As an example, a primary metric for evaluating smoking cessation interventions is the
reduction of smoking point prevalence for any given intervention, relative to usual care.
We anticipate, given the fundamental differences among them, that classes of
interventions (e.g., psychotherapeutic, pharmacologic, telephone counseling,
informational, etc.) are not exchangeable and, hence, would require separate metaanalytic
models. Within classes, however, we may be able to partially pool a subset of
studies, conditional on a suite of covariates that, properly modeled, would allow use to
consider them exchangeable (conditionally independent). These covariates may include
factors such as geographic location, which might cause variation in outcomes indirectly
as the consequence of several unmeasured variables (e.g., cultural, legal, economic);
these are typically best handled using random effects in a multilevel model, provided that
there are sufficient numbers of groups in each case. Other more direct associations may
be modeled as fixed effects, including age or indices of socioeconomic status, if
available. Care must be taken in choosing the membership of each study within a
reasonably small set of intervention classes. It will be important to test the sensitivity of
our meta-analytic models to misclassification error or pooling of studies into the same
class for analytic purposes when they ought not to
As an example, a primary metric for evaluating smoking cessation interventions is thereduction of smoking point prevalence for any given intervention, relative to usual care.We anticipate, given the fundamental differences among them, that classes ofinterventions (e.g., psychotherapeutic, pharmacologic, telephone counseling,informational, etc.) are not exchangeable and, hence, would require separate metaanalyticmodels. Within classes, however, we may be able to partially pool a subset ofstudies, conditional on a suite of covariates that, properly modeled, would allow use toconsider them exchangeable (conditionally independent). These covariates may includefactors such as geographic location, which might cause variation in outcomes indirectlyas the consequence of several unmeasured variables (e.g., cultural, legal, economic);these are typically best handled using random effects in a multilevel model, provided thatthere are sufficient numbers of groups in each case. Other more direct associations maybe modeled as fixed effects, including age or indices of socioeconomic status, ifavailable. Care must be taken in choosing the membership of each study within areasonably small set of intervention classes. It will be important to test the sensitivity ofour meta-analytic models to misclassification error or pooling of studies into the sameclass for analytic purposes when they ought not to
การแปล กรุณารอสักครู่..