RISK MANAGEMENT
clinical action, such as large variations in haemoglobin, platelet count, or coagulation parameters. In transfusion medicine, for example, unexplained differences in blood count parameters between current and previous samples may trigger cross-departmental checks to confirm patient identity and sample integrity.
Implementation approaches range
from fixed percentage or absolute difference thresholds to more advanced reference change values (RCVs) that incorporate both analytical and biological variation. Using RCVs adds scientific rigour by ensuring that only changes unlikely to be due to natural variability are flagged for review. The literature, including work by Frenkel, Farrance, and Badrick, supports the growing role of patient-based QC in routine monitoring, while authors such as Dimech and Vincini have shown that in low-volume or one-at-a-time assays – where traditional lot verification is impractical – delta checks and RCVs can serve as alternative monitoring strategies. However, conventional delta check systems have recognised limitations. A national survey in China found that potassium, glucose, creatinine, haemoglobin, platelet count, and white cell count were the most common analytes used for delta checks, but alerts were often triggered by legitimate clinical changes such as treatment effects or physiological variation, leading to high false-positive rates and increased manual workload.6
This ‘alert fatigue’
can undermine the ability to identify true errors in a timely manner. Emerging machine learning (ML)
approaches have demonstrated clear performance improvements over traditional methods. In haematology testing, a deep learning model based on a Deep Belief Network achieved an area under the ROC curve of 0.977, outperforming three widely used statistical methods in detecting artificially introduced sample mix-ups by analysing multiple parameters together rather than single analytes in isolation.7
In tumour marker testing, a deep neural network
Aspect Purpose Uses
Methods Limitations Summary
Compare current vs prior results to detect implausible changes; part of patient-based QC. Drift/shift detection, mislabelling/WBIT; valuable for low-volume assays. Fixed/absolute change, % change, RCVs, ML-based models. False positives from legitimate changes; alert fatigue.
Best practice Tailor thresholds to test/setting; integrate with IQC, autoverification, FMEA. Table 2. Summary of delta check implementation in the clinical laboratory.
WWW.PATHOLOGYINPRACTICE.COM SEPTEMBER 2025
model showed superior and more consistent detection of misidentification errors across alpha-fetoprotein, CA 19-9, CA 125, carcinoembryonic antigen, and prostate-specific antigen, compared with both random forest algorithms and traditional delta percent change or RCV methods.8
The clinical context is critical. For tumour markers, delta check thresholds tailored to specific settings – such as inpatient, outpatient, or health check- up populations –significantly reduced unnecessary investigations while maintaining very high negative predictive values (>0.99).9
This context-specific
approach aligns with the emphasis of ISO 22367 on proportionality of controls to the risk of harm, ensuring that meaningful changes are investigated without overburdening laboratory workflows (Table 2).
Delta checks are also integral to validated autoverification systems, acting alongside other safeguards such as IQC review, instrument flags, and critical value checks. When embedded in a structured autoverification algorithm and evaluated using Failure Modes and Effects Analysis (FMEA), delta check processes have been shown to contribute to negligible residual risk, with risk priority numbers approaching zero.10 To be effective, delta check systems must be designed with consideration of the clinical purpose of the test, the expected biological stability of the measurand, and the patient population. Thresholds must be set to balance sensitivity and specificity: overly narrow limits generate excessive false positives, while overly wide limits risk delayed detection of clinically important changes. Investigations of alerts should follow a
defined and documented process in line with ISO 15189 requirements for managing quality incidents. Incorporated into an integrated monitoring programme alongside IQC, lot verification, and EQA, delta checks provide a patient-centred safety net for detecting potential error. By evolving from fixed, one-size-fits-all thresholds to adaptive, data-driven models – particularly those leveraging ML – laboratories can enhance their ability to detect clinically significant errors promptly and reliably, minimising the risk of patient harm while maintaining operational efficiency.
Reference range monitoring Monitoring shifts in the distribution of patient results relative to the reference range is a valuable external analytical signal. ISO 15189:2022 requires laboratories to review the validity of reference intervals and decision limits, while ISO 22367:2020 emphasises the importance of detecting changes that could affect the clinical interpretation of results. Tracking patient result trends over time can highlight analytical drift, calibration bias, or pre-analytical changes that may not be detected by IQC alone. In high-volume testing, running patient means or medians is a common method of tracking stability. This approach works best when the sample population is consistent and representative of the test’s clinical use. However, if the population mix changes – such as an influx of results from a critical care unit or a screening programme – these shifts can influence the mean or median independently of analytical performance. For this reason, patient selection criteria must be carefully defined, and trends must be interpreted with knowledge of any concurrent changes in the patient population. Drift toward the upper or lower limits
of the reference range can have important clinical consequences. A small shift in bias may push a proportion of results over a decision limit, leading to changes in diagnosis, treatment, or monitoring frequency. Conversely, a shift away from the limit may mask abnormal results, delaying intervention. When such shifts are detected, a structured review should be carried out, including checks for recent
31
The consequences of inadequate lot verification can be significant. Operationally, an unrecognised bias may lead to an increase in reflex or repeat testing
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84
