BOOKS OF RtfiDIfGS - PAHO/WHO

GREENLAND ET AL.
3. Cornfleld J, Haenszel WH. Some aspects of retrospective
studies. J Chron Dis 1960;11:523.
4. Mantel N, Haenszel WH. Statistical aspects of
the analysis of data from retrospective studies of disease.
Natl Cancer lnst Monogr 1959;22:719.
5. Neutra RR, Drolette ME. Estimating exposurespecific
rates from case-control studies using Bayes'
theorem. Am J Epidemiol 1978;108:214.
6. Prentice RL, Pyke R. Logistie disease incidence
models and case-control studies. Biometrika
1979;66:403.
7. Schlesselman JJ. Samnle size requirements in
cohort and case-control studies of disease. Am J
Epidemiol 1974;99:381.
8. Dom HF. Some problems arising in prospective
and retrospecti ve studies of the etiology of disease. N
Engl J Med 1959;261:571.
Appendix
MEDIcL CARE
9. Kleinbaum DG, Morgenstern H, Kupper LL.
Selection bias in epidemiologic studies. Am J
Epidemiol (in preas).
10. Sackett DL. Bias in analytic research. J Chron
Dis 1979;32:51.
11. Cole P. The evolving case-control study. J
Chron Dis 1979;32:15.
12. Feinstein AR. Clinical biostatistics. St. Louis:
Mosby, 1977.
13. Greenland S, Neutra RR. Control of confound.
ing in technology assessment. Int J Epidemiol (in
press).
14. Miettinen OS. Efficacy of therapeutic
practice-will epidemiology provide the answers?
In: Melmon KL, ed. Drug therapeutics. Elsevier:
New York, 1980.
The rates given in Tables 1 and 2 were computed from the casecontrol
data using Bayes' theorem. A detailed and general discussion of
this method is given in reference 5. Briefly, suppose we know the
overall rate of the outcome, P(D), the probability of exposure among
cases, P(E/D), and the probability of exposure among noncases, P(E I
D). Letting P(D) = 1 - P(D), Bayes' theorem states that the outcome
rate among exposed persons, P(D/E), is given by
P(D 1 E) = P(E I D)P(D)
P(E 1 D)P(D) + P(E D)PD)
In our NICU study, the total size of the target population (unit A + unit
B) was equal to the number of cases (61) plus twice the number of
controls (2 x 150)or 361; the ital number ofcases over the study period
was 61; and so P(D) = 61/361 = 0.169 and P(OD) = 0.831. To illustrate
Bayes' theorem, consider computation of the rate in the first row of
Table 1. Here the exposure is "unit A," P(E 1 D) = 43/61 = 0.705, and
P(E [D) is estimated as 78/150 = 0.520. Thus, using Bayes' theorem
P(D 1 E), the rate in unit A, is estimated as
or 21.6 per cent.
- 107 -
P(D | E) = 0.705(0.169) = .216
0.705 (0.169) + 0.520 (0.831)