The Search Exploration page allows you to generate search terms based on central concepts of your nest. Enter your Population, Interventions (and Comparators), and Outcomes (PICO) to generate a Boolean query that can be used on PubMed, and explore potential search results' abstracts, key terminology, and topics of interest to help construct your final search strategy. When complete, proceed to Lit Search and Run your Search.

When you create a nest, you will have the option to add details about the Population, Interventions (and comparators), and Outcomes (PICO).

Imagine you are asking the Research Question: How do atypical antipsychotics (I) impact the quality of life and disability (O) in adults with schizophrenia (P)?

1. First, enter the Population (patients with schizophrenia), Interventions (atypical antipsychotics), and Outcomes (disability and quality of life).
2. Enter each element, identify whether it is a Population, Intervention, or Outcome, and then hit the Add button (upper right in the image below).
3. Optionally, leave a note.
4. Continue adding terms until you have enough terms to identify your specific topic area (Recommended: 2 to 5 terms per category).

When you have completed the tasks above, you will have unstructured terms of interest, but no grouping of terms and no information yet populated to the Abstracts, RoboPICO, Topic Modeling, and Keywords.

To proceed, click on “PICO detail” in the right to view, add, and group together PICO elements.

1. Drag and drop the Group box from the right-side panel to the respective PICO section.
2. Group together sets of populations, interventions, or outcomes. This group will function like the OR operator in search strings. In this example, specific types of antipsychotic medications are grouped together

To populate Abstracts, RoboPICO, Topic Modeling, and Keywords:

1. In the PICO Detail, select “Update Exploration Records” in the bottom right hand corner.
2. Review the query string to ensure that it is structured as intended.
3. Run the exploratory search. This may take a minute, since this will run a pre-search of PubMed.

When you update search exploration, the references in your nest will not change. The records and data returned from this step will only be used for Search Exploration.

There are several ways to explore the references pulled by your exploratory search. The goal of this process should be to expand or refine your search terms so that they return as many records of interest while limiting irrelevant results.

Iterative Refinement: As you use the tools outlined below to add, remove, or restructure your terms, you should periodically re-run “Update Search Exploration”, as this is only run manually, and not automatically updated.

Skim through abstracts retrieved through your search exploration terms.

The purpose of reviewing Abstracts should be to:

1. Confirm that your search is returning relevant records, and
2. Identify the key terms found in relevant records and adding them to your PICO terms.

Browse commonly-mentioned Populations, Interventions, and Outcomes from abstracts and titles. PICO elements are identified by RoboPICO, which is an open source fork of the models offered in RobotReviewer.

The purpose of RoboPICO should be to:

1. Identify the most common topics of underlying abstracts, and
2. Identify terms that you should add to your PICO (which you can populate to “Add a PICO Element” by clicking on the relevant row).

Clicking a row in the chart also initiates a strict MeSH lookup on the PICO element; not all extracted PICOs will correspond to MeSHs, but expect approximately half of lookups to succeed. In the event of a failed lookup, MeSH and Google search linkouts are offered.

Explore topics that appear most frequently among the references. References may belong to one or more topics, or none. Clicking a topic shows references belonging to the topic.

These topics are derived from a Carrot2 clustering of abstract text to identify semantically similar concepts and group them.

Explore the frequency of word combinations (n-grams) from abstracts or indexed descriptors (MeSH and/or journal keywords). While n-grams are presented in order of occurence in abstract text, descriptors are inherently unordered. Clicking a row shows unique references containing the combination.

You should continue iteratively refining your query until your Abstract, RoboPICO, Topics, and/or Keywords contain mostly relevant content and minimal extraneous records or concepts.

It is useful to check the total number of records returned from your query. To do so, view the number of records in the bottom right of the Preview screen, where you can also click to view the query on PubMed:

There is no 'right' number of records, but a good rule of thumb is to return:

1. A maximum of 200-300 records, with
2. An anticipated inclusion rate at or above 20%.

The reasoning behind this recommendation is that you want to capture relevant records while minimizing extra work during Screening (remember, your team will need to manually Screen all results from your Search!). You can always add new/expanded Searches as you proceed, so achieving a high rate of inclusion should be prioritized over comprehensiveness.

When you are finished with Search Exploration, proceed to review, test, and finalize your search string in Lit Search!