The Glucontrol trial [26] randomised patients into two groups: Group A and Group B. Group A received intensive insulin therapy and Group B received conventional insulin therapy, with target ranges of 4.4-6.1 mmol/L and 7.8-10 mmol/L, respectively. Insulin was administered as a continuous intravenous (IV) infusion. Hourly blood glucose (BG) measurements were recorded when the glycemic level was not within the target range. Otherwise, 2-hourly measurements were taken in the case of limited variation of glycemia, defined as less than a 50% change from the previous glycaemia in 2-hour range. Finally, 4-hourly measurements were taken when the glycemic level was less than 50% of the highest glycemia of the four last hours. If other BG measurements were taken, they were not recorded and did not result in changes to the insulin infusion rate.
[Some details are omitted]
In this study, data was used from 350 patients (175 in each arm) treated using the Glucontrol protocol at CHU de Liege, Belgium, between March 2004 and April 2005. Thus, the Glucontrol data used in this study is from only one centre out of the full study [26]. The selection criteria for patients used in this analysis to generate virtual patients with sufficient data density [15,16,27] are shown in Figure 1. Patients were eliminated from the analysis if they received no insulin for their entire stay (per protocol), had less than 5 BG measurements or received little or no (recorded) carbohydrate administration (in any form) for more than 48 hours of their stay.
[Some details are omitted]
Patients in Group A were slightly older than Group B. However, there were no significant differences in sex, weight, BMI, severity of illness as measured by APACHE II score or initial BG level. Group B received less insulin and more carbohydrate, in alignment with its higher glycemic target.
[Some details are omitted]
Validation Analysis
This study performs three major forms of validation using virtual trials. These three tests provide both per-patient and cohort-wide validation of this in silico approach.
[Some details are omitted]
Results
[Some details are omitted]
Discussion
[Some details are omitted]
The virtual trials approach here treats each group as being treated differently, including the carbohydrate and glucose infusions administered. These infusions were patient-specific and specified based on local and individual clinician standards, rather than per a protocol of any type. Thus, they were kept for each patient. As a result, Glucontrol B patients with the higher target had 2.6× higher glucose administration, which in cross validation was offset by 3.2× more insulin in the virtual trials. Differences in insulin rates between per protocol (as the cross validation was done) and per actual measurement rates makes these differences almost equal at 2.6× higher glucose administration and 2.4× greater insulin required to achieve the almost identical glycemic outcomes. Hence, the patients display similar overall insulin sensitivity, and the virtual trials took independently treated, matched patients and achieved the same outcome despite different initial treatments in the clinical data used to create the virtual patient. More specifically, nutritional treatment differences, within reason, did not affect or influence the results outside of expectations.
More importantly, the relatively small differences show the strength of model-fitted insulin sensitivity as a description of patient metabolic state, rather than as a therapy-specific parameter value. Other causes for remaining differences may also be a function of remaining model approximations or errors. As noted, inter-patient variability in some fixed model parameters is at least one cause of model limitations and errors. However, the limited glucose data with no added or real time insulin data limits the ability to uniquely identify these parameters [27,32].
Finally, this paper shows the potential for TGC protocols to be readily optimised and implemented using model based TGC. The low prediction errors indicate an ability to minimize the risk of hypoglycaemia as well as provide tight control. Even though some TGC clinical trials have not achieved any benefit from TGC [12,38], only 2 protocols have been first optimized with virtual trials [11,17,21]. Both delivered safe, effective TGC with reduced or zero hypoglycaemia.
Conclusions
This paper presented the analysis and validation of an in silico virtual patient and model-based virtual trials methodology. The validation approach, as presented, is readily generalized. It takes advantage of a set of independent clinical data comprised of two clinically matched cohorts treated with two different TGC protocols with two different glycemic targets. Three main conclusions can be drawn:
• Self validation indicated a clinically insignificant error in these virtual patient methods due to model and/or clinical compliance. They also showed the impact of some non-compliance independent of model error.
• Cross validation clearly showed that the virtual patient methods and models enabled by patient-specific SI(t) profiles are effective and the assumption that the SI(t) profiles are independent of the clinical inputs used to generate them holds.
• Thus, the virtual patients and in silico virtual trial methods presented are validated in their ability to accurately simulate, in advance, the clinical results of an independent TGC protocol, directly enabling rapid design and optimisation of safe and effective TGC protocols with high confidence of clinical success.
Overall, this study further shows the potential and capability of model-based, data driven in silico methods to aid protocol design, as well as the potential for models to provide accurate, safe and effective real-time TGC.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
JGC, FS, GS, ALC and JL conceived and developed the models and this analysis. FS, SP and ALC did most of the computational analysis with input from JGC, CGP, TD and KTM. J-CP supplied the data and Glucontrol protocol information. JGC, FS, ALC drafted the manuscript primarily although all authors made contributions. All authors approved the final manuscript.
Acknowledgements
Financial support provided by:
Aaron LE COMPTE: New Zealand Tertiary Education Commission and NZ Foundation for Research Science and Technology Post-Doctoral Fellowship Grant
Jessica LIN: NZ Foundation for Research Science and Technology Post-Doctoral Fellowship Grant
Sophie PENNING: FNRS (Fonds Nationale de la Recherche Scientifique) Research Fellow
Katherine MOORHEAD: University of Liege Post-Doctoral Fellowship Grant
References
1. Capes SE, Hunt D, Malmberg K, Gerstein HC: Stress hyperglycaemia and increased risk of death after myocardial infarction in patients with and without diabetes: a systematic overview. Lancet 2000, 355:773-778.
2. Krinsley JS: Association between hyperglycemia and increased hospital mortality in a heterogeneous population of critically ill patients. Mayo Clin Proc 2003, 78:1471-1478.
3. McCowen KC, Malhotra A, Bistrian BR: Stress-induced hyperglycemia. Crit Care Clin 2001, 17:107-124.
4. Mizock BA: Alterations in fuel metabolism in critical illness: hyperglycaemia. Best Pract Res Clin Endocrinol Metab 2001, 15:533-551.
5. Krinsley JS: Glycemic variability: a strong independent predictor of mortality in critically ill patients. Crit Care Med 2008, 36:3008-3013.
6. Ali NA, O'Brien JM Jr, Dungan K, Phillips G, Marsh CB, Lemeshow S, Connors AF Jr, Preiser JC: Glucose variability and mortality in patients with sepsis. Crit Care Med 2008, 36:2316-2321.
7. Egi M, Bellomo R, Stachowski E, French CJ, Hart G: Variability of blood glucose concentration and short-term mortality in critically ill patients. Anesthesiology 2006, 105:244-252.
8. Van den Berghe G, Wilmer A, Hermans G, Meersseman W, Wouters PJ, Milants I, Van Wijngaerden E, Bobbaers H, Bouillon R: Intensive Insulin Therapy in the Medical ICU. N Engl J Med 2006, 354:449-461.
9. Van den Berghe G, Wouters P, Weekers F, Verwaest C, Bruyninckx F, Schetz M, Vlasselaers D, Ferdinande P, Lauwers P, Bouillon R: Intensive insulin therapy in the critically ill patients. N Engl J Med 2001, 345:1359-1367.
10. Krinsley JS: Effect of an intensive glucose management protocol on the mortality of critically ill adult patients. Mayo Clin Proc 2004, 79:992-1000.
[References 11 – 38 are omitted]
Abstract of this article
Background
In-silico virtual patients and trials offer significant advantages in cost, time and safety for designing effective tight glycemic control (TGC) protocols. However, no such method has fully validated the independence of virtual patients (or resulting clinical trial predictions) from the data used to create them. This study uses matched cohorts from a TGC clinical trial to validate virtual patients and in-silico virtual trial models and methods.
Methods
Data from a 211 patient subset of the Glucontrol trial in Liege, Belgium. Glucontrol-A (N = 142) targeted 4.4-6.1 mmol/L and Glucontrol-B (N = 69) targeted 7.8-10.0 mmol/L. Cohorts were matched by APACHE II score, initial BG, age, weight, BMI and sex (p > 0.25). Virtual patients are created by fitting a clinically validated model to clinical data, yielding time varying insulin sensitivity profiles (SI(t)) that drives in-silico patients.
Model fit and intra-patient (forward) prediction errors are used to validate individual in-silico virtual patients. Self-validation (tests A protocol on Group-A virtual patients; and B protocol on B virtual patients) and cross-validation (tests A protocol on Group-B virtual patients; and B protocol on A virtual patients) are used in comparison to clinical data to assess ability to predict clinical trial results.
Results
Model fit errors were small (<0.25%) for all patients, indicating model fitness. Median forward prediction errors were: 4.3, 2.8 and 3.5% for Group-A, Group-B and Overall (A+B), indicating individual virtual patients were accurate representations of real patients. SI and its variability were similar between cohorts indicating they were metabolically similar.
Self and cross validation results were within 1-10% of the clinical data for both Group-A and Group-B. Self-validation indicated clinically insignificant errors due to model and/or clinical compliance. Cross-validation clearly showed that virtual patients enabled by identified patient-specific SI(t) profiles can accurately predict the performance of independent and different TGC protocols.
Conclusions
This study fully validates these virtual patients and in silico virtual trial methods, and clearly shows they can accurately simulate, in advance, the clinical results of a TGC protocol, enabling rapid in silico protocol design and optimization. These outcomes provide the first rigorous validation of a virtual in-silico patient and virtual trials methodology.
International Journal of Robotics Research, 175(6):482-487, 1998.
