Tuesday, February 9, 2010

How to write a literature review using only the web

This morning we were discussing cloud computing and the next big thing where everything gets done in the so called cloud. In the "cloud", data are hosted in servers elsewhere, available all the time for downloading to machines or downloading may not be necessary as all software and software implementations to handle data will be readily available and one can work as often as one wishes, wherever, whenver, whatever. That discussion brought into mind if we can conduct a systematic review of literature entirely on the web using say only Google Docs, some online reference management website, and pretty much that.

Why is that a desirable thing to do?
For one, if a framework of conducting simple evidence based reviews can be developed that can be done entirely off the cloud or let's say off the web, that frees up resources that are entirely local on someone's computer. Besides, developing systematic reviews is entirely a team work, so more distributed work essentially means faster turnover. That said, I wanted to give it a shot.

The plan is this.

In one hour, I am going to run a simple review by framing a research question, running a search, gathering essential studies, summarizing the key points and writing up the review with nothing other than an online word processor such as Google Docs.

The question:
Is there any association between low birth weight (birth weight less than 2500 grams) and adult heart attacks (acute myocardial infarction)?

OK, it's a random question, but at least a question that has two simplicities:
a) A fairly well defined "exposure" variable if you will
b) And a fairly straightforward health outcomes that we are going to consider

The challenge is to answer this question with a write up or review within a limited time. The tools that we are going to use:
1) Google Docs (Google document or word processor; indeed any other online word processor would do)
2) Google Spreadsheets (or any other online spreadsheet application)
3) Mendeley reference manager (reference management could also have been conducted by Zotero, and more about that later)
4) Pubmed Central (because pubmed central allows free access to full text articles. If there are other implementations, they can be used as well). We add to that Google Scholar

Other than these four fully integrated online services nothing else will be used to develop an answer to the above question.

To test the feasibility, we start the review process thus:

Research question:
Is there a valid association between low birth weight and acute myocardial infarction in adults?

Search terms:
"low birth weight"
"pre terms"
"premature infant"
"acute myocardial infarction"
"heart attack"
"myocardial infarct"
Restrict the search terms to:
English language
all articles that got published within the last five years
all totally freely available articles
We are also not going to include studies that are case studies or case series
We are going to present the results in the following hierarchy:
Systematic review
Randomized trial if any
Cohort studies
Case control studies
Cross sectional survey
Case series

We are not going to conduct a formal meta analysis but just narrative analysis

Let's get going. The plan will be something like this.

1) Set up a document in Google Docs (* more comments about this later)
2) Set up a spreadsheet
3) Set up a data abstraction form in Google Spreadsheets
4) Create a collection in Citeulike
5) Create a collection in Mendeley
6) Run a search in Scholar Google, and pubmed
7) Export the results to Mendeley
8) Get the PDFs after discarding the excludable articles
9) Summarize and annotate the PDFs
10) Write the results

Ten tasks, let's say two hours. Can this be done? Can this task be distributed among a number of participants?

If this is a model and if this works, then this process can be used to speed up the process of evidence generation remarkably fast and perhaps develop a plan for analysis of high throughput literature database.