A Neural-network-based approach to personalize insulin bolus calculation using continuous glucose monitoring
Open accessAbstract
Background: In type 1 diabetes (T1D) therapy, the calculation of the meal insulin bolus is performed according to a standard formula (SF) exploiting carbohydrate intake, carbohydrate-to-insulin ratio, correction factor, insulin on board, and target glucose. Recently, some approaches were proposed to account for preprandial glucose rate of change (ROC) in the SF, including those by Scheiner and by Pettus and Edelman. Here, the aim is to develop a new approach, based on neural networks (NN), to optimize and personalize the bolus calculation using continuous glucose monitoring information and some easily accessible patient parameters. Method: The UVa/Padova T1D Simulator was used to simulate data of 100 virtual adults in a single-meal noise-free scenario with different conditions in terms of meal amount and preprandial blood glucose and ROC values. An NN was trained to learn the optimal insulin dose using the SF parameters, ROC, body weight, insulin pump basal infusion rate and insulin sensitivity as features. The performance of the NN for meal bolus calculation was assessed by blood glucose risk index (BGRI) and compared to the methods by Scheiner and by Pettus and Edelman. Results: The NN approach brings to a small but statistically significant ( P < .001) reduction of BGRI value, equal to 0.37, 0.23, and 0.20 versus SF, Scheiner, and Pettus and Edelman, respectively. Conclusion: This preliminary study showed the potentiality of using NNs for the personalization and optimization of the meal insulin bolus calculation. Future work will deal with more realistic scenarios including technological and physiological/behavioral sources of variability.
AI snapshot
Written from the paper’s full text.
TL;DR
This in silico study developed a neural-network corrector (NNC) to personalize meal insulin bolus calculation using CGM-derived and patient-specific features. In 100 virtual adults under noise-free single-meal conditions, NNC improved BGRI versus standard formula and two literature ROC-correction methods, though errors sometimes worsened glycemic outcomes.
Key findings
- NNC produced a small but statistically significant median BGRI reduction of 0.37 [0.02, 1.51] versus SF, 0.23 [0.01, 0.99] versus Scheiner, and 0.20 [0.01, 0.95] versus Pettus and Edelman, all P < .001.
- NNC approximated the optimal correction closely: median ΔBGRI for OPT versus NNC was 0 [–0.02, 0.01], P = .30, while SF versus OPT showed a median improvement of 0.40 [0.04, 1.56].
- In representative cases, adult#26 had BGRI(SF)=14.27, BGRI(OPT)=BGRI(NNC)=12.10, BGRI(SC)=13.59, BGRI(PE)=13.07; adult#59 had BGRI(SF)=4.01, BGRI(OPT)=BGRI(NNC)=3.26, BGRI(SC)=3.27, BGRI(PE)=3.39.
- The paper reports that in several scenarios NNC error led to worse BGRI than SF, and that there was no correlation between the optimal correction and the actual insulin bolus amount, suggesting fixed ROC corrections may be suboptimal.
Methods
- Data were generated with the UVa/Padova T1D Simulator for 100 virtual adults in a single-meal noise-free scenario, with meal amounts of 50–100 g CHO, preprandial BG levels of 60, 70, 80, 100, 150, and 250 mg/dL, and ROC values of –2, –1, 1, and 2 mg/dL/min, yielding 144 conditions per subject and a final dataset of 9963 records.
- The optimal correction was identified by simulating 41 equally spaced correction values and selecting the one minimizing BGRI; NNC was a feedforward fully connected NN with three hidden layers, dropout layers, 10 inputs (GC, ROC, IOB, CR, CF, GT, Ib, BW, VC, CHO), RMSprop training, TPE hyperparameter optimization, 80/20 train/test split, and 5-fold cross-validation.
- Glycemic outcome was assessed by BGRI, with between-method differences tested by nonparametric multiway ANOVA followed by Dunn’s post hoc test at 0.1% significance.
Limitations
- Stated by the paper: this is a preliminary in silico, noise-free, single-meal study in 100 virtual adults; future work is needed with more realistic scenarios including technological and physiological/behavioral variability, and NNC error can worsen BGRI in several scenarios.
- Stated by the paper: the NNC’s domain of validity is restricted because optimal correction of the standard formula may require all possible physiological classes/typologies during training, and limitations are intrinsically related to the strict constraints of the original SF structure.
- Evident from the text: there was no real-patient or clinical validation, no CGM sensor error/noise, no SMBG error, no behavioral variability, only adult virtual subjects, and ROC = 0 and ±3 mg/dL/min were excluded; only NNs were evaluated among possible nonlinear machine learning methods.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
0 of 6 cited sources checked · 0 supported · 0 flagged
- Findingown data§ Abstract
In the in silico noise-free single-meal study on 100 virtual adults, the neural-network-based approach (NNC) reduces the blood glucose risk index (BGRI) compared with the standard formula (SF) by a median of 0.37 (P < .001).
“The NN approach brings to a small but statistically significant (P < .001) reduction of BGRI value, equal to 0.37, 0.23, and 0.20 versus SF, Scheiner, and Pettus and Edelman, respectively.”
- Also stated in: § Results · plain
“SF-OPT SF-NNC OPT-NNC SC-NNC PE-NNC ΔBGRI 0.40 [0.04, 1.56] 0.37 [0.02, 1.51] 0 [–0.02, 0.01] 0.23 [0.01, 0.99] 0.20 [0.01, 0.95] P value <.001 <.001 .30 <.001 <.001”
Quote in the paperSubstantial claim - Also stated in: § Results · plain
- Findingown data§ Abstract
In the in silico noise-free single-meal study on 100 virtual adults, the NNC reduces BGRI compared with the Pettus and Edelman method (PE) by a median of 0.20 (P < .001).
“The NN approach brings to a small but statistically significant (P < .001) reduction of BGRI value, equal to 0.37, 0.23, and 0.20 versus SF, Scheiner, and Pettus and Edelman, respectively.”
- Also stated in: § Results · plain
“Overall, according to Figure 3, NNC introduces a small but statistical significant improvement (P < .001) equal to 0.23 and 0.20 if compared with SC and PE, respectively (see Table 1).”
Quote in the paperSubstantial claim - Also stated in: § Results · plain
- Findingown data§ Results
Using the optimal correction (OPT), there is a statistically significant median BGRI improvement of 0.40 compared with SF (P < .001).
“In particular, using the optimal correction, we obtained a statistically significant (P < .001) median BGRI improvement of 0.40 (see Table 1).”
- Also stated in: § Results · plain
“SF-OPT SF-NNC OPT-NNC SC-NNC PE-NNC ΔBGRI 0.40 [0.04, 1.56] 0.37 [0.02, 1.51] 0 [–0.02, 0.01] 0.23 [0.01, 0.99] 0.20 [0.01, 0.95] P value <.001 <.001 .30 <.001 <.001”
Quote in the paperSubstantial claim - Also stated in: § Results · plain
- Findingown data§ Results
There is no statistically significant difference in BGRI between OPT and NNC (median difference equal to 0, P = .30).
“No statistically significant difference between the two distributions is found (median difference equal to 0, P = .30).”
- Also stated in: § Results · plain
“SF-OPT SF-NNC OPT-NNC SC-NNC PE-NNC ΔBGRI 0.40 [0.04, 1.56] 0.37 [0.02, 1.51] 0 [–0.02, 0.01] 0.23 [0.01, 0.99] 0.20 [0.01, 0.95] P value <.001 <.001 .30 <.001 <.001”
Quote in the paperSubstantial claim - Also stated in: § Results · plain
- Findingown data§ Abstract
In the in silico noise-free single-meal study on 100 virtual adults, the NNC reduces BGRI compared with the Scheiner method (SC) by a median of 0.23 (P < .001).
“The NN approach brings to a small but statistically significant (P < .001) reduction of BGRI value, equal to 0.37, 0.23, and 0.20 versus SF, Scheiner, and Pettus and Edelman, respectively.”
- Also stated in: § Results · plain
“Overall, according to Figure 3, NNC introduces a small but statistical significant improvement (P < .001) equal to 0.23 and 0.20 if compared with SC and PE, respectively (see Table 1).”
Quote in the paperSubstantial claim - Also stated in: § Results · plain
- Findingown data§ Results
There is no correlation between the optimal correction F and the actual insulin bolus amount.
“Indeed, we found that there is no correlation between and the actual insulin bolus amount.”
Quote in the paperSubstantial claim - Findingown data§ Results
In the representative example of adult#59 (Gc = 80, ROC = 2, CHO = 80), NNC achieves better BGRI compared with SF, SC, and PE.
“Qualitatively equal results are obtained in the bottom panel case (adult#59, Gc = 80, ROC = 2, and CHO = 80). In particular, the NNC estimated value () achieves better BGRI compared with SF, SC and PE.”
- Also stated in: § Results · plain
“In particular, the obtained values of BGRI are: BGRI(SF) = 4.01, BGRI(OPT) = BGRI(NNC) = 3.26, BGRI(SC) = 3.27, and BGRI(PE) = 3.39.”
Quote in the paperNot a substantial claim - Also stated in: § Results · plain
- Interpretationown data§ Results
Building a tool able to correct optimally SF seems possible only if, during the training phase, all the possible physiological classes/typologies of the patients at hand are available.
“Indeed, an accurate analysis of the results of our simulations showed that building a tool able to correct optimally SF seems possible only if, during the training phase, all the possible physiological classes/typologies of the patients at hand are available.”
Quote in the paperSubstantial claim - Premisecitation§ Results
The optimal modulation of insulin bolus is strongly related to preprandial conditions and individual parameters of the patient.
“Both intuition and evidence provided by the in silico studies21 suggest that the optimal modulation of insulin bolus is strongly related to preprandial conditions and individual parameters of the patient.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [21] 21. Marturano F, Cappon G, Vettoretti M, Facchinetti A, Sparacino G. In silico assessment of literature methods to adjust insulin bolus dose according to CGM trend. In: 17th Annual Diabetes Technology Meeting (DTM), Bethesda, MD: DTM; 2017. [Google Scholar]
Not checked in the free run.
- Interpretationown data§ Results
The NNC estimates the optimal target correction values with good precision.
“Notably, the median ΔBGRI and interquartile range is almost the same as in SF versus OPT, meaning that the NNC estimates the optimal target values with good precision.”
- Also stated in: § Results · plain
“As further proof of the capability of the NNC in estimating the optimal correction values and improving the glycemic outcomes, in Figure 3 we report the distribution of the difference between BGRI in OPT and NNC.”
Quote in the paperSubstantial claim - Also stated in: § Results · plain
- Interpretationown data§ Results
The effectiveness of the optimal correction to be applied to the SF is related to the structure of the SF itself.
“The effectiveness of the optimal correction to be applied to the SF is, not surprisingly, related to the structure of the SF itself.”
Quote in the paperSubstantial claim - Interpretationreasoning§ Results
Correcting SF by fixed values as defined by SC and PE can be, in general, suboptimal.
“This consideration suggests that correcting SF by a fixed values as defined by and can be, in general, suboptimal.”
Quote in the paperSubstantial claim - Interpretationcitation§ Abstract
Applying the same correction for ROC to equation (1) for all preprandial BG conditions and equally for all individuals can be somewhat simplistic.
“These results suggest that applying the same correction for ROC to equation (1) for all the preprandial BG conditions and in a way equal for all individuals can be somewhat simplistic.”
Quote in the paperSubstantial claim - Study design and conductreasoning§ Methods
The statistical analysis accounted for dependencies between observations within subjects by using a nonparametric multiway ANOVA with virtual patient and correction method as factors, followed by Dunn's post hoc test.
“Finally, to statistically evaluate the between-methods differences, we performed a nonparametric multiway ANOVA, explicitly considering the virtual patient and the associated correction method as factors, followed by a multicomparison using Dunn’s post hoc test with 0.1% significance level.”
- Also stated in: § Methods · plain
“The reason for choosing such a statistical test is that the test set could contain traces coming from the same virtual patient (same physiology, but different initial conditions). Therefore, we need to perform the appropriate statistical analysis to be able to consider dependencies between observations within subjects.”
Quote in the paperSubstantial claim - Also stated in: § Methods · plain
- Premisecitation§ Abstract
Scheiner and Pettus and Edelman developed empirical formulas (SC and PE) to modify the insulin bolus amount computed by SF according to the ROC arrow value.
“Specifically, Scheiner10 and Pettus and Edelman11 developed some empirical formulas, hereafter indicated by SC and PE, respectively, to modify the insulin bolus amount computed in equation (1) according to the ROC arrow value, that is, a graphical indication of the magnitude and the direction of glucose changing displayed in most of currently commercialized CGM devices.”
Quote in the paperSubstantial claim0 of 2 sources checked- cites Scheiner10
Not checked in the free run.
- cites Pettus and Edelman11
Not checked in the free run.
- May combine two claims. 6
- Classified as probably not a substantial claim. 3
May combine two claims.
Findingown data§ ResultsIn the representative example of adult#26 (Gc = 250, ROC = -1, CHO = 100), NNC avoids hypoglycemia and keeps BG greater than 70 mg/dL, while SF, SC, and PE lead to severe hypoglycemia.
“In the top panel case (adult#26, Gc = 250, ROC = –1 and CHO = 100) the NNC achieves significantly better results compared to SF, SC and PE being able to avoid hypoglycemia.”
- Also stated in: § Results · plain
“Notably, while SF, SC and PE lead adult#26 to severe hypoglycemia, NNC allows to keep BG greater than 70 mg/dL significantly reducing BGRI.”
Quote in the paperSubstantial claim- Also stated in: § Results · plain
May combine two claims.
Interpretationown data§ ResultsIn several scenarios, the NNC error is relevant and, compared to SF, leads to worse BGRI values because the model errs in approximating OPT.
“However, Figure 3 points out that, in several scenarios, the NNC error is relevant and, compared to SF, lead to worse BGRI values.”
- Also stated in: § Results · plain
“In particular, BG control worsens when using NNC to estimate due to the error the model makes in approximating OPT.”
Quote in the paperSubstantial claim- Also stated in: § Results · plain
May combine two claims.
Premisecitation§ AbstractIn the preliminary study by Marturano et al, SC and PE did not perform effectively for all subjects; for a subset of tested subjects, a deterioration of glycemic control was observed.
“In addition, these methods did not perform effectively for all the subjects, but for a subset of the tested subjects a deterioration of glycemic control was observed.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Marturano et al21
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductcitation§ MethodsDropout layers were interposed between hidden layers to prevent overfitting and improve the generalization capability of NNC.
“In addition, two dropout layers have been interposed between each hidden layer to prevent overfitting and improve the generalization capability of NNC.24”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites [24] 24. Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salakhutdinov R. Dropout: a simple way to prevent neural networks from overfitting. J Machine Learning Res. 2014;15:1929-1958. [Google Scholar]
Not checked in the free run.
Classified as probably not a substantial claim. May combine two claims.
Premisecitation§ AbstractA preliminary study by Marturano et al showed that SC and PE are on average more effective than SF, but their performance is strongly dependent on the specific preprandial condition in terms of BG level and ROC value.
“In this context, a preliminary study by Marturano et al21 evaluated these two methodologies in an in-silico, noise-free environment showing that, despite being on average more effective than the SF, their performance is strongly dependent on the specific preprandial condition in terms of BG level and ROC value.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Marturano et al21
Not checked in the free run.
Probably not a substantial claim: may not pass S3 (load-bearing).
Study design and conductown data§ MethodsHyper-parameter performance was assessed in a 5-fold cross-validation setting.
“Then, for a given set of hyper-parameters h, block B assesses its performance in a 5-fold cross-validation setting.”
Quote in the paperNot a substantial claimMay combine two claims.
Interpretationown data§ ResultsIn an in silico study of 100 virtual subjects in noise-free conditions, the new NN-based method outperforms SF, SC, and PE in terms of BGRI over 24 hours, and is close to the optimum determined by exhaustive search.
“An in silico study performed in 100 virtual subjects in noise-free conditions, showed that, in terms of glycemic outcomes measured as BGRI over 24 hours, the new method outperforms the literature approaches SF, SC, and PE, and is close to the optimum determined by exhaustive search.”
Quote in the paperSubstantial claim
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The NN-based bolus calculator achieved significantly lower blood glucose risk index values than the standard formula.
“The results showed that the NN-based BC achieved significantly lower BGRI values than did the SF.”
From Advanced Diabetes Management Using Artificial Intelligence and Continuous Glucose Monitoring Sensors (2020) · cites this paper as [37]
Cappon et al. used a large training set of simulated data to train a neural network that predicts the optimal value for X to correct Gc in the standard formula.
“In particular, Cappon et al. [37] used a large training set of simulated data to train a neural network (NN) that predicts the optimal value for X, i.e., the value of X that optimizes post-prandial glucose control.”
From Advanced Diabetes Management Using Artificial Intelligence and Continuous Glucose Monitoring Sensors (2020) · cites this paper as [37]
Although on average the NN-based bolus calculator improved glycemic control compared to the standard formula, in a non-negligible number of cases the dose provided by the NN-based bolus calculator led to worse glycemic control than the standard formula.
“Although, on average, the NN-based BC did improve BG control when compared to the SF, in a non-negligible number of cases the dose provided by the NN-based BC led to worse glycemic control than did the SF.”
From Advanced Diabetes Management Using Artificial Intelligence and Continuous Glucose Monitoring Sensors (2020) · cites this paper as [37]
The neural network hyper-parameters were tuned using a five-fold cross-validation procedure.
“The NN hyper-parameters (i.e., the number of neurons per layer and the activation function type) were tuned using a five-fold CV procedure, in other words, the training set was divided into five groups of data, four of which were used for training the NN and the remaining one used as a validation set to compute the mean squared error (MSE).”
From Advanced Diabetes Management Using Artificial Intelligence and Continuous Glucose Monitoring Sensors (2020) · cites this paper as [37]
Cites
(2014) · Cited by 33,419 · see citing works
Classification and Diagnosis of Diabetes
Fajans, S. S. (1990) · Medical Entomology and Zoology · Cited by 2,356 · see citing works
Advances in Neural Information Processing Systems 19 (NIPS 2006)
Simonotto, E., Whalley, H. C., Lawrie, S. M., Murray, L. M., McGonigle, D. J., & Storkey, A. (2006) · Cited by 1,651 · see citing works
2. Classification and Diagnosis of Diabetes
American Diabetes Association (2017) · Diabetes Care · doi:10.2337/dc17-s005 · Cited by 1,603 · see citing works
Nonlinear model predictive control of glucose concentration in subjects with type 1 diabetes
Hovorka, R., Canonico, V., Chassin, L. J., Haueter, U., Massi-Benedetti, M., Federici, M. O., Pieber, T. R., Schaller, H. C., Schaupp, L., Vering, T., et al. (2004) · Physiological Measurement · doi:10.1088/0967-3334/25/4/010 · Cited by 1,274 · see citing works
The UVA/PADOVA Type 1 Diabetes Simulator: New Features
Man, C. D., Micheletto, F., Lv, D., Breton, M., Kovatchev, B., & Cobelli, C. (2014) · Journal of Diabetes Science and Technology · doi:10.1177/1932296813514502 · Cited by 791 · see citing works
<i>In Silico</i> Preclinical Trials: A Proof of Concept in Closed-Loop Control of Type 1 Diabetes
Kovatchev, B. P., Breton, M., Dalla Man, C., & Cobelli, C. (2009) · Journal of Diabetes Science and Technology · doi:10.1177/193229680900300106 · Cited by 760 · see citing works
Diabetes: Models, Signals, and Control
Cobelli, C., Dalla Man, C., Sparacino, G., Magni, L., De Nicolao, G., & Kovatchev, B. P. (2009) · IEEE Reviews in Biomedical Engineering · doi:10.1109/rbme.2009.2036073 · Cited by 511 · see citing works
Self-citationProceedings of the 19th IFAC World Congress
(2014) · IFAC Proceedings Volumes · doi:10.3182/20140824-6-za-1003 · Cited by 426 · see citing works
Simulation Environment to Evaluate Closed-Loop Insulin Delivery Systems in Type 1 Diabetes
Wilinska, M. E., Chassin, L. J., Acerini, C. L., Allen, J. M., Dunger, D. B., & Hovorka, R. (2010) · Journal of Diabetes Science and Technology · doi:10.1177/193229681000400117 · Cited by 244 · see citing works
Wearable Continuous Glucose Monitoring Sensors: A Revolution in Diabetes Treatment
Cappon, G., Acciaroli, G., Vettoretti, M., Facchinetti, A., & Sparacino, G. (2017) · Electronics · doi:10.3390/electronics6030065 · Cited by 242 · see citing works
Self-citationContinuous glucose monitoring sensors: past, present and future algorithmic challenges
Facchinetti, A. (2016) · Sensors · doi:10.3390/s16122093 · Cited by 176 · see citing works
Self-citationSchmidt, S., & Nørgaard, K. (2014) · Journal of Diabetes Science and Technology · doi:10.1177/1932296814532906 · Cited by 170 · see citing works
Clinical accuracy of a continuous glucose monitoring system with an advanced algorithm
Bailey, T. S., Chang, A., & Christiansen, M. (2015) · Journal of Diabetes Science and Technology · doi:10.1177/1932296814559746 · Cited by 168 · see citing works
Kanderian, S. S., Weinzimer, S., Voskanyan, G., & Steil, G. (2009) · Journal of Diabetes Science and Technology · doi:10.1177/193229680900300508 · Cited by 165 · see citing works
Davidson, P. C., Hebblewhite, H. R., Steed, R. D., & Bode, B. W. (2008) · Endocrine Practice · doi:10.4158/ep.14.9.1095 · Cited by 153 · see citing works
Gross, T. M., Kayne, D. M., King, A. B., Rother, C., & Juth, S. E. (2003) · Diabetes Technology & Therapeutics · doi:10.1089/152091503765691848 · Cited by 144 · see citing works
Pettus, J., & Edelman, S. V. (2016) · Journal of Diabetes Science and Technology · doi:10.1177/1932296816663747 · Cited by 120 · see citing works
Circadian Variability of Insulin Sensitivity: Physiological Input for In Silico Artificial Pancreas
Visentin, R., Dalla Man, C., Kudva, Y. C., Basu, A., & Cobelli, C. (2015) · Diabetes Technology & Therapeutics · doi:10.1089/dia.2014.0192 · Cited by 113 · see citing works
Vettoretti, M., Facchinetti, A., Sparacino, G., & Cobelli, C. (2017) · IEEE Transactions on Biomedical Engineering · doi:10.1109/tbme.2017.2746340 · Cited by 110 · see citing works
Self-citationAccuracy and Acceptability of the 6-Day Enlite Continuous Subcutaneous Glucose Sensor
Bailey, T. S., Ahmann, A., Brazg, R., Christiansen, M. P., Garg, S. K., Watkins, E., Welsh, J. B., & Lee, S. W. (2014) · Diabetes Technology & Therapeutics · doi:10.1089/dia.2013.0222 · Cited by 100 · see citing works
Facchinetti, A., Del Favero, S., Sparacino, G., & Cobelli, C. (2015) · Medical & Biological Engineering & Computing · doi:10.1007/s11517-014-1226-y · Cited by 80 · see citing works
Self-citationAre risk indices derived from CGM interchangeable with SMBG-Based indices?
Fabris, C., Patek, S. D., & Breton, M. D. (2016) · Journal of Diabetes Science and Technology · doi:10.1177/1932296815599177 · Cited by 52 · see citing works
Regulation Catches Up to Reality: Nonadjunctive Use of Continuous Glucose Monitoring Data
Edelman, S. V. (2017) · Journal of Diabetes Science and Technology · doi:10.1177/1932296816667749 · Cited by 51 · see citing works
Prediction Methods for Blood Glucose Concentration
Kirchsteiger, H., Jørgensen, J. B., Renard, É., & del Re, L. (2015) · Lecture notes in bioengineering · doi:10.1007/978-3-319-25913-0 · Cited by 48 · see citing works
Acciaroli, G., Vettoretti, M., Facchinetti, A., Sparacino, G., & Cobelli, C. (2017) · IEEE Transactions on Biomedical Engineering · doi:10.1109/tbme.2017.2706974 · Cited by 35 · see citing works
Self-citationThe Current Status of Bolus Calculator Decision-Support Software
Klonof, D. C. (2012) · Journal of Diabetes Science and Technology · doi:10.1177/193229681200600501 · Cited by 32 · see citing works
Acciaroli, G., Vettoretti, M., Facchinetti, A., Sparacino, G., & Cobelli, C. (2016) · Diabetes Technology & Therapeutics · doi:10.1089/dia.2016.0088 · Cited by 22 · see citing works
Self-citationA Model of Self-Monitoring Blood Glucose Measurement Error
Vettoretti, M., Facchinetti, A., Sparacino, G., & Cobelli, C. (2017) · Journal of Diabetes Science and Technology · doi:10.1177/1932296817698498 · Cited by 18 · see citing works
Self-citationPredicting Insulin Treatment Scenarios with the Net Effect Method: Domain of Validity
Vettoretti, M., Facchinetti, A., Sparacino, G., & Cobelli, C. (2016) · Diabetes Technology & Therapeutics · doi:10.1089/dia.2016.0148 · Cited by 12 · see citing works
Self-citation
Flagged first, then by citations
Research works printed but not linked: 5
- Reiterer F, Proceedings of the 9th IFAC Symposium on Biological and Medical Systems (BMS 2015), 2015, 231
- Scheiner G., 2015, Practical CGM: Improving Patient Outcomes Through Continuous Glucose Monitoring
- Marturano F, 2017, 17th Annual Diabetes Technology Meeting (DTM)
- Srivastava N, 2014, J Machine Learning Res, 15, 1929
- Bergstra J, Advances in Neural Information Processing Systems 24 (NIPS 2011), 2011, 2546
Cited by
Sorted by citations
Artificial intelligence for diabetes management and decision support: Literature review
Contreras, I., & Vehi, J. (2018) · Journal of Medical Internet Research · doi:10.2196/10775 · Cited by 560 · see citing works
Cappon, G., Vettoretti, M., Sparacino, G., & Facchinetti, A. (2019) · Diabetes & Metabolism Journal · doi:10.4093/dmj.2019.0121 · Cited by 477 · see citing works
Advanced Diabetes Management Using Artificial Intelligence and Continuous Glucose Monitoring Sensors
Vettoretti, M., Cappon, G., Facchinetti, A., & Sparacino, G. (2020) · Sensors · doi:10.3390/s20143870 · Cited by 176 · see citing works
Self-citationContinuous glucose monitoring: Current use in diabetes management and possible future applications
Vettoretti, M., Cappon, G., Acciaroli, G., Facchinetti, A., & Sparacino, G. (2018) · Journal of Diabetes Science and Technology · doi:10.1177/1932296818774078 · Cited by 96 · see citing works
Machine-Learning Based Model to Improve Insulin Bolus Calculation in Type 1 Diabetes Therapy
Noaro, G., Cappon, G., Vettoretti, M., Sparacino, G., Favero, S. D., & Facchinetti, A. (2021) · IEEE Transactions on Biomedical Engineering · doi:10.1109/tbme.2020.3004031 · Cited by 57 · see citing works
Self-citationCappon, G., Marturano, F., Vettoretti, M., Facchinetti, A., & Sparacino, G. (2019) · Journal of Diabetes Science and Technology · doi:10.1177/1932296818777524 · Cited by 18 · see citing works
Self-citation
- Authors
- Cappon, G., Vettoretti, M., Marturano, F., Facchinetti, A., & Sparacino, G.
- Affiliations
- University of Padua (IT)
- Corresponding author
- Sparacino, Giovanni
- Venue
- Journal of Diabetes Science and Technology
- Publisher
- SAGE Publications
- Published
- 2018-03
- Volume
- 12
- Issue
- 2
- Pages
- 265-272
- ISSN
- 1932-2968 (ISSN-L)
- PMC
- PMC5851237
- OpenAlex
- W2792894037
- Semantic Scholar
- 08c5b43f3e61ca1e907edb402366fef19d9963b5
- Cited by
- 62
- References
- 30
- Citations by year
- 2026: 2 · 2025: 4 · 2024: 4 · 2023: 10 · 2022: 7 · 2021: 7 · 2020: 12 · 2019: 10 · 2018: 6
- FWCI
- 6.96
- Type
- journal article
- Language
- en
- Open access
- green
- europepmc.org
- Locations
- Journal of Diabetes Science and Technology (publishedVersion)
- Europe PMC (PubMed Central) (publishedVersion)
- PubMed Central (publishedVersion)
- PubMed (publishedVersion)
- Padua Research Archive (University of Padova) (submittedVersion)
- Topics
- Health Sciences › Medicine › Endocrinology, Diabetes and Metabolism › Diabetes Management and Research
- Health Sciences › Medicine › Endocrinology, Diabetes and Metabolism › Diabetes Management and Education
- Health Sciences › Medicine › Endocrinology, Diabetes and Metabolism › Diabetes Treatment and Management
- Keywords
- type 1 diabetes, continuous glucose monitoring, artificial neural networks, standard formula, insulin sensitivity, body weight
- MeSH terms
- Blood Glucose, Diabetes Mellitus, Type 1, Humans, Hypoglycemic Agents, Insulin, Blood Glucose Self-Monitoring, Neural Networks, Computer, Datasets as Topic
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
