How to define exposure in pharmacoepidemiology study using administrative database
Source: How to define exposure in pharmacoepidemiology study using administrative database   Publish Time: 2013-01-06 05:27   3173 Views   Size:  16px  14px  12px
How to appropriately define exposure is one of difficulties in pharmacoepidemiology study using administrative database.

Author: Xuanqian Xie

How to appropriately define exposure is one of difficulties in pharmacoepidemiology study using administrative database. Once a patient fills in a prescription, a record will be added in the database. We presume that patients strictly follow the prescription. Some researchers categorize the medication use into current use, previous use and remote use, such as these in Juurlink et al. 2009. But, actually there are no well accepted definitions for these three categories, and more importantly, these 3 categories are still not precise enough. For example, one patient may take only one dose of a medication, whereas another patient may take this medication for in a long duration. The two scenarios are clearly different, but those 2 patients can be categorized in the same class, according to these definitions.

The scenarios can be more complex. For example, patients may discontinue and reuse a drug a number of times (how to define the time window?). Patients can be a switcher due to not toleration of other medications (this consideration is difficult to be verified). Patients are likely to simultaneously use a number of different medications for the one indication (other medications may be associated with the outcomes, too). The OTC information is usually unknown (but, the OTC drug is effective, too). Therefore, even under careful considerations, usually there is apparent heterogeneity in the administrative database research.

Furthermore, the effect of medication, including the side effects, can be observed immediately, or after a long duration.  When we create the time window of an intervention, we need think about the duration of drug intake, as well as lag period and persistence period. When the onset of effect is acute (such as drug interactions), it is ideal condition for the researches using the administrative database, although I understand that some researchers use the administrative database to examine the cumulative risk, such as drugs for cancer therapy.  We can use the time-dependent Cox model to estimate the acute effects of a drug (i.e. adverse effects), or drug interactions. The time-dependent variable in Cox model provides a solution of defining the exposure.

 

 

Reference:

Juurlink DN, Gomes T, Ko DT, et al. A population-based study of the drug interaction between proton pump inhibitors and clopidogrel. CMAJ. 2009; 180(7):713-8.

 

 

A few lines of SAS program of examining drug interactions using time-dependent Cox model were list below.  

 

/* Time-dependent Cox model*/

proc phreg data=inter.t_dep_f_2010 ;

 

     model cohort_d*event(0) = ppi_clop_ ppi_ clop_  none_/ties=exact rl;   

     array ppi_clop (*) ppi_clop_1-ppi_clop_91;

            ppi_clop_=ppi_clop[cohort_d];

     array ppi (*) ppi_1-ppi_91;

            ppi_=ppi[cohort_d];

     array clop (*) clop_1-clop_91;

            clop_=clop[cohort_d];

     array none (*) none_1-none_91;

            none_=none[cohort_d];

run;

 

 

proc phreg data=inter.t_dep_f_2010 ;

 

     model cohort_d*event(0) = ppi_clop_ ppi_ clop_ none_  

       Co_hype    co_lip    co_hefa    co_ishe    co_cere    co_pvs

 

       co_arrh    co_glau    co_blind    co_renal  Co_dia /ties=exact rl;   

     array ppi_clop (*) ppi_clop_1-ppi_clop_91;

            ppi_clop_=ppi_clop[cohort_d];

     array ppi (*) ppi_1-ppi_91;

            ppi_=ppi[cohort_d];

     array clop (*) clop_1-clop_91;

            clop_=clop[cohort_d];

     array none (*) none_1-none_91;

            none_=none[cohort_d];

run;