A system and method for organization and analysis of complex and dynamically interactive time series is disclosed. One example comprises a processor based system for relational analysis of physiologic signals for providing early recognition of catastrophic and pathologic events such as pathophysiologic divergence. The processor is programmed to identify pathophysiologic divergence of at least one of first and second physiologic parameters in relationship to the other and to output an indication of the divergence.
An object-based method of iterative relational processing waveform fragments in the time domain is described wherein each more complex waveform object inherits the characteristics of the waveform objects from which it is derived. The first physiologic parameter can be the amplitude and frequency of the variation in chest wall impedance or nasal pressure and the second parameter can be a measure or indication of the arterial oxygen saturation.
CROSS REFERENCE TO RELATED APPLICATIONSThis application claims priority from U.S. Provisional Application No. 60/291,691 filed on May 17, 2001, claims priority from U.S.
Provisional Application No. 60/291,687 filed on May 17, 2001, and is a continuation-in-part of U.S.
Patent application Ser. 09/628,655 filed Jul. 28, 2000 (now U.S. 6,609,016), which is a continuation-in-part of U.S. Patent application Ser. 09/115,226 filed Jul. 14, 1998 (now U.S.
6,223,064), which is a continuation-in-part of U.S. Patent application Ser.
08/789,460 filed Jan. 27, 1997 (now U.S. Field of the InventionThe present disclosure relates to an object-based system for the organization, analysis, and recognition of complex timed processes and the analysis and integration of time-series outputs of data sets and particularly physiologic data sets, and to the evaluation of the financial and physiologic datasets and the determination of relationships between them.2. BackgroundThe analysis of time-series data is widely used to characterize the behavior of a system. The following four general categories of approaches are commonly applied to achieve characterization of such a system and these provide a general background for the present disclosure. The approaches are illustrative both in their conceptualization, application, and limitations.The first such approach represents a form of mathematical reductionism of the complexity through the application of a cascade of rules based on an anticipated relationship between the time-series output and a given set of system mechanisms.
In this approach the operative mechanisms, data set characteristics, and intruding artifact are a priori defined to the best extent possible. Then a set of rules is applied to characterize and analyze the data set based on predicted relationships between the data set and the systems being characterized. Such systems often include cascading branches of decision-based algorithms, the complexity of which increase greatly in the presence of multiple interactive mechanisms. The reductionism approach is severely limited by the uncertainty and complexity, which rapidly emerges when a cascade of rules is applied to a highly interactive data set, when the signal to noise ratio is low, and/or when multiple data sets generated by complex and dynamically interactive systems are evaluated.
These methods become inordinately more cumbersome as the complexity and number of time-series increases. In addition the subtlety of the interactive and dynamic relationships along and between datasets and the variations associated with the technique or tools of data collection often makes the cascading rules very difficult to a priori define.The failure of simplification the analysis through mathematical reductionism to adequately characterize the complex systems generating such data sets, led to the perception that this failure resulted from specific limitations of a particular data format (usually the time domain format).
In other words, the time-series was perceived to contain sufficient information to characterize the system but, it was thought, that the recognition of this information required reformatting into a different mathematical representation, which emphasized other hidden components which were specific for certain important system characteristics. This approach is exemplified by frequency processing methods, which reformat the time-series into frequency components, such as its sine components or wavelets, with the hope that patterns of specific frequency relationships within the system will emerge to be recognized. While often uncovering considerable useful information, this approach remains quite limited when applied to highly complex and interactive systems, because many complex relationships are poorly characterized by their frequency components, and it is often difficult to relate an output derived from frequency-based primitives to specific mechanisms operative within the system. In other words, the advantages associated with mathematically defined linkages between system mechanisms and the rules based analysis provided by reductionism is reduced by the data reformatting process for the purpose of frequency based signal processing as, for example, is provided by Fourier or wavelet transforms.A third approach seeks to identify the patterns or relationships by repetitively reprocessing the time-series with a set of general comparative rules or by statistical processing.
As with the data reformatting approach, the utility of this method in isolation (as embodied in neural network based analysis), is severely limited by dissociation of the output from the complex and interactive operative mechanisms, which define the output. With such processing, the relevant scope and characterization of the relationships of the output to the actual behavior of the dynamic interactions of the system is often quite limited. This limits the applicability of such processing in environments wherein the characterization of behavior of the system as a function by the output may be as important as the actual output values themselves.A fourth approach has been to apply chaotic processing to the time-series. Again, like conventional signal processing, this method is applied with the expectation that some predictive pattern will emerge to be recognized.
This technique shares several of the limitations noted for both frequency and statistical based data reformatting. In addition, as will be discussed, the application of this type of processing to physiologic signals is limited by redundant and interactive higher control which greatly limits the progression of the system to a state of uncontrolled chaotic behavior. Such systems operate in environments of substantial interactive control until the development of a severe disease state, a point at which the diagnostic information provided by processing often has less adjective utility relevant timely intervention.The human physiologic system derives a large array of time-series outputs, which have substantial relevance when monitored over a finite time interval. The human can be considered the prototypic complex interactive system. These interactions and the mechanisms defining them have been the subject of intense research for over one hundred years and most of this work has been performed in the time domain.
For this reason any approach toward the characterization of such a system needs to consider the value of engaging the body of knowledge, which relates to these mechanisms. This has been one of the reasons that reductionism has predominated in the analysis of physiologic signals. 5,765,563 to Vander Schaff, U.S. 5,803,066 to Rapoport, and U.S. 6,138,675 to Berthon-Jones show such simple cascade decision systems for processing physiologic signals. 5,751,911 to Goldman shows a real-time waveform analysis system, which utilizes neural networks to perform various stages of the analysis. 6,144,877 to Depetrillo shows a processor-based method for determining statistical information for time-series data and for detecting a biological condition of a biological system from the statistical information.
5,782,240 and 5,730,144 to Katz show a system that applies chaos analyzers, which generate a time-series, vector representation of each monitored function and apply chaotic processing to identify certain events. All of these systems are deficient in that they are not able to adequately organize, order and analyze the true state of dynamic interaction operative in the generation of these signals.Critical illness is one example of a dynamic timed process, which is poorly characterized by the above-noted conventional methods. When human physiologic stability is under threat, it is maintained by a complex array of interactive physiologic systems, which control the critical time dependent process of oxygen delivery to the organism. Each system (e.g. Respiratory, cardiac or vascular) has multiple biochemical and/or mechanical controls, which operate together in a predictable manner to optimize oxygen delivery under conditions of threat.
For example an increased oxygen requirement during infection causes the patient to increase oxygen delivery by lowering lung carbon dioxide through hyperventilation and the fall in carbon dioxide then causes the hemoglobin molecule to increase its affinity for oxygen thereby further enhancing oxygen delivery. In addition to the basic control of a single system, other systems interact with the originally affected system to produce a predictable pattern of response. For example, in the presence of infection, the cardiac system interacts with the respiratory system such that both the stroke volume and heart rate increase. In addition, the vascular system may respond with a reduction in arterial tone and an increase in venous tone, thereby both reducing impedance to the flow of oxygen to the tissues and shifting more blood into the arterial compartment.Each system generally also has a plurality of predicable compensation responses to adjust for pathologic alteration or injury to the system and these responses interact between systems. For example the development of infectious injury to the lung will result in an increase in volume of ventilated gas to compensate for the loss of functional surface area. This increase in ventilation can then induce a synergistic increase in both stroke volume and heart rate.Finally, a pathologic process altering one system will generally also induce an alteration in one or more other systems and these processes are all time dependent.
Sub acute or acute life threatening conditions such as sepsis, pulmonary embolism, or hemorrhage generally affect the systems in cascades or predictable sequences which may have a time course range from as little as 20 seconds or more than 72 hours. For example, the brief development of airway collapse induces a fall in oxygen saturation, which then causes a compensatory hyperventilation response, which causes a rise in heart rate over as little as 20-30 seconds. An infection, on the other hand, has a more prolonged time course inducing a rise in respiration rate, a rise in heart rate, and then a progressive fall in oxygen saturation, a fall in respiration rate, and a finally a terminal fall in heart rate often over a course of 48-72 hours.It can be seen therefore that each disease process engaging the organism causes the induction of a complex and interactive time-series of pathophysiologic perturbation and compensation. At the onset of the disease (such as early in the course of infection) the degree of physiologic change may be very slight and limited to one or two variables. As a disease progresses both the magnitude of perturbation and the number of systems involved increases. In addition to inducing a predictable range of perturbation, a particular disease process generally produces a specific range of progression and pattern of evolution as a function of injury, compensation, and system interaction. Furthermore, this multi-system complexity, which can be induced by initial pathologic involvement of a single system, is greatly magnified when a plurality of pathologic processes is present.Despite the fact that these conditions represent some of the most important adversities affecting human beings, these pathologic processes are poorly characterized by even the most sophisticated of conventional monitors, which greatly oversimplify the processing and outputs.
Perhaps this is due to the fact that this interactive complexity overwhelmed the developers of substantially all of the conventional physiologic signal-processing methods in the same way that it overwhelms the physicians and nurses at the bedside everyday. Hospital critical care patient monitors have generally been applied as warning devices upon threshold breach of specific critical parameters with the focus on the balance between timely warning of a potentially life threatening threshold breach and the mitigation of false alarms.
However, during the pivotal time, early in the process of the evolution of critical illness, the compensatory responses limit the change in primary critical variables so that the user, monitoring these parameters in isolation, is often given a false sense of security. For this reason it cannot be enough to recognize and warn of the occurrence of a respiratory arrest, or hypotension, or hypoxia, or of a particular type of cardiac arrhythmia. To truly engage and characterize the processes present, a patient monitor must have capability to properly analyze, organize, and output in a quickly and easily understood format the true interactive state of critical illness. As discussed below, it is one of the purposes of the present disclosure to provide such a monitor.
SUMMARY OF DISCLOSED EMBODIMENTSThe present disclosure provides a system and method, which provide comprehensive organization and analysis of interactive complexity along and between pluralities of time-series. One embodiment of the present disclosure includes an objects-based method of iterative relational processing of time-series fragments or their derivatives along and between corresponding time-series. The system then applies an iterative comparison process of those fragments along and between a plurality time-series. In this way, the relationship of a wide range of characteristics of substantially any dynamic occurrence in one time-series can be compared to the same or other characteristics of substantially any dynamic occurrence along another portion of the same time-series or any of the processed corresponding time-series.According to the present disclosure, a first time-series is processed to render a time-series first level derived from sequential time-series segments of the first series. The time-series first level is stored in a relational database, object database or object-relational database.
The first time-series level is processed to render a second time-series level derived from the sequential time-series component of the first time-series level and these are stored in the relational database, object database or object-relational database. Additional levels are then derived as desired.
The compositions of sequential time-series, which make up the first and second levels, are determined by the definitions selected for the respective segments from which each level is derived. Each time-series fragment is represented as a time-series object, and each more complex time-series object inherits the more basic characteristics of time-series objects from which they are derived.The time course of sub acute and acute critical illness to point of death is highly variable and can range from 24-72 hours with toxic shock, to as little as 30 seconds with neonatal apnea. The present inventors recognized that, regardless of its time course, such a pathological occurrence will have a particular “conformation”, which according to the present disclosure can be represented spatially by an object based processing system and method as a particular object or time-series of objects, as a function of the specific progression of the interactive components for the purpose of both processing, and animation. The present inventors also recognized that the development of such a processing system would be capable of organizing and analyzing the inordinate degree of dynamic complexity associated with the output from the biologic systems through the automatic incorporation of these time-series outputs into a highly organized relational, layered, object-based data structure. Finally, the inventors further recognized that because of the potentially rapid time course of these illnesses and the irreversible endpoint, that patient care monitors must provide a quickly and easily understood output, which gives the medical personnel a simplified and succinct analysis of these complex relationships that accurately reflects the interactive complexity faced by the patient's physiologic systems.It has been suggested that the development of periodicity in a human physiologic system represents a simplification of that system. This concept is based on the perception that human interactive physiologic systems operate in an environment of chaos and that a partial loss of control simplifies the relationships, allowing simpler periodic relationships to emerge.
However, there is considerable reason to believe that this is not the case. Patients entering an environment of lower partial pressure of oxygen, as at altitude, will develop periodicity of ventilation. This does not indicate a general simplification of the system but rather, one proposed operative mechanism for the emergence of this new pattern is that the pattern reflects the uncovering of a preexisting dynamic relationship between two controllers, which now, together determine ventilation in this new environment. At sea level the controller responding to oxygen was subordinate to the controller responding to carbon dioxide so that the periodicity was absent. This simple illustration serves to demonstrate the critical linkage between patient outputs and higher control and the criticality of comprehensively comparing dynamic relationships along and between signals to achieve a true picture of the operative physiology. While periodicities are, at times, clearly pathologic, their development in biologic systems, rather than a manifestation of simplification of physiological behavior, than the engagement of new, often represents the engagement of more rudimentary layers of protection of a particular organ function or range built into the control system. This illustration further demonstrates that a given physiologic signal, when monitored in isolation, may appear to exhibit totally unpredictable and chaotic behavior, but when considered in mathematical or graphical relation (as in phase space) to a plurality of corresponding interactive signals, and to the interactive control mechanisms of those corresponding signals, the behavior of the original, chaotic appearing, signal often becomes much more explicable.In an example, consider a timed plot of oxygen saturation (SPO2) under heavy sedation during sleep.
This state is often associated with a loss of the maintenance of a narrow control range of ventilation during sleep and with the loss of stability of the airway so that a plot of the oxygen saturation, in the presence of such deep sedation, shows a highly variable pattern, which often appears grossly unpredictable, with sustained falls in oxygen saturation intermixed with rapid falls and often seemingly random rapid corrections. However, there are definable limits or ranges of the signal, and generally definable patterns, which are definable within the background of a now highly variable SPO2 signal. It may be temping to define this behavior statistically or by a chaotic processor in the hope of defining some emerging patterns as a function of the mathematical behavior of that signal. However, when analyzed with the partial pressure of CO2, the minute ventilation, and a plot of EEG activity the oxygen saturation values are seen as a subordinate signal to the airflow which is now being controlled by a dysfunctional control process, which process is being salvaged by a more coarse and rudimentary survival response mechanism. The apparently chaotic behavior is now seen as driven by a complex but predictable sequence of a plurality of dynamic interactive relationships between corresponding signals and the forces impacting them. Therefore, in the presence of a pathophysiologic process, the behavior and ranges of any given signal are optimally defined by the dynamic patterns of the interactive behavior of corresponding signals and their respective dynamic ranges.A biologic system actually exploits the chaotic output of simple nonlinear relationships by defining control ranges, which are affected by variations in corresponding signals.
This produces a great degree in diversity of dynamic physiologic response, which is beneficial in that it may favor survival of a particular subgroup, in the presence of a certain type of pathophysiologic threat. The present inventors noted that, while this diversity imparts greater complexity, this complexity can be ordered by the application of an iterative microprocessor system, which defines a given signal as a function of a range “dynamic normality”. According to one embodiment of the present disclosure, each signal is defined as a function of its own dynamic range (and in relation to a predicted control range) and as a function of contemporaneously relevant relationships of the dynamic ranges of other corresponding signals (with respect to their respective control ranges).In one embodiment, the present disclosure comprises a system and method for organizing and analyzing multiple time-series of parameters generated by a patient (as during critical illness) and outputting this analysis in readily understandable format. The preferred system is capable of simultaneously processing dynamic time-series of physiologic relationships in real time at multiple levels along each parameter and across multiple levels of different parameters. The present embodiments provide this level of interactive analysis specifically to match the complexity occurring during a pathologic occurrence. More specifically the present embodiments provide an analysis system and method, which analyses the true dynamic state of a biologic system and the interactive primary and compensatory perturbations defining that state. During health, the output of physiologic systems are maintained within tight variances.
As will be discussed, using the signal processing system of the present the extent to which the signals are held within these tight variances are characterized as a function of their dynamic ranges of variance and the signals are further characterized as a function of their dynamic relationships along the time-series within a given signal and between a plurality of additional corresponding signals. As will be learned by the following disclosure, the optimal monitor of the human physiologic state during critical illness must be capable of analyzing time-series relationships along and between a plurality signals with the similar degree of analytic complexity as is operative in the biologic systems controlling the interactive responses which are inducing those signals, and capable of outputting an indication based on the analysis in a readily understandable format. In the preferred embodiment this is provided as a dynamic format such as a two-dimensional or three-dimensional object animation, the configuration of which is related to the analysis output. The configurations of the animation changes with the analysis output, as this output changes over time in relation to changes in the patient's physiologic state.
Faber Castell Mathema
The animation thereby provides a succinct and dynamic summary rendering which organizes the complexity of the interactive components of the output so that they can be more readily understood and used at the bedside and for the purpose of patient management and education of medical staff relevant the application of time-series analysis in the assessment of disease.