Laboratories can accurately detect IV fluid contamination in basic metabolic panel (BMP) results using machine learning, even without curating expertly labeled training data, according to a recent report (Clin Chem 2023; doi: 10.1093/clinchem/hvad207).
Detecting IV fluid contamination is universally difficult for clinical laboratories. Contamination commonly causes preanalytical errors that can delay or misguide treatment decisions and potentially lead to patient harm.
Detecting contamination now relies on delta checks that require prior results or manual technologist intervention vulnerable to human error. Supervised machine learning may help detect contamination but requires expert-labeled training data and classifiers that can detect only anomolies for which they are specifically trained. Proposed solutions involving supervised learning algorithms have historically required a substantial investment to curate expertly labeled training data.
In response, researchers trained and tested a Uniform Manifold Approximation and Projection (UMAP) model using a combination of real patient data and simulated IV fluid contamination on a total of 25,747,291 BMP results from 312,721 patients with data in a laboratory information system.
To provide an objective metric for classification, the researchers derived and assessed an “enrichment score,” comparing researchers’ usual workflow using expert chart review to UMAP predictions.
UMAP embeddings showed outliers suspicious for IV fluid contamination when compared with the simulated contamination’s embeddings. At a flag rate of 3 per 1,000 results, the positive predictive value was adjudicated to be 0.78 from 100 consecutive positive predictions. Of these, 58 were previously undetected by current clinical workflows, with 49 BMPs displaying 56 critical results.
The researchers said they focused on IV fluid contamination as a proof of principle and called for future studies of the UMAP model involving other body fluids, common laboratory errors like mislabeled specimens or hemolysis, and incorporation of other common laboratory panels, such as complete blood count and liver function panels.
OBESITY AND SMOKING LINKED TO BLOOD CANCER PRECURSOR DETECTION
Mass spectrometry (MS)-detected monoclonal gammopathies are associated with a broader range of modifiable multiple myeloma risk factors than what has been previously identified (Blood Adv 2024; doi: 10.1182/bloodadvances.2023010843).
Individuals with obesity are more likely to have monoclonal gammopathy of undetermined significance (MGUS), a benign blood condition that often precedes multiple myeloma, the paper noted.
MGUS has few known risk factors, and the emergence of MS for the detection of MGUS has provided new opportunities to determine risk factors.
Researchers conducted a screening study and an exposure survey of 2,628 individuals at elevated risk for multiple myeloma. The researchers screened their samples using MS and categorized samples with monoclonal proteins (M-proteins) at concentrations greater than 0.2 g/L as MS-MGUS. Using multivariable logistic models, the researchers evaluated associations between exposures and outcomes.
Elevated body mass index (BMI) and smoking were associated with all MS-positive cases.
After controlling for age, sex, race, education, and income, the team found that obesity was associated with 73% higher odds of having MGUS, compared with nonobese individuals.
This association remained unchanged when accounting for physical activity. However, highly active individuals were less likely to have MGUS even after adjusting for BMI class. Samples from individuals with high physical activity — defined as more than 73.5 metabolic equivalent (MET)-hours/week versus fewer than 10.5 MET-hours/week — had a decreased likelihood of MS-MGUS (OR=0.45, 95% CI= 0.24 to 0.80, P=.009).
Individuals who reported heavy smoking and short sleep (less than 6 hours per night) were more likely to have detectable levels of MGUS.
The researchers said that the presence of MGUS serves as a warning to monitor for the condition developing into more critical conditions.
POTENTIAL GESTATIONAL DIABETES MARKERS HIGHLIGHTED
A recent analysis concludes that pregnancy insulin resistance, or hypertriglyceridemia, may be useful in gestational diabetes mellitus (GDM) risk stratification (Nat Commun 2023; doi: 10.1038/s43856-023-00393-8).
Perinatal outcomes vary for women with GDM. Because the precise factors beyond glycemic status that may refine GDM diagnosis remain unclear, researchers conducted a systematic review and meta-analysis of potential precision markers for GDM.
The researchers searched for precision markers in several categories. These include maternal anthropometrics, clinical/sociocultural factors, nonglycemic biochemical markers, genetics/genomics or other “omics,” and fetal biometry. Afterwards, the researchers conducted posthoc meta-analyses of a subset of studies with data on the association of maternal body mass index with larger than average offspring or large-for-gestational age (LGA).
Unsurprisingly, the researchers found that high maternal weight is a risk factor for offspring born larger for their gestational age. Women with GDM and above-average weight/obesity versus women with GDM and healthy BMI are at higher risk of larger-than-average offspring (OR 2.65; 95% CI 1.91, 3.68), and LGA (OR 2.23; 95% CI 2.00, 2.49).
Researchers found other promising markers. Lipids and insulin resistance/secretion indices were the most studied nonglycemic biochemical markers, with increased triglycerides and insulin resistance generally associated with greater risk of offspring macrosomia or LGA. Most studies examining lipids in association with adverse perinatal outcomes have measured a standard lipid panel that includes total cholesterol, LDL and HDL cholesterol, and triglycerides. Half of the studies reported that higher triglycerides, independent of BMI, were associated with larger than average offspring or LGA, with fewer studies finding that higher LDL or lower HDL was associated with neonatal size.
Not all studies included in the review reported positive associations, and many factors, such as differences in timing of blood collection and variability in the distribution of characteristics across studies, could explain inconsistencies. Few studies examined the joint effects of multiple lipid subclasses.
The researchers noted gaps in the literature. These include inadequate data to determine whether a predominant defect in insulin secretion without excess insulin resistance is related to adverse perinatal outcomes, associations between adipokines and adverse perinatal outcomes among women with GDM, and genetic markers that predict adverse outcomes.