20. Cox Regression
Bewick, Cheek & Ball (2004)
ï The P values indicate that the difference between treatments was bordering
on statistical significance, whereas there was strong evidence that age was
associated with length of survival.
ï The coefficient for treatment, â1.887, is the logarithm of the hazard ratio for a
patient given treatment 1 compared with a patient given treatment 2 of the
same age. The exponential (antilog) of this value is 0.152, indicating that a
person receiving treatment 1 is 0.152 times as likely to die at any time as a
patient receiving treatment 2
ï That is, the risk associated with treatment 1 appears to be much lower.
ï However, the confidence interval contains 1, indicating that there may be no
difference in risk associated with the two treatments.
21. Cox Regression
Bewick, Cheek & Ball (2004)
ï Using the KaplanâMeier (log rank) test, the P value for the difference between
treatments was 0.032, whereas using Coxâs regression, and including age as an
explanatory variable, the corresponding P value was 0.052.
ï This is not a substantial change and still suggests that a difference between
treatments is likely. In this case age is clearly an important explanatory
variable and should be included in the analysis.
ï The exponential of the coefficient for age, 1.247, indicates that a patient 1
year older than another patient, both being given the same treatment, has an
increased risk for dying, by a factor of 1.247. Note that, in this case, the
confidence interval does not contain 1, indicating the statistical significance of
age.
22. Assumptions for Cox Regression
1. Censoring is unrelated to prognosis
2. Proportional hazards model: hazard at time t
for a patient in one group is proportional to the
hazard at time t for a patient in the second
group
23. Prognosis Checklist for Validity
ï Was a defined, representative sample of patients
assembled at a common (usually early) point in the course
of their disease?
ï Was patient follow-up sufficiently long and complete?
ï Were outcome criteria objective and unbiased (e.g. applied
in a âblindâ fashion)?
ï If subgroups with different prognoses are identified, did
adjustment for important prognostic factors take place?
24. Prognosis Checklist for Results
ï What are the results?
ï How likely are the outcome events over time?
ï Survival curves (Kaplan-Meier)
ï Prognostic factors
ï HR
ï How precise are the prognostic estimates?
ï Confidence intervals
25. Prognosis Checklist for Applicability
ï Can I apply this valid, important evidence about prognosis
to my patient?
ï Is my patient so different to those in the study that the
results cannot apply?
ï Will results lead directly to selecting or avoiding a
treatment?
ï Will this evidence make a clinically important impact on
my conclusions about what to offer to tell my patients
28. Prognosis Example
P: In elderly men with liver cirrhosis,
I: does smoking cessation and alcohol abstinence
C: compared to do nothing
O: improve survival (live longer)?
29. Prognosis Example
ï āļāļīāļĄāļāđ search terms āđāļ PubMed Clinical Queries āļāļąāļāļāļĩāđ
5-year survival cirrhosis alcohol smoking
ï āđāļĨāļ·āļāļ Category: Prognosis āļāļ 5 āļāļāļāļ§āļēāļĄ
ï āđāļĨāļ·āļāļāļāļāļāļ§āļēāļĄāļāļāļ Pessione et al: Five-year survival predictive factors
in patients with excessive alcohol intake and cirrhosis. Effect of
alcoholic hepatitis, smoking and abstinence. Liver Int. 2003
Feb;23(1):45-53.