The bean counters are missing some beans
There is a recently released report from the Society of Actuaries entitled the Economic Measurement of Medical Errors. It's nearly 300 pages long and includes data on catheter-associated UTI, central line associated bloodstream infection, and surgical site infections. It is based entirely on administrative claims data, which are notoriously inaccurate for healthcare-associated infections. See Kurt Stevenson's paper on this topic here.
As I looked through the report, I noted that the numbers looked quite odd based on my familiarity with the literature. So I compared the SOA report data to CDC's estimates on the burden of HAIs in the US (Klevens et al) and Eli's review of the literature on attributable cost in the table below.
Now there a number of caveats to point out:
As I looked through the report, I noted that the numbers looked quite odd based on my familiarity with the literature. So I compared the SOA report data to CDC's estimates on the burden of HAIs in the US (Klevens et al) and Eli's review of the literature on attributable cost in the table below.
Estimated annual number of cases | Attributable cost/case | |||
SOA Report | CDC | SOA Report | Perencevich | |
CA-UTI | 9,080 | 561,667 | $32,820 | $1,257 |
CLABSI | 3,679 | 248,678 | $110,462 | $18,462 |
Now there a number of caveats to point out:
- The SOA reports the estimated number of cases due to error; to convert from number of cases to cases due to error they multiplied the number of cases by 0.95. Therefore, I divided the "error" cases by 0.95 to convert back to number of cases (Does anyone believe that 95% of CA-UTI cases are due to error, that is, preventable???)
- Eli used 2005 dollars for cost data, and SOA used 2008
- SOA used 2008 claims data, and CDC (Klevens et al) used 1990-2002 NNIS data and National Hospital Discharge Survey 2002 data
- CDC data appear to included non-device associated infections, though we know that the vast majority of UTIs and BSIs are device related
However, despite the differences I note, the SOA data seem hugely flawed. They appear to vastly underestimate the frequency of HAIs, while substantially overestimating the attributable costs. I suspect the SOA report will be widely quoted, so take a look at it and be prepared!
Nice post. I suspect that the SOA analysis like other claims-based studies suffers from a type of temporal bias. We typically talk about temporal bias in cross-sectional or risk-factor studies when we can't determine if the suspected exposure preceded the outcome.
ReplyDeleteIn this case, administrative claims data doesn't contain the exact time of the infection/"exposure", so it's likely that many of the costs or "outcomes" actually manifest prior to the infection. In this case, it is likely that many of the costs associated with UTIs were secondary to the patients extended length of stay prior to the UTIs. UTIs are just a marker for sicker patients with long hospital stays, which are a risk factor for UTIs.
While the SOA claims that UTI->long stays -> costs, it is likely that the causal pathway that predominates is long stay->costs->UTI.
Another thing to look at is whether the cost estimate passes the giggle test. This isn't scientific, but is it really likely that a CAUTI costs as much as a CABG procedure? Thus, the $32k estimate is quite laughable. I would say that the SOA should have consulted with clinicians, but many clinical papers in widely read journals, like CID, make this mistake monthly.