Thursday, September 12, 2013

Predicting Survival After CPR: Can we GO-FAR?

JAMA Internal Medicine (JIM, you'll always be Archives of Internal Medicine to me) has published a study of an outcome prediction tool for in-hospital cardiac arrest. Specifically, a tool to predict the rates of neurologically intact survival after an in-hospital arrest. The developers of the tool call it GO-FAR (Good Outcome Following Attempted Resuscitation).

One could imagine ways of re-titling it FORe-GO, if one chose to spend one's time imagine such things.

This is the best tool of its kind that I've seen, although one needs to be very clear about what it's actually predicting.

The paper mostly describes, in detail, the derivation and validation of the tool. I won't belabor this - they did a good job of it, used a large dataset, created a model, tested and validated it to get the best receiver operator curve, etc. The data come from the massive, US-based, 'Get With the Guidelines Registry' (which used to be called the National Registry of Cardiopulmonary Resuscitation). The GWGR collects standardized data on in-hospital cardiac arrests from 366 hospitals in the US (all types and regions of hospitals) (the national rate of CPR survival to discharge of 18% which most of us are aware of comes from the GWGR). The GO-FAR data come from arrests between 2007-2009 (51,000 patients).

Worth belaboring, because this is critically important in understanding if one chooses to use this as a tool to help patients understand CPR outcomes, is how they defined a good outcome: a Cerebral Performance Category (CPC) of 1. CPC of 1 means:

"The patient is conscious, alert, and able to work but might have mild neurologic or psychological deficits (such as mild dysphagia or minor cranial nerve abnormalities). Patients with a CPC score of 2 have moderate cerebral disability and are able to live independently and work in a sheltered environment. Disabilities may include hemiplegia, seizures, ataxia, dysphagia, or permanent memory or mental changes. Patients with CPC scores of 3 through 5 progress through severe cerebral disability, coma or vegetative state, and finally brain death." 
This is important because while I'll venture to claim that nearly everyone would agree CPC scores of 3-5 are dismal outcomes, I think there could be a substantial number of people for whom a CPC of 2 would be acceptable. Not welcomed, not a 'good outcome' (which is, granted, what the researchers here are trying to predict), but better than death for some. So to be clear, the GO-FAR tool predicts rates of survival to hospital discharge with a CPC of 1, every other outcome from a CPC of 2 to death are lumped together as bad outcomes.

GO-FAR is being explicitly developed to help inform discussions at the time of hospital admission, so they deliberately excluded patient/CPR characteristics which predict outcomes but wouldn't be available necessarily to the admitting clinician (such as initial rhythm after arrest, site of arrest as someone may be admitted to the floor then transferred to the ICU prior to arresting). The characteristics which survived their analysis and were included in the final index are below. They very nicely also mentioned what the overall survival to discharge with good outcome was in all of these categories (remember, this is not overall survival, this is survival with a CPC of 1). The overall survival with good outcome rate for the entire dataset (all-comers) is 10%. 
  • Neurologically intact/minimal deficits at time of admission - CPC of 1 (this predicted a better outcome; everything else here predicted worse outcome) (18%)
  • Major trauma (reason for admission) (6%)
  • Acute stroke (reason for admission) (3.7%)
  • Metastatic solid tumor or any hematologic malignancy (5.2%)
  • Septicemia (basically they mean active bacteremia here; not the sepsis syndrome) (3.6%)
  • Medical non-cardiac diagnosis (reason for admission; ie, patients admitted with cardiac conditions did better) (5.6%)
  • Hepatic insufficiency (bilirubin greater than 2mg/dl or AST more than 2 times the upper limit of normal) (4.4%)
  • Admitted from a skilled nursing facility (3.2%)
  • Hypotension or hypoperfusion (5.9%)
  • Renal insufficiency (creatinine over 2mg/dl) or dialysis (6.4%)
  • Pneumonia (5.2%)
  • Age over 70; the older the worse the outcome (10.2% 70-74 down to 4.5% for over 85 years).
Basically these categories are assigned points, and depending on your score, GO-FAR assigns you a percentage rate of a likely outcome (e.g, 'dismal' less than 1%; 'terrible' 1-3%; 'average' 3-15%; or 'above average greater than 15%). The receiver-operator curve for the final model was 0.78 which is good.

The authors point out that one of their worries with this prediction tool is that it is most likely to, if anything, overestimate good outcomes because all the data come from patients for whom CPR was actually attempted (e.g., patients who had DNR orders who were probably sicker as a group than the full code patients are not reflected in these data, by definition, since they didn't have resuscitative attempts.) 

The authors note they are making an on-line GO-FAR calculator, but it's not yet available (the paper's only been e-published and one wonders if they are trying to get it done in time for the print publication). 

The tool is clearly being created as a clinical tool for patient education, particularly at the time of hospitalization. Clearly, if one chooses to use this, it's important to know that it's aimed at answering the question "What are my chances of coming out pretty much normal?" (not what are the chances of mere survival, which are probably very roughly double the 'good outcome' rates). I obviously need to see the calculator, but can imagine using it to inform my discussions with patients. 

In working with residents, I frequently sense a lot of angst about not feeling facile with CPR outcome data, and what I have historically told them (without getting much sense that they've found this helpful) is that the national survival rates are 18%, about half are neurologically intact; and the patients for whom you personally worry about, think a DNR order is medically indicated because such an intervention is so unlikely to help - the good outcome rates are much worse. The GO-FAR tool is likely going to be a big step in clarifying just how much worse those outcomes are. Perhaps the residents will have less angst with this than my imprecise (but accurate, dammit) hand-waving.

Which brings me to the accompanying editorial, which is one of the best things I've read all year, and has very little to do with the GO-FAR paper itself.

Essentially it is a plea for two (interrelated) things.

One, is that we in medicine, hospitals, etc. need to stop treating CPR as something special (i.e., as a default action, as something that is a genuine medical option 'even when its therapeutic potential is remote.') It's basically the only thing in medicine we treat this way, like its a sort of human right, as opposed to a complicated medical-procedural intervention with indications and contraindications, good reasons to do it, and good reasons not to. Ie, a patient having a cardiac arrest is not the sole indication for attempting CPR. 

The second recommendation is that, in this context, hospitals start looking at CPR attempts from a QI standpoint beyond the technical aspects (time to initiate compressions, how closely the attempt followed guidelines, etc). Important stuff, but inadequate. The editoralist argues quality assurance committees must start looking at questions of whether there was an indication for CPR in the first place, if not, why was it offered/why didn't the patient have a DNR order, why weren't people discussing this with the patient/family, etc. 

Is anyone aware of QI committees looking at these aspects of resuscitation efforts in their hospitals?

ResearchBlogging.org
Ebell MH, Jang W, Shen Y, and Geocadin RG (2013). Development and Validation of the Good Outcome Following Attempted Resuscitation (GO-FAR) Score to Predict Neurologically Intact Survival After In-Hospital Cardiopulmonary Resuscitation. JAMA Internal Medicine PMID: 24018585

Pallimed | Blogger Template adapted from Mash2 by Bloggermint