|SYMPOSIUM: DATA ANALYSIS AND TREND IDENTIFICATION IN THE INTENSIVE CARE UNIT
|Year : 2017 | Volume
| Issue : 3 | Page : 142-146
Republication: Introducing the glucogram – Description of a novel technique to quantify clinical significance of acute hyperglycemic events
Stanislaw P Stawicki1, Dara P Schuster2, J Felix Liu3, Jyoti Kamal3, B Selnur Erdal3, Anthony T Gerlach4, Melissa L Whitmill1, David E Lindsey1, Yalaunda M Thomas1, Claire V Murphy4, Steven M Steinberg1, Charles H Cook1
1 Department of Surgery, The Ohio State University Medical Center, Division of Critical Care, Trauma, and Burn, Columbus, OH, USA
2 The Ohio State University Medical Center, Division of Endocrinology, Diabetes and Metabolism, Columbus, OH, USA
3 Department of Biomedical Informatics, The Ohio State University Medical Center, Columbus, OH, USA
4 The Ohio State University Medical Center, Columbus, OH, USA
|Date of Web Publication||21-Apr-2017|
Stanislaw P Stawicki
Department of Research and Innovation, St. Luke's University Health Network, EW2 Research Administration, 801 Ostrum Street, Bethlehem, PA 18015
Source of Support: None, Conflict of Interest: None
The importance of hyperglycemia in the Intensive Care Unit (ICU) is well established. However, little is known regarding the clinical predictive value of acute hyperglycemic events. This report describes a graphical model that quantifies the correlation between momentum/stochastic indicators, acute hyperglycemia, and clinical events in chronic ICU patients. The model is based on previously described principles of graphical representations of biomedical parameter data. We hypothesize that acute hyperglycemic events are significantly associated with major clinical events and that the model described herein helps to better characterize and quantify this important relationship.
The following core competencies are addressed in this article: Medical knowledge, Patient care, Practice-based learning and improvement, Systems-based practice.
Republished with permission from: Stawicki SP, Schuster D, Liu JF, Kamal J, Erdal S, Gerlach AT, Whitmill ML, Lindsey DE, Thomas YM, Murphy C, Steinberg SM. Introducing the glucogram: Description of a novel technique to quantify clinical significance of acute hyperglycemic events. OPUS 12 Scientist. 2009;3:2-5.
Keywords: Acute hyperglycemic events, glucogram, glycemic control, graphical representation, momentum indicator, predictive modeling, stochastic indicator, Surgical Intensive Care Unit
|How to cite this article:|
Stawicki SP, Schuster DP, Liu J F, Kamal J, Erdal B S, Gerlach AT, Whitmill ML, Lindsey DE, Thomas YM, Murphy CV, Steinberg SM, Cook CH. Republication: Introducing the glucogram – Description of a novel technique to quantify clinical significance of acute hyperglycemic events. Int J Acad Med 2017;3, Suppl S1:142-6
|How to cite this URL:|
Stawicki SP, Schuster DP, Liu J F, Kamal J, Erdal B S, Gerlach AT, Whitmill ML, Lindsey DE, Thomas YM, Murphy CV, Steinberg SM, Cook CH. Republication: Introducing the glucogram – Description of a novel technique to quantify clinical significance of acute hyperglycemic events. Int J Acad Med [serial online] 2017 [cited 2020 Oct 30];3, Suppl S1:142-6. Available from: https://www.ijam-web.org/text.asp?2017/3/3/142/204951
| Introduction|| |
The importance of hyperglycemia in the Intensive Care Unit (ICU) is well established., However, little is known regarding the clinical predictive value of acute hyperglycemic events. This manuscript describes a graphical model that quantifies the correlation between momentum/stochastic indicators (StIRs), acute hyperglycemia, and clinical events in chronic ICU patients. The model is based on previously described principles of graphical representations of biomedical parameter data.,, We hypothesize that acute hyperglycemic events are significantly associated with major clinical events and that the model described herein helps to better characterize and quantify this important relationship.
| Methods|| |
This report describes the use of a high-resolution computerized graphical representation of glucose levels, along with associated momentum and StIRs, to help quantify the relationship between periods of acute hyperglycemia and associated clinical events in the ICU.
Detailed medical record review of a critically ill patient who remained in the ICU for more than 30 days was performed. Glucose measurements were grouped into 12 h epochs, with initial, maximal, minimal, and final values incorporated for each time period. Sequential bar graphs [Figure 1] were plotted, along with momentum and StIRs [Figure 2], glucogram]. Correlations between clinical events and positive indicator status (i.e., indicator “spike”) were then made. Indicator designation status is described in detail below. Clinical events were divided into major [Table 1] and minor [Table 2]. Indicator spike correlating with a major event (s) constituted a positive pair. Lack of indicator spike corresponding to minor (control) event was considered a negative pair.
|Figure 1: Schematic representation of the open-high-low-close bar graph format used in this report. Note that the difference between the high and the low value for each epoch, when averaged over a certain time period, could provide a measure of overall glycemic variability for that time period|
Click here to view
|Figure 2: An example of glucogram. Red squares represent major clinical events correctly matched with indicator spikes. Green circles represent minor clinical events that were correctly correlated with lack of indicator spike. In terms of mismatches, there were three instances of a major event not correlating with an indicator spike (empty black squares) and two gray circles representing positive indicator spike incorrectly correlating with minor events. The uppermost window shows the momentum indicator. In this case, valid momentum indicator spikes were considered to have values of 200% or greater (shaded in yellow). The middle window shows the stochastic indicator. Values of 60 or above were considered to represent positive indicator spikes (shaded in yellow). The bottom window shows the glucose levels represented in the open-high-low-close format, the mean value (moving average), and the upper/lower envelopes of glucose ranges|
Click here to view
The designation of clinical event as major or minor (control) was based on the nature of the clinical event as well as the overall clinical picture associated with that clinical event, as determined by the physician who reviewed the clinical chart (SPS). Each event subgroup (major versus minor) was then broken down into specific subclasses of clinical events. Details regarding the grouping and classification of clinical events utilized in the current study can be seen in [Table 1] and [Table 2].
| Data Organization|| |
Clinical variables were obtained retrospectively from the case patient's glucose dataset. Glucose measurements were recorded over a range of time frequencies of every 60 min to every 8 h for the patient's entire ICU stay. Clinical information was then transformed into the open-high-low-close (OHLC) format [Figure 1]. To properly apply and utilize the OHLC, the data had to be arranged into equal time intervals or epochs. For the purposes of this study, data were arranged into 12 h epochs. For each epoch, the opening value (the first value in the epoch), the high and low values, as well as the closing value (the last value in the epoch) were recorded and graphed.
| The Momentum Indicator|| |
The momentum indicator (MoIR) is a leading indicator (i.e., forward-looking indicator) that measures the rate of change in a particular parameter and is used to detect rapid changes in trends as well as likely trend reversal points. High MoIR readings tend to occur when a trend is at its strongest. Lower MoIR readings are generally found at the start or end of a trend or between trends. Consequently, the traditional interpretation of the MoIR is based on the assumption that when the indicator begins to rise following a trough, it is considered to be the beginning of an ascending trend. Conversely, when the MoIR peaks and begins to descend, it is considered to be the beginning of a descending trend.
The value of the MoIR is determined by taking the current value of the parameter being measured and dividing it by values of the same parameter from a predetermined number (i.e., x) of previous epochs. The resulting value is then multiplied by 100 to convert the output into percentage terms. For example, if the glucose value is 150 during the current time period and was 75 during the previous epoch, then the current MoIR value will be 200% (i.e., 150/75 × 100%).
Of note, peak and trough levels are set individually for each patient and depend on the baseline behavior of the MoIR indicator over the past cycles for that particular patient. In other words, one patient's peak MoIR value of 150 may not be “equal” to another patient's peak MoIR of 250 and depends on the relative MoIR values. Thus, it is based on overall volatility level for that patient. The value of MoIR is usually expressed as a percentage, with values that move above and below 100%.
| The Stochastic Indicator|| |
The StIR is a technical indicator that compares the closing value of a parameter with a range of the same parameter's values for a predetermined time period.,,, It was previously noted that when a parameter's value is rising, it tends to have closing values near the high value for that time period and a falling parameter closes near its low.,,,
The StIR is displayed as two lines. The first (or main) line is called %K and is calculated using the high, low, and closing values from each respective epoch. The secondary line (%D) is a moving average (Mav) of %K. The formula for %K is %K = 100 × [(C − LXclose)/(HX − LX)], where C is the most recent closing value, LX is the lowest low for the last x time periods, and HX is highest high for the same X time periods. %D is a smoothed representation of the %K line. Most often, three time periods (X = 3) are used. The %D formula is %D = 100 × (HY/LY), where HY = the Y-period sum of C − LX and LY is the y-period sum of HX − LX.
The StIR is plotted on a chart with values ranging from 0 to 100 for a specified time period. Much like with Mavs, the sensitivity of StIR increases with shorter time spans. In general, readings above 80 are considered “strong” and indicate that the trend is nearing its highs. Readings below 20 are classified as “weak” and indicate that the trend is nearing its lows.
| Parameter Bands|| |
Parameter bands are lines plotted around the parameter's bar graph structure to form an envelope and are characterized by their fixed percentage (%F) amounts above and below the mean value of the variable [Figure 2]. Of special interest is the behavior of the variable under investigation near the edges of the envelope, which may be helpful in identifying trend reversals.
The mathematical procedure of envelope creation is simple. It consists of plotting the Mav for the particular security, followed by calculation of the upper and lower bands (LBs)., The upper band (UB) is calculated by adding a %F of value to the Mav (UB = Mav + Mav × %F). The LB is calculated similarly by subtracting a %F of value from the Mav (LB = Mav − Mav × %F). The Mav, UB, and LB are then plotted. The number of epochs over which the Mav is averaged, as well as the %F value, are subject to the preference of the person analyzing the data.
| The Glucogram|| |
The graph below [Figure 2] represents an example of glucogram. It is the representation of glucoses for a patient with severe pancreatitis (APACHE II score of 19). The patient had a total of 27 major clinical events. Of those, 24 were associated with a positive indicator spike(s) and three did not have an associated indicator spike. In addition, there were two instances where an indicator spike did not coincide with any clinical event and five instances of minor events correctly correlating with the absence of indicator spike(s). Considering these findings, the sensitivity for major clinical events was 88.9% and specificity was 71.4%.
| Discussion|| |
The use of various indicators as described in this article provides a novel way of describing and interpreting biomedical (vital sign, input/output, pressure-based, and laboratory) data.,, Consistent with previously published experience, when a trend was present, indicators tended to demonstrate it well. When no trend was present, indicators tended to “wander around” until the next trend was identified.,,
There are several potential advantages to using financial analysis tools in biomedical applications. First, financial indicators are well described, have been used for quite some time in nonbiomedical applications, and are relatively easy to learn, understand, and interpret. Second, these indicators provide a method of objectively evaluating clinical parameters and scenarios for which traditionally subjective interpretations were used (i.e., quantifying the clinical importance of clinically observed trends such as acute hyperglycemic spikes). Third, financial indicators provide a way of standardizing clinical trend interpretation and potentially minimizing interobserver variability in interpreting such trends. In fact, by allowing greater interobserver agreement, the latter advantage may ultimately contribute to the clinical success of these techniques.
At the present time, while it appears to be a very promising research tool for analysis of biomedical data trends, the authors do not recommend using the glucogram as the sole predictor of significant clinical events. Instead, one should always take the overall patient clinical picture into consideration before making any diagnostic or therapeutic determinations. As more data emerge regarding the clinical utility of the glucogram, its clinical applications are likely to grow in scope.
While graphs and trends within individual variables (i.e., glucose, white blood cell count, heart rate, or hemoglobin) may be valuable from the standpoint of clarity of the graphical representation of each individual variable, the most useful clinical information will most likely come from indexing of multiple variables., In such arrangement, multiple biomedical variables are used to create a composite index. Based on certain assumptions, this composite index would then be used to estimate the physiologic economy of the patient., Fundamental principles of the proposed model have been previously described in detail and are beyond the scope of this discussion.,
Directions for future research in this area include (a) multiparametric studies; (b) incorporating parametric corrections for therapeutic interventions – i.e., accounting for insulin dosage; (c) incorporating other types of indicators into the model [Table 3]; (d) prospective evaluation of the most successful retrospective parameter model combination(s); and (e) design of devices and/or software packages that allow real-time analysis of various biomedical parameters.
|Table 3: Examples of various indicator candidates for further biomedical parameter modeling design and analysis|
Click here to view
| Conclusions|| |
Minimizing the subjective component of patient data interpretation and maximizing the objective component may provide a better way of assessing patients and when correlated with clinical data, may have useful adjunctive confirmatory or possibly even predictive value. Momentum and/or StIR spikes on glucograms appear to have reasonable sensitivity for major clinical events. Specificity appears to be somewhat lower. Future research in this area is certainly warranted, with a focus on absolute deviation from normal values, prospective data collection, device design, time to achieve normoglycemia, and multiparameter modeling.
Justifications for re-publishing this scholarly content include: (a) The phasing out of the original publication after a formal merger of OPUS 12 Scientist with the International Journal of Academic Medicine and (b) Wider dissemination of the research outcome(s) and the associated scientific knowledge.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Finney SJ, Zekveld C, Elia A, Evans TW. Glucose control and mortality in critically ill patients. JAMA 2003;290:2041-7.
Lacherade JC, Jabre P, Bastuji-Garin S, Grimaldi D, Fangio P, Théron V, et al.
Failure to achieve glycemic control despite intensive insulin therapy in a medical ICU: Incidence and influence on ICU mortality. Intensive Care Med 2007;33:814-21.
Stawicki SP. Application of financial analysis techniques to vital sign data: A novel method of trend interpretation in the Intensive Care Unit. OPUS 12 Sci 2007;1:14-6.
Stawicki SP. Application of financial analysis techniques to clinical laboratory data: A novel method of trend interpretation in the Intensive Care Unit. OPUS 12 Sci 2007;1:1-4.
Stawicki SP. Financial analysis techniques in clinical practice: From 'micro' to 'macro'. OPUS 12 Sci 2008;2:3-9.
Lane GC. Lane's Stochastics. Technical Analysis of Stocks and Commodities Magazine. 1984;2:87-90.
Bollinger JA. Bollinger on Bollinger Bands. New York: McGraw-Hill Trade; 2001.
Stawicki SP, Prosciak MP. The pulmonary artery catheter in 2008 – A (finally) maturing modality? OPUS 12 Sci 2008;2:5-9.
[Figure 1], [Figure 2]
[Table 1], [Table 2], [Table 3]