Key words: Aortic aneurysm, abdominal; cost-benefit analysis; risk factors; screening

Introduction

Early detection is defined as " the identification of a disorder before symptoms or signs become apparent." 1 The term screening is often used as synonymous with early detection. However, case-finding, which is " the detection of disease by means of tests or procedures that are undertaken by health workers on patients who are consulting for unrelated symptoms,"1 is also an important form of early detection, and, for some disorders, it is a more appropriate method of early detection than population screening. In this paper screening implies either form of early detection.

The mid-century enthusiasm for screening for chronic diseases was eventually tempered with realism. For a disease to be considered as a candidate for screening, it was recognized that it should be of reasonably high prevalence, have a high case-fatality rate or produce major disability, be treatable and have a preclinical phase during which it could be detected using a safe, sensitive and specific test, acceptable to the target population.2

There seems to be little doubt that abdominal aortic aneurysm (AAA) meets these criteria. Although there are problems in interpreting routine statistics, the age-adjusted mortality and morbidity from AAA is not declining, as is the case for heart disease and stroke.3-8 The numbers of patients with AAA will increase as the population ages. Screening programs have found that about 5% of elderly people have previously undiagnosed AAA, an estimate derived from several studies using a variety of definitions including the one used in this study (> or = 3.0 cm).9

The case-fatality rate for ruptured AAA is virtually 100% unless treated promptly. A recent large Canadian study found that, even if treated, the short-term fatality rate for ruptured AAA was 49%, while that for patients treated electively prior to rupture was 5%. From the same study, the five-year survival rate for patients who survived surgical repair of ruptured AAA was estimated to be 26%, compared to 68% following elective surgery.10,11 AAA has a well-defined preclinical phase during which it is readily diagnosed by ultrasound. The latter has high sensitivity and specificity for detecting the presence of an aneurysm,12 though the precision of ultrasound estimates of the extent of aneurysmal dilatation has been questioned.13

The fact that a disease meets the above criteria is a necessary, but not sufficient, condition for the introduction of a screening program. It must also be demonstrated that such a program is effective, meaning that it does more good than harm to those to whom it is offered, as well as cost-effective, meaning that money spent on it would not be better spent in some other way.14,15 To date there is no evidence from controlled studies that screening for AAA is effective, though one such study is under way.16 The cost-effectiveness of screening for AAA has also been challenged.17,18

The cost-effectiveness of a screening program might be enhanced by focusing it on people with higher than average risk of harbouring latent disease, i.e. selective screening. Of course, all screening programs are selective to some degree, for example, with respect to age and sex. Since ruptured AAA is rare below the age of 55, it seems reasonable to focus screening on the " elderly" population, aged 55 and over. The male-to-female risk ratio for death from ruptured AAA at ages 55 and over is about 3.5,8 and 45% of people in that age group are male.

Using the theory of selective screening (see Appendix), it is easy to calculate that 74% of the total reduction in mortality produced by screening all the population aged 55 and over could be achieved by restricting the program to men.19 In other words, nearly three quarters of the potential benefit would be obtained by screening less than half the elderly population, a considerable increase in cost-effectiveness; thus this type of strategy has been advocated.20

A further step might be to restrict the screening to elderly men who smoke. In a British screening study, 42% of the newly detected aneurysms were found among the 26% of men who smoked.21 We have estimated that 32% of the potential benefit of screening could be achieved by restricting screening to the 12% of the elderly population who are male smokers.19 Although this would increase the cost-effectiveness of screening, many people feel that a screening program should aim at achieving more than one third of the potential benefit. In this paper we explore the increased cost-effectiveness that might be obtained by adding further risk factors to the selection process.

Methods

The data used in the calculations come from a case-control study of the risk factors for AAA and intermittent claudication, which has been described elsewhere.22 We use here the results for 78 men attending hospitals in Ottawa, Ontario, for treatment of unruptured AAA and for 99 male controls who were attending the same hospitals for treatment of conditions other than cardiovascular disease, cancer or diabetes. From the results of a structured interview, physical examination and blood tests, the following risk factors were selected: age (years), cigarette smoking (pack-years), blood pressure (1 if blood pressure greater than 140/90 or history of treatment for hypertension, 0 otherwise), history of heart disease (1 if yes, 0 otherwise), body mass index (weight in kg divided by height in m²) and serum high-density lipoprotein (mmol/L).

Logistic regression analysis was used to derive the risk of AAA for an individual based on his risk factors.23 This risk was calculated for each of the controls, who were assumed to be a representative sample of the elderly male population. For each individual the potential benefit of screening was calculated by multiplying his risk of developing AAA by his expectation of life, estimated from the 1991 abridged life table for Canadian men.24 The controls were then ranked with respect to potential benefit (those with lower risk having greater potential benefit), and the proportion of the total benefit that would be obtained by restricting screening to a given proportion of the population with the greatest potential benefit was calculated by summation. These calculations were made using logistic regressions based on (1) age only; (2) age plus each of the other risk factors in turn; and (3) all the risk factors together.

Although the risk equations are derived from a case-control study rather than a cohort study, we believe that it is appropriate to use them to calculate the proportion of total benefit in this way (see Appendix). Because of missing data, the number of subjects included in the regression equations varied. However, for consistency, the 95 control subjects with complete data on all the risk factors were used in the different calculations of potential benefit.

Results

The logistic regression equations used in the calculations are shown in Table 1. With all risk factors included, the regression coefficients were significantly different from zero for every factor except for high blood pressure. The coefficients for body mass index and high-density lipoprotein were negative, indicating a significant protective effect. The implications of these findings have been discussed elsewhere.22

Table 2 shows the percentage of the population of elderly men that it would be necessary to examine to achieve a given percentage of the potential benefit from examining them all. For example, 80% of the total possible benefit could be achieved if 52% of elderly men were examined, using a risk function based on age alone; 17% if all six risk factors were taken into account in the selection process. Adding a single risk factor to age in the equation makes little difference to the proportion of the population that must be screened, except in the case of pack-years of smoking. The fraction of the population that would be needed for any level of benefit using age and smoking lies approximately midway between the value using age alone and that using all the risk factors.

TABLE 2 Proportion of population of elderly men needed to achieve a given proportion of total benefit
Percentage of total benefit	PERCENTAGE OF POPULATION NEEDED USING:a						All Risk Factors
	Age Only	Age plus:
		Cig	BP	HHD	BMI	HDL
50	22	10	20	18	21	17	5
60	29	15	26	25	28	22	8
70	40	24	35	34	38	30	11
80	52	35	48	45	51	41	17
90	68	52	66	63	68	56	29
a See Table 1 for definitions of risk factors

Discussion
Based on these calculations, selective screening for AAA seems to be a promising strategy. However, this conclusion must be tentative for several reasons. First, the case-control study used to derive the risk equations was small and was hospital-based. Nevertheless, the results are consistent with those of other studies.25,26 It is also possible that the controls may not be truly representative of the general population of elderly men, although this would be less of a problem if a case-finding approach were used to screen men seeking medical care and if the distribution of the controls with respect to age, education and smoking status were similar to that of elderly men in Ontario. Secondly, the risk of diagnosed disease may not be the same as the risk of latent disease, since risk factors may influence both the progression of disease and/or the likelihood of it being discovered by chance. Thirdly, men with high risk of cardiovascular disease have a lower than average expectation of life. Fourthly, the controls were not screened and some of them might have had AAA.

Some of these uncertainties could be resolved by a demonstration project in which risk factors are included in a screening program, thus providing empirical validation of the predictions, at least in terms of the numbers of cases detected. If validated, the approach could then be incorporated into a randomized controlled study in which a criterion of entry would be a given level of the risk factor score.

References

1. Spitzer WO. The scientific admissibility of evidence on the effectiveness of preventive interventions. In: Goldbloom RB, Lawrence RS, eds. Preventing disease. Beyond the rhetoric. London: Springer-Verlag, 1990;1-4.

2. Wilson JMG, Jungner G. Principles and practice of screening for disease. Geneva: World Health Organization, 1968.

3. Melton III LJ, Bickerstaff LK, Hollier LH, Van Peenen HJ, Lie JT, Paierolo PC, et al. Changing incidence of abdominal aortic aneurysms: a population-based study. Am J Epidemiol 1984;120:379-86.

4. Castleden WM, Mercer JC. Abdominal aortic aneurysms in Western Australia: descriptive epidemiology and patterns of rupture. Br J Surg 1985;72:109-12.

5. Lilienfeld DE, Gunderson PD, Sprafka JM, Vargas C. Epidemiology of aortic aneurysms: I. Mortality trends in the United States 1951 to 1981. Atherosclerosis 1987;7:637-43.

6. Fowkes FGR, MacIntyre CCA, Ruckley CV. Increasing incidence of aortic aneurysms in England and Wales. Br Med J 1989;298:33-5.

7. Semenciw R, Morrison H, Wigle D, Cole W, Hill G. Recent trends in morbidity and mortality rates for abdominal aortic aneurysms. Can J Public Health 1992;83:274-6.

8. Millar WJ, Cole CW, Hill GB. Trends in mortality and hospital morbidity due to abdominal aortic aneurysms, Canada, 1972-90. Health Reports 1995;7:19-27.

9. Pleumeekers HJCM, Hoes AW, van der Does E, Van Urk H, Grobbee DE. Epidemiology of abdominal aortic aneursysms. Eur J Vasc Surg 1994;8:119-28.

10. Johnston KW. Ruptured abdominal aortic aneurysm: six-year follow-up results of a multicenter prospective study. J Vasc Surg 1994;19:888-900.

11. Johnston KW. Non-ruptured abdominal aortic aneurysm: six-year follow-up results from the multicentre Canadian aneurysm study. J Vasc Surg 1994;20:163-70.

12. Bluth EI. Ultrasound of the abdominal aorta. Arch Intern Med 1984;144:377-80.

13. Ellis M, Powell JT, Place J, Mills S, Wolfe JNH, Boultbee J, et al. The limitations of ultrasound in surveillance of small abdominal aortic aneurysms. In: Greenhalgh RM, Mannic JA, Powell JT. The cause and management of aneurysms. London: WB Saunders Company, 1990:117-21.

14. Spitzer WO (Chairman). Report of the Task Force on the Periodic Health Examination. Can Med Assoc J 1979;121:1193-254.

15. Bergqvist D, Jendteg S, Lindgren B. Standards for the cost-benefit approach to vascular surgery. Acta Chir Scand Suppl 1990;555:105-10.

16. Scott RAP, Wilson NM, Ashton HA, Kay DN. Abdominal aortic aneurysm in 4237 screened patients: prevalence, development and management over 6 years. Br J Surg 1991;78:1122-5.

17. Frame PS, Frybaack DG, Patterson C. Screening for abdominal aortic aneurysms in men 60-80 years. Ann Intern Med 1993;119:411-6.

18. Mason JM, Wakeman AP, Drummond MF, Crump BJ. Population screening for abdominal aortic aneurysm: do the benefits outweigh the costs? J Public Health Med 1993;15:154-60.

19. Hakama M, Pukkala E, Saastomoinen F. Selective screening: theory and practice based on high-risk groups of cervical cancer. J Epidemiol Comm Health 1979;33:257-61.

20. Collin J. Screening for abdominal aortic aneurysm. In: Hill GB, Koumanakos D, Anderson L, eds. Proceedings of the Workshop on the Control of Abdominal Aortic Aneurysm; 1994 May 25-26; Aylmer (Que). Chronic Dis Can 1994;15(4 Suppl):S28.

21. O'Kelly TJ, Heather BP. General practice-based population screening for abdominal aortic aneurysms: a pilot study. Br J Surg 1989;76:479-80.

22. Cole CW, Hill GB, Bouchard AG, Moher D, Farzad E, Rody K, et al. Atherosclerotic risk factors in abdominal aortic aneurysm and peripheral vascular occlusive disease. Chronic Dis Can 1994;15(4):120-2.

23. Cox DR. Analysis of binary data. London: Methuen, 1970.

24. Statistics Canada. Abridged life tables, 1991. Unpublished tables.

25. Reed D, Reed C, Stemmerman G, Hyashi T. Are aortic aneurysms caused by atherosclerosis? Circulation 1992;85:205-11.

26. Strachan DP. Predictors of death from aortic aneurysm among middle-aged men: the Whitehall Study. Br J Surg 1991;78:401-4.

For example, if the risk factor is being male, the relative risk of AAA for males is R = 3.5, and P = 0.45 of the elderly population is male, then S = 0.45 x 3.5 / (0.45 x 2.5 + 1) = 0.74, and 74% of the disease would be detected by screening 45% of the elderly population.

In this paper we extend this approach to the case where there is a risk equation based on several risk factors, some discrete and some continuous, estimated by logistic regression. If data from a cohort study had been available, then the logistic regression equation would have estimated the risk of disease directly. The calculation of the potential benefit is a straightforward summation for individual controls with risks above a given cut-point as a proportion of the total for all controls. Using data from a case-control study, the logistic regression equation does not provide an absolute measure of the risk for an individual, and a more indirect argument is needed.

Let

Zi = the vector of risk factor values for control Ci, i = 1...N

Zo = the vector of risk factor values for an arbitrary member of the target population, Co

B = the vector of coefficients for the risk factors in the logistic regression equation

Ri= the risk of disease for Ci relative to that for Co

Then

Ri = exp{B(Zi - Zo)}, assuming the disease is rare.

In the present application we weight the risks RiPo by the expectation of life of each individual, Ei, and the expression for SM becomes

It is not difficult to show that the expression for SM is invariant with respect to the choice of Co.

___________________________________
Author References
C William Cole, Ottawa General Hospital, 501 Smyth Road, Ottawa, Ontario K1H 8L6
Gerry B Hill, Department of Epidemiology and Community Medicine, University of Ottawa, Ottawa, Ontario
Wayne J Millar, Health Statistics Division, Statistics Canada, Ottawa, Ontario
Andreas Laupacis, Clinical Epidemiology Unit, Loeb Research Institute, Ottawa Civic Hospital, Ottawa, Ontario
K Wayne Johnston, Division of Vascular Surgery, University of Toronto, Toronto, Ontario

[Table of Contents] [Next]