Cox Proportional Hazards Model for Censored Data

This example shows how to construct a Cox proportional hazards model, and assess the significance of the predictor variables.

Step 1. Load sample data.

Load the sample data.

load readmissiontimes

The response variable is ReadmissionTime, which shows the readmission times for 100 patients. The predictor variables are Age, Sex, Weight, and the smoking status of each patient, Smoker. 1 indicates the patient is a smoker, and 0 indicates that the patient does not smoke. The column vector Censored has the censorship information for each patient, where 1 indicates censored data, and 0 indicates the exact readmission times are observed. This is simulated data.

Step 2. Fit Cox proportional hazards function.

Fit a Cox proportional hazard function with the variable Sex as the predictor variable, taking the censoring into account.

X = Sex;
[b,logl,H,stats] = coxphfit(X,ReadmissionTime,'censoring',Censored);

Assess the statistical significance of the term Sex.

stats

stats = struct with fields:
                    covb: 0.1016
                    beta: -1.7642
                      se: 0.3188
                       z: -5.5335
                       p: 3.1392e-08
                   csres: [100x1 double]
                  devres: [100x1 double]
                 martres: [100x1 double]
                  schres: [100x1 double]
                 sschres: [100x1 double]
                  scores: [100x1 double]
                 sscores: [100x1 double]
    LikelihoodRatioTestP: 5.9825e-09

The $$p$$-value, p, indicates that the term Sex is statistically significant.

Save the loglikelihood value with a different name. You will use this to assess the significance of the extended models.

loglSex = logl

loglSex = 
-262.1365

Step 3. Add Age and Weight to the model.

Fit a Cox proportional hazards model with the variables Sex, Age, and Weight.

X = [Sex Age Weight];
[b,logl,H,stats] = coxphfit(X,ReadmissionTime,'censoring',Censored);

Assess the significance of the terms.

stats.beta

ans = 3×1

   -0.5441
    0.0143
    0.0250

stats.p

ans = 3×1

    0.4953
    0.3842
    0.0960

None of the terms, adjusted for others, is statistically significant.

Assess the significance of the terms using the log likelihood ratio. You can assess the significance of the new model using the likelihood ratio statistic. First find the difference between the log-likelihood statistic of the model without the terms Age and Weight and the log-likelihood of the model with Sex, Age, and Weight.

-2*[loglSex - logl]

ans = 
3.6705

Now, compute the $$p$$-value for the likelihood ratio statistic. The likelihood ratio statistic has a Chi-square distribution with a degrees of freedom equal to the number of predictor variables being assessed. In this case, the degrees of freedom is 2.

p = 1 - cdf('chi2',3.6705,2)

p = 
0.1596

The $$p$$-value of 0.1596 indicates that the terms Age and Weight are not statistically significant, given the term Sex in the model.

Step 4. Add Smoker to the model.

Fit a Cox proportional hazards model with the variables Sex and Smoker.

X = [Sex Smoker];
[b,logl,H,stats] = coxphfit(X,ReadmissionTime,...
'censoring',Censored);

Assess the significance of the terms in the model.

stats.p

ans = 2×1

    0.0000
    0.0148

Compare this model to the first model where Sex is the only term.

 -2*[loglSex - logl]

ans = 
5.5789

Compute the $$p$$-value for the likelihood ratio statistic. The likelihood ratio statistic has a Chi-square distribution with a degree of freedom of 1.

p = 1 - cdf('chi2',5.5789,1)

p = 
0.0182

The $$p$$-value of 0.0182 indicates that Sex and Smoker are statistically significant given the other is in the model. The model with Sex and Smoker is a better fit compared to the model with only Sex.

Request the coefficient estimates.

 stats.beta

ans = 2×1

   -1.7165
    0.6338

The default baseline is the mean of X, so the final model for the hazard ratio is

$$HR=\frac{h_X(t)}{h_{\underset{}{\overset{‾}{X}}}(t)}=\exp [{\beta }_s(X_s-{\underset{}{\overset{‾}{X}}}_s)+{\beta }_{\alpha }(X_{\alpha }-{\underset{}{\overset{‾}{X}}}_{\alpha })].$$

Fit a Cox proportional hazards model with a baseline of 0.

X = [Sex Smoker];
[b,logl,H,stats] = coxphfit(X,ReadmissionTime,...
'censoring',Censored,'baseline',0);

The model for the hazard ratio is

$$HR=\frac{h_X(t)}{h_0(t)}=\exp [{\beta }_s{X}_s+{\beta }_{\alpha }{X}_{\alpha }].$$

Request the coefficient estimates.

 stats.beta

ans = 2×1

   -1.7165
    0.6338

The coefficients are not affected, but the hazard rate differs from when the baseline is the mean of X.