As pharmaceutical and healthcare scientists, product managers, and marketers, we often need to understand the big picture in our field, asking questions like “what are the results across all published Alzheimer’s studies?” or “what are all of the comparisons against budesonide?” We call the answers to these questions a landscape analysis because they present the entire field within a single view. Landscapes can be analyzed from either a disease or therapeutic focus, as shown with our Alzheimer’s and budesonide examples.
Such landscape analyses are important for a host of tasks such as:
- finding differentiation amongst competing therapies
- performing “gap” analysis
- guiding trial design criteria
- uncovering new indications
- Or simply getting up-to-speed for a disease or around a particular therapy
Today, the process of finding, reading, analyzing, indexing, storing and summarizing all of the known literature for a single disease area (or therapy focus) is no small feat. Based on our previous analysis we found doing a single meta-analysis itself takes a long time and a lot of money. And generating a landscape is akin to producing lots and lots of these! (Or, at least, curating lots of results – sometimes hundreds or even thousands of them).
So, given all of that effort, how much does it cost to do a landscape by hand? Here we attempt to address that question.
As with our previous analysis, there are two questions to ask: (1) how much effort is involved, and (2) how much does that effort cost.
So we first tackle how much effort is involved in creating a landscape. We will focus on disease landscapes, since those have more general applicability (e.g., more people would care about say, asthma, than any particular asthma treatment)
To get a ballpark scope of how much goes into creating a disease landscape, we first needed a mechanism to pick which diseases to analyze. For that, we turned to the World Health Organization (WHO), which created the following chart of the top 10 causes of death globally . This chart sheds light upon which fatal causes we could address that would have a large impact. Of these top 10 causes, 9 are diseases (the 10th, sadly, involve traffic accidents) so we will use this set of diseases for our analysis.
As in our previous work, we will use PubMed searches as a proxy for the amount of effort involved. In particular, we will count how many meta analyses have been done for each of these diseases, over the last five years. This gives us a sense of the work involved for each disease landscape. The table below shows our PubMed query, along with the number of papers returned for each.
|Disease||Studies last 5 years||PubMed Query (date range: 1/1/2013 – 12/31/2017)|
|Ischemic Heart Disease||2,146||(“meta-analysis”[pt]) AND (“ischemic heart disease”)|
|Stroke||2,548||(“meta-analysis”[pt]) AND (stroke)|
|Lower Respiratory Infection||974||(“meta-analysis”[pt]) AND (“respiratory infection”)|
|Chronic Obstructive Pulmonary Disease||410||(“meta-analysis”[pt]) AND (“chronic obstructive pulmonary disease”)|
|Trachea, Broncus, Lung Cancers||1,471||(“meta-analysis”[pt]) AND (“lung cancer”)|
|Diabetes mellitus||2,429||(“meta-analysis”[pt]) AND (“diabetes mellitus”)|
|Alzheimer disease||545||(“meta-analysis”[pt]) AND (“Alzheimer disease”)|
|Diarrhoel disease||510||(“meta-analysis”[pt]) AND (“diarrhea”)|
|Tuberculosis||420||(“meta-analysis”[pt]) AND (“tuberculosis”)|
Of course, there are two big issues with this analysis: first, our queries may not actually return what we want. For instance, an article returned for (“meta-analysis”[pt]) AND (“tuberculosis”) may be a meta-analysis that merely mentions tuberculosis (say as a side effect outcome) but it’s main focus is something else. Second, not all of these meta-analyses could be worthy of analysis. So, to be extra conservative, let’s assume only 10% of these are truly representative of the effort to generate the landscape. That is still roughly 127 analyses to build a landscape!
Now, for our second question, how much the landscape costs, we can reuse our previous result. There we estimated each study costs roughly $138,000 dollars. So, plugging that in ($138,000 x 127) means that if each entry in a landscape were a full meta-analysis, it would cost $17.5M to produce. Of course, not all of the results in a landscape require full meta-analysis, since many may be the only result to answer a particular question (e.g., the only paper to compare two specific therapies, measuring a particular outcome). So, let’s assume 1/3rd actually do. Even then, we estimate that total cost to produce a landscape would be about $5,800,000!
Landscape analysis is one of the primary uses of Evid Science – our tool can generate therapy and disease landscapes, in seconds, all using published results (see the figure below). Here we show a therapy landscape around budesonide – this is a scrolling page; so here we show just the first 4 results (out of 150+).
While we don’t claim to have fully automated the landscape process, we are getting there. So if you find yourself in need of generating one landscape (or multiple!), and don’t want to spend $5,800,000 to do it, let us know. We would be happy to chat.