Precision partners: AI in diagnostic labs for clinical decision support

Clinical decision support (CDS) systems can streamline the diagnostic process by assessing the appropriateness of ordered tests, identifying critical values, and minimizing treatment delays through automated decision-making support for clinicians. CDS has had a longstanding presence in diagnostics, exemplified by features like reflex testing and autoverification, which have aimed to optimize laboratory workflows and improve diagnostic outcomes. However, diagnostics are on the precipice of a significant shift in the landscape of CDS, marked by the emergence of a new generation of systems leveraging artificial intelligence (AI) technologies. This new generation of systems could improve laboratory testing by introducing advanced capabilities for data analysis, pattern recognition, and patient-centric interpretations. By harnessing the power of AI, next-generation CDS systems promise to deliver more sophisticated insights, enable faster and more accurate diagnoses, and enhance the overall quality of patient care. Here, we explore the potential impact of AI-backed CDS on diagnostic testing and emphasize its capacity for patient-centric interpretations to provide insights into how these advancements will reshape the field of laboratory medicine and improve healthcare outcomes.

Applying AI decisioning in lab workflows

Recent studies have shown that up to 70% of laboratory-related errors occur in the preanalytical phase of laboratory testing.1 AI-based CDS systems have shown significant potential in the pre-analytical phase. For instance, a neural network algorithm effectively detected clots that could skew coagulation test results.2 Similarly, a recent study revealed that a machine learning (ML) algorithm can detect mislabeled samples with 92.1% accuracy, surpassing human performance by approximately 14%.3 Another application in the pre-analytical phase is assessing serum quality based on hemolysis, icterus, and lipemia (HIL). A recent study used convolutional neural networkbased deep learning models to analyze sample images to determine their HIL status and by extension their quality.4 Similar models are being integrated into vision systems on diagnostic instruments to detect HIL errors in real-time, preventing costly analytical cycles from being wasted on unsuitable samples.

One of the most critical roles of clinical laboratories is the test report verification process at the post-analytical phase. This process identifies potential errors before releasing test results. To streamline this, clinical laboratories historically employed techniques, such as autoverification, which use computer-based algorithms to evaluate and validate test results by following rule-based predefined criteria. Autoverification algorithms utilize data filters based on instrument error and quality control flags, serum indices, critical values, result range limits, delta checks, and consistency checks. These rules detect abnormal results and verify them for possible errors before release. They fall into two categories: comparing test results with predefined limits for the test type or comparing them with prior results for the same patient to ensure consistency.

Autoverification rules are very time consuming to implement and vary between laboratories. In addition, reports that are labeled ‘invalid’ by an autoverification system, will need manual verification. The promise of an AI-based autoverification system was demonstrated in a Chinese hospital where the AI autoverification system outperformed the traditional rule-based system by reducing the number of invalid reports by 80% thus reducing the need for manual intervention and workload.5

Similar to autoverification, reflex testing typically relies on simple “if-then” rules; for example, thyroid stimulating hormone (TSH) followed by reflex testing for free thyroxine (FT4) if TSH is above or below established criteria. However, this limits their scope since most test ordering decisions involve more complexity than a simple rule will allow. With the emergence of electronic medical records and AI, it appears plausible that future laboratory services could include an advanced reflex testing system with a wider scope and greater impact that can take multiple data points into consideration. In a recent study, the concept of smart reflex testing using an ML model was introduced where upon getting a complete blood count (CBC) result, the model predicted the need for ferritin testing to confirm anemia.6 The model performed moderately well in predicting ferritin testing need and demonstrated greater suitability to reflex testing than rule-based approaches.

Applying AI decisioning to lab diagnostics data

The integration of CDS systems that use AI to address various questions in healthcare using lab diagnostics data in the post-analytical phase has shown promising applications. CDS systems can help in the interpretation of multidimensional test results by providing likelihood scores for potential diseases, improving risk stratification, predicting disease outcomes, and providing differential diagnosis. These technologies can play a critical role in monitoring organ functions by tracking blood results and trends and facilitating early and accurate disease diagnosis. Furthermore, by predicting disease progression and patient outcomes, AI/ML can empower healthcare providers to make well-informed decisions regarding patient care and intervention. The use-cases cover a diverse range of medical conditions, including but not limited to infectious diseases, such as sepsis and COVID,7,8 cardiovascular diseases,9 cancer diagnostics,10 liver diseases,11 kidney diseases,12 and autoimmune diseases.13

Among all the use-cases of integration of ML into laboratory medicine for developing a CDS solution, early sepsis identification and outcome prediction is undergoing rapid advancement.14 These solutions, however, have not yet found wide clinical adoption as various challenges like population variability in biomarkers, limited access to datasets due to privacy issues, lack of external validations, and lack of large datasets with sufficient parameters to robustly train decisioning algorithms persist and prevent well-functioning algorithms from being created. Despite such challenges, some systems have overcome these hurdles. Recently, Prenosis has received a de novo FDA clearance for its AI/ML-based sepsis prediction algorithm ImmunoScore,TM which can help in assessing the risk of having sepsis or progressing to sepsis within 24 hours of patient assessment. Similarly, Smart Blood Analytics has released the CE-marked SBAS Software, which employs advanced ML algorithms to determine the most likely diagnoses based solely on an individual's blood test results.

The adoption of AI to lab data is in its early stages, but it is likely to grow considering the trajectory of similar diagnostic advancements in imaging applications. An example of the use of AI in diagnostic imaging can be found in the hematology space from Scopio Labs (Tel Aviv-Yafo, Israel). The interpretation of white blood cell (WBC) differentials, red blood cell (RBC) morphology, and platelet estimation is done using AI from a full-field peripheral blood smear image made on FDA-approved systems, thus reducing turnaround time for sample review by 60%.15

The FDA-approved list16 of AI/ML-enabled medical devices is dominated by radiology solutions, with 87% of such devices approved in 2022 belonging to this category. AI has impacted all aspects of radiology from staff scheduling and picking the correct scan settings to the imaging process itself, i.e., interpretation and identification of features in an image.17 Top diagnostics imaging manufacturers have incorporated AI in their imaging software to improve accuracy and efficiency of imaging, such as enabling shorter imaging time in MRI and reducing radiation doses in PET scanning. On the CDS front, there are multiple companies who serve different use cases like lung cancer (Aidence), breast cancer (DeepHealth), brain cancer (Cortechs.ai), and hemorrhage and aneurysm detection (Viz Neuro™). AI supported solutions can also enhance the quality of care in remote areas. An example is a solution from Qure.ai, which in conjunction with a portable X-ray, can be used to screen for tuberculosis from chest X-ray scans.18

Conclusion

While modern laboratories have significantly improved their efficiency and accuracy in processing samples and delivering results by utilizing the latest advancements in the pre-analytical and analytical space, the selection of tests and interpretation of results by clinicians still largely relies on manual cognitive processes, leaving room for human error and variability. It is highly likely that in the future CDS systems will also provide tailored recommendations of an optimal set of tests to aid clinicians in confirming the most probable diagnosis, ruling out improbable ones, and assessing the risks associated with lethal diseases. In addition, lab test interpretation, personalized analytics reporting, and tailored diagnostics based on age, gender, and co-morbidities will become more prevalent in the clinical field.19 This ensures that the correct diagnosis and therapy are provided to the patient rather than a one-for-all system, which may not fit the patient’s history, symptoms, and diagnosis.

It is also worth mentioning that a vision for the future involves integrative diagnostics, where data from clinical laboratories, radiology, pathology, and electronic health records will be aggregated and contextualized by CDS systems for clinical actions.20 Although AI-driven CDS systems are still in their early stages, the potential exists to significantly transform diagnostics and enhance patient care. Given the rapid advancements in the field, the beneficial impact of AI on laboratory medicine is not merely a distant aspiration but rather a present and tangible reality.

References

1.            Najat D. Prevalence of pre-analytical errors in clinical chemistry diagnostic labs in Sulaimani City of Iraqi Kurdistan. PLoS One. 2017;12(1):e0170211. doi:10.1371/journal.pone.0170211.

2.            Fang K, Dong Z, Chen X, et al. Using machine learning to identify clotted specimens in coagulation testing. Clin Chem Lab Med. 2021;59(7):1289-1297. doi:10.1515/cclm-2021-0081.

3.            Farrell CJ. Identifying mislabelled samples: Machine learning models exceed human performance. Ann Clin Biochem. 2021;58(6):650-652. doi:10.1177/00045632211032991.

4.            Yang C, Li D, Sun D, et al. A deep learning-based system for assessment of serum quality using sample images. Clin Chim Acta. 2022;531:254-260. doi:10.1016/j.cca.2022.04.010.

5.            Wang H, Wang H, Zhang J, et al. Using machine learning to develop an autoverification system in a clinical biochemistry laboratory. Clin Chem Lab Med. 2020;59(5):883-891. doi:10.1515/cclm-2020-0716.

6.            McDermott M, Dighe A, Szolovits P, Luo Y, Baron J. Using machine learning to develop smart reflex testing protocols. J Am Med Inform Assoc. 2024;31(2):416-425. doi:10.1093/jamia/ocad187.

7.            Carobene A, Milella F, Famiglini L, Cabitza F. How is test laboratory data used and characterised by machine learning models? A systematic review of diagnostic and prognostic models developed for COVID-19 patients using only laboratory data. Clin Chem Lab Med. 2022;60(12):1887-1901. doi:10.1515/cclm-2022-0182.

8.            Islam KR, Prithula J, Kumar J, et al. Machine learning-based early prediction of sepsis using electronic health records: A systematic review. J Clin Med. 2023;12(17):5658. doi:10.3390/jcm12175658.

9.            Mohd Faizal AS, Thevarajah TM, Khor SM, Chang SW. A review of risk prediction models in cardiovascular disease: conventional approach vs. artificial intelligent approach. Comput Methods Programs Biomed. 2021;207:106190. doi:10.1016/j.cmpb.2021.106190.

10.         Swanson K, Wu E, Zhang A, Alizadeh AA, Zou J. From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment. Cell. 2023;186(8):1772-1791. doi:10.1016/j.cell.2023.01.035.

11.         Altaf I, Butt MA, Zaman M, et al. Disease Detection and Prediction Using the Liver Function Test Data: A Review of Machine Learning Algorithms. In: International Conference on Innovative Computing and Communications Advances in Intelligent Systems and Computing. Springer; 2022.

12.         Sanmarchi F, Fanconi C, Golinelli D, et al. Predict, diagnose, and treat chronic kidney disease with machine learning: a systematic literature review. J Nephrol. 2023;36(4):1101-1117. doi:10.1007/s40620-023-01573-4.

13.         Stafford IS, Kellermann M, Mossotto E, et al. A systematic review of the applications of artificial intelligence and machine learning in autoimmune diseases. NPJ Digit Med. 2020;3:30. doi:10.1038/s41746-020-0229-3.

14.         G A, K L N, M S AS. Improving sepsis classification performance with artificial intelligence algorithms: A comprehensive overview of healthcare applications. J Crit Care. 2024;83:154815. doi:10.1016/j.jcrc.2024.154815.

15.         Katz BZ, Feldman MD, Tessema M, et al. Evaluation of Scopio Labs X100 Full Field PBS: The first high-resolution full field viewing of peripheral blood specimens combined with artificial intelligence-based morphological analysis. Int J Lab Hematol. 2021;43(6):1408-1416. doi:10.1111/ijlh.13681.

16.         Center for Devices, Radiological Health. Artificial intelligence and machine learning (AI/ML)-enabled medical devices. U.S. Food and Drug Administration. August 7, 2024. Accessed November 13, 2024. https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices.

17.         Najjar R. Redefining radiology: A review of artificial intelligence integration in medical imaging. Diagnostics (Basel). 2023;13(17):2760. doi:10.3390/diagnostics13172760.

18.         John S, Usman S, Rahman T  Creswell M. Comparing tuberculosis symptom screening to chest X-ray with artificial intelligence in an active case finding campaign in Northeast Nigeria. BMC Global and Public Health. 2023;(17).

19.         Velev J, LeBien J, Roche-Lima A. Unsupervised machine learning method for indirect estimation of reference intervals for chronic kidney disease in the Puerto Rican population. Sci Rep. 2023;13(1):17198. doi:10.1038/s41598-023-43830-3.

20.         Beauchamp NJ, Bryan RN, Bui MM, et al. Integrative diagnostics: The time is now-A report from the International Society for Strategic Studies in Radiology. J Am Coll Radiol. 2023;20(4):455-466. doi:10.1016/j.jacr.2022.11.015.