Epidemiologists, medical and public health professionals use the terms 'primary prevention', 'secondary prevention', and even 'tertiary prevention'. What is the difference between primary and secondary prevention of disease? What is an example of tertiary prevention?

Let's try to apply these terms (you'll find that it is not as easy as it is defined)...

Is prevention of disease a worthwhile effort? Danaei, Ding et al. (2009) assessed the effects of 12 modifiable risk factors on mortality in the U.S. These authors estimated that tobacco smoking and high blood pressure were responsible for 467,000 and 395,000 deaths respectively, in 2005, accounting for about one in every five or six deaths among U.S. adults. Overweight obesity and physical inactivity each accounted for 1/10 deaths. If the U.S. population was filled with active nonsmokers with controlled blood pressure who were not overweight, perhaps as many as 4/10 deaths would be averted each year! These substantial numbers support the importance of epidemiological studies applied to the prevention of disease, as opposed to simply identifying causative factors.

If a large proportion of deaths is associated with preventable risk factors, should resources be allocated to eliminating these risk factors? In the face of competing demands and finite resources, what preventative measures have the greatest impact? Which services should be the focus of clinical practice improvement and national policies and programs? Masciosek et al rank 25 evidence-based clinical primary and secondary preventative services based on their relative value to the U.S. population (as of 2004 information). The measures used for the rankings were a clinically preventable burden (CPB) and cost-effectiveness (CE). CPB was defined as the disease, injury, and premature death that would be prevented if the service were delivered at recommended intervals to a U.S. birth cohort, expressed as quality-adjusted life years. CE was defined as the average net cost per QALY gained by offering the preventative service. The Partnership for Prevention lists the rankings and provides supporting evidence on the Annals of Family medicine website. Clicking on a preventative service in the rankings brings up mortality rates, incidence rates, and risk factor prevalence by sub-population when those data are available.

In this course, we are characterizing early detection and screening as secondary prevention. Classic examples include mammography to detect breast cancer, Pap smears to detect cervical cancer, fasting blood glucose to detect diabetes, PSA's to detect prostate cancer, etc.

Let's look at Gordis' map of the natural history of disease, (from Gordis L. Epidemiology. Philadelphia: Saunders and Company, 1996). The biological onset of disease is followed by clinical symptoms, then diagnosis, and therapy until there is an outcome.

We can label phases in this process. From the onset of disease until clinical symptoms occur is the pre-clinical phase. The individual has the disease but doesn't know it. The clinical phase is the latter part of the process, from the occurrence of clinical symptoms through therapy.

Within the preclinical phase, there may be an interval between the onset of the disease and the occurrence of clinical symptoms during which disease can be detected with certain tests. This is called a detectable pre-clinical phase. If treatment is more effective when the disease is in the preclinical stage, screening for disease during the detectable pre-clinical phase offers an advantage.

The gain from screening for disease is the difference between the time a disease would have been diagnosed by clinical symptoms and when it is detected with a screening procedure. This is the lead time.

What is the Objective of Screening?

• To improve the quality of life or to reduce the morbidity and mortality for an individual, by applying effective treatment to disease or increased risk at an early stage, when treatment is more effective than if it were applied at a later stage.

To meet this objective, we seek to identify diseased or at-risk individuals at an asymptomatic stage. This helps us to separate individuals into populations such as:

• diseased vs non-diseased
• at-risk vs not at-risk

However, screening tests are not 100% accurate at classifying individuals. As a result, the distributions are not completely separated. For example, consider the distributions of blood sugar in diabetics and non-diabetics as depicted below. How would you set the cutpoint for a test of blood sugar to indicate diabetes? If you choose a low value, people with normal blood sugar will be included among the diabetics; if you select a high value, some diabetics will be included with the normals.

We must also recognize that screening is only useful when we are assured that treatment is more effective at the earliest stages of the disease.

If these results are from a population-based study, prevalence can be calculated as follows:

Let's see how this works out with some numbers...

Now, let's change the prevalence.

Using the same test in a population with a higher prevalence increases positive predictive value. Conversely, increased prevalence results in decreased negative predictive value. When considering predictive values of diagnostic or screening tests, recognize the influence of the prevalence of the disease. The figure below depicts the relationship between disease prevalence and predictive value in a test with 95% sensitivity and 95% specificity:

Relationship between disease prevalence and predictive value in a test with 95% sensitivity and 85% specificity.
(From Mausner JS, Kramer S: Mausner and Bahn Epidemiology: An Introductory Text. Philadelphia, WB Saunders, 1985, p. 221.)

What is a good test in a population? Actually, all tests have advantages and disadvantages, such that no test is perfect. There is no free lunch in disease screening and early detection.

We started with 10,000 people, 315 of these we have labeled as positive out of the total of 500 that have diabetes. Net sensitivity for the series is 63%, 315 out of 500. Net specificity includes the 7600 persons correctly identified as negative with the first test plus the 1710 individuals who were ruled out with the second test divided by the total of those without diabetes, 9500, for a net specificity of 98%.

The net specificity is much higher by using the two tests in a series than by just using the first test in a population with a prevalence of 5%. A significant advantage is gained from performing the simple test upfront, identifying individuals who are positive, and following up in this group with the more complex and costly test.

Tests have false positives and false negatives. But there are risks associated with screening also. Recently, risks of screening are being given more attention.

• Delayed intervention
• Disregard of early signs or symptoms

Bias results from a problem with the methods of a study that can't be corrected in an analysis. We can adjust for the effects of confounders in an analysis. For example, we can calculate adjusted rates, but we can't correct for biases.

There are two common types of biases

Common biases in screening include:

• Lead time (information bias)- the systematic error of apparent increased survival from detecting disease in an early stage
• Length (information bias) - the systematic error from detecting disease with a long latency or pre-clinical period
• Referral/Volunteer bias (selection bias) - the systematic error in detecting disease in persons who have the propensity to seek healthcare
• Detection (information bias) - the detection of insignificant disease

Let's take a look at a chart we saw earlier. Here the disease starts in 1985, is diagnosed in 1992 and the person dies of that disease in 1995. How long is his survival? Three years.

Now we institute an effective screening program. The disease starts in 1985 and is detected by the screening program in 1989. The person dies of the disease in 1995. How long was the survival? Six years. Screening seems to have increased their survival time, correct?

You have also noted that in either situation, it is 10 years from the time the disease started until the person dies. If our measure is survival time, we can easily produce a lead-time bias. In this example, there is actually no benefit of the screening process, in terms of survival. The person still died in 1995. They know about the disease for three years longer; that is the effect of the screening. This example demonstrates a lead-time bias of three years.

An effective screening program for a life-threatening disease should extend life. Screening studies with an outcome of survival time are subject to lead-time bias that can favor the screening process when there is no actual benefit to the program.

Another way to represent the lead-time bias is on a survival curve:

Gordis, L. Epidemiology. Philadelphia: Saunders and Company, 1996

In this graph, at time zero 100% of the people are alive. After five years, 30% are alive. We have instituted a screening program that detects disease one year earlier. In this case, five years following the screening diagnosis gets us out to a 50% survival rate for individuals five years following the screening diagnosis. We have a lead time bias of a 20% increased survival to five years. All that has been done here is to move the diagnosis back one year!

Instead, we need to look at the mortality rates from the disease, i.e. the mortality rates in the exposed group and the non-exposed group. Mortality rates are the "gold standard" for measuring the effect of early screening and treatment, not survival time.

Let's use this graph to consider the effect of length bias:

Disease onset is at zero and each line represents an individual. The bottom person, for instance, has a very slow growth rate of disease. The top line, with the steepest slope, represents someone with an aggressive disease. This person has rapid growth and dies (D). The individuals with slower growth lived to the point where they get screened (S).

A screening initiative is more likely to detect slow-growing diseases. There is not much that a researcher can do about this type of bias other than to realize it is likely to occur.

Breast Cancer Screening - HIP Experience

Data from the Health Insurance Program (HIP) in New York are represented as rates of death in 10,000 women per year, from all causes and from cardiovascular (CV) disease causes in the table below:

The control group has death rates of 54 and 25/10,000. The experimental group, women who were screened, were recruited in two different ways:

1.  women who volunteered, after being asked to participate
2.  women who did not volunteer to be screened, after being asked to participate

The state was able to record the death rates for each of these groups. The women who volunteered for screening had much lower rates of death, both all causes and cardiovascular causes. The women who volunteer appear to be healthier. The women who were offered to participate in the program, but refused, have the highest death rates.

If a screening study does not include a randomized process for selection, volunteers for the study are likely to be in better health than the general population. Thus, in evaluating a screening study, consider how the subjects were recruited. Were they volunteers or were they randomized into screening or no-screening groups?

Can we design a screening study without these biases?

For lead time bias – use mortality rather than survival rates.

A randomized clinical trial design can reduce biases:

• For length time bias – count all outcomes regardless of the method of detection
• For volunteer bias – count all outcomes regardless of the group; follow up with those who refuse to get outcomes

A randomized (experimental) trial prevents many design problems but is rare and expensive for evaluating a screening test. Randomized screening trials do allow estimation of lead-time bias. If accepted practice is to treat a disease early, which is likely for any screening effort, is it ethical to randomize subjects to a non-screening arm?

A non-randomized study is subject to selection biases as discussed earlier: volunteers are likely to be healthier than those not screened; length bias (the natural courses of the detected cases have longer pre-clinical phases than those cases not detected by screening) and lead-time bias (additional 'survival' time is a result of earlier detection, not longer lifespan). A non-randomized design may be a case-control study in which cases (that would have been detected without screening) and controls (cases in all stages) are retrospectively compared concerning prior screening experience.

A randomized screening trial is depicted on the left in the figure below. Individuals are randomized into either screening or no screening arms of the trial. When the disease is detected, treatment begins. Outcomes are compared between these two arms. This design tests the effect of screening.

Another type of study is represented on the right in the figure below. All are screened and followed. If an early disease is found, individuals are randomized into early treatment or no treatment until symptoms present (usual treatment). Then, measure outcomes. In this study design, what is really tested is the efficacy of treating early disease. Does treating early disease improve outcomes?

It seems a little hard to understand how the second design can meet the requirements of an IRB today. If you find individuals with the disease, how do you make an ethical argument for not treating all? Only if the necessity of early treatment is questioned. Prostate cancer is a good example because there is controversy as to whether it is more effective to treat men early or adopt a posture of watchful waiting.

Multiple Observations or Observers

Do different people look at the same results, independently and consistently arrive at the same conclusions?

Reliability can be assessed in various ways:

• Intrasubject (multiple screening tests) - means, averages; paired t-tests
• Inter-observer or inter-instrument (multiple observers or instruments)
  • The dichotomous outcome with paired samples
  • Percent agreement = a / (a + b + c)
  • Kappa statistic (test agreement, not quantify agreement)
  • McNemar’s test - non-parametric test of agreement of paired samples
• Continuous outcome
  • Differences in paired measurements
  • Coefficient of variation

This 2 x 2 table below depicts inter-observer reliability data. Each observer makes a judgment. If inter-observer reliability is high, the greatest proportion of their decisions are in cells a and d, but some of them are going to end up in cells b and c.

We can use percent agreement, a Kappa statistic, or McNemar's test to assess such data.

To summarize, all of the following will be considered when making decisions about screening programs or guidelines. Are these factors considered in the ongoing discussions about mammography guidelines?

Patient/Community/Population Considerations

• Acceptability/desirability by patient and community
• The historical precedent for screening
• The sufficient burden of disease

Epidemiological/Medical Considerations

• Different populations have a distinct and different distributions of disease
• The natural history of disease supports early detection
• Availability of ‘adequate’ screening tests
• Availability of and compliance with ‘effective’ treatment and follow up
• Acceptability by the provider and health care system

Resource Considerations

• Will of the political and healthcare system
• Availability of resources - funds, services, transportation,…
• Cost-effectiveness of screening

• Barrett A, Irwig L, Glasziou P et al. User’s guide to the medical literature XVII. How to use guidelines and recommendations about screening. JAMA 1999;281(21):2029-2034.

• Coffield AB et. Priorities among recommended clinical preventive services. Am J Prev Med. 2001 Jul;21(1):1-9.

• Danaei G, Ding EL, Mozaffarian D, Taylor B, Rehm J, et al. The Preventable Causes of Death in the United States: Comparative Risk Assessment of Dietary, Lifestyle, and Metabolic Risk Factors. PLoS Med 2009 6(4): e1000058. doi:10.1371/journal.pmed.1000058.

• Ghandur-Mnaymneh L, Raub WA, Sridar KS, et al. The accuracy of the histological classification of lung carcinoma and its reproducibility: a study of 75 archival cases of adenosquamous carcinoma. Cancer Invest 11:641, 1993.

• Gordis, L. Epidemiology. Philadelphia: Saunders and Company, 1996. Chapter 4.

• Kramer BS, Brown ML, Prorok PC, Potosky AL, Gohagan JK. Prostate cancer screening: what we know and what we need to know. Ann Int Med 1993;119:914-923.

• Maciosek, MV, Coffield, A.B., Edwards, N.M., Flottemesch, T.J., Goodman, M.J., Solberg L.I. Priorities Among Effective Clinical Preventive Services: Results of a Systematic Review and Analysis American Journal of Preventive Medicine, Volume 31, Issue 1, July 2006, Pages 52-61,
  Michael V

• McGinnis JM, Foege WH. JAMA 1993; 270(18);2207-12.

• Wilkinson I. Illinois legislature repeals requirement for prenuptual AIDS tests. NY Times. June 25, 1989.

• Winawer SJ, Fletcher RH, Millar L, et al. Colorectal cancer screening: clinical guidelines and rationale. Gastroenterology 1997;112:594-642.