Imaging

Sensitivity

Sensitivity describes how often the imaging test would give a “positive result” in a patient with cancer (i.e., true-positive finding). Ideally, the test precisely detects and locates one or multiple cancers in a given patient. Thus

Sensitivity can be calculated on a per-patient basis or a per–malignant lesion basis. The per-patient basis most commonly is used in screening studies for early diagnosis, whereas the per-lesion basis may be used in patients expected to have multiple sites of tumor. Per-lesion detection analyses can be misleading because they can be heavily biased by a single patient's results if that patient has multiple tumor foci.

It can sometimes be difficult to judge how “good” a test is by reading the literature. Sensitivity is supposed to be substantially independent of study population composition, but as discussed in the next section, certain imaging tests may be insensitive for some very early-stage disease, but very, very sensitive for more advanced disease. Each imaging test has a limit of detection threshold below which tumors cannot be detected because they are not distinguishable from the background tissues. Thus the patient population and very often the tumor burden and average tumor size can make a difference in the sensitivity of a test for detection of cancer. Virtually all noninvasive imaging tests are less sensitive for small-volume disease than for large-volume disease. For example, if an imaging test is used in a patient population in which patients have advanced disease before seeking medical attention (e.g., they are symptomatic at presentation), the imaging test may have far greater sensitivity than if it were used in patients with earlier-stage, smaller tumors. For example, positron emission tomography (PET) with Fluorine-18 ( ¹⁸ F)–fluorodeoxyglucose (FDG) has been reported to be more than 90% sensitive for detecting metastatic melanoma, but it is less than 20% sensitive in detecting early (low tumor volume) nodal metastases of melanoma at initial surgical resection. Mammography has higher sensitivity in women with more radiolucent than radiodense breasts. A test with high sensitivity has a low number of false-negative results. The false-negative fraction usually is expressed as 1 – sensitivity (in this case, sensitivity being rated on a 0–1 scale).

Specificity

Specificity is the frequency with which a test result is negative if no disease is present, or the true-negative ratio. As a percentage, specificity is

Specificity can be calculated on a per-patient, per-lesion, or per-region basis. The per-patient calculations commonly are performed in the screening setting. They also can be done per region of the body (e.g., Is the liver free of tumor? Are the draining lymph nodes free of tumor?). It is technically difficult and sometimes impossible to know exactly how many tumor foci are present because this depends on the reference gold standard. It is not possible to perform “whole-body” biopsies antemortem, so some very small tumor foci may not be known to be present when disease is diagnosed. Specificity can be affected substantially if the imaging test is used in a population that has a characteristic that can result in false-positive results for the imaging test. For example, inflammatory and infectious lung disease, such as active tuberculosis (TB) or sarcoidosis, if present in a patient population, can result in false-positive findings on PET or computed tomography (CT) scans or other imaging methods. In this situation, the specificity of FDG-PET, and likely of CT, for staging the mediastinum for cancer would vary. Thus the specificity of PET for assessing mediastinal lymph nodes may be much lower in areas of the world with endogenous TB than in developed areas without it. Therefore an imaging test that is very useful in one part of the world may be far less useful in another part of the world. A highly specific test has a low frequency of false-positive results (i.e., a low frequency of positive test results in the patient population that does not have the disease). The ideal imaging test has both high sensitivity and high specificity, although none of our current imaging tests have perfect sensitivity and specificity.

Screening Concepts and Challenges

Screening programs for cancer often have taken the form of laboratory tests such as the Papanicolaou (Pap) smear, or, more recently, blood tests for tumor markers. The success of the Pap smear in reducing mortality rates from cervical cancer is incontrovertible. The use of imaging in screening for cancer is an example of success and is of considerable interest, but also is a source of considerable controversy. As discussed in detail in the chapters on breast cancer and lung cancer, screening mammography programs have been shown to be capable of saving lives in women older than 50 years. These programs also may save lives in women 40 to 50 years of age, but the data are less compelling.

Studies have been initiated in which CT scanning is used in an attempt to detect early lung cancer. Screening high-risk populations with CT imaging was proven to reduce lung cancer–specific mortality in the National Lung Screening Trial (NLST). This followed results from the Early Lung Cancer Action Project (ELCAP), a large study screening patients at increased risk of lung cancer with low-dose CT, which reported promising results in 1999. The ELCAP showed that lung cancers are detected at a smaller size and that patients whose cancers are detected by screening live longer following diagnosis than patients whose tumors are not detected by screening. Whether this translates into longer-term survival for the screened population remains unclear. The ELCAP study further evaluated 31,567 asymptomatic persons at risk for lung cancer with use of low-dose CT from 1993 through 2005, and from 1994 through 2005; 27,456 repeated screenings were performed. A diagnosis of lung cancer was made in 484 participants based on screening. Of particular note, 412 patients (85%) had clinical stage I lung cancer. Ten-year survival approached 90% in this group. This study demonstrated that annual spiral CT screening can detect lung cancer that is curable.

The results of another large randomized trial of lung cancer screening, the NLST, were reported in 2011. In this trial, 53,454 persons at high risk for lung cancer were enrolled from over 30 US sites. Study participants were randomly assigned to undergo three annual screenings with either low-dose CT (26,722 participants) or single-view posteroanterior chest radiography (26,732); 24.2% and 6.9%, respectively, of the CT-screened and chest radiography–screened groups had positive screening studies at some point. There was a high false-positive screening rate: 96.4% of the positive screening results in the low-dose CT group and 94.5% in the radiography group. The incidence of lung cancer was significantly higher (1060 versus 941 cancers) in the low-dose CT group, as compared with the chest radiography group, There were 247 deaths from lung cancer per 100,000 person-years in the low-dose CT group and 309 deaths per 100,000 person-years in the radiography group, a relative reduction in mortality from lung cancer with low-dose CT screening of 20.0% (95% CI, 6.8–26.7; P = .004). The rate of death from any cause was reduced in the low-dose CT group as compared with the radiography group by 6.7% (95% CI, 1.2–13.6; P = .02).

This exciting trial has raised some controversies: the vast majority of small pulmonary nodules identified are not malignant, the costs of the screening and of the medical care for incidentally detected lesions are substantial, and a substantial radiation burden, which in principle could be carcinogenic, is delivered to patients undergoing the screening. However, there is considerable hope that screening programs can achieve a reduction in lung cancer mortality as the radiation is given to patients later in their lives, when risks of radiation-induced cancers are reduced. Indeed, the evidence supporting lung cancer screening with CT was of sufficiently high quality to be included as a covered benefit in the United States under insurance policies consistent with the Affordable Care Act (ACA).

Based on the findings of the NLST, and the approval by US Medicare of screening, many centers have offered lung cancer screening clinics for high-risk patients. A risk, however, for lower-risk patients is that screening may have an unacceptably high false-positive rate. It is important to note that the promising results of NLST are for a higher-risk population, and extrapolations of benefit to lower-risk populations may be highly problematic.

As discussed later in this chapter, CT colonography holds excellent promise as a screening method for colon cancer, as an alternative to optical colonoscopy, and is increasingly being paid for by health insurers based on its accuracy.

Other areas in which screening by noninvasive imaging has been performed include colorectal cancer, where virtual colonoscopy can be used to look for early colon cancers, and in the pelvis in women who are at risk for ovarian cancer. More recently, screening CT centers offering a virtual evaluation of the entire body have become available, and even more recently, magnetic resonance imaging (MRI) and PET screening have been offered in some locales. The growth of these centers has been driven emotionally and economically; these tests are not yet well founded scientifically in the screening setting. In some cultures, screening imaging is done more commonly than in others. The risk of false positives in patients at low cancer risk is substantial.

Screening carries challenges, risks, and costs that are beyond the scope of this overview chapter. However, several key points apply to all screening approaches, including those using noninvasive imaging. These points include (1) whether a screening program is reasonable to consider; (2) lead-time bias; (3) length bias; (4) the overall economic cost implications of screening, especially the costs of investigating false-positive results, and (5) the risk that radiation used in screening may include subsequent cancers.

The requirements that a screening program must meet to be considered “reasonable” may differ substantially based on the specific society's values and a specific individual's perception of risk. However, in general, the following characteristics are important for cancer screening:

• The cancer must have a considerable public health effect.

• The disease must have an asymptomatic period in which detection by imaging is possible.

• A therapeutic intervention that should lead to better survival or quality of life must be available.

• The prevalence of the disease must be sufficient in the population being screened to justify screening (especially the cost). Low prevalence of disease lowers the positive predictive value of positive scans.

• Medical, surgical, or other treatment must be available for the early-stage cancer identified by screening.

• The screening test itself must not cause disease at a significant rate.

• There must be a high likelihood that the patients in whom early cancer is identified by image-based screening will go on to undergo a suitable therapeutic intervention.

Furthermore, the imaging test itself must be acceptable to patients (in terms of level of discomfort, cost, and radiation burden), and it must be sufficiently sensitive to identify cancer often and sufficiently specific to minimize false-positive results. Finally, the costs of the screening process, and attendant costs related to false-positive results, must be compatible with the society's or the individual's economic resources, and the screening procedure must pose little or no risk to the patient.

Another important consideration in screening programs is lead-time bias. This concept, simply stated, indicates that if the natural history of a disease is unchanged but the diagnosis is made earlier in the course of the illness, the apparent survival will be improved. For example, let us assume that a tumor has a 6-year natural history from its beginning until the death of the patient, and that treatment is ineffective. The disease might become clinically detectable after 4 years and lead to death in 6 years, a 2-year survival after diagnosis. With screening, if the tumor is detected 3 years after the onset of disease and no improvement in treatment occurs, then the survival in the screened population would appear to increase from 2 to 3 years after diagnosis. This illusion of improved survival in the screened population is a considerable concern and can lead to inappropriate enthusiasm for screening programs.

Another important consideration in screening is the possibility of length bias. This is a more complex concept, but it may be related to the types of cancer that can be detected by screening programs. A possibility is that very rapidly growing and presumably highly lethal cancers are less likely to be detected by annual screening programs, whereas more slowly growing cancers, which have an intrinsically better prognosis, may be detected more frequently by screening. In fact, some of the early cancers discovered by screening may not be biologically relevant at all. If so, the patients with cancers identified in the screened population could appear to have a better survival than the patients with cancers identified in the unscreened population. This concern exists for prostate cancer screening by prostate-specific antigen (PSA), for example, wherein there are major concerns regarding overdiagnosis of slow-growing, and presumed indolent, cancers that might not need surgery. This concern also exists for breast cancer, wherein some argue that a small to a more substantial portion of the cancers diagnosed would not have proven harmful to patients had they been left undiagnosed and untreated. That said, we do not fully know which cancers need no treatment, and an issue existing with “overdiagnosis” is that it really is “overtreatment.” Better screening tests that identify biologically relevant cancers would be ideal.

A third factor is the selection bias that is difficult to control for in retrospective observational studies: how the patients and their referring physicians determined to have a scan performed. Selection bias occurs when there are unintended differences between the groups observed that, although they are associated with the variable used to sort the groups—for example, exposure in case-control studies and outcome in cohort studies—affect measurement of the study variable. For instance, in a case-control study of the effects on disease-specific mortality of a particular screening program, investigators would examine records from patients who have died from the disease in question versus those who have not, then determine the rates of the screening intervention in these two populations.

It is possible that those likeliest to have sought screening were the ones at highest risk for the disease. Results for the screening's effect on mortality could be underestimated in such a scenario. Selection bias also can work in the opposite direction, wherein the high-risk or poor-prognosis individuals are less likely to seek screening. With the relative absence of large randomized prospective trials, as is commonly the case when evaluating imaging as a screening tool, one may be tempted to base conclusions on the findings of cohort or case-control studies. However, the most accurate conclusions would be drawn from a prospective randomized study design.

Collectively, lead-time bias and length bias and, at times, selection bias can make screening programs appear to improve the survival of patients with cancer. Because of these major biases intrinsic to screening, very large, randomized, studies are required to show that overall cancer-specific mortality (and ideally mortality from all causes) declines as a result of screening programs.

All-cause mortality is a critical parameter. If the treatment of a presumed tumor discovered by a screening imaging test carries with it a risk of death or morbidity, the screening program could lower cancer-specific mortality but not all-cause mortality. If the screened population is very young, late adverse effects of screening may be difficult to detect, such as slightly increased risks of cancer due to radiation.

The advent of imaging and other screening tools that uncover a tumor long before it is symptomatic also brings up the need for biomarker discovery to help physicians to determine whether to treat what has been discovered through screening. The experience with PSA screening in prostate cancer serves to illustrate the point that not all cancers identified by screening eventually lead to death. This is likely also the case for some early-stage breast cancers of low proliferative rate and potential. Discovering and validating biomarkers that can help to distinguish reliably between lethal and nonlethal tumor types will be of great help in reducing unnecessary treatment in patients with less active disease. Identifying and phenotyping tumors with the greatest risk for progression is of great importance because it is clear that not all cancers carry the same risk of mortality if left untreated.

Screening Costs

The determination of whether a screening program is valuable to society is often, in part, based on its cost and benefit. The concept of quality-adjusted life-years (QALYs) often is applied. This concept is defined as the economic cost to society required to result in 1 additional year of good-quality life for a member of the society. In many Western countries, a figure of $50,000 has been considered a useful guide (although many affluent countries are willing to spend far more per QALY), with QALYs costing less than this amount considered cost-effective. Such a guideline, however, does not necessarily apply when individuals make their own determinations as to whether to pay for a screening test. For example, it is reasonable to expect that those with greater disposable income would be willing to pay more per QALY than those with less disposable income. Thus it can be difficult to generalize about the cost efficacy of screening procedures.

Even when people choose to undergo a screening test at their own expense, considerable costs can be transferred to society as a result of the screening program. For example, if a screening test has a high rate of false-positive results, a substantial number of follow-up biopsies or procedures will be performed and can cost a great deal of money. Such costs can dramatically raise the total cost per QALY. Invasive procedures also can increase the likelihood of morbidity or death as a result of additional investigations. To determine the true cost per QALY associated with screening, these additional costs and risks must be considered. Thus screening remains an area of great promise, but also of considerable controversy.

Screening beyond breast imaging must overcome major hurdles before it is likely to be accepted. There has been considerable progress for both virtual colonoscopy as an alternative to optical colonoscopy and for CT scanning of the lungs in patients at high risk for lung cancer. In both these populations, the Centers for Medicare and Medicaid Services (CMS) pays for screening, as do ACA-compatible private insurance policies. It is almost certainly the case, however, that in high-risk patient groups, such as those with a family history of cancer or major carcinogen exposure with a high penetrance or early onset of disease, screening may prove of greater value than in the general population. As an example, MRI screening in patients at high risk for breast cancer is increasingly accepted as appropriate.

Size of Detectable Lesions

Noninvasive imaging methods in humans cannot enable detection and localization of a single malignant cell, although flow cytometric methods are very sensitive for finding a very few cancer cells in a patient's blood, and there is a great interest in assays for circulating tumor cells and for tumor DNA and RNA in the blood. Imaging methods are improving, however, and detection of a much smaller number of cells is possible in small-animal models. It has been estimated that by the time a tumor reaches 3 to 5 mm in diameter, which is the lower limit in size for detection by the best current noninvasive methods in humans, the tumor has undergone more than 25 doublings and contains 0.1 to 1 billion cells, depending on their size. In contrast, a cytologist on a very good day or with use of flow cytometric methods may be able to identify a single cell as malignant by using a microscope.

Realistically, even for histologic assessment of malignancy, typically a group of tumor cells must be present before cancer is diagnosed. However, light microscopy and more sensitive techniques such as immunohistochemistry and polymerase chain reaction studies mean that pathologic techniques potentially will be more sensitive than imaging methods. One very important proviso is that for light microscopy pathologic methods to be effective, actual examination of the malignant cells is required; the sample containing the tumor must be cut appropriately and viewed under a microscope. This task may be impossible, because the 8-µm sections that are used for pathologic examination typically are taken only from a small portion of a tumor or lymph node, whereas most of the tumor or node will be unexamined (e.g., to assess a 1-cm node by using 8-µm-thick sections, approximately 125 slices would be required). This large number of sections is not typically obtained.

Despite the markedly superior sensitivity of histologic methods over noninvasive imaging, there is a major sampling error issue for histologic sampling, and, paradoxically, imaging in some diseases potentially may be more sensitive than histologic examination for cancer. This situation can arise in mammography, in which small tumor foci can be seen on the mammogram but can be missed on cytologic sampling, possibly because of sampling error. When tumors are imaged with noninvasive methods, the entire tumor is visualized, not just a small portion. So, paradoxically, imaging, despite limited resolution, can be more sensitive than cytology or pathology. However, if exhaustive and thorough sampling is performed, with microscopic examination of tissue there usually is greater sensitivity for tumor than there is with current noninvasive imaging techniques.

With the advent of serum protein tumor markers, circulating tumor cell assays, and DNA- and RNA-based methods, it is possible that imaging will not always be a primary tool for cancer detection. These other methods may be more sensitive than imaging, although imaging has the distinct advantage of both detection and physical localization of tumor foci. In addition, noninvasive imaging can detect heterogeneity in cancers that cannot be identified with circulating cells or DNA. Such information is potentially more “actionable” medically than simply detecting whether tumor is present or absent.