Maria Lorubbio
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Maria Lorubbio has a total of 116 co-authors. Here are the top co-authors by number of shared publications:
- Agostino Ognibene
Department of Laboratory Medicine , Azienda South-East Tuscany, San Donato Hospital , Arezzo , Italy
9 shared publications - Carolina Pieroni
DIESSE Diagnostica Senese SpA, 53035 Monteriggioni, Italy
3 shared publications - Daniela Diamanti
DIESSE Diagnostica Senese SpA, 53035 Monteriggioni, Italy
3 shared publications Agostino Ognibene
Chemical-Clinical Analysis Laboratory, San Donato Hospital, Via Pietro Nenni, 20/22, 52100, Arezzo, Italy
2 shared publicationsAlessandra Fanelli
General Laboratory , Careggi University Hospital , Florence , Italy
2 shared publications- Andrea Campagner
University of Milano-Bicocca
2 shared publications
Latest Publications
- Open Access
- 9 October 2026
- Published by Walter de Gruyter GmbH in Advances in Laboratory Medicine / Avances en Medicina de Laboratorio
Automatic microscopic count and characterization of cerebrospinal fluid cells: an alternative to standardized analysis
Objectives: Cerebrospinal fluid (CSF) analysis is important for the diagnosis and follow-up of central nervous system diseases. In this study, we propose a cellular analysis of CSF by digital optical microscopy (OM) using a semi-automatic instrument (SediMAX Lite, Menarini Diagnostic) that evaluates cells and produces digital images. Methods: A total of 110 CSF samples were analyzed (70 from the General Laboratory of the Careggi University Hospital in Florence and 40 from the Clinical-Chemical Analysis Laboratory of the San Donato Hospital in Arezzo). Manual OM cell counts were performed in a manual OM counting chamber. For semi-automatic cell counting, the SediMAX Lite was used, which captured 15 digital images for each sample (treated and untreated with Turk staining), stored in the instrument database. Method comparison and instrument evaluation were performed. Results: The agreement between the OM manual and SediMAX Lite for RBC showed a kappa=0.80, while for WBC count was kappa=0.86. The Passing-Bablok regression slope, in the comparison of WBC counts between the two methods in samples non-pretreated with Türk’s fluid, was 0.94 (95 % CI 0.86 to 1.00) and intercept was 1.00 (95 % CI, 0.00–1.00). The Bland-Altman bias was 4.39 cells/µL (95 % CI, −4.27 to 13.05), lower limit bias was −82.89 cells/µL (95 % CI, −97.89 to −67.89) and upper limit was 91.68 cells/µL (95 % CI, 76.68 to 106.68). Comparison of WBC counts between the two methods in samples pretreated with Türk’s fluid showed at the Passing-Bablok regression slope was 1.29 (95 % CI 1.16 to 1.44) and intercept was 1.00 (95 % CI, 0.56–1.00). The Bland-Altman bias was −22.12 cells/µL (95 % CI, −36.92 to −7.32), lower limit bias was −171.24 cells/µL (95 % CI, −196.87 to −145.61) and upper limit was 127.00 cells/µL (95 % CI, 101.38 to 152.63). Conclusions: SediMAX Lite shows quantitative agreement is limited, particularly in low cellularity samples. The system may serve as a supportive tool for standardized digital imaging and qualitative identification of pleocytosis in laboratories requiring continuous availability.
- 9 October 2026
- Published by Walter de Gruyter GmbH in Advances in Laboratory Medicine / Avances en Medicina de Laboratorio
Open AccessAutomatic microscopic count and characterization of cerebrospinal fluid cells: an alternative to standardized analysis
Objectives: Cerebrospinal fluid (CSF) analysis is important for the diagnosis and follow-up of central nervous system diseases. In this study, we propose a cellular analysis of CSF by digital optical microscopy (OM) using a semi-automatic instrument (SediMAX Lite, Menarini Diagnostic) that evaluates cells and produces digital images. Methods: A total of 110 CSF samples were analyzed (70 from the General Laboratory of the Careggi University Hospital in Florence and 40 from the Clinical-Chemical Analysis Laboratory of the San Donato Hospital in Arezzo). Manual OM cell counts were performed in a manual OM counting chamber. For semi-automatic cell counting, the SediMAX Lite was used, which captured 15 digital images for each sample (treated and untreated with Turk staining), stored in the instrument database. Method comparison and instrument evaluation were performed. Results: The agreement between the OM manual and SediMAX Lite for RBC showed a kappa=0.80, while for WBC count was kappa=0.86. The Passing-Bablok regression slope, in the comparison of WBC counts between the two methods in samples non-pretreated with Türk’s fluid, was 0.94 (95 % CI 0.86 to 1.00) and intercept was 1.00 (95 % CI, 0.00–1.00). The Bland-Altman bias was 4.39 cells/µL (95 % CI, −4.27 to 13.05), lower limit bias was −82.89 cells/µL (95 % CI, −97.89 to −67.89) and upper limit was 91.68 cells/µL (95 % CI, 76.68 to 106.68). Comparison of WBC counts between the two methods in samples pretreated with Türk’s fluid showed at the Passing-Bablok regression slope was 1.29 (95 % CI 1.16 to 1.44) and intercept was 1.00 (95 % CI, 0.56–1.00). The Bland-Altman bias was −22.12 cells/µL (95 % CI, −36.92 to −7.32), lower limit bias was −171.24 cells/µL (95 % CI, −196.87 to −145.61) and upper limit was 127.00 cells/µL (95 % CI, 101.38 to 152.63). Conclusions: SediMAX Lite shows quantitative agreement is limited, particularly in low cellularity samples. The system may serve as a supportive tool for standardized digital imaging and qualitative identification of pleocytosis in laboratories requiring continuous availability.
- Open Access
- 21 August 2025
- Published by MDPI AG in Preprints.org
Changing in Blood Cells for Infection Detection: The Significance of “Monocyte Distribution Width” in Sepsis Prevention
Background/Objectives: Blood is often the first organ to show changes during local or systemic infection, but common definitions of sepsis rarely include CBC findings in risk assessment tools. Infection remains the main cause of sepsis, which Sepsis-3 defines as life-threatening organ dysfunction from a dysregulated immune response. Changes in circulating monocytes—such as altered MDW and increased inflammation—can be seen in sepsis and SARS-COV-2. This study evaluates MDW as a blood marker for detecting infections and compares it with traditional markers for risk stratification. Methods: From December 2021 to December 2023, 608 adults with suspected infection or sepsis were enrolled at San Donato Hospital's Emergency Department in Arezzo. All patients were admitted and monitored in assigned wards, with destination, history, clinical, and lab data collected during their stay and follow-up. Results: Monocyte Distribution Width (MDW) detects bloodstream infection with high efficacy (sensitivity 92%, specificity 85%, and negative predictive value 95% at ER admission). MDW also stratifies risk for infections that may progress to sepsis—including urinary, respiratory, post-surgical, and wound infections—with sensitivity, specificity, and negative predictive value around 70–72%. Conclusions: Blood is usually the first organ to show changes during infection, with blood cell morphology and CBC parameters affected by bacterial, viral, or fungal pathogens. Techniques like flow cytometry and impedance measurements provide Cell Population Data (CPD), such as MDW, which support diagnostic algorithms for infection or sepsis. CPDs are well-suited for AI analysis, and the extensive information from CBCs should be integrated into both diagnosis and treatment.
- 21 August 2025
- Published by MDPI AG in Preprints.org
Open AccessChanging in Blood Cells for Infection Detection: The Significance of “Monocyte Distribution Width” in Sepsis Prevention
Background/Objectives: Blood is often the first organ to show changes during local or systemic infection, but common definitions of sepsis rarely include CBC findings in risk assessment tools. Infection remains the main cause of sepsis, which Sepsis-3 defines as life-threatening organ dysfunction from a dysregulated immune response. Changes in circulating monocytes—such as altered MDW and increased inflammation—can be seen in sepsis and SARS-COV-2. This study evaluates MDW as a blood marker for detecting infections and compares it with traditional markers for risk stratification. Methods: From December 2021 to December 2023, 608 adults with suspected infection or sepsis were enrolled at San Donato Hospital's Emergency Department in Arezzo. All patients were admitted and monitored in assigned wards, with destination, history, clinical, and lab data collected during their stay and follow-up. Results: Monocyte Distribution Width (MDW) detects bloodstream infection with high efficacy (sensitivity 92%, specificity 85%, and negative predictive value 95% at ER admission). MDW also stratifies risk for infections that may progress to sepsis—including urinary, respiratory, post-surgical, and wound infections—with sensitivity, specificity, and negative predictive value around 70–72%. Conclusions: Blood is usually the first organ to show changes during infection, with blood cell morphology and CBC parameters affected by bacterial, viral, or fungal pathogens. Techniques like flow cytometry and impedance measurements provide Cell Population Data (CPD), such as MDW, which support diagnostic algorithms for infection or sepsis. CPDs are well-suited for AI analysis, and the extensive information from CBCs should be integrated into both diagnosis and treatment.
- Open Access
- 26 April 2025
- Published by MDPI AG in Diagnostics
- Vol. 15 (9) , 1101
Erythrocyte Sedimentation Rate Reference Intervals Determined via VES-MATIC 5 and CUBE 30 Touch with Respect to the Westergren Method
Objectives: The erythrocyte sedimentation rate (ESR) is a diagnostic test that is employed worldwide to assess a patient’s inflammatory status. Like all laboratory tests, it requires reference intervals (RIs) to support clinical decision making and facilitate accurate diagnosis. In this study, we aimed to generate RIs for the automatic analyzers VES-MATIC 5 (VM5) and CUBE 30 touch (C30T) compared to the gold standard method. Methods: A total of 989 (presumably healthy) participants from Arezzo Hospital in Italy were enrolled. The ESR RIs were established according to CLSI for all three methods. Results: The analysis pointed out significant differences between women and men and age-related increases in ESRs obtained via all three analytical methods. The average and median values resulting from VM5 and C30T were, respectively, 1 mm/h smaller and higher than the gold standard. The RIs were calculated based on three clusters: the first pertained to patients aged ≥ 18 but ≤ 49 years; the second pertained to patients aged ≥ 50 but ≤ 69 years; the last comprised patients aged ≥ 70 years. Due to the clear overlap between these ranges and the statistical analysis, we identified only one range for females ≥ 18 years (Westergren: 1–22 mm/h; VM5: 1–22 mm/h; C30T: 1–25 mm/h). For the male participants, two separate RIs were proposed: one for those aged ≥ 18 but < 69 years (Westergren: 1–14 mm/h; VM5: 1–14 mm/h; C30T: 1–18 mm/h) and one for those aged 70 years or above (Westergren: 1–22 mm/h; VM5: 1–23 mm/h; C30T: 1–29 mm/h). Conclusions: The proposed RI for automated analyzers C30T and VM5 agreed with the reference method and can be adopted to measure ESRs within EDTA blood samples.
- 26 April 2025
- Published by MDPI AG in Diagnostics
- Vol. 15 (9) , 1101
Open AccessErythrocyte Sedimentation Rate Reference Intervals Determined via VES-MATIC 5 and CUBE 30 Touch with Respect to the Westergren Method
Objectives: The erythrocyte sedimentation rate (ESR) is a diagnostic test that is employed worldwide to assess a patient’s inflammatory status. Like all laboratory tests, it requires reference intervals (RIs) to support clinical decision making and facilitate accurate diagnosis. In this study, we aimed to generate RIs for the automatic analyzers VES-MATIC 5 (VM5) and CUBE 30 touch (C30T) compared to the gold standard method. Methods: A total of 989 (presumably healthy) participants from Arezzo Hospital in Italy were enrolled. The ESR RIs were established according to CLSI for all three methods. Results: The analysis pointed out significant differences between women and men and age-related increases in ESRs obtained via all three analytical methods. The average and median values resulting from VM5 and C30T were, respectively, 1 mm/h smaller and higher than the gold standard. The RIs were calculated based on three clusters: the first pertained to patients aged ≥ 18 but ≤ 49 years; the second pertained to patients aged ≥ 50 but ≤ 69 years; the last comprised patients aged ≥ 70 years. Due to the clear overlap between these ranges and the statistical analysis, we identified only one range for females ≥ 18 years (Westergren: 1–22 mm/h; VM5: 1–22 mm/h; C30T: 1–25 mm/h). For the male participants, two separate RIs were proposed: one for those aged ≥ 18 but < 69 years (Westergren: 1–14 mm/h; VM5: 1–14 mm/h; C30T: 1–18 mm/h) and one for those aged 70 years or above (Westergren: 1–22 mm/h; VM5: 1–23 mm/h; C30T: 1–29 mm/h). Conclusions: The proposed RI for automated analyzers C30T and VM5 agreed with the reference method and can be adopted to measure ESRs within EDTA blood samples.
- Open Access
- 26 February 2025
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 27, e55492
Complete Blood Count and Monocyte Distribution Width–Based Machine Learning Algorithms for Sepsis Detection: Multicentric Development and External Validation Study
Background: Sepsis is an organ dysfunction caused by a dysregulated host response to infection. Early detection is fundamental to improving the patient outcome. Laboratory medicine can play a crucial role by providing biomarkers whose alteration can be detected before the onset of clinical signs and symptoms. In particular, the relevance of monocyte distribution width (MDW) as a sepsis biomarker has emerged in the previous decade. However, despite encouraging results, MDW has poor sensitivity and positive predictive value when compared to other biomarkers. Objective: This study aims to investigate the use of machine learning (ML) to overcome the limitations mentioned earlier by combining different parameters and therefore improving sepsis detection. However, making ML models function in clinical practice may be problematic, as their performance may suffer when deployed in contexts other than the research environment. In fact, even widely used commercially available models have been demonstrated to generalize poorly in out-of-distribution scenarios. Methods: In this multicentric study, we developed ML models whose intended use is the early detection of sepsis on the basis of MDW and complete blood count parameters. In total, data from 6 patient cohorts (encompassing 5344 patients) collected at 5 different Italian hospitals were used to train and externally validate ML models. The models were trained on a patient cohort encompassing patients enrolled at the emergency department, and it was externally validated on 5 different cohorts encompassing patients enrolled at both the emergency department and the intensive care unit. The cohorts were selected to exhibit a variety of data distribution shifts compared to the training set, including label, covariate, and missing data shifts, enabling a conservative validation of the developed models. To improve generalizability and robustness to different types of distribution shifts, the developed ML models combine traditional methodologies with advanced techniques inspired by controllable artificial intelligence (AI), namely cautious classification, which gives the ML models the ability to abstain from making predictions, and explainable AI, which provides health operators with useful information about the models’ functioning. Results: The developed models achieved good performance on the internal validation (area under the receiver operating characteristic curve between 0.91 and 0.98), as well as consistent generalization performance across the external validation datasets (area under the receiver operating characteristic curve between 0.75 and 0.95), outperforming baseline biomarkers and state-of-the-art ML models for sepsis detection. Controllable AI techniques were further able to improve performance and were used to derive an interpretable set of diagnostic rules. Conclusions: Our findings demonstrate how controllable AI approaches based on complete blood count and MDW may be used for the early detection of sepsis while also demonstrating how the proposed methodology can be used to develop ML models that are more resistant to different types of data distribution shifts.
- 26 February 2025
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 27, e55492
Open AccessComplete Blood Count and Monocyte Distribution Width–Based Machine Learning Algorithms for Sepsis Detection: Multicentric Development and External Validation Study
Background: Sepsis is an organ dysfunction caused by a dysregulated host response to infection. Early detection is fundamental to improving the patient outcome. Laboratory medicine can play a crucial role by providing biomarkers whose alteration can be detected before the onset of clinical signs and symptoms. In particular, the relevance of monocyte distribution width (MDW) as a sepsis biomarker has emerged in the previous decade. However, despite encouraging results, MDW has poor sensitivity and positive predictive value when compared to other biomarkers. Objective: This study aims to investigate the use of machine learning (ML) to overcome the limitations mentioned earlier by combining different parameters and therefore improving sepsis detection. However, making ML models function in clinical practice may be problematic, as their performance may suffer when deployed in contexts other than the research environment. In fact, even widely used commercially available models have been demonstrated to generalize poorly in out-of-distribution scenarios. Methods: In this multicentric study, we developed ML models whose intended use is the early detection of sepsis on the basis of MDW and complete blood count parameters. In total, data from 6 patient cohorts (encompassing 5344 patients) collected at 5 different Italian hospitals were used to train and externally validate ML models. The models were trained on a patient cohort encompassing patients enrolled at the emergency department, and it was externally validated on 5 different cohorts encompassing patients enrolled at both the emergency department and the intensive care unit. The cohorts were selected to exhibit a variety of data distribution shifts compared to the training set, including label, covariate, and missing data shifts, enabling a conservative validation of the developed models. To improve generalizability and robustness to different types of distribution shifts, the developed ML models combine traditional methodologies with advanced techniques inspired by controllable artificial intelligence (AI), namely cautious classification, which gives the ML models the ability to abstain from making predictions, and explainable AI, which provides health operators with useful information about the models’ functioning. Results: The developed models achieved good performance on the internal validation (area under the receiver operating characteristic curve between 0.91 and 0.98), as well as consistent generalization performance across the external validation datasets (area under the receiver operating characteristic curve between 0.75 and 0.95), outperforming baseline biomarkers and state-of-the-art ML models for sepsis detection. Controllable AI techniques were further able to improve performance and were used to derive an interpretable set of diagnostic rules. Conclusions: Our findings demonstrate how controllable AI approaches based on complete blood count and MDW may be used for the early detection of sepsis while also demonstrating how the proposed methodology can be used to develop ML models that are more resistant to different types of data distribution shifts.
- Open Access
- 11 September 2024
- Published by MDPI AG in Diagnostics
- Vol. 14 (18) , 2011
Performance Evaluation of Automated Erythrocyte Sedimentation Rate (ESR) Analyzers in a Multicentric Study
Background: Erythrocyte Sedimentation Rate (ESR) is an easy test used to diagnose and monitor inflammatory and infectious diseases. The aim of this study was the evaluation of the performance of three ESR automated analyzers, VES-MATIC 5, CUBE 30 TOUCH, and MINI-CUBE, involving four Italian polyclinics in Rome, Siena, Como, and Arezzo, as well as inter-site variability assessment to detect possible device-dependent and operator-dependent influences. Methods: Accuracy analysis was carried out by analyzing the same samples with all three instruments and comparing them with the Westergren method. Precision was assessed with quality control material through intra-run and inter-run precision. Repeatability was estimated by reanalyzing fresh blood samples belonging to three ESR ranges (low, intermediate, and high) six times. Results: The results showed a strong correlation (Spearman coefficients R2) between the manual method and VES-MATIC 5 (0.978), CUBE 30 TOUCH (0.981), and MINI-CUBE (0.974). The accuracy of all clinics was excellent, with coefficients of variation (CVs) of less than 10% for all instruments. Repeatability confirmed an excellent level for all ESR ranges, with CVs below 10%. Conclusions: The study proved that all three automated instruments offer optimal performance for accuracy and precision and are suitable for both large and small facilities without influences of the laboratory environment.
- 11 September 2024
- Published by MDPI AG in Diagnostics
- Vol. 14 (18) , 2011
Open AccessPerformance Evaluation of Automated Erythrocyte Sedimentation Rate (ESR) Analyzers in a Multicentric Study
Background: Erythrocyte Sedimentation Rate (ESR) is an easy test used to diagnose and monitor inflammatory and infectious diseases. The aim of this study was the evaluation of the performance of three ESR automated analyzers, VES-MATIC 5, CUBE 30 TOUCH, and MINI-CUBE, involving four Italian polyclinics in Rome, Siena, Como, and Arezzo, as well as inter-site variability assessment to detect possible device-dependent and operator-dependent influences. Methods: Accuracy analysis was carried out by analyzing the same samples with all three instruments and comparing them with the Westergren method. Precision was assessed with quality control material through intra-run and inter-run precision. Repeatability was estimated by reanalyzing fresh blood samples belonging to three ESR ranges (low, intermediate, and high) six times. Results: The results showed a strong correlation (Spearman coefficients R2) between the manual method and VES-MATIC 5 (0.978), CUBE 30 TOUCH (0.981), and MINI-CUBE (0.974). The accuracy of all clinics was excellent, with coefficients of variation (CVs) of less than 10% for all instruments. Repeatability confirmed an excellent level for all ESR ranges, with CVs below 10%. Conclusions: The study proved that all three automated instruments offer optimal performance for accuracy and precision and are suitable for both large and small facilities without influences of the laboratory environment.
- Open Access
- 6 March 2024
- Published by MDPI AG in Diagnostics
- Vol. 14 (5) , 557
Evaluation of Stability and Accuracy Compared to the Westergren Method of ESR Samples Analyzed at VES-MATIC 5
The Erythrocyte Sedimentation Rate (ESR) is a diagnostic estimator of systemic inflammation as a reflection of acute phase proteins circulating in the blood. The purpose of this manuscript is to evaluate the blood stability at room temperature (RT) and at 4 °C to avoid ESR diagnostic errors, as well as the accuracy of the VES-MATIC 5 analyzer. The ESR stability evaluation at RT for 24 h (4 h “T1”, 6 h “T2”, 8 h “T3”, 10 h “T4”, 24 h “T5”) and at 4 °C (24 h, 36 h, 48 h) was carried out using 635 total samples, starting with T0 (2 h of venipuncture). For method comparison, 164 patients were analyzed using VES-MATIC 5 and then the Westergren reference method. The sample at RT is established by a significant gradual decrease in correlation R = 0.99 (T0 vs. T1), R = 0.97 (T0 vs. T2), R = 0.92 (T0 vs. T3), R = 0.87 (T0 vs. T4), and R = 0.40 (T0 vs. T5). The stability at 4 °C after 24 h, 36 h, and 48 h showed a regression of R = 0.99, R = 0.97, and R = 0.95, respectively. Therefore, ESR measurements on RT samples beyond 6 h after collection cannot be carried out, but the ESR can be measured until 36 h for samples stored at 4 °C. Moreover, the VES-MATIC 5 accuracy performance compared to the Westergren method (R = 0.96) is confirmed.
- 6 March 2024
- Published by MDPI AG in Diagnostics
- Vol. 14 (5) , 557
Open AccessEvaluation of Stability and Accuracy Compared to the Westergren Method of ESR Samples Analyzed at VES-MATIC 5
The Erythrocyte Sedimentation Rate (ESR) is a diagnostic estimator of systemic inflammation as a reflection of acute phase proteins circulating in the blood. The purpose of this manuscript is to evaluate the blood stability at room temperature (RT) and at 4 °C to avoid ESR diagnostic errors, as well as the accuracy of the VES-MATIC 5 analyzer. The ESR stability evaluation at RT for 24 h (4 h “T1”, 6 h “T2”, 8 h “T3”, 10 h “T4”, 24 h “T5”) and at 4 °C (24 h, 36 h, 48 h) was carried out using 635 total samples, starting with T0 (2 h of venipuncture). For method comparison, 164 patients were analyzed using VES-MATIC 5 and then the Westergren reference method. The sample at RT is established by a significant gradual decrease in correlation R = 0.99 (T0 vs. T1), R = 0.97 (T0 vs. T2), R = 0.92 (T0 vs. T3), R = 0.87 (T0 vs. T4), and R = 0.40 (T0 vs. T5). The stability at 4 °C after 24 h, 36 h, and 48 h showed a regression of R = 0.99, R = 0.97, and R = 0.95, respectively. Therefore, ESR measurements on RT samples beyond 6 h after collection cannot be carried out, but the ESR can be measured until 36 h for samples stored at 4 °C. Moreover, the VES-MATIC 5 accuracy performance compared to the Westergren method (R = 0.96) is confirmed.
- Open Access
- 15 December 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Complete Blood Count and Monocyte Distribution Width–Based Machine Learning Algorithms for Sepsis Detection: Multicentric Development and External Validation Study (Preprint)
BACKGROUND Sepsis is an organ dysfunction caused by a dysregulated host response to infection. Early detection is fundamental to improving the patient outcome. Laboratory medicine can play a crucial role by providing biomarkers whose alteration can be detected before the onset of clinical signs and symptoms. In particular, the relevance of monocyte distribution width (MDW) as a sepsis biomarker has emerged in the previous decade. However, despite encouraging results, MDW has poor sensitivity and positive predictive value when compared to other biomarkers. OBJECTIVE This study aims to investigate the use of machine learning (ML) to overcome the limitations mentioned earlier by combining different parameters and therefore improving sepsis detection. However, making ML models function in clinical practice may be problematic, as their performance may suffer when deployed in contexts other than the research environment. In fact, even widely used commercially available models have been demonstrated to generalize poorly in out-of-distribution scenarios. METHODS In this multicentric study, we developed ML models whose intended use is the early detection of sepsis on the basis of MDW and complete blood count parameters. In total, data from 6 patient cohorts (encompassing 5344 patients) collected at 5 different Italian hospitals were used to train and externally validate ML models. The models were trained on a patient cohort encompassing patients enrolled at the emergency department, and it was externally validated on 5 different cohorts encompassing patients enrolled at both the emergency department and the intensive care unit. The cohorts were selected to exhibit a variety of data distribution shifts compared to the training set, including label, covariate, and missing data shifts, enabling a conservative validation of the developed models. To improve generalizability and robustness to different types of distribution shifts, the developed ML models combine traditional methodologies with advanced techniques inspired by controllable artificial intelligence (AI), namely cautious classification, which gives the ML models the ability to abstain from making predictions, and explainable AI, which provides health operators with useful information about the models’ functioning. RESULTS The developed models achieved good performance on the internal validation (area under the receiver operating characteristic curve between 0.91 and 0.98), as well as consistent generalization performance across the external validation datasets (area under the receiver operating characteristic curve between 0.75 and 0.95), outperforming baseline biomarkers and state-of-the-art ML models for sepsis detection. Controllable AI techniques were further able to improve performance and were used to derive an interpretable set of diagnostic rules. CONCLUSIONS Our findings demonstrate how controllable AI approaches based on complete blood count and MDW may be used for the early detection of sepsis while also demonstrating how the proposed methodology can be used to develop ML models that are more resistant to different types of data distribution shifts.
- 15 December 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessComplete Blood Count and Monocyte Distribution Width–Based Machine Learning Algorithms for Sepsis Detection: Multicentric Development and External Validation Study (Preprint)
BACKGROUND Sepsis is an organ dysfunction caused by a dysregulated host response to infection. Early detection is fundamental to improving the patient outcome. Laboratory medicine can play a crucial role by providing biomarkers whose alteration can be detected before the onset of clinical signs and symptoms. In particular, the relevance of monocyte distribution width (MDW) as a sepsis biomarker has emerged in the previous decade. However, despite encouraging results, MDW has poor sensitivity and positive predictive value when compared to other biomarkers. OBJECTIVE This study aims to investigate the use of machine learning (ML) to overcome the limitations mentioned earlier by combining different parameters and therefore improving sepsis detection. However, making ML models function in clinical practice may be problematic, as their performance may suffer when deployed in contexts other than the research environment. In fact, even widely used commercially available models have been demonstrated to generalize poorly in out-of-distribution scenarios. METHODS In this multicentric study, we developed ML models whose intended use is the early detection of sepsis on the basis of MDW and complete blood count parameters. In total, data from 6 patient cohorts (encompassing 5344 patients) collected at 5 different Italian hospitals were used to train and externally validate ML models. The models were trained on a patient cohort encompassing patients enrolled at the emergency department, and it was externally validated on 5 different cohorts encompassing patients enrolled at both the emergency department and the intensive care unit. The cohorts were selected to exhibit a variety of data distribution shifts compared to the training set, including label, covariate, and missing data shifts, enabling a conservative validation of the developed models. To improve generalizability and robustness to different types of distribution shifts, the developed ML models combine traditional methodologies with advanced techniques inspired by controllable artificial intelligence (AI), namely cautious classification, which gives the ML models the ability to abstain from making predictions, and explainable AI, which provides health operators with useful information about the models’ functioning. RESULTS The developed models achieved good performance on the internal validation (area under the receiver operating characteristic curve between 0.91 and 0.98), as well as consistent generalization performance across the external validation datasets (area under the receiver operating characteristic curve between 0.75 and 0.95), outperforming baseline biomarkers and state-of-the-art ML models for sepsis detection. Controllable AI techniques were further able to improve performance and were used to derive an interpretable set of diagnostic rules. CONCLUSIONS Our findings demonstrate how controllable AI approaches based on complete blood count and MDW may be used for the early detection of sepsis while also demonstrating how the proposed methodology can be used to develop ML models that are more resistant to different types of data distribution shifts.
- Open Access
- 28 March 2023
- Published by MDPI AG in Biomedicines
- Vol. 11 (4) , 1046
Novel Biomarkers for Early Detection of Acute Kidney Injury and Prediction of Long-Term Kidney Function Decline after Partial Nephrectomy
Background: Identifying acute kidney injury (AKI) within few hours of onset is certainly helpful. However, early prediction of a long-term eGFR decline may be an even more important goal. Our aim was to identify and compare serum [creatinine, kineticGFR, cystatin C, neutrophil gelatinase–associated lipocalin (NGAL)] and urinary (NephroCheck, NGAL, proteinuria, albuminuria, acantocytes at urinary sediment) predictors of AKI that might efficiently predict long-term GFR decline after robotic Nephron-Spearing Surgery (rNSS). Methods: Monocentric prospective observational study. Patients scheduled for rNSS for suspected localized Renal Cell Carcinoma from May 2017 to October 2017 were enrolled. Samples were collected preoperatively and postoperatively (timepoints: 4 h, 10 h, 24 h, 48 h), while kidney function was re-assessed up to 24 months. Results: 38 patients were included; 16 (42%) developed clinical AKI. The eGFR decline at 24 months was more pronounced after postoperative AKI (−20.75 vs. −7.20, p < 0.0001). KineticGFR at 4 h (p = 0.008) and NephroCheck at 10 h (p = 0.001) were, at multivariable linear regression analysis, efficient predictors of post-operative AKI and long-term eGFR decline if compared to creatinine (R2 0.33 vs. 0.04). Conclusions: NephroCheck and kineticGFR have emerged as promising noninvasive, accurate, and early biomarkers of postoperative AKI and long-term GFR decline after rNSS. Combining NephroCheck and kineticGFR in clinical practice would allow to identify high risk of postoperative AKI and long-term GFR decline as early as 10 h after surgery.
- 28 March 2023
- Published by MDPI AG in Biomedicines
- Vol. 11 (4) , 1046
Open AccessNovel Biomarkers for Early Detection of Acute Kidney Injury and Prediction of Long-Term Kidney Function Decline after Partial Nephrectomy
Background: Identifying acute kidney injury (AKI) within few hours of onset is certainly helpful. However, early prediction of a long-term eGFR decline may be an even more important goal. Our aim was to identify and compare serum [creatinine, kineticGFR, cystatin C, neutrophil gelatinase–associated lipocalin (NGAL)] and urinary (NephroCheck, NGAL, proteinuria, albuminuria, acantocytes at urinary sediment) predictors of AKI that might efficiently predict long-term GFR decline after robotic Nephron-Spearing Surgery (rNSS). Methods: Monocentric prospective observational study. Patients scheduled for rNSS for suspected localized Renal Cell Carcinoma from May 2017 to October 2017 were enrolled. Samples were collected preoperatively and postoperatively (timepoints: 4 h, 10 h, 24 h, 48 h), while kidney function was re-assessed up to 24 months. Results: 38 patients were included; 16 (42%) developed clinical AKI. The eGFR decline at 24 months was more pronounced after postoperative AKI (−20.75 vs. −7.20, p < 0.0001). KineticGFR at 4 h (p = 0.008) and NephroCheck at 10 h (p = 0.001) were, at multivariable linear regression analysis, efficient predictors of post-operative AKI and long-term eGFR decline if compared to creatinine (R2 0.33 vs. 0.04). Conclusions: NephroCheck and kineticGFR have emerged as promising noninvasive, accurate, and early biomarkers of postoperative AKI and long-term GFR decline after rNSS. Combining NephroCheck and kineticGFR in clinical practice would allow to identify high risk of postoperative AKI and long-term GFR decline as early as 10 h after surgery.
- Open Access
Carriers of ADAMTS13 Rare Variants Are at High Risk of Life-Threatening COVID-19
Thrombosis of small and large vessels is reported as a key player in COVID-19 severity. However, host genetic determinants of this susceptibility are still unclear. Congenital Thrombotic Thrombocytopenic Purpura is a severe autosomal recessive disorder characterized by uncleaved ultra-large vWF and thrombotic microangiopathy, frequently triggered by infections. Carriers are reported to be asymptomatic. Exome analysis of about 3000 SARS-CoV-2 infected subjects of different severities, belonging to the GEN-COVID cohort, revealed the specific role of vWF cleaving enzyme ADAMTS13 (A disintegrin-like and metalloprotease with thrombospondin type 1 motif, 13). We report here that ultra-rare variants in a heterozygous state lead to a rare form of COVID-19 characterized by hyper-inflammation signs, which segregates in families as an autosomal dominant disorder conditioned by SARS-CoV-2 infection, sex, and age. This has clinical relevance due to the availability of drugs such as Caplacizumab, which inhibits vWF–platelet interaction, and Crizanlizumab, which, by inhibiting P-selectin binding to its ligands, prevents leukocyte recruitment and platelet aggregation at the site of vascular damage.
Open AccessCarriers of ADAMTS13 Rare Variants Are at High Risk of Life-Threatening COVID-19
Thrombosis of small and large vessels is reported as a key player in COVID-19 severity. However, host genetic determinants of this susceptibility are still unclear. Congenital Thrombotic Thrombocytopenic Purpura is a severe autosomal recessive disorder characterized by uncleaved ultra-large vWF and thrombotic microangiopathy, frequently triggered by infections. Carriers are reported to be asymptomatic. Exome analysis of about 3000 SARS-CoV-2 infected subjects of different severities, belonging to the GEN-COVID cohort, revealed the specific role of vWF cleaving enzyme ADAMTS13 (A disintegrin-like and metalloprotease with thrombospondin type 1 motif, 13). We report here that ultra-rare variants in a heterozygous state lead to a rare form of COVID-19 characterized by hyper-inflammation signs, which segregates in families as an autosomal dominant disorder conditioned by SARS-CoV-2 infection, sex, and age. This has clinical relevance due to the availability of drugs such as Caplacizumab, which inhibits vWF–platelet interaction, and Crizanlizumab, which, by inhibiting P-selectin binding to its ligands, prevents leukocyte recruitment and platelet aggregation at the site of vascular damage.
- Open Access
- 14 January 2022
- Published by Elsevier in Practical Laboratory Medicine
- Vol. 29, e00264
Isolated case of platelet satellitism around white blood cells and phagocytosis by neutrophils and monocytes
- 14 January 2022
- Published by Elsevier in Practical Laboratory Medicine
- Vol. 29, e00264
Open AccessIsolated case of platelet satellitism around white blood cells and phagocytosis by neutrophils and monocytes