medTurk (inspired by Amazon's Mechanical Turk) supports clinical research by using the ingenuity of humans to extract information from unstructured clinical notes.
It's software (i.e., a web server) you host privately. It lets curators answer questions in parallel analogous to the way Amazon Mechanical Turk operates. The answers are then downloadable in CSV for analysis using your favorite software (e.g., R, Excel).
Suppose you have clinical notes for cancer survivors and you are interested in every patient's cancer diagnosis. medTurk permits the creation of a questionnaire that contains questions such as: Was this patient diagnosed with leukemia? or Was this patient diagnosed with neuroblastoma? . You may then assign answer choices such as Yes, No, and Not Sure. Lastly, you assign keywords such as neuroblastoma, ALL, or leukemia. medTurk uses these keywords to locate passages of text that are probably relevant to answering the question. After curators are assigned for this particular application, a curator may login and be presented with a question as shown below:
A curator may view all relevant passages of text over time (all specific to a particular patient) to answer the question. Upon answering, the curator is randomly presented with another question. The status field in this screenshot indicates how many more questions must be answered (in this case, 0 of 100 have been answered).
At any time, an admin of medTurk may download answered data in CSV and obtain a file such as shown below:
- Questionnaires are shareable: once they're created, they are downloadable and shareable with others
- Standardizes the process of data collection
- If a patient lacks keyword triggers in their notes relevant to the question, that question will not be generated for that patient
Please visit our wiki page to get started.