FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W1995442846> ?p ?o ?g. }

• W1995442846 endingPage "710" @default.
• W1995442846 startingPage "707" @default.
• W1995442846 abstract "The radar operator sits hunched over the luminous screen in the darkened radar room. The pale green dot was at first so dim it was barely visible, but with each sweep of the display it grew brighter. Is it simply a bird or an enemy aircraft? Anti-aircraft defenses will reveal the position of the ship. Waiting only reduces the chances of a successful defense. A decision has to be made. In this issue of Anesthesia & Analgesia, Funk et al.1 review the minimally invasive monitoring tools available to assess cardiac output (CO). In their introduction, the authors state: “The value of patient monitors to real-time decision making during patient care is especially important when caring for critically ill patients.” Just like the radar operator, anesthesia providers continually observe monitor displays and use the information to make decisions that have an impact on morbidity and mortality. Fortunately, for the radar operators, the value of radar systems to support decision making is well documented. In contrast, the value of medical devices to guide clinical care decisions is not well documented. Despite the lack of an obvious relationship between a physician and a radar operator, the methodology called receiver operating characteristic (ROC) analysis, which was developed to document radar performance, can be applied to studies of CO monitoring. Before discussing ROC analysis, it is useful to address the limitations of the current approaches to evaluating CO monitors. The introduction of the pulmonary artery catheter established thermodilution CO measurement as a widely used patient monitor for critically ill patients, although more recent studies have questioned the benefit of pulmonary artery catheterization in light of its cost and associated risks.2 Nevertheless, much effort has been expended to develop devices for measuring CO and to compare them with the reference standard of thermodilution. If CO measurement is so desirable when caring for critically ill patients, why is it that tools to measure this variable have not been more widely accepted into clinical care? There is certainly no shortage of studies evaluating these technologies as evidenced by the references in the review by Funk et al. Part of the answer lies in obstacles to acceptance, such as cost or familiarity. Beyond that, there is an understandable reticence to rely upon a new device to guide decisions in critically ill patients. When evaluating the role of new CO devices in clinical care, the fundamental question is whether the new device can replace thermodilution CO measurement as a guide to clinical decisions. Despite the large number of studies evaluating new CO devices, few, if any, answer this fundamental question. The typical study concurrently measures CO with the new device and with thermodilution in a study population. These concurrent measurements are then compared statistically. The study population selected is often patients or volunteers who are not critically ill, or do not have significant hemodynamic derangement. Studying these patients may help to gain familiarity with a device but does not help us to understand how the device might be used to guide decisions when caring for critically ill patients. Even if an appropriate patient population is selected, the statistical tools typically used in these studies are inadequate to support conclusions about the utility of the device to guide decision making. Two common statistical approaches used to compare methods for measuring CO are linear regression and correlation. These are inadequate analytical tools for assessing agreement between two methods. As pointed out by Altman and Bland,3 linear regression and correlation do not provide any insight into the agreement between the measurements or the ability to substitute one measurement for another (unless the paired data happen to lie on the line of identity which is unlikely). Correlation and linear regression only indicate the degree to which the measures being compared change in concert with one another, and says nothing about whether they are interchangeable. Bland and Altman advocated describing the agreement between two measures by calculating the mean and standard deviation of their differences over the range of values studied. The standard deviation of the differences is a measure of the limits of agreement. If the differences are normally distributed, then 95% of the differences will fall within (±) 2 standard deviations of the mean difference. The differences can be viewed graphically by plotting the difference between the measurements against the average of the measurements (Fig. 1). Lines drawn for the mean ± 2 standard deviations of the differences indicate the limits of agreement for the population studied.Figure 1.: Bland Altman plot comparing the differences between cardiac output measurements for two devices. The x axis is the average of the two measures.When comparing a new method of CO measurement to thermodilution, the x axis on the Bland Altman plot is the average of the two measurements because thermodilution is not a calibrated measurement standard, and the true CO is not known with any measurable certainty. In method comparison studies, if the reference method has a known accuracy and precision (e.g., aortic flowprobe), then the x axis can be the reference measurements themselves, rather than the average of the two measurements. In this case, the Bland Altman approach provides insight into the accuracy of the test technique. However, very few clinical measurements are sufficiently accurate to qualify as a true reference value, in which case the x axis is the average of the two measurements and one cannot make any conclusions about accuracy. One can only gain insight into the bias and limits of agreement between the two methods being compared. Critchley and Critchley4 rationalized how the limits of agreement derived according to Bland and Altman can be used to compare test and reference methods given the inherent inaccuracy of the reference methods. The example described in their article assumed an inherent accuracy for thermodilution CO of ±20% and a similar accuracy of the test method, yielding a combined limit of agreement of ±28%. Selecting a threshold of 28%–30% as the clinically acceptable limit of agreement is a simplification that makes assumptions about the accuracy of thermodilution and does not consider the impact on decision making. If the true CO is 3 L/min, a measured value of 30% less is 2.1 L/min, which could lead to a very different management decision than if the measured value is 30% larger, or 3.9 L/min. Furthermore, the example in their article assumed an inherent accuracy of ±20% for thermodilution and the test method, whereas the error-gram in the same article demonstrated that the combined percentage error can be more than 30% (Fig. 2). Critchley and Critchley are to be congratulated for exposing the limitation of the Bland Altman analysis for determining limits of agreement, but their results cannot be used to define the “acceptable” limits of agreement.Figure 2.: Error-gram indicating the combined percentage error (right-hand axis) of the limits of agreement given the inherent accuracy of both the test and the reference methods of cardiac output measurement. Isolines indicate the percentage error of the reference method. Reproduced with permission from Critchely LAH, Critchley JAJH. A meta-analysis of studies using bias and precision statistics. Journal of Clinical Monitoring and Computing. 1999:15, 85.Altman and Bland5 correctly observe: “How far apart measurements can be without causing difficulties is a question of judgment.” Judgment is critical. Bland Altman analysis is a descriptive statistical methodology. Calculation of bias and limits of agreement simply describes the data in summary fashion. There is no relationship to the clinical context. To understand the implications of the measurements for clinical decision making, one needs to compare the decisions that would result from using the new method compared with the reference method. The “judgment” about whether the observed limits of agreement are clinically acceptable is a matter of opinion. The opinion of the physician faced with a dying patient may be quite different than the opinion of an enthusiastic investigator or inventor. There are other problems inherent in using the Bland Altman approach to determine whether a new measurement technique can replace an existing technology. Bias and limits of agreement are typically calculated for the entire range of measurements obtained in a study population. CO measurement is, however, most likely to influence patient management at the extremes of CO and, in particular, at low COs. Measurements obtained when comparing methods are obtained over a range of COs and both bias and limits of agreement will vary depending upon the range of values for which they are calculated (Table 1). Limits of agreement that are meaningful for decision making relate to CO values most likely to require a diagnostic or therapeutic intervention, i.e., the extremes of CO.Table 1: Bias and Limits of Agreement (±2 sd) Calculated for Data Presented in Figure 1The Bland and Altman approach to comparing two methods of measurement has become the generally accepted approach to comparing CO measurement devices. This approach allows us to describe the observed limits of agreement, but whether or not these limits of agreement are acceptable for clinical care becomes a matter of opinion. Are there analytical tools available that can more rigorously answer the question about the utility of a given tool to guide decision making? ROC ANALYSIS The answer to this question lies in the early days of radar development when statistical tools were used that have subsequently been used to guide studies on medical decision making. Radar engineers needed to evaluate the sensitivity and specificity of their systems to detect enemy aircraft. A system that was too sensitive would detect too many objects (false positive) and waste munitions shooting birds out of the sky. A system that was too specific (false negatives) would fail to detect an attacking aircraft, with obvious consequences. The ROC* curve was developed as a method for evaluating sensitivity and specificity and determining the quality of the tool and the optimal decision point (Fig. 3). By plotting the true-positive rate (sensitivity) against the false-positive rate (1- specificity) for a test at different decision points, the quality of the test can be determined. Useful tests generate plots in the upper left hand quadrant of the ROC curve and become reliable tests for making decisions.6 More importantly, the actual performance of the test for making a decision is disclosed. Wikipedia has an excellent review of the ROC.†Figure 3.: Sample receiver operating characteristic (ROC) plot comparing different clinical tools for predicting a patient’s response to fluid administration. Plots located in the upper left-hand quadrant of the graph have better sensitivity and specificity. ΔPP = pulse pressure variation; CVP = central venous pressure; PCWP = pulmonary capillary wedge pressure; CI = cardiac index. Used with permission from Cannesson et al.7To develop an ROC curve, one must define the threshold for making a decision and then calculate from the observed measurements the true-positive rate and false-positive rate. By plotting the true and false-positive rates at different decision thresholds using the same measurement tool, a curve is generated. A typical use of ROC analysis would be to evaluate a new diagnostic test and answer two questions: 1) Is the test any good? and 2) What is the right threshold value for decision making? ROC analysis can also be used to compare the quality of different methods for the same test. This approach has been used recently to compare the ability of different monitoring methods to predict fluid responsiveness.7 ROC analysis has the potential to compare the impact on clinical decision making of different methods of measuring CO. If the CO threshold at which a clinical decision would change is defined, it becomes possible to determine the sensitivity and specificity of the measurements from each device. ROC analysis can then be used to compare the utility of each device to guide decision making based upon that threshold. By comparing the ROC plots for each device, the impact of each device on clinical decision making can be understood. Although the technologies reviewed by Funk et al. are used to varying degrees in clinical practice, none have become widely adopted despite the stated importance of CO monitoring in the care of critically ill patients. In the concluding paragraph of their review, the authors state that “With an increasing number of clinical studies being published … their use should continue to gain popularity.” Although the Bland Altman approach is a useful tool to describe the limits of agreement between two methods of measurement, conclusions about the acceptability of the limits of agreement are a matter of opinion, not science. Analytical methods for comparing CO measurement techniques need to move beyond this approach to provide insight into the role of the technology in clinical decision making. Patient populations in which CO monitoring is important need to be studied within the range of values that would dictate the need for clinical intervention. ROC analysis can be used to gain insight into the impact of a tool on medical decision making. The impact of those decisions on patient outcome are the ultimate question and will require even more sophisticated protocols in larger patient populations." @default.
• W1995442846 created "2016-06-24" @default.
• W1995442846 creator A5049504137 @default.
• W1995442846 date "2009-03-01" @default.
• W1995442846 modified "2023-10-16" @default.
• W1995442846 title "Is It a Bird? Is It a Plane? The Role of Patient Monitors in Medical Decision Making" @default.
• W1995442846 cites W1586655573 @default.
• W1995442846 cites W2006101344 @default.
• W1995442846 cites W2133750711 @default.
• W1995442846 cites W2147246192 @default.
• W1995442846 cites W2171467831 @default.
• W1995442846 doi "https://doi.org/10.1213/ane.0b013e318196c7b6" @default.
• W1995442846 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/19224773" @default.
• W1995442846 hasPublicationYear "2009" @default.
• W1995442846 type Work @default.
• W1995442846 sameAs 1995442846 @default.
• W1995442846 citedByCount "38" @default.
• W1995442846 countsByYear W19954428462012 @default.
• W1995442846 countsByYear W19954428462013 @default.
• W1995442846 countsByYear W19954428462014 @default.
• W1995442846 countsByYear W19954428462015 @default.
• W1995442846 countsByYear W19954428462016 @default.
• W1995442846 countsByYear W19954428462017 @default.
• W1995442846 countsByYear W19954428462018 @default.
• W1995442846 countsByYear W19954428462019 @default.
• W1995442846 countsByYear W19954428462021 @default.
• W1995442846 countsByYear W19954428462023 @default.
• W1995442846 crossrefType "journal-article" @default.
• W1995442846 hasAuthorship W1995442846A5049504137 @default.
• W1995442846 hasBestOaLocation W19954428461 @default.
• W1995442846 hasConcept C177713679 @default.
• W1995442846 hasConcept C17825722 @default.
• W1995442846 hasConcept C19527891 @default.
• W1995442846 hasConcept C2524010 @default.
• W1995442846 hasConcept C2983241795 @default.
• W1995442846 hasConcept C2989179672 @default.
• W1995442846 hasConcept C33923547 @default.
• W1995442846 hasConcept C545542383 @default.
• W1995442846 hasConcept C71924100 @default.
• W1995442846 hasConceptScore W1995442846C177713679 @default.
• W1995442846 hasConceptScore W1995442846C17825722 @default.
• W1995442846 hasConceptScore W1995442846C19527891 @default.
• W1995442846 hasConceptScore W1995442846C2524010 @default.
• W1995442846 hasConceptScore W1995442846C2983241795 @default.
• W1995442846 hasConceptScore W1995442846C2989179672 @default.
• W1995442846 hasConceptScore W1995442846C33923547 @default.
• W1995442846 hasConceptScore W1995442846C545542383 @default.
• W1995442846 hasConceptScore W1995442846C71924100 @default.
• W1995442846 hasIssue "3" @default.
• W1995442846 hasLocation W19954428461 @default.
• W1995442846 hasLocation W19954428462 @default.
• W1995442846 hasLocation W19954428463 @default.
• W1995442846 hasOpenAccess W1995442846 @default.
• W1995442846 hasPrimaryLocation W19954428461 @default.
• W1995442846 hasRelatedWork W2324745759 @default.
• W1995442846 hasRelatedWork W2423093072 @default.
• W1995442846 hasRelatedWork W2463337598 @default.
• W1995442846 hasRelatedWork W2596471058 @default.
• W1995442846 hasRelatedWork W2924416622 @default.
• W1995442846 hasRelatedWork W2978161189 @default.
• W1995442846 hasRelatedWork W3176990581 @default.
• W1995442846 hasRelatedWork W4220750537 @default.
• W1995442846 hasRelatedWork W4298363913 @default.
• W1995442846 hasRelatedWork W4299297278 @default.
• W1995442846 hasVolume "108" @default.
• W1995442846 isParatext "false" @default.
• W1995442846 isRetracted "false" @default.
• W1995442846 magId "1995442846" @default.
• W1995442846 workType "article" @default.