The order in which filters are applied impacts on the results

When using Cohort Discovery, you can select filters from the various tabs in any order that you like. For example, you can select a Diagnosis followed by a time frame and then a demographic filter.

However, the filters are always applied in a specific order, beginning with the Scope selections.

For non-Scope filters, the order of application has no impact on the number of individuals identified by the criteria. For example, the number of people who are female and had a diagnosis of depression is the same as the number of people who had a diagnosis of depression and who are female.

When Scope criteria are combined with other filters, it DOES make a difference. For example, the number of patients who have a diagnosis of depression who were admitted to Hospital A is different from the number of people admitted to Hospital A with a diagnosis of depression. This is because in order for the latter to be true, the diagnosis must have been recorded in the context of an admission to Hospital A, whereas in the former case it could have applied at any time. Hence:

(A)

Has a diagnosis of depression (ever) + Admitted to Hospital A (ever)

is not the same as:

(B)

Admitted to Hospital A (ever) + Has a diagnosis of depression (during an admission to Hospital A)

Cohort Discovery will always apply the two filters in order (B), regardless of the order in which you make the selections.

This does not mean, however, that order (A) cannot be achieved using Atmolytics. To achieve the effect of example (A) you would need to use one of the following approaches:

• Create two cohorts, one with a diagnosis of depression and the other with an admission to Hospital A.
• Use the Compare or merge feature to identify the intersection to which both apply.
• Create a subgroup of the intersection of the two cohorts.

or:

• Create a cohort with a diagnosis of depression.
• Select Hospital A as the data source in report creation.
• Only include admissions in which the filters apply (in which case only admissions to Hospital A will be included).

Which of the above approaches is best will depend upon whether it is critical to your purposes that the cohort only includes people to whom both filters apply; or whether it is sufficient that the data reported upon pertains only to contexts in which both filters apply.