Specifying drugs for logistic regression

For a logistic regression run, you define the number of predictor values (that is, how many drugs) to include in computations for each of the events you select for the run. You also define the minimum number of reports in which a selected event must occur in combination with a drug for that drug to be included. Optionally, you can supply a list of drug values to include in the run.

Using these values, the Empirica Signal application determines what drugs to include for each event in the run, including any explicitly specified drug values and selecting additional drugs that occur in combination with each run event frequently, based on the defined total number of drugs and the minimum number of occurrences.  

• If you explicitly specify fewer drug values than the defined total number, the application finds additional drugs in the configuration to make up the difference. The application uses the defined minimum to limit the possible set of drugs to include in the run to those found most frequently in combination with each of the selected run events. For each event, only drugs found in combination at least that number of times are included.
• Each of the drug values that you explicitly supply are included in the run, even if they do not meet the specified minimum for a given run event. (However, the drug+event combination must occur at least once to be included.) If you supply more drug values than the defined total number to include in the run, the Empirica Signal application orders the drug values alphabetically and includes only the first <total number> from the list.

For example, assume that you specified a single event, Rash, for the logistic regression run. You keep the default, 10, as the minimum number of cases in which the drugs must occur, and you explicitly include DrugD. With the following drugs in the data configuration, and occurrences of Rash and each drug, the run computes scores for Rash and the shaded drugs only:

Oracle recommends that you include drugs that occur most frequently with an event, as computational biases can result if they are left out. For more information, see Logistic regression computations.

1.         On the Select Drugs page, enter the minimum number of cases in which a drug must occur in combination with an event to be included in the logistic regression run. This minimum must be 5 or greater. The default  is 10.

2.         Enter the maximum number of drugs to include in the run. The default is 100.

Note: When defining the number of drugs, keep in mind that the run computes a logistic regression for each predictor value. The more drugs you include as predictors, the longer the LR run takes.

3.         You can also specify drugs of interest to include in the run (even if the number of reports with both the event and the drug is less than the minimum specified, but greater than 0).

To select event terms from the hierarchy browser, click Select <hierarchy> Terms.

This option appears if your user preference, Enable Adverse Event Hierarchy Browser, is checked. (For more information, see Drug and event hierarchies.)

or

To select values from a list, click Select Available Values.

or

To refer to a saved list of terms, including custom terms, click Select Saved List.

or

To look up a drug name, click Trade/Generic Lookup. You can find the generic name for a drug (if you only know its trade name) or vice versa. (This link is available if the data configuration's source data included trade name-generic name mapping.)

If you created a custom term, it is included by default in this list of specified drug values. If you remove it from this list, your custom term is not be included in run results.

4.         To include the computation of interaction scores in the logistic regression run, check Compute interactions.

When an analysis of predictor interactions is requested, the Empirica Signal application generates a table that has a row for every possible triple of the form (item1, item2, response), where each item can be either a drug or a covariate category that is in the model specification for the particular response.

The purpose of each row of the table is to examine the count, N, of times the response event was present across all cases where both item1 and item2 are present, and to compare it to the predicted count, E, for those cases that has been estimated by the standard or extended logistic regression computation for that response. The table enables you to compare the observed and fitted response frequencies when both items are present against the individual LROR values for each of the two predictor items.

5.         Click Next.

The Run Options page appears.

6.         Continue to define run options.

Note: Run size varies depending on the parameters you define. A message displays if your specifications will result in a very large run. You can reduce the number of drugs to include in the run on this page, or click Back to reduce the number of events to include.