference mammograms were almost exclusively film-screen.
One shortcoming of our study, because of the limitations of our mammography reporting and tracking software, was our inability to extract information on the overall breast density of our patient population. As a result, we are unable to determine whether our increased cancer detection rate is due to a higher than normal population of patients with dense breasts. Also, our database does not allow us to detect how many cancers were detected in patients we saw for the first time during audit 2.
We believe this study is important because it shows the superiority of digital imaging over film-screen. It is also important because it is the only study of which we are aware that confirms a transient increase in callback rate during the film-to-digital transition. We hope that our results will stimulate other small community-based imaging centers and practices to perform and publish similar audits so that more data on the experience of conversion to digital mammography can be shared with the radiology community. We also hope that this study encourages others to have databases more sophisticated than ours so that more detailed information may be extracted. Before performing the audits for this study, we did not recognize the importance and utility of having such detailed information at hand. As a result, we are in the process of upgrading our reporting and tracking systems to allow us to perform more detailed analysis on our performance.
Address correspondence to F. S. Vernacchia ([email protected]).
| References | |
|---|
