Pharma funded research alternatives

[psychnpgirl]

With the rise of Big Data, do you think it would be possible to form some kind of de-identifying, HIPAA-compliant software cloudware thing that all psychotropic prescribers could be given for free by the NIH to process all their EHR notes and look for patterns of response to all our medications and compile it in a national database to assess response to medications? I'm not saying it would replace randomized control trials or meta-analyses, but could add another dimension to the available research.

---

Members don't see this ad.

 

[WisNeuro]

Considering how much needless documentation already eats up any clinical providers time, not even just prescribing providers, I see a major roadblock. How do you account for the extra time throughout the day to require them to engage in data entry when their schedules are usually packed out to the gills to hit productivity targets?

Before that, you'd need to train/educate them to do the data entry. Each entered variable would need to be operationally defined within narrow parameters. A large database loses it's helpfulness when people are defining the variables any of a thousand possible ways. This would likely be one of the messiest datasets imaginable.

I could see it working, kind of, in a single academic medical center type setting, but nationally, this would be a nightmare. Just from a logistical standpoint. This is coming from experience on several multi-site clinical research projects with extensive funding. This would be a hugely expensive and laborious undertaking. Like tens of billions expensive.

And, from the title, pharma has no reason to put any money into this. At all. Pharma studies focus on the internal validity aspects of research, they really don't care much about the external validity, especially early on. They want to design a study to show that their drug is the most efficacious thing out there. Mostly through data manipulation and inappropriate study and statistical design. Why would they help fund a project that would essentially show smaller effect sizes than their trials data?

 

[psychnpgirl]

I visualize no extra data entry. more charting would be totally unrealistic and I would refuse. simply analysis same way google searches and such are analyzed.

My point is to not have any Pharma funding - to create an alternative/complementary research mode to avoid drug company bias. my point is exactly that - to assess external validity.

how much do data analysis projects at marketing firms cost? tens of billions sound like a lot.

 

[WisNeuro]

Well then, the title should have been "Alternatives to Pharma funded research" then. So, are you talking about a program that scrapes data from EMRs into a database? You're still running into the variable problem of things being defined in different ways within and between patients and providers even if you get past the biggest logistical hurdle of security and HIPAA compliance with the undertaking.

 

[WisNeuro]

Or, are you talking about creating one, single EMR that everyone uses? Which, unless we go to a single-payer system, will never happen in our lifetime.

 

[dl2dp2]

There are lots of existing databases of "Medium Data" of existing randomized trials that's already conducted. Many of them are public. You are welcome to mine them.

The bottleneck in research isn't the availability of data. It's the availability of the funding that pays for individuals who have the expertise to extract value from the data, and secondarily, the availability of the individuals who have the expertise (though through the laws of supply and demand, these two things are connected).

 

[WisNeuro]

The bottleneck in research isn't the availability of data. It's the availability of the funding that pays for individuals who have the expertise to extract value from the data, and secondarily, the availability of the individuals who have the expertise (though through the laws of supply and demand, these two things are connected).

I will say that in many areas, there is a problem in the availability of good data.

 

[psychnpgirl]

My title is a misnomer then. Sorry.

I agree- there is an absence of unbiased data. I wouldn’t include any trial data in this, only real world epidemiologic datapoints, to avoid any possible trial bias.

I see it as analogous this to the algorithms Google uses to determine what ads to show us. They too run into the issue of different terms and definitions but still create useful tho imperfect results.

I should have said complementary, not alternative. A way of analyzing data that is completely epidemiologic and not lab based in nature, to avoid any risk of bias.

Perhaps an easier though less useful approach would be mining the forums of Erowid, Crazymeds, Social anxiety message boards, bipolar message boards, for responses to medications posted by random online folks. Or sending out surveys to populations to ask if they’ve ever been on meds, which ones, and how do they interpret the effects. Wouldn’t be like an RCT but could be of value.

Big data analysis just seems like it could be useful in empowering the field to rely less on Pharma funded research.

 

[WisNeuro]

I should have said complementary, not alternative. A way of analyzing data that is completely epidemiologic and not lab based in nature, to avoid any risk of bias.

It's still biased by the system that a provider works in and how it handles such things. I've worked in many hospitals over the years and they all document things differently. The data is inherently biased by its lack of uniformity and the system which sets up the prescribed parameters.

Perhaps an easier though less useful approach would be mining the forums of Erowid, Crazymeds, Social anxiety message boards, bipolar message boards, for responses to medications posted by random online folks. Or sending out surveys to populations to ask if they’ve ever been on meds, which ones, and how do they interpret the effects. Wouldn’t be like an RCT but could be of value.

You run into huge issues of response bias in this one. The online message board would lead to huge outlier problems. This is also present in the surveying. We know from previous work that in these kinds of situations, you get extremely low response from people not having any problems, and much higher response from those that do have problems. The data you get back will be enormously skewed. Additionally, how are you controlling for all of the other myriad numbers of factors that could be responsible for what the patient feels are adverse effects of medications (somaticization, comorbid health conditions, poor nutrition, poor health maintenance, comorbid psych disorders, etc)?

I don't mean to piss in your Corn Flakes here, it's just that there are huge hurdles to such an undertaking. And, with a fairly large undertaking, you'd have to clear out those hurdles and show why such a study design could give you either newer/better/more comprehensive/more useful data that what is already out there. Big Data is all fine and such, but any data gathering undertaking still has to contend with the GIGO problem. If you don't want to rely on pharma funded research, an easier avenue would be to lobby to stop the year over year of declines of publicly funded research.

 

[psychnpgirl]

Totally hear you. One thing though is that on a massive enough scale since all meds would be subject to the same reporting and systems biases you outline , would that on some level not control for bias? Since all are biased in the same way and the sheer number of numbers would give a snapshot of the real world landscape thus providing a new complementary source of external validity? Correct my logic if it’s wrong..

 

[WisNeuro]

Totally hear you. One thing though is that on a massive enough scale since all meds would be subject to the same reporting and systems biases you outline , would that on some level not control for bias? Since all are biased in the same way and the sheer number of numbers would give a snapshot of the real world landscape thus providing a new complementary source of external validity? Correct my logic if it’s wrong..

No, because the error that is introduced is not systematic in nature across the system. The data would not be biased in all the same way, it would be biased in different ways at different sites and even between different providers. In data analysis, you are ok with some systematic variance because you can control for it, if the variance is not systematic, you run into many more problems with analyzing the data.

 

[psychnpgirl]

That makes sense to me. So why is epidemiology even a thing then? How do ever control for variance on a population scale?

 

[WisNeuro]

That makes sense to me. So why is epidemiology even a thing then? How do ever control for variance on a population scale?

It's messy data, which is why it should always be interpreted with caution. Someone who does epidemiological research can explain it much better than I (who has only worked on projects ranging from 50-2000ish) but, one thing in epidemiological studies is that you try to constrain your variables of interest within set parameters. You carefully define the variables you are collecting so that they are similar, rather than leaving them up to the interpretations and systematic variations of what would in this case be thousands of people "entering data."

 

[psychnpgirl]

SO maybe the thing I’m describing could be done as an epidemiologic study within a few different major Va hospitals.

 

[birchswing]

I don't think messy data poses a problem long-term. Google makes sense of messy data, such as predicting the flu with various searches. At some point I think AI computing will be able to make more and more sense out of everything we've collectively dumped onto the Internet. And adding spurious data and using an algorithm to account for it can create necessary privacy protections (Differential privacy - Wikipedia). The bottleneck I would see is getting organizations to hand their data over (and many like my own psychiatrist don't even use EHR). It seems like something that would have to be part of a massive regulatory change like HIPAA or the ACA, to require providers to give access to differentially privatized data for public use. I'm not sure if they have a motivation to otherwise. But a place like Stanford, for example, might do it just for their own patients.

 

[WisNeuro]

SO maybe the thing I’m describing could be done as an epidemiologic study within a few different major Va hospitals.

Theoretically, easier. But the VA is still a heterogeneous system. And, the patient population could be argued as fundamentally different than the general population, reducing generalizability and limiting interpretation.

I don't think messy data poses a problem long-term.

Messy data is always a problem, in any kind of research. GIGO.

 

[dl2dp2]

SO maybe the thing I’m describing could be done as an epidemiologic study within a few different major Va hospitals.

There are several major "Big Data" in mental health efforts sponsored by the NIH, and there are also at least a handful of pharma related efforts representing collaborations between tech companies and pharma companies in these areas. Insurance companies and care providers (i.e. Kaiser) are also very interested in this area. If you are interested in working in this area I would encourage you to read more about it. This is probably one of the most active areas of research in mental health in the coming decades.

As you can see, your questions are not easy to answer because they induce more questions. It's like answering "How big is the universe and can we use Big Data to study it?" It's fairly meaningless as a question--more of a tagline.

 

[dl2dp2]

I will say that in many areas, there is a problem in the availability of good data.

That's a separate issue in my mind. To collect good data, you need EVEN MORE money. There is enough good data out there to occupy the time of lots of data science experts. The problem is we can't afford them. There are a lot of high quality clinical research data that's released and very much underutilized.

 

[psychnpgirl]

Cool I can’t decide what degree I should get to study these things. I’m currently a psych NP. Should I get a PhD or MPH?

 

[dl2dp2]

Cool I can’t decide what degree I should get to study these things. I’m currently a psych NP. Should I get a PhD or MPH?

PhD in computer science or bioinformatics.

 

[birchswing]

Messy data is always a problem, in any kind of research. GIGO.

I've been wistfully hopeful about the power of AI ever since watching Her. Not for the purpose of a romantic companion. I just think it would be neat if there could be order made out of the large virtual garbage dump that is the Internet where so many things are stored in so many differently contextualized ways, sometimes written to be misinformation or in sarcasm, little of it semantically tagged. And while the movie Her was about a virtual romantic companion, the part that interested me was that the AI could wade through all that garbage and make coherent sense out of it, even enough sense to . . . [the conclusion of the movie].