CLN - The Sample

Machine Learning-based Cyst Fluid Analysis Bests Standard-of-Care in Distinguishing Benign, High-risk Pancreatic Cysts

The Sample: September 2019

In a proof-of-concept study a test that uses machine learning and incorporates data from clinical, molecular, and imaging features of pancreatic cysts outperformed standard-of-care pathology analysis in identifying benign cysts that required no further follow-up, those that needed ongoing monitoring, and those that were high risk and needed surgical removal (Sci Transl Med 2019;11:eaav4772). The findings, if validated prospectively, could significantly reduce the number of unnecessary surgeries in patients with pancreatic cysts, according to the researchers.

As many as 8% of people older than age 70 develop pancreatic cysts, but only a fraction of cysts progress to cancer. Yet because current tests don’t distinguish precancerous cysts very well from low-risk ones, “essentially all people diagnosed with a cyst are followed long-term,” the researchers noted in a statement, and many may have surgery “that in hindsight may not have been necessary.”

The study involved 16 centers worldwide and 862 patients with pancreatic cysts who had cyst fluid analysis and cyst removal surgery. Cyst fluid analysis from a training set of 436 patients revealed four types of molecular abnormalities, including: mutations in 11 genes associated with specific cyst types; loss of heterozygosity in chromosome regions containing tumor suppressor genes; aneuploidy; and the protein markers carcinoembryonic antigen and vascular endothelial growth factor A. From this data the researchers worked out molecular features associated with benign cysts, mucin-producing cysts that had malignant potential, and malignant cysts.

The researchers then used a machine learning algorithm to combine these molecular features with clinical and imaging data to test “millions of combinations of features to predict the correct treatment pathway with the highest sensitivity and specificity,” according to the researchers.

Next, the investigators validated the test—dubbed CompCyst—in a set of data from 426 patients and compared its performance against standard pathology assessment in discerning the three categories of cysts. CompCyst outperformed standard pathology in identifying these cysts’ types but it did so most strikingly with benign cysts, correctly identifying 60.4% of cases that did not need follow-up versus 18.9%. CompCyst identified 48.6% of cases that should have been monitored and 90.8% that needed surgery versus 34.3% and 88.8% identified by pathology assessments, respectively.

Had CompCyst been used prospectively in this cohort of patients, surgery would have been avoided in 60% of 193 who had unnecessary procedures to remove their cysts.

Substantial, Persistent Quality Gaps Found in Chronic Kidney Disease Care

Patients with diagnosed chronic kidney disease (CKD) experienced “substantial and persistent gaps in quality of care” according to an analysis of a national cross-sectional study of outpatient visits (Clin J Am Soc Nephrol 2019; The findings from National Ambulatory Medical Care Surveys in 2006-2008 and 2012-2014 of 7,099 unweighted visits weighted to represent national estimates point to an “urgent need for CKD-specific quality measures and implementation of quality improvement interventions,” according to the authors.

The investigators found that nearly half of individuals with a diagnosis of CKD (46%) in the 2006-2008 survey had uncontrolled hypertension with blood pressure >130/80 mmHg. This percentage had barely moved in the 2012-2014 survey, when 48% had blood pressure >140/90 mmHg.

Figures for other quality measures the researchers considered also were low and showed little-to-no improvement between the two survey periods. Less than one-third of participants at least age 50 were taking statins (29% in 2006-2008; 31% in 2012-2014); and less than half were on angiotensin receptor blocker or angiotensin-converting enzyme inhibitor therapy (45% in 2006-2008; 36% in 2012-2014). HbA1c test results were only available in the 2012-2014 survey but 40% of individuals had values >7%.

In addition to the lack of dedicated CKD-specific quality metrics, the authors cited other potential causes of suboptimal care in individuals living with CKD, including low rates of referral to nephrologists and lack of information about the quality of care in nephrology clinics.

AACC has endorsed the National Kidney Foundation’s laboratory engagement plan, aimed at better diagnosing CKD through use of the Kidney Profile Test, among other initiatives. A recent Q&A addressed optimal use of biomarkers for CKD (Clin Chem 2019; doi:10.1373/clinchem.2018.29907).

Immune Marker Changes Signal Progression of MGUS to Multiple Myeloma

A prospective study investigating alterations in serum immune markers in patients with stable versus progressive monoclonal gammopathy of undetermined significance (MGUS) found that majorities of high-risk MGUS (53%) and low- or intermediate-risk MGUS (70%) converted to multiple myeloma (MM) within 5 years (JAMA Oncol 2019; doi:10.1001/jamaoncol.2019.1568). This finding suggests that the risk of MGUS progression is more complex and fluid than the 0.5% to 1.0% annual risk of progression generally ascribed to these populations and supports annual blood testing and risk assessment for patients with MGUS or light-chain MGUS, according to the authors.

This cohort study from the Prostate, Lung, Colorectal, and Cancer Screening Trial included 187 participants diagnosed with progressing MGUS and 498 with stable MGUS. The authors analyzed baseline and all serial samples obtained before participants’ disease progressed.

Risk factors most associated with progressive MGUS included: skewed (<0.1 or >10) serum free light chains (FLC) ratio, ≥15 g/L monoclonal spike, severe immunoparesis with ≥2 suppressed uninvolved immunoglobulins (Ig), and IgA isotype, with adjusted odds ratios (ORs) of 46.4, 23.5, 19.1, and 1.80, respectively. Risk factors associated with progressive light-chain MGUS included severe immunoparesis and skewed serum FLC ratio, with adjusted ORs of 48.6, and 44.0, respectively.