Association, causation, and the worship of surrogates

I am constantly amazed at how many smart people in medicine and in medical leadership or policy positions fail to grasp the difference between association and causation, and end up focused on a surrogate rather than the issue.

Consider pneumonia. Defined as a disease of the lungs that is characterized by inflammation and consolidation of lung tissue followed by resolution, and is characterized by fever, chills, cough, and difficulty in breathing, it comes in many shapes and sizes. It can be asymptomatic, mild, severe, or fatal. It can be acute or chronic. It can be bacterial, viral,  fungal, chemical, allergic, localized or diffuse.  One could easily develop a severity scoring system for pneumonia using degree of fever, frequency of cough, and low oxygen levels. This Pneumonia Severity Score (PSS) would accurately reflect the severity of the pneumonia, but no reasonable clinician would confuse the severity score (or its components) with the cause of the pneumonia. It simply reflects the underlying process.

It would be absurd to grade the quality of pneumonia treatment by looking at improvement in the severity score. A clinician could easily improve the severity score by administering oxygen, narcotics to suppress cough, and ibuprofen for fever.  This would give the institution or clinician good metrics and allow them to ‘earn’ pay-for-performance (P4P) bonuses. There is a problem with this approach: it would not change outcomes. People would die or recover from their pneumonia at the same rate.  That’s because the PSS is a surrogate that is associated with pneumonia severity, and not a causative factor. Focusing on the surrogate at best is foolish. At worst, it would incentivize clinicians to ignore the real causation and underlying pathologic processes and result in worse outcomes and deaths.

Think this is far fetched?  It’s not.  In fact, this how many current quality incentives are structured. 

The A1c in diabetes is the poster child for confusing association and causation, and focusing on a surrogate rather than the actual problem. There is no question that the A1c reflects average blood sugars, that better control of diabetes is associated with lower A1cs, and that lower A1cs are associated with better outcomes.  There is also no question that the A1c can be useful for the patient and clinician to track how well they are doing at lowering average sugars. 

However, the A1c is like our hypothetical PSS for pneumonia.  It reflects but does not cause control. While there is evidence that some interventions which lower A1c also reduce complications, there is no evidence that lowering the A1c - in and of itself - reduces complications or improves outcomes. In fact, there is  clear evidence that lowering A1cs too much causes worse outcomes. It may well be that the ease of lowering the A1c with treatment reflects the severity of the disease, its propensity to cause complications, or the patient’s engagement in treatment, rather than the skill of the clinician or the quality of the treatment.  A QI or P4P program that rewards clinicians or institutions based on lower A1cs risks increasing the use of drugs that lower sugars but worsen outcomes, doses that lower sugars but cause complications like hypoglycemia, the use of multiple drugs or complicated regimens that result in drug interactions or worsen quality of life. 

A similar problem exists for hypertension, where quality incentives and P4P programs target a specific BP as the marker for good care and the ticket to incentives. Beta blockers lower blood pressure pretty well but do not reduce the risk of heart attacks or strokes, so one can achieve good metrics without providing good care.

Focusing narrowly on the A1c or on specific BP targets is simple but stupid. It is tempting because quality is hard to define, and harder to measure. When you have pneumonia, you do not want your clinician to treat your fever and cough and ignore the bacteria causing your illness. For the same reasons, you do not want your clinician to treat your A1c or systolic blood pressure. You want them to assess you and your illness and help you find the most appropriate treatment - and not be distracted by worries about his salary being impacted by poorly chosen surrogates on a report card.


Links to more on this topic::