Background
In Mexico, since 2006, breast cancer is the leading cause of death from cancer in women [
1] with epidemiological and demographic transitions contributing to an increasing trend in rates. Approximately 90 % of breast cancer cases are diagnosed at an advanced stage, which in part is due to the lack of an organized screening program, limited access to mammography and shortage of radiologists to interpret screening mammography, particularly in rural areas [
2]. These conditions may contribute to delays in diagnosis and less successful treatment [
2‐
5].
The Mexican Official Norm for breast cancer (NOM-041-SSA2-2011) recommends mammography every two years for healthy women ages 40–69 years and provides the guidelines for the breast cancer national program to achieve greater coverage and quality [
6]. However, the results of the recent Mexican National Health Survey, showed that only 17.2 % and 29.4 % of women between 40 and 49 years and between 50 and 69 years, had a mammogram within the previous 2 years, respectively [
7]. Only 291 radiologists participate in the screening mammography program in Mexico, among whom 260 focus exclusively on breast imaging. With these current infrastructure and human resources, it will clearly not be possible to increase the coverage up to 70 % as suggested by the World Health Organization (WHO) [
8], given that there are close to 14 million women eligible for screening [
2,
3,
9].
Since the mid-1980s, there has been a growing literature suggesting that shortages of radiologists could be overcome, and costs reduced if radiographers (also known as radiologic technologists) [
10,
11], or mid-level practitioners (physician assistants or nurse practitioners) could interpret mammograms and serve as first readers, determining the presence or absence of abnormal images in cases that warrant further evaluation by a radiologist. [
12] In 1995 in the United Kingdom (UK), there was both a shortage and falling recruitment rates of radiologists, and therefore the possibility of radiographers as readers was explored in order to maintain double reading [
13]. Since then, several studies have demonstrated radiographers’ ability to identify abnormalities on screening mammograms [
14] albeit with higher false positive rates, but similar sensitivity compared with radiologists [
15‐
19]. Furthermore, trained radiographers have been shown to perform well as pre-readers in a clinical setting [
20].
To our knowledge there are no studies examining the potential for radiographers or physician extenders to contribute to mammography interpretation in Latin America. Given Mexico’s public health policy goals for breast cancer screening and the shortage of radiologists to interpret mammograms, we designed a study to evaluate the performance of 15 radiographers in the interpretation of mammograms after a 6 months training period in a screening setting. Additionally, using the results of the study, we modeled the effects on costs and performance of two hypothetical screening scenarios; both were compared with the status quo in which the standard practice is a single reading by a radiologist. We conducted this study to determine the feasibility of having radiographers as first readers, in order to consider a strategy for integrating non-radiologist readers into mammography screening in countries that, face the problem of shortages of radiologists with specialization and interest in mammography.
Discussion
Our study used a test set to evaluate the effect of a 6-month screening mammography interpretation training program for radiographers. The median sensitivity was 73.3 %, but was achieved at the expense of a high percentage of false positives (49.5 %) on a test set enriched with abnormal examinations with an expert recall rate of 26.4 % (29/110). Our results are consistent with other studies evaluating performance on consecutive series of patients or test set studies where sensitivity rates have ranged from 73 % to 90 % [
33]. Recall rates tend to be higher in test set situations than one expects to find in screening conditions [
34]. In addition, this test set was developed to be challenging, especially for the non-cancer cases. For example, the clinical sensitivity obtained from U.S. radiologists on the same films read in clinical practice, was 86.7 % (13/15) while the false positives rate for the test set films was 38.9 % (37/95) (personal communication: Andy Bogart BCSC Statistical Coordinating Center).
Among the advantages of our study was that training took place in a center far from the radiographers’ work environment, enabling them to dedicate themselves exclusively to learning to interpret mammograms. The test was conducted in a computer lab and we ensured that radiographers did not communicate with each other during the exam. The interpretations were sent directly to an external evaluator (BCSC) so that neither the researchers nor the radiographers knew the results of the assessment at the time of examination. In our study, we also measured sensitivity by treating recalled biopsy-proven benign cases as true positives when the abnormality was judged by expert radiologists to warrant recall (median = 75.9; IQR: 65.5–86.2; data not shown). This is a reasonable and clinically relevant approach for both radiologists and radiographers, since some screening exams, although eventually determined to be benign, must be recalled due to the suspicious nature of the abnormality. Since non-radiologist readers may be expected to interpret exams with a lower threshold for suspicion compared with radiologists, this approach may be even more appropriate for non-radiologists [
15,
35]. Finally, not knowing the number/proportion of cancers in the test set, or the fraction of non-cancers for which recall was expected, prevented the examination of cases with a “counting down” approach to the identification of abnormal exams.
Our investigation has some limitations. In many centers in Mexico, mammography examinations are performed using full-field digital mammography units and the radiographers were trained utilizing digital mammography images, while the test set was constructed with digitized analog images, which may have affected the radiographers’ performance. Radiologists who trained the radiographers may have varied in their ability to accurately interpret screening mammograms, such as it has been previously observed [
31,
36,
37]. The participants in this study achieved the stated performance levels with relatively low levels of overall training and experience in the field, compared with some other settings in which radiographers read mammograms. For example, in the U.K., radiographers are all initially trained to a bachelor’s degree standard, and to work in breast screening they must undertake a master’s level course of approximately one year’s duration including the reading of 1500 to 2000 mammograms with feedback [
38]. In contrast, an average Mexican radiographer studies 2–3 years after junior high-school, and rarely is exposed to curricula specifically related to breast cancer screening (e.g., only 3 out of 27 schools analyzed), which they typically will receive as on-the-job training once they begin working. Further, in the present study the radiographer read fewer mammograms (mean = 770; SD 174), and only received feedback in class on a subsample of the homework.
The radiologists’ specificity used for the mathematical model is not directly comparable with the radiographers’ specificity obtained in our study, given that the conditions from which the measures were derived were different. If the evaluation had been comparable, i.e., U.S. test set, radiologists likely would have a higher false positive rate, and would be more expensive (Scenario A, Radiologist would cost US $19, 061 total cost per month assuming a mean value of sensitivity = 73.3 and specificity = 53.7, data not shown vs. US $17, 019 (Table
5)). An important difference between scenarios A and C, the implications of which are not quantified in our results, is that scenario C could increase access to mammography screening by increasing the number of film readers, reducing the screening load of the radiologists, and providing greater time for the radiologist to devote to evaluating difficult and abnormal screening exams. In a setting where there are too few radiologists to achieve recommended screening goals, scenario C offers a potential solution. The model presented in this study does not provide sufficient evidence for the alternative scenarios, but provides a first estimate of how these scenarios would compare to usual practice. Further, we would expect that radiographers’ accuracy, both sensitivity and specificity, would improve with continuing experience and training, thus steadily improving the cost-effectiveness. Lastly, the accuracy of mammography interpretation by Mexican radiologists measured in an earlier study was based on films interpreted with a film viewer rather than digitized images using a computer screen as was used in our study, and no information is available about the breast cancer lesion types and difficulty in that earlier evaluation used for the Mexican radiologist’s evaluation [
31]. Finally, we had no baseline measure of performance for the radiographers and no comparison group, so we are not able to directly measure the effect of the training program. However, radiographers in Mexico do not have a formal role in the interpretation of mammograms, and given that was their first experience, we believe our results are a reasonable proxy of the effect of training.
Radiographers are good non-radiologist candidates for the interpretation of mammograms because of their considerable experience with breast imaging, professional dedication [
18], and because they work under the supervision of a radiologist. Sumkin et al. showed that even without undergoing additional training, technologists classified screening mammograms at a reasonable level of accuracy [
19]. In addition, when radiographers participate in the interpretation of mammograms it contributes to increased realization of the importance of producing high quality mammographic images [
18], and their satisfaction at work [
35]. Besides radiographers, other health professionals such as physician assistants, nurse practitioners, and general practitioners are worthy of consideration as candidates where there are shortages of radiologists, provided that they have adequate initial training and supervision, on-going training and evaluation, and can perform at pre-set target levels determined for the program. There also is evidence that radiologists [
39,
40], and other specialists [
41,
42], will accept other health professionals performing services that traditionally only have been performed by them if there has been formal training, and there are access problems, such as shortages of specialists in rural areas [
43,
44].
Compared with radiologists, radiographers or physician assistants have achieved similar sensitivity after initial training, although generally with higher false positive rates in screening settings and on test sets [
33]. A test-set likely still underestimates clinical specificity performance because the participants would have known that a recall decision in the test would not carry a cost (e.g., unnecessary procedures and psychological effects) for a real woman [
45]. Investigators have noted that it is realistic to anticipate that specificity would improve with additional training and experience to the equivalent of radiologists reading screening mammograms. [
12,
16,
46,
47] Evidence from mature programs that have included radiographers in the interpretation of mammograms, such as the U.K. National Health Services Breast Screening Program (NHSBSP), confirms that both sensitivity and specificity are similar among radiographers and radiologists [
48,
49]. Improvement in accuracy also has been observed in the learning curves of radiologists involved in breast imaging [
50].
Investigations focused on the ability of radiographers and other non-radiologists to interpret mammograms typically have taken place in settings where there was not an acute shortage of radiologists [
33,
51], although consideration of the potential for non-radiologists to play a role in the interpretation of mammograms usually has been motivated by affordability, anticipated personnel shortages, and the pressure of a growing number of women invited to screening due to demographic change and program expansion. In the U.K., for example, increasing workloads led to interest in training radiographers to reduce the time demands on radiologists while maintaining the programmatic commitment to double reading. Presently radiographers contribute to a significant fraction of screening interpretations in the NHSBSP, and the evidence indicates that there are no significant differences in the interpretative accuracy of radiographers and radiologists [
48,
49].
While pre-reading by radiographers has been proposed as an alternative to the interpretation of mammograms solely by radiologists in a screening setting [
17] it has not been supported by others [
12] due, in part, to the risk of missing lesions [
33]. However, it has to be acknowledged that radiologists also do not achieve perfect sensitivity in practice. Some false negatives are not visible in retrospect, and even the most skilled radiologist does not detect all breast cancers. To consider the potential for radiographers as first readers, they must be able to achieve similar, not necessarily superior, screening sensitivity in detecting cancers compared with radiologists, and the evidence consistently supports that with adequate training they do achieve that benchmark. Indeed, in some U.K. practices, radiographers are paired for double reading, and radiologists only interpret non-concordant exams [
51].
While the ability to achieve the same sensitivity as radiologists is important, there are numerous options to achieve that goal. After training, a radiographer could be paired with a radiologist or experienced radiographer in a program of double reading and periodic proficiency testing with enriched tests sets until program leaders were satisfied that the radiographer’s performance was reliable. To assure confidence in their performance, periodic proficiency testing could be required for a period of time after completion of training, and regular medical audits afterwards. A program could follow the U.K. model and have all exams double read by radiographers, with discordant interpretations referred by a radiologist. Alternatively, a program could accept lower specificity as a way to reach the goal of high sensitivity. Radiographers also could be entirely or initially limited to reading mammograms only from women without significant breast density, leaving more difficult cases for radiologists. Each of these options reduces the amount of radiologist time in the interpretation of screening exams, for which the large majority will be normal, while assuring equivalent accuracy. A skills-mix model such as this allows the physician to focus their time, which is scarce, on supervision, refereeing discordant cases, and diagnostic evaluation of abnormal test results and women who present with symptoms. Still, while scientific evidence and the U.K. experience leaves little doubt that radiographers can perform effectively as interpreters of screening mammograms, the process of their integration into a screening program requires adherence to high standards, and careful implementation in order to assure the confidence of policy makers, radiologists, and the public.
Combining the expertise and skills of both a radiographer and a radiologist in the interpretation of screening mammograms could be an efficient alternative to the traditional model where radiologists are responsible for all screening and diagnostic mammography, especially in a setting where there is a shortage of radiologists or where growing imaging needs will eventually exceed available specialty resources. Although the model does not provide sufficient evidence for other alternative scenarios, our results suggest that taking advantage of the high sensitivity of interpretations by radiographers and high specificity of radiologists could result in an efficient strategy for screening mammography. This would imply a different use of radiologists’ time and a more rapid delivery of positive results to patients by letting the radiographers taking care of obvious interpretations, and triage those that warrant evaluation by the radiologist. Improved training of radiographers and practice could improve these results so radiographers would not likely be generating additional procedures beyond what the radiologists would generate if they were reading as single readers.
While scientific evidence and the U.K. experience leaves little doubt that radiographers can perform effectively as interpreters of screening mammograms, the implementation of a mixed skills program faces numerous challenges. Costs must be considered in the design of training programs, including the potential for enhanced salaries. There also is the requirement for implementation of regulations regarding the additional radiographer’s responsibility to undertake mammographic image interpretation. In this study, the mathematical model to assess the costs and outcomes of screening mammography interpretation, by radiologists and radiographers, was based on cases in which abnormalities were detected. Going further, it would be desirable to perform a cost-effectiveness study to estimate the cost of breast cancer screening under different scenarios of personnel involved in interpretation with an emphasis on deaths averted from breast cancer or life years saved, which is the ultimate goal of screening. Where shortages of radiologists exist, there is a need to determine whether there is an adequate pool of qualified radiographers, and whether recruiting them to be readers would create personnel shortages of radiographers. There likely would be a need to determine the training needs and costs, and compare the performance of non-radiographers as interpreters. There also is the need to determine how many non-radiologists are needed, and the volume of examinations they would be expected to interpret. Above all else, the process of their integration into a screening program requires adherence to high standards, and careful implementation in order to assure the confidence of policy makers, radiologists, and the public.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
All authors made substantial contributions to conception and design, analysis, and interpretation of data, and critical review of the manuscript. GTM was involved in the study conception and design, conduction of the study, collection and assembly of data, data analysis and interpretation, manuscript writing and manuscript approval. RAS was involved in the study conception and design, evaluation design, provision of images for evaluation, data analysis and interpretation, manuscript writing and manuscript approval. MLCF was involved in the training program design and the training and evaluation of the radiographers. AB was involved in the evaluation design, provision of images for evaluation, collection and assembly of data, data analysis and interpretation, manuscript writing and manuscript approval. LMM was involved in collection and assembly of data, data analysis and interpretation and manuscript writing. DLM and KK were involved in the evaluation design, provision of images for evaluation, data analysis and interpretation, manuscript writing and manuscript approval. COO was involved in the study design, conduction of the study and the training program design. EMV was involved in the conduction of the study and the training and evaluation of the radiographers. AALl was involved in the study design, conduction of the study, and manuscript writing. SBA was involved in the designing of the mathematical decision tree model, analysis, interpretation and manuscript writing. GSG was involved in the designing of the mathematical decision tree model, analysis, interpretation and manuscript writing. OGMM was involved in the study conception and design and manuscript writing. SRUS was involved in the study design and manuscript writing. ELP was involved in the study conception and design, data interpretation, manuscript writing and manuscript approval. MHA was involved in the study conception and design, data interpretation, manuscript writing and manuscript approval. All authors read and approved the final manuscript.