Monthly Archives: September 2014

Changing my mind on SES Risk Adjustment

I’m sorry I haven’t had a chance to blog in a while – I took a new job as the Director of the Harvard Global Health Institute and it has completely consumed my life.  I’ve decided it’s time to stop whining and start writing again, and I’m leading off with a piece about adjusting for socioeconomic status. It’s pretty controversial – and a topic where I have changed my mind.  I used to be against it – but having spent some more time thinking about it, it’s the right thing to do under specific circumstances.  This blog is about how I came to change my mind – and the data that got me there.

Changing my mind on SES Risk Adjustment

We recently had a readmission – a straightforward case, really.  Mr. Jones, a 64 year-old homeless veteran, intermittently took his diabetes medications and would often run out.  He had recently been discharged from our hospital (a VA hospital) after admission for hyperglycemia.  The discharging team had been meticulous in their care.  At the time of discharge, they had simplified his medication regimen, called him at his shelter to check in a few days later, and set up a primary care appointment.  They had done basically everything, short of finding Mr. Jones an apartment.

Ten days later, Mr. Jones was back — readmitted with a blood glucose of 600, severely dehydrated and in kidney failure.  His medications had been stolen at the shelter, he reported, and he’d never made it to his primary care appointment.  And then it was too late, and he was back in the hospital.

The following afternoon, I spoke with one of the best statisticians at Harvard, Alan Zaslavsky, about the case.  This is why we need to adjust quality measures for socioeconomic status (SES), he said.  I’m worried, I said. Hospitals shouldn’t get credit for providing bad care to poor patients.  Mr. Jones had a real readmission – and the hospital should own up to it.  Adjusting for SES, I worried, might create a lower standard of care for poor patients and thus, create the “soft bigotry of low expectations” that perpetuates disparities.  But Alan made me wonder: would it really?

To adjust or not to adjust?

Because of Alan’s prompting, I re-examined my assumptions about adjustment for SES. As he walked me through the data, I concluded that the issue of adjustment was far more nuanced than I had appreciated.

Here’s the key: effective socio-economic adjustment doesn’t reward providers for giving bad care to poor patients. It just ensures that they aren’t penalized for taking care of more of them. In my clinical example, if people like Mr. Jones had a higher readmission rate, adjusting for SES wouldn’t give hospitals credit for lower quality care to poor patients.  Done right, it would give credit to hospitals for having more poor patients, and that’s an important difference.  Consider three scenarios of hospital performance on a readmission rates (modified from our JAMA piece).


In scenario 1 and 2, let’s assume that patients are readmitted 20% of the time on average, whether or not they’re poor.  In scenario 1, Hospital A (a safety-net hospital) has higher readmission rates for everyone.  They may have more poor patients, but their readmission rate is high for both poor and non-poor patients.  So, compared to Hospital B, they look worse in unadjusted and adjusted scores.  Adjustment doesn’t help.

In scenario 2, Hospital A has higher readmission rates for its poor patients and therefore has an overall readmission rate of 25%.  Hospital B doesn’t suffer from readmitting its poor patients too often – hence its readmission rate is 20%.  In this case, safety-net hospitals look worse than Hospital B in both unadjusted and adjusted analyses.  Again, adjustment doesn’t help.

In scenario 3, Hospital A and B both struggle with readmissions for their poor patients – as does the rest of the country.  The only thing that differentiates Hospital A from Hospital B is the proportion of poor patients in the hospital.  In this case, adjustment makes a big difference.  By adjusting, we account for the different proportions of poor patients between Hospital A and B.  Adjustment ensures that organizations are judged by how well they care for their patients, not by how many poor patients they have.

One Size Does Not Fit All

The debate about whether to adjust for socioeconomic status needs to be far more nuanced than it has been to date.  Specifically, we must recognize that quality measurement has multiple purposes, and we need to think about each one when deciding whether to adjust or not.  If the goal is transparency –letting patients know how they are likely to fare – then the best approach is stratified data. In scenario 3 (where adjustment makes a difference) a poor patient will do about as well at both hospitals – and unadjusted numbers are misleading, because they tell poor patients that hospital B is better.  If Hospital B has a larger co-pay or is out-of-network, you have done real harm by pushing a patient to a more expensive place that doesn’t provide better care.


To push hospitals to improve quality, unadjusted numbers are best.  In all three scenarios, Hospital A should be more motivated to get better than Hospital B because for its patients, it tends to have worse performance.  But in each scenario, the hospitals need stratified data. Without it they will have no idea where to target their efforts.

For penalties, we should use adjusted data.  It will make no difference in scenarios 1 and 2.  But, in scenario 3, it makes little sense to penalize the safety net hospital compared to other hospitals just for taking care of more poor patients.  That’s not a smart policy.  Penalties for bad care for poor patients?  Sure.  Penalties just for caring for more poor patients?  Not so sure.

 A way forward

The bottom line is that the care of poor patients is not evenly distributed across all U.S. hospitals.  Some hospitals have a lot more patients like Mr. Jones than others have.  And caring for people like him, who are homeless and without a social network, is challenging.  None of us are very good at it.  Why penalize the safety-net hospitals just for taking care of more poor patients?

Given the concern that safety-net hospitals may be disproportionately penalized, a bi-partisan group of Senators (3 Democrats and 3 Republicans) has signed on to a bill that would require CMS to account for SES when it doles out penalties for the HRRP (Senate Bill 2501).  It’s an excellent start.

Adjusting for SES is an acknowledgement that medicine is not the only factor – and indeed may be a relatively minor factor – in health outcomes. For Mr. Jones, homelessness and poverty clearly contributed to his readmission to the hospital. Bad medical care did not. We should have no qualms penalizing safety-net hospitals for providing sub-standard care.  But we just shouldn’t penalize them simply because they have more poor patients.