AACT Database Schema

What were the primary considerations when designing the database?

When designing the database, we tried to balance the following objectives:

• Present data exactly as it exists in ClinicalTrials.gov.
• Make the information as easy to understand & analyze as possible.
• Use consistent names and structures throughout the database ot make it predictable and minimize uncertainty.
• Provide value-added attributes, identify them as such, and keep them separate from the raw ClinicalTrials.gov content. (For example the Calculated_Values table contains data elements that were derived from existing data.)

Naming Conventions

• Table names are all plural. (ie. studies, facilities, interventions, etc.)
• Column names are all singular. (ie. description, phase, name, etc.)
• Table/column names derived from multiple words are delimited with underscores. (ie. mesh_term, study_first_submitted_date, number_of_groups, etc.)
• Case (upper vs lower) is not relevant since PostgreSQL ignores case. Studies, STUDIES and studies all represent the same table and can be used interchangeably.
• Information about study design entered into ClinicalTrials.gov during registration is stored in AACT tables prefixed with Design_ to distinguish it from the results data. For example, the Design_Groups table contains registry information about anticipated participant groups, whereas the Result_Groups table contains information that was entered after the study has completed to describe actual participant groups. Design_Outcomes contains information about the outcomes to be measured and Outcomes contains info about the actual outcomes reported when the study completed.
• Where possible, tables & columns are given fully qualified names; abbreviations are avoided. (ie. description rather than desc; category rather than ctgry)
• Unnecessary and duplicate verbiage is avoided. For example: Studies.source instead of Studies.study_source
• Columns that end with _id represent foreign keys. The prefix to the _id suffix is always the singular name of the parent table to which the child table is related. These foreign keys always link to the id column of the parent table.

  Child_Table.parent_table_id = Parent_Tables.id

  For example, a row in Facility_Contacts links to it’s facility through the facility_id column.

  Facility_Contacts.facility_id = Facilities.id

Structural Conventions

• Every table has an nct_id column to link rows to its related study in the Studies table. All study-related data can be linked directly to the Studies table via the nct_id. (Note: The schema diagram omits several of the lines that represent relationships to Studies. This was done to avoid appearing complex and confusing. Relationships to the Studies table can be assumed since every table includes the NCT ID.)

Studies.nct_id = Outcomes.nct_id will link outcomes to their related study.

• Every table has the primary key: id. (Studies is the one exception since it's primary key is the unique study identifier assigned by ClinicalTrials.gov: nct_id.)
• Columns that end with _date contain date-type values.
• Columns that contain month/year dates are saved as character strings in a column with a _month_year suffix. A date-type estimate of the value (using the 1st of the month as the 'day') is stored in an adjacent column with the _date suffix. (This applies to date values in the Studies table.)
• Derived/calculated values are stored in the Calculated_Values table.

While we tried to rigorously adhere to these conventions, reality occasionally failed to cooperate, so compromises were made and exceptions to these rules exist. For example, to limit duplicate verbiage, we preferred the table name References over Study_References, however the word 'References' is a PostgreSQL reserved word and cannot be used as a table name, so Study_References it is.

How are arms/groups identified?

Considerable thought went into how to present arm and group information to facilitate analysis by simplifying naming and data structures while retaining data fidelity. NLM defines groups/arms this way:

• Arm: A pre-specified group or subgroup of participant(s) in a clinical trial assigned to receive specific intervention(s) (or no intervention) according to a protocol.
• Group: The predefined participant groups (cohorts) to be studied, corresponding to Number of Groups specified under Study Design (for single-group studies.

In short, observational studies use the term ‘groups’; interventional studies use ‘arms’, though for the purpose of analysis, they both refer to the same thing. Because 'group' is more intuitive to the general public, AACT standardized on the term 'group(s)' and does not use the term 'arms'.

Participant Groups: Registry vs Results

When a study is registered in ClinicalTrials.gov, information is entered about how the study defines participant groups. In AACT, this information is stored in the Design_Groups table, while info about actual groups that is entered after the study has completed is stored in the Result_Groups table. (AACT has not attempted to link data between these 2 tables.)

Result information, for the most part, is organized in ClinicalTrials.gov by participant group. Result_Contacts & Result_Agreements are the only result tables not associated with groups. This section describes how AACT has structured group-related results data.

AACT provides four general categories of result information:

• Participant Flow (Milestones & Drop/Withdrawals)
• Baselines
• Outcomes
• Reported Events

The Result_Groups table represents an aggregate list of all groups associated with these result types. All result tables (Outcomes, Outcome_Counts, Baseline_Measures, Reported_Events, etc.) relate to Result_Groups via the foreign key result_group_id.

For example, Outcomes.result_group_id = Result_Groups.id.

ClinicalTrials.gov assigns an identifier to each group/result that is unique within the study. The identifier includes two leading characters that represent the type of result (BG for Baseline, OG for Outcomes, EG for Reported Event, and FG for Participant Flow) followed by a three digit number starting with 000 that uniquely identifies the group in that context. To illustrate assume that study NCT001 had 2 groups: experimental & control, and reported multiple baseline measures, outcome measures, reported events and milestone/drop-withdrawals for each group. The following table illustrates how the Result_Groups table organizes the group information received from ClinicalTrials.gov in this case:

Notice that the integer in the code provided by ClinicalTrials.gov (ctgov_group_code) is often the same for one group across the different result types, but this is not always the case. In the example above, BG000, EG000 & FG000 all represent the 'experimental group', so you're tempted to think that '000' equates to to the 'experimental group' for this study, however for Outcomes, OG000 represents the control group. In short, the number in the ctgov_group_code often links the same group across all result types in a study, but for about 25% of studies, this is not the case, so it can't be counted on to indicate this relationship.

Information about dates

When clinicaltrials.gov started tracking studies, it only tracked the month & year for some fields and not the day for several date values including start date, completion date, primary completion date and verification date. Because day is not provided for some studies, AACT stores these dates in the studies table as character type rather than date type values. Character type dates are of limited utility in an analytic database because they can’t be used to perform standard date calculations such as determining study duration or the average number of months for someone to report results or identify studies registered before/after a certain date.

We provide 2 columns in the Studies table for each date element:

1. A character-type column that displays the value exactly as it was received from ClinicalTrials.gov
2. A date-type column that can be used for date calculations.

If the date received from ClinicalTrials.gov has only month/year, in order to convert the string to a date, it is assigned the first day of the month. For example, a study with start date “June 2014” will have “June 2014” in the start_month_year column and “06/01/14” in the start_date column.