This is a cross-post, plus commentary at end, starting with an item today by SPM member Alexandra Albin (@MsAxolotl, a frequent patient herself), from her blog: My life in the Bush of Doctors. It arose from a boiling discussion on the SPM member listserv. To participate in the SPM listserv, join here.
The views in Alex’s post are hers, and mine are mine; neither is a policy of the SPM. — e-PatientDave
First, Alex’s cross-post:
Today, on my listserv group at the Society for Participatory Medicine, the question was raised about the NY Times Well Blog post by Tara Pope on FICO’s new Medication Adherence Score. I was also inspired by @epatientDave pointing out that @nyceve1 had a blog post this morning FICO-scoring-millions-of-Americans-on-medication-compliance .
And yes I am angry…because if you look FICOs website here is what they say about their new tool, that is easy to implement. I recommend that any self respecting patient contact them, because they are not in the business of helping patients, they really seem targeted to the pharma companies, and “Care Organizations” and “Health Care Payers” (I don’t think they are talking about us).
Check out the tabs and pdfs for some fun reading…this is a dangerous tool that doesn’t really measure true compliance, i.e. whether someone took the medication. [It only monitors whether you paid for a prescription that a doctor ordered. – Ed.] What if I picked up the medication, or opened the glowing cap? As someone on the #SPM listserv pointed out, “We will not be able to fix compliance problems if the denominator is as wrong as the numerator!!”
The words that rub me the wrong way are highlighted in bold. Nor do they really say what data they are pulling…only the following: “third party data sources, name and address, prescription claims when available.”
So, yes, I have got my cynical hat on….and my take away is (unless we revolt) pharma companies are given tools so they can better target their marketing budget and relationships can be further leveraged with doctors … and, enable deeper relationships with insurance companies who can have further control of what medications are on their formularies, or potentially use the data to develop additional criteria (of course they will use pretty words to make it seem fine) for dropping coverage for patients due to a poor FICO score.
It is bad enough that pharma companies sell their drugs direct to patients, and take doctor offices out to lunch…Anyone see the movie “Love and other Drugs”?
So, I sent them an email. Let’s see what they say…..and you can too on their contact page or call US (toll free): +1 888 342 6336
Here is information from the PDF on the FICO site: (Cynic Hat on)….. [emphasis added] Oh, and it’s easy to implement….
FICO® Medication Adherence Score is a powerful tool for predicting individual consumer’s likelihood of adhering to a drug regimen over the next year. This fully HIPAA-compliant solution helps brands identify patients at highest risk for non- compliance, direct marketing tactics where they have the greatest impact on medication adherence and health outcomes, and maximize the return on the consumer marketing budget. While pharmaceutical marketers typically rely on self-reported adherence data to identify non- compliance, FICO can score an entire patient database or list quickly and efficiently using only an individual’s name and address. Many pharmaceutical companies help address this challenge through a variety of consumer-directed programs. FICO can boost the effectiveness of these programs through a revolutionary data-driven approach to identifying a patient’s propensity toward medication adherence. Using the same world- class predictive analytics used to create the FICO® Score, the FICO® Medication Adherence Score accurately predicts an individual’s adherence propensity using a wide array of third-party data sources commonly used by direct marketers in a variety of industries.
FICO® Medication Adherence Score leverages a patient’s prescription claims history when available and pulls on other third-party data sources when no other information is present. The result is a powerful and versatile score that can be applied universally across a patient base to predict each patient‘s adherence over the next 12 months. This tool enables care organizations to gauge the right level of action across the patient base to optimize care, case and utilization programs— setting a universal baseline assessment on which survey results or other information can be overlaid if desired/present. The result is a powerful and versatile score that can be applied universally across a patient base to predict each patient‘s adherence over the next 12 months. This tool enables care organizations to gauge the right level of action across the patient base to optimize care, case and utilization programs—
The Medication Adherence Score is available for common chronic conditions, including diabetes, asthma, high cholesterol, hypertension and depression. Harnessing the predictive power of multiple, rich third-party data sources, Medication Adherence Score improves the effectiveness of all intervention targeting efforts.
Recent FICO research has shown that third- party data sources can effectively identify drug adherence propensity and can enhance the precision of models using claims data only. While some of these predictors, such as age and gender, are known to be associated with disease prevalence and adherence trends, FICO has unlocked the predictive power of other data sources, such as retail purchase behavior, geo-credit profiles and income/wealth indicators. The result is a powerful assessment tool that works across a prospective, new or existing patient base with minimal information requirements.
Learn how your organization can benefit from the most advanced analysis solution for predicting medication adherence. Email us …
End of Alex’s post
Commentary by Dave –
This may win my all-time prize for Worst Use of Medical Data.
I’ve written before about inept use of poorly chosen data, with nasty consequences for anyone who believes the data without checking its origin, but this scheme is shot through with logical errors, making it a particularly unreliable measure of what it claims to measure.
For one thing, it has no idea whether you actually took the pill; it only knows whether you bought the prescription. So anyone who wanted to game the system (to protect their “credit score”) could just buy the stuff and throw it out.
And that would drive up healthcare costs, costing insurance companies for the purchase of products that were not actually used.
Could this be a scam to maximize prescription sales, by putting a FICO gun to every patient’s head? I’m sure it’s not, but if you were to design one, it would look a lot like this.
But let’s be real – what patient has money to throw away, just to defeat a credit score? More common is patients at the poor end of the scale, which @nyceve1 cited in her Daily Kos post. She notes, among other things, the Commonwealth Fund’s data on the extraordinary payment problems experienced by US patients. In short, a lot of people in the US – far more than any other country, 10x more than in the UK – simply can’t afford their prescribed medications. To construe this as non-compliant behavior – an “adherence risk” – is simply false logic.
Or, perhaps FICO believes they’re justified in giving a lower rating to poor people and the uninsured? You want to kick ’em while they’re down, Fair Isaac? (FICO is Fair Isaac Corporation.)
Some have wondered whether this score won’t be used by insurers to price differentially, just like the interest rates on loans that vary depending on your FICO score today. Wouldn’t that be a lovely screw for the poor: “Buy the scripts, or we’ll charge more for your insurance. Your choice. (We don’t care whether you take it – just buy it.)” If regulations don’t already exist to block that predatory behavior, let’s hope HHS will promptly create them.
Finally, the Kos post links to one by analytics geek @MedicalQuack Barbara Duck, which cites books like insurance whistle blower Wendell Potter’s Deadly Spin and Charles Seifer’s Proofiness, the Dark Side of Mathematical Deception.
My bottom line: if data is going to be used to measure something, it better be valid data, and it better truly measure what it thinks it’s measuring – or it’s just not fair to use it against someone. (And the buyers of the service will be defrauded.)
As Barbara Duck said in her post: has Wall Street not taught us about credit ratings and how those numbers get skewed?
This is completely absurd. I’m amazed too that people can claim this is a “HIPAA-compliant” solution. The HIPAA privacy rules list both name and address as items that make information “identifiable” and matching it with pharmacy records sure doesn’t fit that bill.
It’s appalling to me to think that we now have FICO involved in our health care. While I recognize the challenges of medication compliance, there are so many reasons for someone not taking a prescription that aren’t considered here. These kind of decisions and discussions should be left to those who should be in charge of a patient’s health care–the physician and the patient.
Thanks, Dave, for pointing out this serious issue. I’ll be adding a blog post on it today and will link here as well.
I am against this idea more than 100%. My prescription history should be just between me and my doctor. I don’t need or want any kind of score that tells me or anyone else if I am adherant with my medications. The doctor or PA is the one that should be asking that.
I think there are inaccuracies in this account, and that it’s a bit over the top. I, too, am concerned about this and any attempt to profile specific individuals based on data that “suggests” behaviors. But we should be clear about what FICO is actually doing here.
For instant, there’s an editorial insertion above that says “It only monitors whether you paid for a prescription that a doctor ordered.” However FICO doesn’t appear to be monitoring, but predicting. What they’re predicting is not whether you paid for the prescription, but whether you’re at risk for non-compliance, i.e. whether you’re less likely to follow a prescribed treatment. The professed purpose for this is not sinister on the face of it: supposely the predictive data will enable targeting of those who will most likely fail to take medications as prescribed “to optimize care, case, and utilization programs.” Lack of adherence or compliance is an acknowledged problem: http://www.enotes.com/public-health-encyclopedia/adherence-compliance-behavior.
I personally think that physicians over-prescribe, and I’m opposed to aggressive marketing of prescription drugs, so there are larger relevant questions we could address. But I think it’s important to be accurate.
Re the HIPAA question, note that FICO doesn’t appear to be using medical data associated with identity, and they’re focusing quite a bit on non-medical data.
So, Jon, I’m open to understanding what we’ve overlooked, but can you be specific?
I’ve honestly tried and I can’t imagine any legitimate algorithm that would predict use in a way that’s fair to all consumers who get rated. And I can’t see a way to justify rating consumers without ensuring that it’s fair, since such info can easily be used against us.
Let’s chew on this – what are we missing here? The FICO data sheet does say that they’ll use claims data where it’s available (i.e. insurance billing history) and other, unspecified info where claims data isn’t. Can anyone have any confidence in a fair use of a pig-in-a-poke rating algorithm?
I’m not trying to be snarky – really wondering. I’ve crunched data, and I can’t see a pathway to a valid and fair rating.
In fact I wish the FICO people would show up and explain why we’re wrong. If I personally was wrong I’m glad to be corrected – I hate going around saying wrong stuff!
(btw, nobody said that lack of “compliance” (ugh) isn’t a problem. The complaints as I hear them are the statistical validity of this rating, and its quite possible misuse.)
I can expand on my issues with the post, focusing particulary on what you yourself posted.
If we assume that what they say is accurate, then they’re not monitoring, they’re predicting. Which means that they neither know whether you took the pill nor bought the prescription: rather, they’re predicting whether you will comply, based on an assortment of factors. Marcela noted that there use of healthcare data was in creating the predictive algorithms, and the data was evidently de-identified.
I don’t see what the FICO gun to the patient’s head would be, this seems to be a stretch. What would the gun be? Forced compliance? Denial of access to drugs legitimately described? I don’t see anything like that suggested in the FICO material. I’m not sure we should be going there without evidence that this would be the case.
Being too poor to afford medication IS an adherence risk, I’m not sure why you would say otherwise. It would be reasonable to assume that lack of income could suggest lack of adherence, so if FICO draws that inference, it’s probably correct. This isn’t necessarily a problem for the insurer: if the patient can’t make the copayment and doesn’t buy the drug, they’re not out the money. What about pharma? Could this as easily mean that they would target these patients with subsidies to help them acquire the drugs? I know some pharma companies do this, but on request.
What does a “lower FICO rating” mean? I didn’t see anything about a ratings system, so I’m not prepared to make documented assumptions.
I just see a lot of conjecture in the post. You acknowledge your cynicism. I can be cynical, too… but I think it’s too soon to make these public assumptions. As patient advocates, we make an informed assessment, as opposed to a knee-jerk reaction.
Great discussion – thanks, Jon. This is demonstrating how people can disagree (perhaps strongly) and point to weaknesses in each other’s arguments and listen, and hear each other.
So here’s what I see on their site:
“Leverages historical prescription data or publicly available third-party data
“The Medication Adherence Score will use a patient’s prescription claims history when available and pull on other third-party data sources when no other information is present.”
– it’s a score, aka (in my view) a rating
– it’s based on your insurance claims history, if they can get their hands on it
Maybe I’m missing something but, it seems obvious to me that if I don’t buy what the doctor ordered – for any reason – FICO will detect it, and it will affect my “credit rating” (my trustworthiness rating) in their system. That’s what I meant, specifically, by the “gun to the head”: “Buy what you’re told, or your rating will be affected.”
Hm, and what about if I do buy the prescription, but buy it at a place like Walgreens where generics are cheaper than my deductible, so it doesn’t show up in insurance claims?
(Now that I think of it, I know a guy who always paid cash for his psychiatric medications because he didn’t want it showing up in the system. I wonder how that would play out here – would FICO erroneously conclude that he was “non-compliant” when he was in fact taking the meds, quite avidly?)
On top of that I just don’t think it’s kosher for them to have a non-transparent rating system, with no ability for a consumer to inspect the data and correct errors.
As we discussed on the SPM member listserv, it reminds me of this summer 2008 Consumer Reports video about the “MIB,” the insurance industry databank that asserts they have no liability for harm caused by errors in their data, and nor do the people who inserted the data.
(It’s not directly relevant to this issue, but the post also documents how bad data and persist: it shows how artfully the MIB sleazebags make it hard to know who put the error in there – they actively erase all records of it.)
Since the MIB charges for the service, and is not responsible for the data’s quality, and it can harm consumers, I say that’s unethical and consumers ought to watch out. Data quality is serious business, especially data about us that’s sold for profit, and especially if we have no right to check its accuracy (and get errors corrected).
And I suggest that those issues are identical risks in the FICO instance. Without transparency and ability to correct errors, it’s business with consumer impact without accountability.
Your turn. :–)
I can accept Jon’s assertion that the post was, “over the top” because I am angry for a number of reasons about how the system works for patients. And, sometimes it is important to get angry. I also strongly disagree with the Supreme Court’s decision to allow Pharma companies data mine our prescription behaviour and somehow, to me at least, they are indirectly related.
The thing that got me rolling was mostly the language on their site and how they came to their conclusion.
From what I can see they based the score on …”data from a large pharmacy benefits manager that provided information for a random sample of nearly 600,000 anonymous patients with diabetes, heart disease and asthma. Using the data set, FICO was able to track the patterns of patients who filled and refilled prescriptions and those who didn’t. The company used the data to identify the variables most associated with medication adherence and developed a risk score on a scale of 0 to 500″
!- correlational analysis is a logical fallacy nor scientifically sound.
2-The score does not address the often case of when patients just buy the med and may not use the whole thing. For instance what happens with antibiotic use? or, when medications don’t work or mix well with other meds.
3-What was the breadth of their data…they say “random sample of nearly 600,000 anonymous patients with diabetes, heart disease and asthma”. Was it a random sample in a particular region? under a certain Insurance company? a single pharmacy brand? WE need some transparency on data sources for something that can have a deleterious impact on patient’s “score”
4-The is no value proposition for the patient. This service is being geared to specific components of the healthcare business that are armed with lobbyists in Washington and with Big Companies, especially Pharma and “health payers”, who have typically not been in the business of protecting/helping patients. When I went to their “connect” tab I was really hoping that meant connecting patients to their service..but no.
5-And, Tara’s headline is loaded: “Keeping Score on How You Take Your Medicine” It’s not about taking it is about buying. Big difference.
If you are arguing it is just predicting medication adherence. what is that saying about the prediction criteria? What is the validity? back to my point #2
Of course medication adherence is a critical issue, but we have to use criteria that are valid. Otherwise your are mixing apple and oranges In my mind we are supposed to be the voice of patients and protect patient rights. The other big boys have enough deep pockets
Please shed some light so I don’t have be so galled.
Sorry that I didn’t respond earlier. I agree the accuracy of correlation is questionable, like so many of the metrics used in sales and marketing.
But I think it’s important to point out that they used correlations from an analysis of data up front to create a profile – but I don’t get the impression that they’re extracting each person’s prescription history to predict adherence for that specific person, which I think would be a worse case.
There’s also the question of why they’re doing this. Supposedly it’s to help the patient adhere – i.e. if you know which patients are likely to have an issue following through on treatment for whatever reason, you can target that patient for followups. This doesn’t seem sinister to me, though I may be missing something.
Kathy…that was a great say…Thanks
Nonadherence is WAY to complicated to just assign a score and think that it can be magically fixed.
I have written a post that discussed the issues with utilization of claim data for measuring adherence.
If nonadherence was this easily solved, it would have been fixed decades ago.
If you want to read my thoughts I posted them on my site at:
FICO Score – How Can You Predict Adherence Based on Claim Data?
great. I was just going to add your post to this discussion. I think your post goes far further down the spectrum of understanding.
Substantial comments were added on this subject six months later on Fred Trotter’s post Data, Damn Data, and Statistics. (See the comments section as well as the post.)