When prevention success stagnates? Treat the patient!

Prevention is paramount – I do believe this, and I know that is so much of what healthcare epidemiologist strive for; however we often become very myopic and focus exclusively on “modifiable risk factors.” It is refreshing to read a nicely done epidemiologic study to illustrate what an impact the infectious disease community can have by improving the way we approach patient treatment. When reading the recent article in JAMA IM by Michihiko Goto and colleagues I was expecting a nice ecologic study showing an impressionistic picture of how improved treatment processes correlate with improved MRSA bacteremia mortality - but our VA colleagues working with big data have painted more of a realistic 

Le Déjeuner sur l’herbe Painting 

by Édouard Manet, 

1863 Musée d’Orsay, Paris

Google Arts & Culture

than an impressionistic picture. Using the VA database, capturing deaths occurring during both the inpatient stay and the post-discharge period, they quantify improved survival at the patient level is driven by improved processes of care for S. aureus bloodstream infection (Association of Evidence-Based Care Processes With Mortality in Staphylococcus aureus Bacteremia at Veterans Health Administration Hospitals, 2003-2014 JAMA IM).

I have been pushing for transitioning efforts to prevent S. aureus (more specifically MRSA) bacteremia to the post-discharge setting, worried that the recent reductions observed among hospital-onset MRSA BSI are not being realized in the post-acute care setting. We’re getting stuck; in fact we have been stuck for a while at preventing community-associated MRSA BSI (See figure). 

Goto and colleagues provide some clarity to preventing deaths related to MRSA BSI (and S. aureus BSI overall) even during periods of prevention stagnation such as we may be in currently. Of note, the incidences of healthcare-associated and hospital-onset MRSA BSI decreased in VHA hospitals between 2003 and 2014, whereas the incidence of CA bacteremia was stable. This is identical to trends illustrated nationally using CDC's EIP data, suggesting the VA analysis may be reflective of what is going on nationally.

Goto utilized the national Veterans Health Administration (VHA) health care system to first determine how to best risk adjust mortality; and then determine the independent effect of each of three pillars of guideline directed processes of care for managing S. aureus bacteremia (SAB): (1) appropriate antibiotic therapy, (2) echocardiography, and (3) consultation with ID specialists. They report lower risk-adjusted mortality among patients with SAB when they received (1), (2), or (3), and there was a nice dose-response relationship between the number of care processes and mortality. They estimate “57.3% of the decrease in risk-adjusted mortality among patients with SAB between 2003 and 2014 could be attributed to increased use of these evidence-based care processes.”

Their paper also sheds some light on the importance of capturing post-discharge data when quantifying mortality related to processes or infections related to the hospital setting!. Risk-adjusted mortality decreased from 23.5% in 2003 to 18.2% in 2014, regardless of MRSA, MSSA, and place of acquisition (Figure).

From Supplemental Figures, Goto et. al.

I was caught by the discrepancy between these mortality rates and the mortality reported by the CDCs Emerging Infections Program invasive MRSA Surveillance, reported around 12%. The latter is limited to in-hospital or 30 day mortality, whichever comes first. Recent data at IDWeek by one Emerging Infection Program site identified another 30% of deaths among patients with MRSA BSI occurred post-discharge. The Goto paper captures these deaths, making their conclusions more realistic. In addition, their risk adjustment for mortality included over 13 comorbidities, timing of infection, and susceptibility. They did an outstanding job of trying to evaluate the relative importance of each care process while accounting for changes/absence/presence of these underlying predictors of mortality (both inpatient and post-discharge). They even did a sensitivity analysis to account for early deaths, before these care processes could occur. Furthermore, the impact of receiving the care processes was similar regardless if the SAB was hospital-onset or community-onset!

The bottom line – following evidence base care processes does save lives. Their analysis support their conclusion that “there is a need for continued implementation of quality improvement initiatives to increase the adoption of these evidence-based care processes for patients with SAB.” Let’s be bold and call these what they are: performance measures! Here we have a potential metric (proportion of SAB receiving said care process), that are closely linked to improved survival, and can be captured electronically in an objective manner (at least in the VA!) improving the reliability of these metrics across facilities. I was glad to find through a google search that IDSA recently (December 2016) urged CMS to move in this direction -- although the details of the status of this proposed measure (#407) are not clear to me.

While we anticipate novel therapeutics for MRSA BSI including immunotherapy adjuvants or vaccines to become available in the not to distant future; quality improvement efforts in the evidence-based processes of care for SAB now will likely improve our patient’s outcome. I hope we can turn our attention here soon -- and have hospitals rewarded for doing so.

Excess Mortality in CRKp: A Non-Randomized (Fortunately) Trial

Understanding the burden of antimicrobial resistance is critically important if we are to appropriately target research and clinical resources. For years, lack of proper estimates of the morbidity, mortality and costs associated with multidrug-resistant bacteria greatly limited the attention paid to these pathogens. This changed with the 2013 CDC Antibiotic Resistance Threats Report which provided the public with the number 23,000.  In the report, carbapenem-resistant Klebsiella pneumoniae (CRKp) was estimated to cause 7,900 infections and 520 deaths per year. But questions remain: Is the 6.5% (520/7900) mortality estimate high or low and how can we estimate the burden of resistance since we can't (fortunately) perform randomized trials where we randomly infect patients?

To answer these important questions, a group of investigators formed The Consortium on Resistance against Carbapenems in K. pneumoniae (CRACKLE) and just published a cohort study in Clinical Microbiology and Infection. This group, of what appears to be 18 Great Lakes hospitals, prospectively collected CRKp BSI (N=90), pneumonia (N=49), and UTI (N=121) isolates along with a control group (N=223) of patients with CRKp urinary tract colonization. The use of patients colonized but not infected with the pathogen as controls is interesting. The authors explain that they chose these controls since "non-infection-associated contribution to overall mortality is relatively larger in patients colonized with CRKp compared with patients colonized with more susceptible organisms, since risk factors for mortality such as chronic and acute illness, overlap with risk factors for CRKp colonization. An estimate of this non-infection-related mortality may be approximated in patients who are colonized, but not infected with CRKp." This is another way of saying that they wanted to isolate the attributable mortality risk of infection, not underlying disease.

The primary outcome was time-to-hospital-mortality from the time of the first positive CRKp culture as calculated by an adjusted hazard ratio using Cox proportional hazard models. I've included the unadjusted outcomes below. The full paper includes separate models and Kaplan-Meier curves for each infection, which don't differ greatly from the unadjusted outcomes.

As you can see, 39% of both BSI and pneumonia patients died or were transferred to hospice care compared to 12% of controls giving an attributable mortality of 27% for CRKp infection. In the Cox models, the adjusted hazard ratio was 2.59 (1.52-4.50) for BSI and 3.44 (1.80-6.48) for pneumonia. In contrast CRKp UTI was protective in both the unadjusted (3% lower mortality) and adjusted (aHR=0.68, p=0.33) analyses. This is further evidence that we need to rethink our definitions and focus on UTI.

Overall, a very nice study that utilized a novel control group of patients colonized but not infected with the organism of interest. It is likely that this approach when coupled with multivariable analysis reduced the effects of measured and unmeasured confounders. And it looks like CDC should increase the attributable mortality from 6.5% in their 2013 report to something a bit higher - say 27%.

H7N9 Mortality

A week ago, in response to some scary reports, I tweeted this:

I wanted to see how my public prediction held up in the H7N9 outbreak after an admittedly short 7 days and approximately 2 weeks after the outbreak gained widespread attention. To do this, I simply plotted the reported mortality rate using the daily case and mortality counts available through the @WHO twitter feed between April 1st and April 16th.

While the case counts have gone from 3 to 63 over this period, reported deaths have risen more slowly from 2 to 14 and the mortality rate has fallen from 67% to 22%. When I made the prediction it was 30%. If I had included data from April 17th, the mortality rate would have fallen further to 20.7%, 17 deaths and 82 cases. I've plotted the daily mortality rate below. Good news, but still a ridiculously virulent virus. Hopefully, as more surveillance specimens return, the mortality rate will continue to fall. Hopefully.

Addendum: Immediately after posting this, several twitter discussions began. To clarify, I know many of the sick patients are still hospitalized in the ICU. Even cases that appear to initially recover could succumb to secondary bacterial infections. Helen Branswell made the important observation that two of the three cases in the recent NEJM report were infected with Carbapenem-resistant Acinetobacter baumannii and at least one of them didn't appear to receive appropriate antibiotic therapy. Hopefully, as recognition increases, so will early diagnosis and early antiviral therapy along with effective antibiotic therapy (if available).

Swine flu mortality estimate

The New York Times reports this morning that CDC has revised its estimate of deaths due to swine flu in the US at 4,000. The revision now will make the swine flu mortality estimate more comparable to the mortality estimate for seasonal flu, which on average is about 36,000 persons yearly in the US.