Sirs,
I have with great interest read Sackett DL and Haynes RB’s timely article
about the architecture of diagnostic research (BMJ 324: 539-41; 2002).
Their classification of questions to be answered into Phase I to Phase IV
questions seems very useful when studying the potential or real diagnostic
value of a physical sign or laboratory test. I do, however, fail to
understand the rationale behind the calculation of predictive values in
Phase II as it is done in Table 2. These patients are said to originate
from four different patient groups, namely normal controls with regard to
the phenomenon studied and three different groups of patients with
coronary artery disease and varying degree of left ventricular
dysfunction. Thus, they cannot represent any hypothetical population with
a prior probability of disease, such as prevalence of disease, before the
test result is known. Could I persuade the authors to elaborate on this
issue and give their reasons for calculating predictive values in a group
of individuals sampled in this way?
F Scheutz, Dental School, Aarhus, Denmark
Competing interests:
No competing interests
20 March 2002
Flemming Scheutz
Senior associate professor
Dental School, University of Aarhus, 8000 Aarhus C, Denmark
Rapid Response:
Should predictive values always be calculated?
Sirs,
I have with great interest read Sackett DL and Haynes RB’s timely article
about the architecture of diagnostic research (BMJ 324: 539-41; 2002).
Their classification of questions to be answered into Phase I to Phase IV
questions seems very useful when studying the potential or real diagnostic
value of a physical sign or laboratory test. I do, however, fail to
understand the rationale behind the calculation of predictive values in
Phase II as it is done in Table 2. These patients are said to originate
from four different patient groups, namely normal controls with regard to
the phenomenon studied and three different groups of patients with
coronary artery disease and varying degree of left ventricular
dysfunction. Thus, they cannot represent any hypothetical population with
a prior probability of disease, such as prevalence of disease, before the
test result is known. Could I persuade the authors to elaborate on this
issue and give their reasons for calculating predictive values in a group
of individuals sampled in this way?
F Scheutz, Dental School, Aarhus, Denmark
Competing interests: No competing interests