To the Editor,
In previous papers, we discussed the rationale of a clinical decision making-based laboratory medicine department, where first-line tests are rapidly performed in a core laboratory and results immediately reported to clinical users, and tests requiring a specialized knowledge are carried out in dedicated sections with the objective of enhancing patient care , . These satellite laboratory sections (e.g., endocrinology, oncology, etc.) performing tests requiring a specialized knowledge may more fruitfully cooperate with care teams for specific medical conditions assuring that their results may effectively work in the correct clinical setting. One of these satellite sections is dealing with the diagnosis of protein disorders, conducting specific investigations for the evaluation and management of important medical conditions, including monoclonal gammopathies and related diseases, by performing protein electrophoresis, together with immunotyping and immunofixation electrophoresis for characterizing monoclonal components in both serum and urine. The visual inspection of serum protein electrophoresis (SPE) patterns and the need for an interpretative rather than numerical report represent limiting factors slowing down the turnaround time (TAT) of the examination process of this specialized section. To improve the process efficiency, expert systems, based on neural networks and decision algorithms, able to automatically identify and quickly report SPE patterns without abnormalities, have been proposed , , .
Our protein diagnostics section serves a network of four hospitals with a daily workload of approximately 200 SPE (approximately 70% from inpatients and the remaining from outpatients). To improve the efficiency of SPE reviewing process, in 2017 we introduced in our workflow the Neurosoft expert system integrated into the software managing SPE patterns (Phoresis Sebia), which automatically review, both qualitatively and quantitatively, SPE profiles analyzed in Protein(e) 6 program of Capillarys 2 instruments (Sebia). Neurosoft is an artificial intelligence-based system composed of six neural networks (one for each protein fraction: albumin, alpha1, alpha2, beta1, beta2, and gamma) designed for both morphological and quantitative interpretation analysis of SPE patterns. Previous experience showed that Neurosoft in its current version can discriminate SPE patterns without abnormalities with a negative predictive value of 100% , . Here we evaluated the contribution that Neurosoft provided in the first two years after its implementation in our laboratory specialized section with special reference to the impact on SPE TAT, defined as the time from sample check-in in the laboratory department (T 0) to the time when results became available to clinical requestors (T r ). We retrieved from the Phoresis software all SPE patterns analysed in the 2016 and 2018–2019 periods (i.e., 2016 – before Neurosoft implementation – and 2018–2019 – after Neurosoft implementation). The 2018–2019 SPE patterns were further subdivided in two groups: group N, including SPE without abnormalities when evaluated by Neurosoft, and group A, including SPE with abnormalities detected by Neurosoft that needed visual inspection by skilled operators and, when judged necessary, further in-depth investigations. The laboratory information system (DNlab, Dedalus) allowed tracing the time of all testing phases of examined samples [from pre-analytical phase (sample check-in and centrifugation), through the analytical phase (total protein measurements and SPE analysis), to the post-analytical phase, consisting of the following steps: Neurosoft validation, visual inspection of abnormal patterns, further investigations, if needed, and validation of those patterns by laboratory professionals, result transfer from Phoresys to the middleware Halia (Dedalus), and final availability of results to requestors] (Figure 1). Statistical analyses were performed using MedCalc v13.0 statistical package. The Mann–Whitney test was used to test differences in TAT between different groups. p values <0.05 were regarded as significant.
Table 1 reports the number of SPE patterns examined in the evaluated periods and corresponding TAT. In the two years after implementation, Neurosoft detected an average of 43.5% of N patterns in performed SPE. This allowed a significant reduction of overall TAT when compared to the 2016 one (from a median of 211 min to a median of 179 min, p<0.0001). As expected, group N has shorter TAT when compared to the group A, the mean difference being 33 min in 2018 and 25 min in 2019, respectively.
|Year||No. of SPE||SPE patterns at Neurosoft||TAT, min|
|Normal||Abnormal||Total||Neurosoft normal||Neurosoft abnormal|
|2018||27,553||12,434 (45.1%)||15,119 (54.9%)||180 (1,386)a||163 (361)b||196 (1,845)|
|2019||32,356||13,545 (41.9%)||18,811 (58.1%)||178 (740)a||162 (288)b||187 (1,452)|
ap<0.0001 vs. 2016. bp<0.0001 vs. Neurosoft abnormal.
As clinicians judge the adequacy of laboratory performance even considering the speed with which results are reported to them, strategies to improve TAT should be undertaken, acting on all phases of the total examination process. Although TAT appears to be less critical for SPE analysis, an efficient management of a large volume of tests may significantly impact on laboratory and staff management; it is therefore a key aspect of the organization of a decision making-based laboratory department . This is the first real-life study evaluating the efficiency of the Neurosoft software to reduce SPE workload through the automatic identification of N patterns by the TAT analysis. In our two-year experience, Neurosoft has identified 43.5% of SPE patterns as N, resulting in an overall average time saving of 32 min in SPE result reporting. Our TAT results were overall good, with validated SPE patterns available to clinicians in average 179 min after sample check-in. This time further decreased to 162.5 min when Neurosoft was able to detect N SPE patterns. It is also noteworthy that all 90th percentile data of TAT were markedly reduced after Neurosoft introduction, reaching in 2019, 288 min, i.e., less than 5 h, for N patterns and no more than 24 h for SPE patterns needing further investigations.
Although further margins of TAT improvement would be possible, e.g. the implementation of a real-time data exchange system from the middleware to the Phoresis software in order to transfer patient’s information data, SPE patterns and numeric results regardless of the physical presence of the operator, our data show that use of Neurosoft represents a valid improvement in the organization of the protein diagnostics section workflow. Its implementation also assures a uniformity of SPE pattern interpretation, which is more difficult to reach when different operators are involved in the first phase of SPE pattern screening .
We acknowledge the enthusiastic contribution of technicians of the ‘Protein diagnostics section’ of our Laboratory Medicine Department to the continuous improvement of the quality of provided service.
Research funding: None declared.
Author contributions: All authors have accepted responsibility for the entire content of this manuscript and approved its submission.
Competing interests: Authors state no conflict of interest.
Ethical approval: Not applicable.
2. Dolci, A, Giavarina, D, Pasqualetti, S, Szőke, D, Panteghini, M. Total laboratory automation: do stat tests still matter? Clin Biochem 2017;50:605–11. https://doi.org/10.1016/j.clinbiochem.2017.04.002.Search in Google Scholar PubMed
3. Jonsson, M, Carlson, J. Computer-supported interpretation of protein profiles after capillary electrophoresis. Clin Chem 2002;48:1084–93. https://doi.org/10.1093/clinchem/48.7.1084.Search in Google Scholar
4. Ognibene, A, Motta, R, Caldini, A, Terreni, A, Dea, ED, Fabris, M, et al.. Artificial neural network-based algorithm for the evaluation of serum protein capillary electrophoresis. Clin Chem Lab Med 2004;42:1451–2. https://doi.org/10.1515/cclm.2004.271.Search in Google Scholar
5. Altinier, S, Sarti, L, Varagnolo, M, Zaninotto, M, Maggini, M, Plebani, M. An expert system for the classification of serum protein electrophoresis patterns. Clin Chem Lab Med 2008;46:1458–63. https://doi.org/10.1515/cclm.2008.284.Search in Google Scholar PubMed
6. Dorizzi, RM, Zanardi, V, Agnoletti, R, Alberelli, A, Babini, A, De Vita, P. Assessment of an expert system for the automated validation of electrophoretic profiles. Clin Lab 2015;61:191–4. https://doi.org/10.7754/clin.lab.2014.140715.Search in Google Scholar PubMed
7. Frattolillo, D, Chiodino, C, Sartori, U, Peruzzi, S, Incerti, SS, Colla, R. Evaluation of an expert system for computer-supported evaluation of serum protein electrophoresis patterns. Biochim Clin 2016;40:316–21.Search in Google Scholar
8. Lou, AH, Elnenaei, MO, Sadek, I, Thompson, S, Crocker, BD, Nassar, B. Evaluation of the impact of a total automation system in a large core laboratory on turnaround time. Clin Biochem 2016;49:1254–8. https://doi.org/10.1016/j.clinbiochem.2016.08.018.Search in Google Scholar PubMed
9. Luraschi, P, Infusino, I, Merlotti, C, Franzini, C. Analytical variation in the measurement of serum monoclonal component by capillary electrophoresis. Clin Chim Acta 2004;349:151–6. https://doi.org/10.1016/j.cccn.2004.06.016.Search in Google Scholar PubMed
© 2021 Francesca Borrillo et al., published by De Gruyter, Berlin/Boston
This work is licensed under the Creative Commons Attribution 4.0 International License.