A Real Time Control Architecture for Continuously Managing Patients in a Care Unit

The monitoring and treatment of patients in a care unit is a complex task in which even the most experienced clinicians can make errors. A hemato-oncology department in which patients undergo chemotherapy asked for a computerized system able to provide intelligent and continuous support in this task. One issue in building such a system is the definition of a control architecture able to manage, in real time, a treatment plan containing prescriptions and protocols in which temporal constraints are expressed in various ways, that is, which supervises the treatment, including controlling the timely execution of prescriptions and suggesting modifications to the plan according to the patient’s evolving condition. The system to solve these issues, called SEPIA, has to manage the dynamic, processes involved in patient care. Its role is to generate, in real time, commands for the patient’s care (execution of tests, administration of drugs) from a plan, and to monitor the patient’s state so that it may propose actions updating the plan. The necessity of an explicit time representation is shown. We propose using a linear time structure towards the past, with precise and absolute dates, open towards the future, and with imprecise and relative dates. Temporal relative scales are introduced to facilitate knowledge representation and access.

MOSAIC: segmenting multiple aligned DNA sequences

MOSAIC is a set of tools for the segmentation of multiple aligned DNA sequences into homogeneous zones. The segmentation is based on the distribution of mutational events along the alignment. As an example, the analysis of one repeated sequence belonging to the subtelomeric regions of the yeast genome is presented.

AVAILABILITY

Free access from ftp://ftp.biomath.jussieu.fr/pub/papers/MOSAIC

Hidden Markov model approach for identifying the modular framework of the protein backbone

The hidden Markov model (HMM) was used to identify recurrent short 3D structural building blocks (SBBs) describing protein backbones, independently of any a priori knowledge. Polypeptide chains are decomposed into a series of short segments defined by their inter-alpha-carbon distances. Basically, the model takes into account the sequentiality of the observed segments and assumes that each one corresponds to one of several possible SBBs. Fitting the model to a database of non-redundant proteins allowed us to decode proteins in terms of 12 distinct SBBs with different roles in protein structure. Some SBBs correspond to classical regular secondary structures. Others correspond to a significant subdivision of their bounding regions previously considered to be a single pattern. The major contribution of the HMM is that this model implicitly takes into account the sequential connections between SBBs and thus describes the most probable pathways by which the blocks are connected to form the framework of the protein structures. Validation of the SBBs code was performed by extracting SBB series repeated in recoding proteins and examining their structural similarities. Preliminary results on the sequence specificity of SBBs suggest promising perspectives for the prediction of SBBs or series of SBBs from the protein sequences.

[Patient-centered consultation of "good practice guidelines": OncoDoc, a decision support system for the management of breast cancer patients]

Beyond considerations of cost-effectiveness, clinical practice guidelines (CPG) can reduce practice variations and thus improve the quality of care. However, despite the proliferation of implemented CPG and their wide diffusion thanks to Internet-based technologies, physicians' compliance with formal standards is weak. Developed according to a document-based paradigm, OncoDoc proposes an original framework for implementing CPG. Domain knowledge has been encoded as a decision tree whose branches are both exclusive and exhaustive. This generic knowledge is operationalized at the point of care by the interactive building, through hypertextual navigation, of a patient-based clinical context leading to specific therapeutic recommendations. OncoDoc has first been applied to the management of breast cancer patients and demonstrated within a full-scale experimentation in a clinical setting a compliance of 80 per cent.

Modelling behavioral syndromes using Bayesian networks

In this paper Bayesian networks modelling is applied to a multidimensional model of depression. The characterization of the probabilistic model exploits expert knowledge to associate latent concentrations of neurotransmitters and symptoms. An evolution perspective is also considered. Specific criteria are introduced to detect the influence of the latent variable on the observation of symptoms. The Bayesian analysis is carried out using Gibbs sampling technique which is implemented in the BUGS software. The estimation phase leads to the selection of symptoms entering into the definition of behavioral syndromes. Results on real data are discussed. The last section deals with simulation experiments. Simulation results confirm our methodological choices. Results of the paper can enlarge to the central problem of the management of latent variables in Bayesian networks modelling.

From text to knowledge: a unifying document-centered view of analyzed medical language

Although medical language processing (MLP) has achieved some success, the actual use and dissemination of data extracted from free text by MLP systems is still very limited. We claim that the adoption of an 'enriched-document' paradigm (or 'document-centered' view) can help to address this issue. We present this paradigm and explain how it can be implemented, then discuss its expected benefits both for end-users and MLP researchers.

A strategy for finding regions of similarity in complete genome sequences

MOTIVATION

Complete genomic sequences will become available in the future. New methods to deal with very large sequences (sizes beyond 100 kb) efficiently are required. One of the main aims of such work is to increase our understanding of genome organization and evolution. This requires studies of the locations of regions of similarity.

RESULTS

We present here a new tool, ASSIRC ('Accelerated Search for SImilarity Regions in Chromosomes'), for finding regions of similarity in genomic sequences. The method involves three steps: (i) identification of short exact chains of fixed size, called 'seeds', common to both sequences, using hashing functions; (ii) extension of these seeds into putative regions of similarity by a 'random walk' procedure; (iii) final selection of regions of similarity by assessing alignments of the putative sequences. We used simulations to estimate the proportion of regions of similarity not detected for particular region sizes, base identity proportions and seed sizes. This approach can be tailored to the user's specifications. We looked for regions of similarity between two yeast chromosomes (V and IX). The efficiency of the approach was compared to those of conventional programs BLAST and FASTA, by assessing CPU time required and the regions of similarity found for the same data set.

AVAILABILITY

Source programs are freely available at the following address: ftp://ftp.biologie.ens. fr/pub/molbio/assirc.tar.gz

CONTACT

vincens@biologie.ens.fr, hazout@urbb.jussieu.fr

Hypertextual navigation operationalizing generic clinical practice guidelines for patient-specific therapeutic decisions

Despite the proliferation of implemented clinical practice guidelines, there is still little evidence of physicians compliance to formal standards. The ONCODOC project proposes a framework for elaborating generic decision support guidelines in a document-based paradigm with a knowledge-based approach. It has been first applied to assist clinicians in the treatment of breast cancer patients. Therapeutic expertise has been encoded as a decision tree. The decision process is driven by the clinician who interactively browses a hypertext version of the decision tree. During the navigation, he incrementally assigns values to decision parameters on the basis of his free interpretation of his patient's condition and thus builds a clinical context leading to patient-specific therapeutic recommendations. These guidelines are distributed on a hospital intranet and are evaluated at the point of care in an oncology department.

Hospitexte: towards a document-based hypertextual electronic medical record

The patient record is a repository for knowledge about a patient. Work in Artificial Intelligence and knowledge representation has evidenced the intrinsic difficulty of formalizing knowledge for computer processing. It is therefore not a surprise that most attempts at computerizing the patient record have only had a limited degree of success or applicability. We claim that this is due to the fact that medicine is an empirical domain, and thus fundamentally resists formalization. Therefore, the only way medical knowledge can be fully expressed is through natural languages which is indeed what clinicians actually use. We proposed and designed an electronic medical record which adheres to this hypothesis and where structured documents play a prominent role.

Bi-dimensional scaling map (BDS-Map): an approach for building large genetic maps

MOTIVATION

The approaches usually used for building large genetic maps consist of dividing the marker set into linkage groups and provide local orders that can be tested by multi-point linkage analysis. To deal with the limitations of these approaches, a strategy taking the marker set into account globally is defined.

RESULTS

The paper presents a new approach called 'Bi-Dimensional Scaling Map (BDS-Map) for inferring marker orders and distances in genetic maps based on the use of an additional dimension orthogonal to the map into which markers are projected. Dynamical forces based on a two-point analysis are applied to tend to optimize the marker locations in space. The efficiency of the approach is exemplified on real data (16 and 70 markers on chromosomes 6 and 2, respectively) and simulated data (50 maps of 70 markers).

[Comparative study of French drug data banks]

We proposed a set of 9 criteria to evaluate computerized French drug data banks currently available on the Minitel, medical software, or the Web. These criteria are used for testing the scientific quality of the information provided in terms of reliability and completeness. They round off the nine criteria proposed in a previous study for sole drug interactions. None of the drug data banks is better than the others on all the selected dimensions. But the methodology presented should allow a particular user with defined needs to weigh up the criteria and optimize his choice.

Evaluating a normalized conceptual representation produced from natural language patient discharge summaries

The Menelas project aimed to produce a normalized conceptual representation from natural language patient discharge summaries. Because of the complex and detailed nature of conceptual representations, evaluating the quality of output of such a system is difficult. We present the method designed to measure the quality of Menelas output, and its application to the state of the French Menelas prototype as of the end of the project. We examine this method in the framework recently proposed by Friedman and Hripcsak. We also propose two conditions which enable to reduce the evaluation preparation workload.

A real time control architecture for continuously managing patients in a care unit

The monitoring and treatment of patients in a care unit is a complex task in which even the most experienced clinicians can make errors. A hemato-oncology department in which patients undergo chemotherapy asked for a computerized system able to provide intelligent and continuous support in this task. One issue in building such a system is the definition of a control architecture able to manage, in real time, a treatment plan containing prescriptions and protocols in which temporal constraints are expressed in various ways, that is, which supervises the treatment, including controlling the timely execution of prescriptions and suggesting modifications to the plan according to the patient's evolving condition. The system to solve these issues, called SEPIA, has to manage the dynamic, processes involved in patient care. Its role is to generate, in real time, commands for the patient's care (execution of tests, administration of drugs) from a plan, and to monitor the patient's state so that it may propose actions updating the plan. The necessity of an explicit time representation is shown. We propose using a linear time structure towards the past, with precise and absolute dates, open towards the future, and with imprecise and relative dates. Temporal relative scales are introduced to facilitate knowledge representation and access.

Control theory as a conceptual framework for intensive care monitoring

Monitoring patients hospitalized in hemato-oncology departments to undergo clinical protocols of therapy is a complex task. The main difficulty arises in the management of the oncology protocol as well as in the management of critical episodes of acute illness which frequently occur due to high toxicity of the used antimitotics. This problem of controlling a patient's condition can be conceptualized within the control theory paradigm as the task of controlling a process whose state changes over time and can deviate unacceptably from a normal range. Upon request of a French clinical department of hemato-oncology, we developed an intelligent patient monitor named SEPIA to assist clinicians in this task. Following the control theory analogy at the level of knowledge bases design, we have modeled the medical knowledge as control information to represent the medical actions, and state information is used as feedback control to characterize the patient's state. After a general presentation of SEPIA, the analogies between control theory actors and SEPIA's components are specified. The article then focuses on state variables representation. The data reduction process, activated as new data are given to the system, is described. A simulated running session is finally presented to illustrate the whole reasoning process.

Issues in the structuring and acquisition of an ontology for medical language understanding

Medical natural language understanding basically aims at representing the contents of medical texts in a formal, conceptual representation. The understanding process itself increasingly relies on a body of domain knowledge, generally expressed in the same conceptual formalism. The design of such a conceptual representation is a key knowledge-acquisition issue. When representing knowledge, the most important point is to ensure that the formal exploitation of the knowledge representation conforms to its meaning in the domain. We examined some methodological and theoretical principles to enforce this conformity. These principles result from our experience in MENELAS, a medical language understanding project.

A multi-lingual architecture for building a normalised conceptual representation from medical language

The overall goal of MENELAS is to provide better access to the information contained in natural language patient discharge summaries (PDSs), through the design and implementation of a prototype able to analyse medical texts. The approach taken by MENELAS is based on the following key principles: (i) to maximise the usefulness of natural language analysis and the usability of its results, the output of natural language analysis must be a normalised conceptual representation of medical information; and (ii) to maximise the reuse of resources, language analysis should be domain-independent and conceptual representation should be language-independent. This paper discusses the results obtained and the issues raised when implementing these principles during the project.

Real time monitoring in the control theory paradigm

Monitoring patients hospitalized in hemato-oncology departments to undergo clinical protocols of therapy is a complex task. The main difficulty arises in the follow-up of the oncology protocol and in the management of critical episodes of acute illness which frequently occur due to the high toxicity of the antimitotics used. This problem can be conceptualized within the control theory paradigm as the task of controlling a process whose state can deviate unacceptably from a normal range. Following the control theory analogy at the level of knowledge bases design, we have modeled the medical knowledge as control information to represent the medical actions, and state information is used as a feedback control to readjust the command.

Structuration and acquisition of medical knowledge. Using UMLS in the conceptual graph formalism

The use of a taxonomy, such as the concept type lattice (CTL) of Conceptual Graphs, is a central structuring piece in a knowledge-based system. The knowledge it contains is constantly used by the system, and its structure provides a guide for the acquisition of other pieces of knowledge. We show how UMLS can be used as a knowledge resource to build a CTL and how the CTL can help the process of acquisition for other kinds of knowledge. We illustrate this method in the context of the MENELAS natural language understanding project.

[Treatment of arterial hypertension in insulin-treated diabetic patients. Change over 3 years (1985-1988)]

In 1985, an assessment of arterial hypertension treatment in insulin-treated diabetic patient gave disappointing results. In 1988, we carried out another study in order to assess the impact of new antihypertensive drugs (angiotensin converting enzyme inhibitors and calcium antagonists) on the management of arterial hypertension and to identify patients in whom strict normal blood pressure control is mandatory. Seven hundred and fifty four patients were selected. The prevalence of arterial hypertension was 38.4 p. 100 (n = 290). Two hundred and thirty five patients (31.2 p. 100) were on antihypertensive treatment: monotherapy: 60.4 p. 100 (n = 142), bitherapy: 30.6 p. 100 (n = 72), tritherapy: 9 p. 100 (n = 21). In descending order of frequency, the following drugs were used: angiotensin converting enzyme inhibitors, calcium antagonists, diuretics, cardio-selective beta-blockers, central acting agents. Blood pressure values significantly decreased (148/83 mmHg, in 1988, vs 157/85 mmHg, in 1985, p less than 0.05). However, 20 p. 100 of the patients still had blood pressure values greater than or equal to 160 and/or 95 mmHg with or without antihypertensive treatment, and on average, blood pressure values remained higher in patients with antihypertensive treatment than in those without (148/83 mmHg vs 131/77 mmHg, p less than 0.001). Patients with urinary albumin excretion above or equal 30 mg/24 h compared to those with normal albuminuria had significant higher values of blood pressure, glycosylated haemoglobin and blood lipids (p less than 0.01). Only 51 p. 100 of these patients, received an antihypertensive treatment. This study emphasizes the difficulty of antihypertensive treatment in insulin-treated diabetic patients and the necessity to improve education in patients with high risk for widespread angiopathy, and particularly those with increased urinary albumin excretion.

[Prevalence and management of arterial hypertension in diabetic patients treated with insulin]

In small series of selected hypertensive diabetic patients the short-term benefit of antihypertensive treatments is well documented. The purpose of this study was to analyse the management of arterial hypertension in a cohort of 612 insulin-treated diabetic out-patients routinely and consecutively examined in a diabetologic clinic between January 1, 1985 and April 1, 1986. The prevalence of arterial hypertension (i.e. patients with blood pressure values greater than or equal to 160 and/or 95 mmHg or on antihypertensive treatment) was 38 per cent (232 patients). One hundred and eighty-two patients (29.7 per cent) received an antihypertensive treatment (one drug 78 per cent, two drugs 17 per cent, three drugs 5 per cent). In decreasing order of frequency these drugs were: beta-blockers, diuretics, central acting agents, angiotensin-converting enzyme inhibitors and calcium antagonists. In treated hypertensive diabetic patients the mean +/- SD systolic blood pressure (157.1 +/- 19.1 mmHg) and diastolic pressure (85.3 +/- 9.3 mmHg) remained higher than in patients without antihypertensive treatment (133.4 +/- 17.2 and 77.8 +/- 8.3 mmHg respectively). One hundred and forty-two diabetic patients still had blood pressure values greater than or equal to 160 and/or 95 mmHg during visits; 92 were on antihypertensive treatment, 50 were untreated. In hypertensive diabetic patients the mean total glycosylated haemoglobin (HbA1) level was higher than in normotensive diabetic patients (9 +/- 1.6 versus 8.6 +/- 1.8; P less than 0.05). Hypertensive insulin-treated diabetics are the most seriously ill patients, they are under inadequate care. Wrong choice of antihypertensive drugs, incorrect goals of blood pressure reduction, lack of information and education to improve compliance were the main reasons for the poor results of this study.

[Testing an expert system for hypertension]

An Expert System (ES) has been connected to a database management system for the management and follow-up of hypertensive patients. The patient data base, called Artemis, contains approximately 18,000 medical records. About 90% of the initial informations used by the ES is contained in the medical records of the Artemis data base. The knowledge base consists of 870 rules. A first group of rules allows the description of knowledge structures (hierachies, graphs and mutual exclusions). The second group consists of production rules which describe the dynamic reasoning of the expert. The inference engine uses a combination of forward and backward chaining. The ES produce diagnostic hypotheses (possible causes of hypertension) and therapeutic suggestions before and after requiring additional information (patient supplementary interrogation, biological or radiological investigations). The evaluation of the diagnostic performance of the ES was made on 40 confirmed cases of secondary hypertension (SH) and 40 cases of essential hypertension (EH). The initial initial diagnosis, just after the forward chaining step, was correct in 17 cases of SH and 32 cases of EH. The final diagnosis proposed after several steps of forward and backward chaining was correct in 37 cases (92%) of SH and 36 (90%) of EH. Averages of 5 (EH) and 8 (SH) questions were formulated by the ES to reach the final diagnosis. The integration of the ES to the database is expected to facilitate the validation of the knowledge base and to enhance its overall acceptability. Whether or not such an integration will be useful and accepted as a complementary tool by physicians remains however an open question.

Alternative approaches to estimation of population pharmacokinetic parameters: comparison with the nonlinear mixed-effect model

Individual pharmacokinetic parameters can be viewed as independent realizations of a random variable. The probability density function of the variable is assumed to be specified by its first two moments (mean vector and covariance matrix), and these moments then characterize the distribution of the parameters in the population. The following methods are presented for estimation of population characteristics from a set of pharmacokinetic measurements in a sample of subjects: The Global Two-Stage Approach (GTS) uses estimates (and their covariances) of individual parameters obtained after separate fitting of each individual's data. The Iterated Two-Stage Approach (ITS) makes the GTS procedure iterative, using refined bayesian estimates of individual parameters at each step. The Nonlinear Filtering Approach (NLF) also relies on individual parameter estimates produced by using an optimal filter on each subject's data. The three methods give exact results (maximum likelihood estimates), as does NONMEM (the Nonlinear Mixed-Effect Model Approach), when the individual pharmacokinetic model is linear with respect to the parameters and when the distributions of the pharmacokinetic parameters and of the measurement noise in the individual data are both multivariate normal. When the individual pharmacokinetic model is statistically nonlinear (the usual case), the methods differ with respect to: (1) their strategy for handling nonlinearity, (2) their ability to deal with any type of data (experimental and/or routine), and (3) their sensitivity to the amplitude of random effects. With regard to computational aspects, both the computer memory storage requirements and the amount of computation required for the GTS approach are much smaller than for the three other methods. Contrasting considerations as well as results of simulations suggest that GTS, ITS, and, in future, NLF may be valuable alternatives to NONMEM or modifications of it for estimation of population characteristics of pharmacokinetic parameters.

A time-lag model for pharmacokinetics of drugs subject to enterohepatic circulation

A two-compartment model with time lag is proposed to describe the pharmacokinetics of drugs subject to enterohepatic circulation. The basic model, including two compartments for body and GI tract, respectively, with elimination occurring from both compartments, was previously proposed. The assumption that the reabsorption of a drug molecule is delayed after its biliary excretion is expressed by the addition of a time lag in the transfer from the first to the second compartment. Computer simulation of the model for intravenous bolus injection and oral intake of the drug was performed through first-order numerical integration. Several qualitative results concerning changes in pharmacokinetics due to modifications in biliary excretion, in reabsorption, or in elimination are identical with predictions using the basic model. However, several qualitative and quantitative results were significantly different. The pharmacokinetics, though remaining linear, are no longer biexponential. Initial decay after intravenous injection was not affected by modifications in reabsorption or elimination from intestine. Predictions based on the time-delay model agree with existing experimental evidence concerning pharmacokinetics of substances undergoing enterohepatic cycling. Delayed recirculation may lead to rebounds in plasma level profiles as well as after intravenous and oral administration. The half-life of the drug is significantly prolonged even when the kinetic processes involved in recirculation remain unchanged.

A non-linear mathematical model for the in vivo determination of Kupffer cells number and rate of phagocytosis of radiocolloids in rats

In order to perform a quantification of the hepatic RES function, a mathematical model of the colloids phagocytosis was constructed and validated in normal and partially hepatectomised rats. The experimental design consisted of the collection of successive blood samples for the measurement of radiocolloids time courses after the injection of different doses of gelatin colloids (0.075--5 mg of gelatin/100 g body wt). The unknown parameters were estimated by method of maximum likelihood using a second order algorithm. A good fit between experimental and simulated data was obtained for a large range of injected doses using a single set of parameters. Comparisons of parameters values between normal and hepatectomised rats were found to be -onsistent with the hepatectomy ratio. This computerised estimation of parameters provides a determination of both the total number of Kupffer cells and the mean time of a complete phagocytosis cycle which cannot be obtained by classical approaches.