There is an interesting paper in Health Affairs on what types of insurances are paying for increased opioid admissions and costs. I had a good discussion on Twitter with a health researcher about data validity due to risk adjustment. I am a bit wary of diagnosis based assessments because there is a major non-random shock to the data set for some classes of admissions that is not applicable to other groups.
In 2012, ~40% of hospitalizations for opioid abuse/dep paid for by Medicaid @Health_Affairshttps://t.co/cY2sFUMK2B pic.twitter.com/pdtBc7GOZC
— Walid Gellad (@walidgellad) June 5, 2016
@bjdickmayhew To be clear, any discharge diagnosis including opioid ICD-9’s included. so not necessarily primary diagnoses. but still a lot.
— Walid Gellad (@walidgellad) June 6, 2016
As an insurance plumber, one of the truisms is that we only get good, complete and complex data from providers when that data triggers money to the providers. This is why directory data tends to be bad. Most claims and claims systems pay on the CPT-4 procedure code. Most claims systems will require at least one ICD-9 or ICD-10 diagnosis code for payment. They can handle dozens of diagnosis codes, but paying a low dollar claim is seldom dependent on what particular diagnosis is submitted.
This matters for risk adjustment.
Medicare Advantage and Medicaid Managed Care Organizations (MCOs) operate in a world of risk adjustment. Medicare Advantage uses the HCC model which is a diagnosis based model. If a person gets a diagnosis of X on a clean claim, then the Medicare Advantage insurer gets a kick payment tied to the Diagnosis X. MCOs often are risk adjusted using similar non-HCC models that are overwhelmingly driven by diagnoses received on a claim. There are some models that take some pharmacy data into consideration. Some states run their risk adjustment programs similar to Medicare where a kick payment is made for each particular diagnosis category submitted, while others use a relative risk revenue neutral model like Exchange.
Risk adjustment where there is significant money at stake creates a strong incentive for risk adjusted insurers to aggressively chase diagnoses. Insurers will create chase lists and pay incentives to providers to close diagnoses. The incentive is to create an exhaustive list of defensible diagnoses. One of the projects that I am familiar with creates chases lists that dump directly into the Electronic Medical Record system of a major hospital and specialist provider group. A patient being discharged will have a list of possible conditions that were not primary treatment conditions that could be coded for it the doctor used that information in any of their decision making processes. Coding education and coding payments can lead to upcoding and fraud because there is money on the line.
Commercial/Employer sponsored insurance and pre-PPACA individual market insurance was not risk adjusted. For low dollar and medium dollar claims as long as the CPT4 codes made sense to the claims system’s logic and there was a valid diagnosis code on the claim, the claim would pay at the regular rate. There is no strong reason in the ESI world for insurers to care too much about the diagnosis coding intensity.
We have some evidence of the impact of risk adjustment on coding intensity as Fee For Service Medicare currently sees its covered lives as coded as 10% healthier than Medicare Advantage membership even though the underlying health status of the two populations are roughly the same.
According to a new paper by Richard Kronick and W. Pete Welch, upcoding by Medicare Advantage plans happens. Big time. This matters because Medicare Advantage (MA) plans are paid more for higher risk score enrollees….
Fee for Service Medicare has no risk adjustment incentive so the coding is lighter than Medicare Advantage.
What does this mean?
I have complete trust in the comparability of the data between ESI, Medicare and Medicaid if the diagnosis is the primary diagnosis that is driving the admission. That data will have some idiosyncratic variance but over a national claims universe, those things should wash out. However, I would be suspicious of the diagnosis data if it occupies the “other diagnoses” slots for Medicare Advantage and MCO Medicaid as those are the places where providers fill in their risk adjustment codes even if there was minimal involvement of those codes for decision making, treatment or evaluation purposes.
Ideally, large studies that attempt to attribute disease conditions to different payer categories would use a combination of CPT-4 procedure codes as those codes drive payment and thus drive clean provider created data and then diagnosis coding to clean the data after the first pass. As more and more of the world becomes risk adjusted through Medicare Advantage and MCO expansion, Exchange/small group SHOP, Medicare FFS Accountable Care Organization (ACO) build out and private sector ACO proliferation, the diagnosis data will become even muddier and less comparable between different payer groupings.
Benw
Maybe you can correct the “other diagnosis” data by taking a cohort of providers from across the payment spectrum and compare their primary vs other diagnosis rates. Then you have a handle to calibrate the inflated diagnosis rates to the primary, supposedly correct rates, and can scale accordingly.
Benw
No wait, you don’t want to correct the data, you just want to apply uncertainties correctly! So you need to know the difference between the “true” diagnosis and what other diagnoses are a billing artifact (i.e. fabrications). You probably can’t do this by comparing primary to other diagnosis rates because the secondary diagnoses are skewed by being applied on top of the primary, I assume. So you have to get a set of providers to cough up the truth of how they bill and the rates in which diagnoses are true vs for billing. Then you can apply a systematic uncertainty based on that to your data!
Richard Mayhew
@Benw: it is not even fabrications.
For instance there is an icd9 v-code for history of a liver transplant. That is relevant to treatment in a lot of scenarios including an annual well pcp visit. Under ESI and FFS there is no strong reason for a provider to put that status code on a claim. In risk adjustment world, the doc is getting pinged and paid to put that v-code on a well visit pcp claim.
It is real, it is there but it is not always coded as it is not always immediately relevant to a problem.
amygdala
My hospital embarked on a huge billing capture program (which carried some sort of Orwellian name like “data integrity”… it was so absurd I’ve repressed it) in which practitioners were “educated” about adding modifiers that were at best fuzzy and nearly always clinically irrelevant to the patient’s various diagnoses.
If it were possible to analyze data from particular facilities before and after the implementation of such programs, that might be a way of estimating effect size of such interventions. Emphasis on estimate, since in addition to the usual problems of historical controls, the ACA itself could make the before and after groups significantly different.
Death certificate data are bad enough, unless it’s based on autopsy, and I’m only more skeptical of coding data. ICD in its various versions and CPT don’t work well for capturing clinical thinking. The ICD-9 codes for stroke, one of the most common reasons patients require acute neurologic care, are ridiculous, reflecting neither current or even historical clinical classification. I got out before ICD-10 came into use, but that degree of granularity will only exacerbate what happened under ICD-9: that clinicians learn a dozen or so diagnosis codes for the conditions they see the most, and use those.
That is all worsened by EHRs that forbid closing a note without using those codes, even on an admission note when it’s not clear what’s going on with the patient. That provisional diagnosis can get carried through even when subsequent workup clearly demonstrates that something else is going on.
And that’s before the “data integrity” people start leaving you notes in the chart “reminding” you to add this or that modifier to an inactive diagnosis that you’re including in your daily notes mostly to remind yourself to avoid interventions for the active problems that will turn one of those currently irrelevant issues into something that could cause the patient grief.
Maybe the sheer volume of data compensates for how messy it is, but from the practitioner side, I’d be concerned about GIGO. Not to say there aren’t some signals the data could provide, but mostly at a pretty macro level.
MomSense
Hi Richard, I have a huge favor/question for you. My friend was diagnosed with stage IV breast cancer. She is transferring her care to Dana Farber but her insurance is a cost sharing silver here in Maine. Dana Farber wants to do their own CTscans, etc and not use the ones done in Maine but insurance won’t approve of them so soon after the first were done.
She received a call asking about MaineCare (Medicaid enrollment) but she doesn’t know who it was who called and there are some language issues. She may have said no thinking her current insurance is enough. I’m thinking that it could potentially be Medicaid picking up what isn’t covered by insurance. There is also the fact that she can’t work now. Anyway, I’m wondering who she should talk to about it. I encouraged her that it could be a really good thing because I know her treatments are only going to get more expensive and I don’t think she can pay them without risking her business.
Richard Mayhew
@MomSense: Most states Medicaid programs have a Breast and Cervical Cancer eligiblity category. There is a very good chance that Medicaid would be the secondary insurance for her to cover co-pays and deductibles that she incurs in her course of treatment for breast cancer.
Have her call them back immediately as this is a very valuable disease carve-out.
MomSense
@Richard Mayhew:
Thanks! That is what I was hoping you would say! I’ll go over and make the call with her.
Richard Mayhew
@amygdala: Nope, the volume does not help, it is GIGO(ish) at best.