1
$\begingroup$

Could you please guide me to tutorials which help to interpret AFT model results (does group A and B have different survival probability or not)? I have done a lot of googling, but there is not much information available.

    ## Call:
## flexsurvreg(formula = Surv(time, status) ~ group + age + sex + 
##     comorbidity, data = data, dist = "genf")
## 
## Estimates: 
##              data mean  est       L95%      U95%      se        exp(est)
## mu                 NA   15.62529  14.95734  16.29325   0.34080        NA
## sigma              NA    2.65714   2.48783   2.83798   0.08926        NA
## Q                  NA   -0.79804  -1.14869  -0.44738   0.17891        NA
## P                  NA    0.53072   0.23235   1.21224   0.22366        NA
## groupB        0.00433    0.44093  -0.49010   1.37197   0.47503   1.55416
## age          78.77030   -0.10538  -0.11199  -0.09877   0.00337   0.89999
## sexm          0.28273   -0.83899  -0.97916  -0.69881   0.07152   0.43215
## comorbidity   1.66034   -0.28185  -0.31697  -0.24674   0.01792   0.75439
##              L95%      U95%    
## mu                 NA        NA
## sigma              NA        NA
## Q                  NA        NA
## P                  NA        NA
## groupB        0.61256   3.94311
## age           0.89406   0.90595
## sexm          0.37563   0.49717
## comorbidity   0.72836   0.78135
## 
## N = 11541,  Events: 3851,  Censored: 7690
## Total time at risk: 4024558
## Log-likelihood = -28894.55, df = 8
## AIC = 57805.09
Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ something beyond the papers and vignettes? $\endgroup$
    – Chris
    Commented Apr 10, 2020 at 19:33
  • $\begingroup$ Please say more about which parts you specifically don't understand: the parameterization of the baseline survival distribution, the coefficient estimates, standard errors and confidence intervals, or what? Note that software-specific issues are off-topic here, but you do seem to have questions of a more general statistical nature. It also might help if you could specify the nature of each of the predictor variables (continuous versus 2-level factors). $\endgroup$
    – EdM
    Commented Apr 10, 2020 at 23:01
  • $\begingroup$ Thank you guys for your comments! Two patient groups, A (n=11000) and B (n=50). Is there any difference in their adjusted survival during 460-day follow up. I adjusted for age (50-104), sex(male, female) and comorbidity (integer, 0-12). My problem is about interpretation as I am only familiar with HRs, ORs, but AFT gives what kind of estimate? Is the following sentence correct: on average, the patients of group B had similar adjusted survival during the 460 day follow-up as those from group A, ?? 1.55 [CI 0.61, 0.3.94]. The “1.55” is not OR, HR, how should I write about it? $\endgroup$
    – st4co4
    Commented Apr 13, 2020 at 9:56

1 Answer 1

Reset to default
0
$\begingroup$

What you're interest in, you can find in the line "groupB" of the table. Estimated parameter is 0.44, its exponential is 1.55, which means that elements in group B survive 1.55 times longer shorter by average, than in group A. This difference, however, is not statistically significant (value 0 is included in the confidence interval [L95%, U95%]).

The 1.55 value is called time ratio (TR) and it's actually quite easy to understand: probability of dying is estimated to come 1.55 faster for group B than for group A, if fixing the other factors (it is an adjusted TR, you may say). As the name itself says, AFT models distribute failure probability over time by accelerating or slowing it among groups. Group B here has its time accelerated.

This means I must correct myself: unfortunately, different statistical packages parametrize effect sizes with opposite signs, so it's easy to get confused between conservative and risk factors. It appears that flexsurvreg uses PH-alike parametrization, so group B survives shorter, not longer.

Improve this answer
$\endgroup$
4
  • $\begingroup$ Thank you Carlo! I am only familiar with HRs, ORs, but AFT gives what kind of estimate? Using your sentence: on average, the patients of group B had similar adjusted survival during the 460 day follow-up as those from group A, ?? 1.55 [CI 0.61, 0.3.94]. The “1.55” is not OR, HR, how should I write about it? $\endgroup$
    – st4co4
    Commented Apr 13, 2020 at 9:59
  • $\begingroup$ expanded the answer $\endgroup$
    – carlo
    Commented Apr 13, 2020 at 11:51
  • $\begingroup$ Carlo, I am super thankful! Is there any possibility that you can check my other post about AFT? My only concern there is if AFT is an acceptable method to use (model fit, log-likelihood, distribution question)? stackoverflow.com/questions/61186413/… $\endgroup$
    – st4co4
    Commented Apr 13, 2020 at 12:52
  • $\begingroup$ I had to move the post, the here now: link $\endgroup$
    – st4co4
    Commented Apr 14, 2020 at 4:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.