Conditional Logistic Regression

Using SASí PROC PHREG Procedure

PHREG information provided by Dr. David Brown

 

This procedure performs conditional logistic regression (CLR) for 1:1, 1:m and n:m matched studies. CLR estimates for 1:1 matched studies may be obtained using the PROC LOGISTIC procedure. However, to obtain CLR estimates for 1:m and n:m matched studies using SAS, the PROC PHREG procedure must be used.

 General model syntax

  proc phreg data =dataset nosummary;

model status*censor(0)= variable(s) of interest

/ties=discrete [or breslow] risklimits;

strata factors matched upon;

 

ESSENTIAL STEPS in using PROC PHREG.

1. The variable representing cases and controls (e.g., CACO) MUST be redefined, or a new variable created (e.g., STATUS) so it has the value 1 for cases and the value 2 for controls. This is required so that the probability of being a case is modeled.

2. Create a variable called CENSOR. This is a survival analysis term for Cox proportional hazards modeling, but it is not necessary to understand its meaning, only to know that in the output, it represents controls. For cases CENSOR must have a value of 1 and for controls it must have a value of 0.

 

 

CACO

STATUS

CENSOR

case =

1

1

1

control =

0

2

0

 

3. You must specify TIES=BRESLOW for 1:m and TIES=DISCRETE for n:m matching.

4. RISKLIMITS will display the lower and upper confidence limits (alpha=0.05 is the default).

5. The STRATA statement identifies the risk sets you created during the matching process. You have two options using the STRATA statement.

i) you may create a risk set identification variable (e.g., RISKSET).

strata riskset;

This results in 1:m matching. Use the SUMMARY option (see note below) to view the individual strata and determine how many cases and controls are represented in each strata.

ii) you may enter each of the matching variables into the strata statement. For example, if you have matched on birth year, race and sex, then

  strata ybirth race sex;

This results in n:m matching, by pooling together matched sets with similar matching criteria. Use the SUMMARY option (see note below) to view the individual strata strata and determine how many cases and controls are represented in each strata.

NOTE: In column headings displayed from the SUMMARY option, cases are referred to as EVENTS and controls are referred to as CENSORED. Cases and controls combined are referred to as TOTAL. 

Test your model - using SUMMARY option

Confirm you are utilizing the correct number subjects in the model and the correct number of matched sets. Use the SUMMARY option above to list each matched risk set as a separate strata (the exception to this occurs for n:m matching where pooled matched sets are combined). The NOSUMMARY option is used to suppress this output.

How to obtain crude estimates

To obtain crude estimates, remove the matching factors form the STRATA statement. You can verify that this is in fact the crude estimate by comparing it to the estimate obtained using the CMH option in PROC FREQ (the output has three estimates, compare using case-control estimate);

proc freq data=dataset;

tables caco* variable of interest / CMH

How to race-, sex-, or variable-specific estimates

To obtain estimates specific for a matching factor, remove the matching factor from the STRATA statement, and include a BY statement using the removed matching factor.


PHREG information provided by Dr. David Brown