A new prediction rule using just three lab tests—urinalysis, absolute neutrophil count (ANC), and serum procalcitonin (PCT) levels—accurately identifies febrile infants 60 days old or younger at low risk for serious bacterial infection (SBI) (JAMA Pediatr 2018; doi:10.1001/jamapediatrics.2019.5501). Once further validated, this rule could obviate unnecessary treatments in such young patients, according to the investigators.
SBIs occur in 8%-13% of febrile babies ≤2 months old. Because the consequences of missing an SBI can be so serious, treatment for these young patients often involves lumbar punctures, broad-spectrum antibiotics, and hospitalizations, all of which have associated risks. Algorithms to evaluate this population lack precision and specificity and have demonstrated “less than ideal accuracy,” according to the researchers.
The study involved 1,826 febrile infants seen at 26 emergency departments in the Pediatric Emergency Care Applied Research Network. All babies had blood and urine culture samples taken; attending clinicians decided whether patients underwent lumbar puncture for cerebrospinal fluid (CSF) testing; ultimately 76% had CSF cultures.
The investigators defined urinary tract infections (UTI) as growth of a single urine pathogen with at least 1,000 cfu/mL for culture obtained via suprapubic aspiration, with higher thresholds for catheterized specimens.
For the prediction rule, the researchers considered variables like the patient’s temperature, duration of fever, and clinician suspicion, but after univariable analysis and recursive partitioning analysis the model retained urinalysis, ANC, and PCT to identify babies at low risk of SBI. In a derivation set of 908 infants, negative urinalysis, ANC ≤4,090/µL, and PCT ≤1.71 ng/mL identified 522 with an SBI risk of 0.4%. In the validation cohort of 913, the prediction rule using these cutoffs had a sensitivity, specificity, and negative predictive value of 97.7%, 60%, and 99.6%, respectively, misclassifying one baby with bacteremia and two with UTIs. It identified all infants who had bacterial meningitis.
CVD Risk Algorithms Perform Similarly After Recalibration
A head-to-head comparison of four commonly used cardiovascular disease (CVD) risk algorithms found that their clinical performance initially “varied substantially” (Eur Heart J 2019;40:621-31). However, after researchers recalibrated the algorithms to account for differences in the risk characteristics of the populations being studied, their clinical performance was “nearly equalized.”
These findings support the notion of using regularly recalibrated risk algorithms in clinical practice, according to the authors. The investigators also suggest that CVD primary prevention guidelines should “shift away from debates about the relative merits of particular risk algorithms and, instead, achieve consensus about the need for more widespread use of any recalibrated algorithm.”
Debate has been ongoing about how well four key CVD risk algorithms—pooled cohort equations (PCE), Systematic Coronary Risk Evaluation (SCORE), Framingham risk score (FRS), and Reynolds risk score (RRS)—capture and predict risk, and guide clinicians in determining the best treatment approaches for their patients.
PCE, recommended by the American College of Cardiology/American Heart Association, and SCORE, recommended by the European Society of Cardiology, as well as FRS and RRS, include common risk inputs but also differ not only in risk factors considered but also their mathematical formulations and the CVD outcomes they employ. A few, relatively small studies have sought to compare the equations. The investigators used individual participant data on 360,737 individuals without CVD in 86 prospective studies from 22 countries to calculate the models’ risk discrimination and calibration, and to project how well they targeted CVD preventive action to clinical need.
Before the recalibration, the four algorithms’ risk discrimination was similar. PCE, SCORE, and FRS overpredicted CVD risk before recalibration on average by 41%, 52%, and 10%, respectively, while RRS underpredicted this risk by 10%, and the algorithms classified as high risk 29%-39% of individuals age 40 or older. Recalibration reduced this proportion to 22%-24% for all algorithms.
Recalibration also lowered the estimated number of people who would need to be treated with statins to prevent one CVD event, from 44-51 to about 38.
HbA1c: Not First Test Choice but More Likely to Lead to Diabetes, Prediabetes Diagnosis
An analysis of screening practices for diabetes and prediabetes since 2010, when the American Diabetes Association (ADA) first recommended HbA1c testing as a screening option, found that while HbA1c testing is used less often for screening than glucose testing (14% versus 86%, respectively) it is more likely to result in a clinical diagnosis (Diabetes Care 2019; doi.org/10.2337/dc17-1726).
The study examined the claims and electronic health records of 12,772 individuals ≥45 years old who had been enrolled in the Blue Care Network of Michigan for 3 consecutive years, who did not have diabetes or take antidiabetic medications, and who received primary care at University of Michigan Health System. The authors conducted a similar study examining screening practices between 1998 and 2000.
The investigators found that 78% of individuals had been screened by any method, versus 69% in their earlier study. Almost all glucose tests were performed as part of chemistry panels, with just 18% performed as standalone tests. Abnormal (≥5.7%) HbA1c results were more likely than glucose tests ≥100 mg/dL but no more likely than glucose tests ≥126 mg/dL to be associated with follow-up visits within 6 months. However, an HbA1c result ≥5.7% was more likely than a glucose test result ≥126 mg/dL to lead to a diagnosis of diabetes or prediabetes within 6 months.
These findings suggest a need for better defined cutpoints to delineate abnormal random glucose tests, according to the authors. ADA’s recommendation that random glucose levels ≥200 mg/dL could be diagnostic of diabetes when accompanied by signs and symptoms of the disease, might have lulled “practitioners into believing that random glucose levels <200 mg/dL are normal.” The authors stated, however, that random glucose levels ≥126 mg/dL and possibly ≥100 mg/dL, even when performed as part of chemistry panels, “deserve follow-up with a definitive diagnostic test, either an HbA1c or fasting glucose tests.”