Adjustment for Risky Patients

When people compare hospitals and doctors, and "control for the differences among patients," the risk adjustment is pitifully poor.

Since risk adjustment is ineffective, hospitals can improve their results by denying care to the patients with the worst conditions (“We can’t help you…”), giving the hospital a better “success” rate. Attention to outcome measures leads to denial of care to the sickest.
​

• “the most assiduous work on risk adjustment has produced tools of only moderate power. The prospects for solving this problem with improved risk adjustment are not promising.[4],[5]” http://medicaring.org/2014/12/16/protecting-hospitals/
• “You can’t get all the right variables on the page” said Berwick, former Medicare Administrator. ​https://jamanetwork.com/journals/jama/fullarticle/2673607

As a very professional and problematic example, Medicare’s adjustment of health condition (HCC) is poor. It explains only 2% to 12% of the total variation actually caused by patient mix (p.65 table 3-22, “r-squared” of version 21).
http://www.cms.gov/Medicare/Health-Plans/MedicareAdvtgSpecRateStats/downloads/evaluation_risk_adj_model_2011.pdf
​
Medicare’s adjustment of patient mix for readmission penalties is also poor. For example, their equations explain 3% of the variation in readmissions among heart failures (p.30), 5% for heart attacks (p.30) and pneumonia (p.29). These percents date from 2008 and have not been updated.
http://www.globe1234.info/medicare/category/medicare-texts
Medicare now shows c-statistics between 0.61 and 0.66 for readmission penalties,
http://www.qualitynet.org/dcs/ContentServer?c=Page&pagename=QnetPublic%2FPage%2FQnetTier4&cid=1219069855841
 The c-statistic has a scale of 0.5 to 1, where 0.5 means their equations do no better than chance, and 1 means their equations are perfect. So some equations are little better than chance, and they still rate hospitals with them. “Models are typically considered reasonable when the C-statistic is higher than 0.7 and strong when C exceeds 0.8”
http://mchp-appserv.cpe.umanitoba.ca/viewDefinition.php?definitionID=104234
​so none of their equations is “reasonable,” and they still charge hospitals hundreds of millions of dollars of penalties with them each year, driving hospitals to reduce admissions among the sickest.
​
The Society of Thoracic Surgeons (STS) has its own risk adjustment.

• https://publicreporting.sts.org/chsd-risk-model
• https://www.sts.org/sites/default/files/documents/Shahian-2018-Risk-Model-Part-1-of-2.pdf
• https://www.sts.org/sites/default/files/documents/Shahian-%202018%20Risk%20Model-Part%202%20of%202.pdf 
• https://ars.els-cdn.com/content/image/1-s2.0-S0003497518303710-mmc1.docx has c-statistics

They give “c-statistics” ranging from 0.616 to 0.826, so some of their equations are not "reasonable," and are little better than chance, but they use them to compare hospitals.