August 2019: ID Diagnostics

How does the Karius next-generation sequencing assay perform in children with suspected infection?  This paper comes from the Heme/Onc and ID divisions of the Children’s Hospital of Chicago, and for what it’s worth none of the authors are affiliated with Karius or disclosed potential conflicts of interest with the company.  If you haven’t heard of Karius or next-gen sequencing (NGS), the idea is that you take a plasma sample from a patient with suspected infection, isolate the cell-free DNA, run PCR on it with a cocktail of primers so broad they ought to amplify anything, and then sequence what you get out and compare that to a library of fungal, bacterial, parasitic, and (DNA) viral pathogens.  In theory, this could be your one-stop shop for pathogen identification.  However, it has four big limitations: (1) it can’t ‘localize the lesion’ to aid in surgical source control, (2) there’s little data to show it can differentiate between infecting and colonizing organisms, (3) it costs an arm and a leg, and (4) there’s still very little real-world validation of the test’s clinical utility.  This paper aims to address the last concern.

The authors retrospectively reviewed their use of the Karius assay over a year and a half.  They classified test results as “clinically relevant” if the treating physician made management decisions based on the test result; if this was unclear by chart review, clinical relevance was determined by two of the Pedi ID physician co-authors, with adjudication by a third Pedi ID specialist in cases of disagreement.  Finally, sensitivity and specificity for the assay were calculated by also considering each patient’s standard laboratory workup and defining a “true positive” as identification of a clinically relevant pathogen by either Kariaus or standard methods.

The authors found 79 patients (aged 6mo to 24 years, 76% immune compromised) who received a total 100 Karius assays during the study period.  Half of these folks were tested out of concern for a fungal or other atypical infection, and a fifth each were tested for acute onset of fever/sepsis and prolonged or recurrent fever, respectively. The Karius assay returned a positive result in 70/100 cases; of these, 37/70 identified a single organisms and the remainder multiple organisms.  The positive Karius assays most often identified bacteria (74%), followed by viruses (36%) and fungi (26%); the authors thought these results clinically relevant in 80% of cases, and calculated the sensitivity and specificity of the Karius assay as 92% and 64% vs 77% and 89% for the standard microbiologic workup (p<0.01 for both comparisons).

Now, spoiler alert: based on personal anecdotal experiences, I am kind of a fan of the Karius assay.  That said, this is a squirrelly outcome definition if I’ve ever seen one.  Defining “clinically relevant” as “the doctors acted on the test result” leaves no room for questioning whether that action was appropriate or beneficial to the patient.  By these criteria, the urine culture would be a “clinically relevant” test for asymptomatic bacteriuria because so many clinicians inappropriately order and prescribe based on it.  A better study design would have been to use your three Pedi ID coauthors to adjudicate whether the Karius assay identified a clinically probable cause of the patient’s infection, and whether it led to improved patient care.  The most frustrating part of this paper is that the authors show in a fantastic table that they were collecting all the information needed to make those calls – they even stated in each case whether they thought Karius could have averted one of the invasive procedures the patient underwent.  Too bad that’s all buried in the supplemental materials. C’mon guys! 31375834

Speaking of NGS, a similar study was published this month out of university hospital in Shanghai.  This team used an in-house NGS system and compared it to standard up for the diagnosis of patients with suspected focal infections. Here, the authors had two clinicians make a clinical diagnosis of infection based on patient data, and two other reasearchers interpret the NGS results and their relevance to the clinically diagnosed type of infection.  The authors made a final diagnosis of definite focal infection, possible focal infection, and non-focal infection based on the combination of clinical and pathogen data.  Of 95 enrolled patients, 67 were diagnosed with a definite focal infection.  The most common organisms were K.pneumoniae (recall that this part of the world has the hypervirulent K.pneumo strains), M. tuberculosis, S. aureus, and P.aeruginosa.  NGS data was more often both positive and compatible with the clinical diagnosis than standard testing (86% vs 45-58%), and agreement between positive NGS and conventional testing was 90%.  So, I do think there’s something to NGS as a diagnostic tool: 31442461

Synovial D-lactate is about as effective as synovial WBC count in diagnosing periprosthetic joint infection.  D-lactate is produced by bacteria but not people (remember D-lactic acidosis is a problem in people with short-gut syndrome, whose GI tracts leave all the calories for bacteria to consume); hence, it’s a promising marker for bacterial infection.   These authors evaluated D-lactate in synovial fluid as a diagnostic tool for PJI.  The study used a prospective cohort design, enrolling adults seen for a pain at the site of a knee, hip, or shoulder prosthesis over a one-year period; they excluded patients in whom the synovial fluid was diluted, <3ml in volume, or had sat out for more than 48hr.  The D-lactate test results were collected strictly for research; treating physicians were not appraised of the results.  The authors used the European Bone and Joint Infection Society criteria for PJI, which use a combination of clinical symptoms, histopathologic findings, synovial fluid studies, or microbiology data - they’re fairly generous as clinical definitions go.

One hundred and forty-eight patients were included in the study; the mean age was 70, half were men, 70% had suspected knee infections, only 14% had early suspected infections (i.e. within 3mo of the original device implantation), and 69% underwent some sort of revision surgery.  Ultimately 44 patients were found to have PJI (48% due to CONS and 22% to S.aureus) and 104 patients had aseptic failure.  Of the synovial fluid tests, WBC count >2000 had the greatest diagnostic yield (sensitivity 80% and specificity 91%; AUC 91%); synovial D-lactate at a cutoff of 1.26 mmol/L was the next best test (sensitivity 86% and specificity 82%; AUC 90%).  When the cases of PIJ were stratified by timing, the D-lactate was more sensitive but less specific (100% and 58%, respective) for early infections.  Maybe I missed it, but I really wish they’d given us the diagnostic parameters of the combination of D-lactate and WBC, and used their data to define D-lactate cutoffs that reduce the false-positive and false-negative rates of the synovial WBC count, respectively.  Because I don’t see anyone setting aside the synovial WBC count anytime soon, and in fact these data suggest it remains the non-microbiologic synovial fluid test of choice. 31125637

When considering bacterial infection in the ED, look to the eosinophils.  So, acute bacterial infections of just about any type cause a leukocytosis with neutrophilia, and in really severe cases margination of immature neutrophils (“bandemia”).  Interestingly, the same infectious stress response causes eosinophilis to plummet.  So, can we use the absolute eosinophil count (AEC) as a marker of bacterial infection?

This study was conducted across several EDs in France.  The authors retrospectively enrolled all patients seen over a six-month period who received diagnoses of pneumonia, pyelonephritis, prostatitis, appendicitis, cholecystitis, or diverticulitis (“infected patients”); as a comparator group, they randomly selected an equal number of patients seen in the ED for other problems (“uninfected patients”).  Finally, they compared the results of the complete blood count and differential between groups and generated diagnostic characteristics for each test component.

The final cohort included total 876 patients (466 infected and uninfected patients each).  Among the infected patients, 29% presented with sepsis, and 3% presented with septic shock.  Compared with the uninfected controls, the infected patients had higher WBC counts (13.5k vs 9.2k/mm3) and absolute neutrophil counts (11.1k vs 6.6k/mm3), and lower eosinophils (59 vs 129/mm3), basophils (33 vs 44/mm3) and lymphocytes (1295 vs 1870/mm3).  In this study, the diagnostically optimal cutoff for leukocytosis was 10380/mm3, which yielded a sensitivity of 74% and specificity of 70% for infection; at a WBC >12k/mm3, the sensitivity dropped to 59% but specificity increased to 82%.  Neutrophilia performed similarly (sensitivity of 74% and specificity of 74%), as did eosinopenia at a cutoff of <50/mm3 (sensitivity 67% specificity 76%).  However, “deep eosinopenia” (i.e. an AEC <10/mm3) was highly specific for infection (sensitivity 35%, specificity 94%; LR+ 6.7).  Basophil, monocyte, and lymphocyte counts were all pretty useless, and CRP actually outperformed all of these CBC parameters, with 80% sensitivity and 86% specificity at a cutoff of 16 mg/L – but that requires a separate test.

Long story short: a very low eosinophil count (<10/mm3) appears highly specific for acute bacterial infection, and can be another clue alongside leukocytosis and neutrophilia.  Just remember that corticosteroids slay eosinophils, so you can’t trust eosinopenia as a maker of infection in a patient on steroids any more than you can trust a leukocytosis.  31119578

Not all Group A Strep causing human disease is Streptococcus pyogenes these days.  Steptococcus dysgalactiae subsp. equisimilis, which is typically a group C strep, appears to have undergone genetic recombination with S.pyogenes and switched to express the group A antigen.  The CDC’s surveillance system identified 35 such isolates from patients with invasive infections between 2015 and 2018, so it seems this is an emerging pathogen. Fun trivia: S.equisimilis is so named because it’s a lot like S.equi, a group C strep that causes “strangles,” an often-fatal submandibular/retropharyngeal necrotizing lymphadentitis that basically causes Ludwig’s angina in horses. 31158071

Procalcitonin-guided duration of therapy does not reduce mortality in severely critically ill patients.  Or does it? I just want to mention this article briefly because it’s on a hot topic but seemed too fishy to be worth a full read and write-up. The authors performed a meta-analysis of 16 RTCs (total n=6452) that compared hospitalized critically ill patients who received either standard or PCT-tailored courses of antibiotics (this could include starting antibiotics based on PCT, stopping based on PCT, or both).  Use of PCT was not associated with lower mortality (RR 0.9 with 95% CI 0.8-1.01 – and, using the raw event rate numbers, about a 0.2% absolute risk reduction in mortality); of course, the other way to look at this is that you can use PCT to potentially reduce antibiotic use in critically ill patients without increasing mortality.

But anyway, what bugged me was a subgroup analysis looking at the antibiotic cessation trials.  The authors write that they found short-term mortality was lower in critically-ill patients receiving PCT-guided vs standard antibiotics if the SOFA score was <8 (RR 0.81 with 95% CI 0.66-0.99) but not if the SOFA score was >8 (RR 0.85 with 95% CI 0.66-1.11).  Wait, what?  How can we conclude that PCT-guided is associated with mortality in one instance but not the other, when the point estimates and confidence bounds are nearly identical, but the sample in one group is half the size and that leads the upper 95% CI bound to cross 1?  This is the statistical equivalent of following the letter of the law but not the spirit.  I think the better way to communicate the authors’ findings is to say that their meta-analysis suggests some (extremely modest absolute risk reduction) mortality benefit for PCT-guided antibiotic therapy in patients with a SOFA <8, and was insufficiently powered to detect a similar degree of benefit among those with a SOFA >8.

Come for the science, stay for the pedantry. 31229612