Tradeoffs in Disease Screening.

I start today’s Dean’s Note noting, once again, that another gun-related tragedy, this time in San Bernardino, has claimed 14 lives, injuring 21 others. The event is made ever so more poignant by its hitting of public health professionals. This issue persists, remaining a preventable public health epidemic, a tragedy that we allow to happen. I have written Dean’s Notes on this, including last week’s Dean’s Note. I also wanted to note that in many ways this falls squarely in the rubric of the argument I made in another Dean’s Note where we continue to find gun-related morbidity and mortality acceptable, when it should be thoroughly unacceptable. What sadness to reflect on more lives lost when we can do so much better. It is perhaps a ray of hope amidst the darkness to read a front-page editorial in the New York Times saying, “It is a moral outrage and a national disgrace that civilians can legally purchase weapons designed specifically to kill people with brutal speed and efficiency. ” I could not agree more. Could this be a piece of the moral outrage that we collectively need to create social momentum to bring about change?

On to today’s note. Screening has been very much in the news lately, with the American Cancer Society (ACS) recently issuing new guidelines for breast cancer screening. Figure 1 below shows some of the new recommended mammography screening guidelines and how they compare with the latest set of guidelines from the US Preventive Services Task Force. Clearly there are some real differences between the new and older guidelines. These changes reflect both new data and emerging directions in the field.

Why, though, is it so difficult for us to determine what screening is worthwhile, and what the optimal guidelines are to optimize health in populations? I take the occasion of the new ACS guidelines to revisit our understanding of screening—a core concept in population health.

The Fundamental Challenge: Early Disease Detection, and Disease Progression

Public health aims to prevent disease and promote health. To that end, screening, with its promise of detecting early markers of pathology and preventing disease, stands to be a core part of the public health armamentarium. However, this rests on the core concept that underlies screening: that we can detect pathology early in such a way that we can alter the progression of disease. And this is where, particularly in the case of some pathologies where our understanding continues to evolve, it becomes less than perfectly clear whether we should be screening or not.

The October 12 issue of Time magazine led with a cover story, “Why Doctors Are Rethinking Breast Cancer Treatment.” The patient at the center of this story, Desiree Basila, was diagnosed with Ductal Carcinoma in Situ (DCIS) at age 52. Her surgeon asked her what she wanted to do about the 5-centimeter long, 2.5-centimer wide tumor found in her right breast, and told her a slot for a mastectomy was open the following week. Her response: “What if I just do nothing?” While at the time of her diagnosis eight years ago, Desiree was making a dramatic choice by choosing hormone therapy and close surveillance over surgery, her choice now appears empirically supported. Recent epidemiologic data on breast cancer screening and survival rates in women is suggesting that overdiagnosis represents one-quarter to one-third of cases identified. In particular, overdiagnosis may be defined as the identification of tumors that would have never led to clinical symptoms. Indeed, a recent article found that a diagnosis of DCIS is associated with a 3.3 percent mortality rate, comparable to the mortality rate from breast cancer in the general population (as noted in the accompanying editorial). Further, among those who received lumpectomy, radiotherapy was not associated with breast-cancer specific mortality at 10 years, suggesting absence of benefit. As stated in the accompanying editorial, “Given the low breast cancer mortality risk, we should stop telling women that DCIS is an emergency and that they should schedule definitive surgery within two weeks of diagnosis.”

The emerging evidence suggesting we should be considering DCIS differently is mirrored by emerging evidence about the benefits of mammography itself.

A large-scale US study found that in the more than three decades since the introduction of breast cancer screening mammography, the rate of early stage detection doubled (112 to 234 cases per 100,000), while the rate of late stage identification decreased by only 8 percent (102 to 94 cases per 100,000). The authors suggest that in 2008, overdiagnosis accounted for 70,000 cases, or 31 percent of all breast cancers diagnosed; in the past 39 years, they estimate that 1.3 million US women have been overdiagnosed. An independent UK review on breast cancer published around the same time came to similar conclusions, estimating that for every case of breast cancer prevented, three cases are overdiagnosed. These data clearly suggest that our use of screening to detect disease early enough to make a difference is limited, and perhaps resulting in more, rather than less harm. Why is this?

Screening, Cutoffs, and the Heart of the Tradeoff

Screening, such as mammography, is the process of using a test or set of tests to determine whether an individual likely has or will likely develop a given disease or health indicator. These tests are based on prior data gathered in comparisons of a screening tool against a gold-standard measure. Screening measures broadly seek to maximize sensitivity and specificity, as exemplified by the use of receiver operating characteristic (ROC) curve (see Figure 2). The top left corner of this figure represents perfect prediction, while the dotted blue line represents a screening test that performs no better than chance. Critically, however, the cutoff choice for a given measure depends on a value judgment regarding the cost of false positives versus false negatives. For example, in the case of HIV, favoring sensitivity (e.g. minimizing false negatives) over specificity (e.g. minimizing false positives) is a good strategy given its transmissibility. Conversely, if the follow-up procedures and treatments are highly expensive and invasive, as in the case of pap smears for cervical cancer, a higher false negative rate may be allowed. Importantly, sensitivity and specificity are characteristics of the test in a particular context. In the case of breast cancer, mammography is frequently highly sensitive, readily picking up breast tissue irregularities, but having very limited specificity—that is, being quite poor at identifying whether these irregularities really represent pathology. This occasions much overdiagnosis—the picking up of irregularities that would, if left untouched, lead to no harm—which is accompanied by further testing with its own burden of cost and harm to patients.

Why Context Matters: How Population Prevalence Determines the Utility of a Test

Positive predictive value (PPV) and negative predictive value (NPV) provide another way to understand test performance in terms that are salient to patients. In contrast to sensitivity and specificity, PPV and NPV are characteristics that pertain to individuals. PPV is the probability of truly having a disease after screening positive, while NPV is the probability of truly not having a disease after screening negative. As the prevalence of a given health indicator goes up, PPV rises, and as the prevalence goes down, PPV drops, even as sensitivity and specificity remain stable. In the case of breast cancer, the relatively low prevalence among women ages 40 to 49 and the accompanying low PPV has figured prominently in the argument to implement mammography screening at 50 and not 40. This argument, that the high level of false positives among this age group cannot justify the risks, was among those put forth in the controversial 2009 US Preventive Services Task Force (USPSTF) recommendation statement on screening for breast cancer; notably, this recommendation was not extended to women with a family history of breast cancer who have a higher prevalence than those without such history.

First Do No Harm: The Unintended Consequences of Unnecessary Screening

So, given this background, why then screen, and when? My colleagues and I cover this topic in our textbook, Epidemiology Matters, where we note that screening measures should be implemented when the indicator of interest is “an important determinant of population health”; when “it can be detected before signs and symptoms appear”; and when the process of screening, early detection, and treatment results in population improvements in morbidity or mortality. Pap smear for cervical cancer is a great example of a screening test that meets all these criteria. In contrast, while breast cancer remains an important determinant of population health, and may be detected before signs and symptoms appear, it is not clear among certain subgroups of women that the process of early detection and treatment results in population improvements in morbidity and mortality. Therefore, simply put, we should not be screening when we cannot, through screening, effect a difference and alter the course of the disease. Doing so introduces the risk of substantial false positives that include attendant morbidity, and potentially iatrogenic mortality. I appreciated this recent summary of the issue from Biller-Andorno and Juni: “From an ethical perspective, a public health program that does not clearly produce more benefits than harms is hard to justify.”

Welch and Passow offer a useful and instructive example about the consequences of unnecessary screening, shown in Figure 3 below.

None of this is particularly easy. Screening appears to be a compelling social good. The endorsement of screening by celebrities has led to phenomena such as the Couric effect, where the endorsement of colonoscopy by a celebrity resulted in a temporary rise in the rates of colonoscopy in the population. In addition, general understanding of the benefits of screening often vary dramatically from the reality, as shown effectively in Figure 4.

In some ways it is easy to argue that “everyone should get screened” for any number of diseases, and we hear this argument not infrequently, including in regards to both breast cancer and prostate cancer. Unfortunately, the limitations of our screening tests for these diseases means that our screening is accompanied by both false positives and false negatives, and that we should weigh the benefits and harms of both before recommending screening for particular populations.

In general, I thought the new ACS guidelines commendable. They represent a judicious effort to grapple with a tough issue and to adopt the best evidence-informed guidelines for when and whom to screen, even accepting that we simply do not know enough about whether to screen or not at some ages. That strikes me as a suitable humble approach to data, and an effort to wrap our brain around complexity in order to recommend optimal screening to the end of preventing as much disease as possible, without doing harm, in populations.

I hope everyone has a terrific week. Until next week.

Warm regards,

Sandro

Sandro Galea, MD, DrPH
Dean and Professor, Boston University School of Public Health
Twitter: @sandrogalea

Acknowledgement: I am grateful for the contributions of Gregory Cohen, MSW, to this Dean’s Note.

Previous Dean’s Notes are archived at: https://www.bu.edu/sph/category/news/deans-notes/