Skip Navigation


Europace Advance Access originally published online on February 10, 2006
Europace 2006 8(3):161-167; doi:10.1093/europace/euj028
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
8/3/161    most recent
euj028v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (1)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Diem, B. H.
Right arrow Articles by Hanrath, P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Diem, B. H.
Right arrow Articles by Hanrath, P.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

© The European Society of Cardiology 2006. All rights reserved. For Permissions, please e-mail: journals.permissions@oxfordjournals.org

ICDS

First multi-centre evaluation of a knowledge-based implant-assistant for implantable cardioverter-defibrillators

Bjoern Henrik Diem1,2, Volker Lang3, Anil Martin Sinha1, Michael Mlynski4, Christoph Stellbrink1,*, Walter Ameling4 and Peter Hanrath1

1 Department of CardiologyUniversity Hospital AachenAachen Germany ; 2 Biotronik GmbH & Co KGBerlin Germany ; 3 MSEILake Oswego, OR USA ; 4 CWAAachen Germany

Manuscript submitted 22 September 2004. Accepted after revision 1 October 2005.

* Corresponding author: Klinik für Kardiologie und Intensivmedizin, Städtische Kliniken Bielefeld, 33604 Bielefeld, Germany. Tel: +49 521 5813401; fax: +49 521 5813499. E-mail address: christoph.stellbrink{at}sk-bielefeld.de


   Abstract
Top
 Abstract
Introduction
 Methods
 Results
 Discussion
 Limitations
 Conclusions
 Acknowledgements
 References
 
Aims Modern implantable cardioverter-defibrillators (ICDs) place increasing demands on the physician, as their complexity requires more and more knowledge and effort in handling them. To overcome this problem an implant-assistant has been developed, which transfers clinical data entered by the physician into a complete set of parameters for programming a dual-chamber ICD (Tachos-DR, Biotronik, Berlin, Germany) at DFT testing (DFT-Prog) and first permanent programming (Perm-Prog) after implant.

Methods and results Routine ICD implantations were initially evaluated by clinical experts at 19 centres in USA and Europe from 178 patient files. The rating of parameters was related to the number of parameters available in each patient. For DFT-Prog, 98.4% of parameter suggestions were identical to experts' expectations, an additional 1.0% were accepted, 0.5% were rejected, and none was considered harmful. This resulted in an overall acceptance of 94.4% of the DFT-Prog. For Perm-Prog, 96.1% of parameters were identical to those advised by experts, an additional 2.4% were accepted, 1.5% rejected, and seven parameters (0.04%) were considered potentially harmful by experts with an overall acceptance of 86.5%. Adaptation of the implant-assistant increased the overall acceptance to 100% for DFT-Prog and 90.6% for first Perm-Prog without any potentially harmful suggestions.

Conclusion The ICD implant-assistant, which allows the physician to programme ICDs directly from clinical data, is a promising method to simplify the programming of modern ICDs.

Key Words: Implantable cardioverter-defibrillator, Implant-assistant, Knowledge-based system, Expert system, Education


   Introduction
Top
 Abstract
 Introduction
Methods
 Results
 Discussion
 Limitations
 Conclusions
 Acknowledgements
 References
 
Modern implantable devices such as implantable cardioverter-defibrillators (ICDs) and pacemakers offer a large variety of therapeutic and diagnostic options to adapt the devices to the specific needs of each patient.1Go,2Go Unfortunately, this flexibility requires adjustment of a large number of parameters, which makes the programming procedure complex and time-consuming. In addition, the physician has to spend an increasing amount of time to keep up with the rapid advances in device technology. One possibility to facilitate both education of the physician and safe programming of the device is a knowledge-based support tool. As the development was initially restricted to support the implant procedure of ICDs, it is called implant-assistant.

Implant-assistant
The implant-assistant allows the physician to enter clinical and arrhythmia data gathered from a patient's history or by non-invasive and invasive studies, e.g. electrophysiological study, in a series of dialogues (Fig. 1). Using these data, the implant-assistant determines the setting of all ICD parameters according to a collection of defined rules, called the knowledge base. The rules are a synopsis of knowledge from different sources: device developers, technical specifications, clinical experts, clinical trials, and other literature and contain the knowledge necessary to adapt and optimize the complete parameter settings of a specific ICD to the patient's needs (Fig. 2). In addition, the implant-assistant provides the physician with explanations why a certain parameter setting is considered optimal and issues warnings for critical device-specific constellations, e.g. low-sensing amplitudes (Fig. 3).


Figure 0281
View larger version (56K):
[in this window]
[in a new window]
  		Figure 1 Screenshot of the implant-assistant: input-dialogue for general data. This screenshot shows the first of six input dialogues of the implant-assistant. The physician can easily choose from common medical parameters, e.g. cardiac disease or medication. Functional parameters like the ejection fraction or the degree of angina and dyspnoea can also be specified. In addition, the system permits entering of data relevant for cardiac resynchronization therapy: optimized electrode position and optimized AV-delay. Note that the implant-assistant does not request technical ICD parameters in its clinical interface.

 


Figure 0282
View larger version (27K):
[in this window]
[in a new window]
  		Figure 2 Flow of information in the implant-assistant. The physician provides the ‘input’ consisting of patient data, test results (e.g. defibrillation threshold) and personal preferences (e.g. low energy cardioversion shock) to the system. The ‘knowledge base’, which represents the combined knowledge of different sources is used to calculate the ‘output’ of the implant-assistant, which consists of an adapted and complete parameter set and the corresponding explanations. The current knowledge base consists of about 1800 conditional rules.

 


Figure 0283
View larger version (16K):
[in this window]
[in a new window]
  		Figure 3 Example for device-specific warnings. This is a screenshot of a warning message to the physician. The implant-assistant identified a special constellation, in this example a measured P-wave amplitude below 1.5 mV. Though a value in this range does not necessarily preclude the use of the atrial signal for pacing, this signal should not be used for detection algorithms such as SMART®, which depends on secure atrial detection.

 
The key features of the implant-assistant are as follows:
  1. The ICD programming is based on a device-independent clinical language, which is more convenient and familiar to most physicians.
  2. The programming of the device is accelerated as the number of clinical input parameters is less than one-third the number of programmable technical ICD parameters.
  3. The safety and accuracy of ICD programming is increased as the appropriateness of the parameter set is cross-checked.
  4. The implant-assistant provides full transparency and educational benefit by explaining all rules and decisions to the physician.
The current version of the implant-assistant supports the implant of four devices (Tachos DR, Tupos LV, Tupos LA, and Deikos A+, all from Biotronik). However, the present study was limited to the evaluation of the suggestions for one dual-chamber ICD with full atrial and ventricular therapy options (Tachos DR).

As we believe that this is the first knowledge-based support tool for ICD programming, a multi-centre, pre-clinical study was carried out to determine safety, quality, and ease of handling of the implant-assistant.


   Methods
Top
 Abstract
 Introduction
 Methods
Results
 Discussion
 Limitations
 Conclusions
 Acknowledgements
 References
 
The implant-assistant's parameter suggestions were evaluated by physicians with long-standing experience in ICD programming at 19 centres in the United States, United Kingdom, and Germany to determine the quality and safety of the implant-assistant. As this was a new technology in the field of ICD, a two-phase study design was chosen. In the first phase each expert entered clinical data from 10 previously anonymized files of patients, who had received a dual-chamber ICD, in the implant-assistant installed on a commercial notebook computer. On the basis of these data, the implant-assistant created parameter suggestions, which were separately evaluated for ventricular DFT testing and the first permanent programme by the expert. These parameter suggestions comprise all programmable parameters of the ICD, including all VT/VF detection and therapy, AT/AF prevention, detection, and therapy, and also bradycardia pacing settings. Different rating systems for each parameter, the parameter sets as a whole and the handling were used and the experts were asked to provide alternative values for suggested parameter settings, which did not meet their expectations. On the basis of these answers, the knowledge base was refined, before the patient cases were recalculated and re-evaluated in the second phase.

Rating systems
The participating physicians attributed one of four categories to each calculated parameter. Category A: identical to physician's own opinion; Category B: not identical, but acceptable for the physician; Category C: outside the physician's acceptable range, but still not harmful; Category D: potentially harmful. Because of the activation and deactivation of ICD therapy zones and other features, the total number of rated parameters may differ from patient to patient. Therefore, the relative frequency of the four categories was determined for each patient and the mean and SD of all patients was computed. In addition to this parameter evaluation, each programme as a whole was rated using one of the following three categories: acceptable, unacceptable, and potentially harmful. The total number of programmes for each category was determined. The physician's opinion on the simplicity of the implant-assistant's handling and the ease of understanding the terminology were quantified by the physician giving a rate between 1 (very easy) and 10 (very difficult). The distribution of all ratings was determined. In addition, the ratings for the first case of all 19 centres were compared with the ratings of the last case of each centre in order to evaluate learning curve effects.

Patient population
A total of 178 files of patients with implanted dual-chamber ICD were included in the first step of the study. At implant, 143 males (80%) and 35 females (20%) with a mean age of 66±13 years presented with a mean ejection fraction of 31±15%. New York Heart Association classes were distributed as follows: class I, 57 patients (32%); class II, 73 (41%); class III, 46 (26%); and class IV, 2 (1%). One hundred and twenty-five patients (70%) suffered from coronary artery disease or ischaemic cardiomyopathy; 89 (50%) had a history of myocardial infarction; dilated cardiomyopathy was present in 78 (44%); valvular disease in 21 (12%); hypertrophic cardiomyopathy in three (2%); and long QT syndrome in three (2%). In the absence of any structural cardiac disease idiopathic arrhythmia was found in three patients (2%).

Ventricular fibrillation was known in 40 patients (22%) prior to ICD implantation, sustained ventricular tachycardia in 112 (63%), and non-sustained ventricular tachycardia in 62 (35%). Intermittent atrial arrhythmias of any kind were known in 66 (37%) and chronic atrial arrhythmia in 3 (2%) of the patients.

Statistical methods
Because of the missing data in some cases in Phase II, a subgroup was created, which contained only the complete cases with data for Phase I and Phase II. The ratings were compared for Phase I between the total group and the subgroup as well as for the subgroup between Phase I and Phase II. The {chi}2 test according to Brandt/Snedecor was used for the distribution of the parameter ratings (4x2 matrix). As no harmful rating was observed for any programme, the regular {chi}2 test could be used to compare the Phase I results between the total group and the subgroup (2x2 matrix). The McNemar test was used to compare the paired programme ratings of the subgroup.


   Results
Top
 Abstract
 Introduction
 Methods
 Results
Discussion
 Limitations
 Conclusions
 Acknowledgements
 References
 
Phase I
For all 178 included patient files, the complete ratings were available for the study. The parameter ratings for DFT testing and for the first permanent programme are displayed in Fig. 4. The combination of Category A (identical value) and Category B (acceptable value) was the measurement of acceptable performance and was 99.4% for DFT testing and 98.5% for the first permanent programme. Despite the high acceptance rates, seven parameters were considered potentially harmful for the first permanent programme. Five of these seven ratings referred to the activation of a second ATP scheme in VT zones, which covered the cycle lengths of observed unstable VTs. The remaining two ratings resulted from a VT-1 zone limit, which was considered too high for patients with severe cardiomyopathy, though only VTs with rates above 200 bpm. were documented.


Figure 0284
View larger version (21K):
[in this window]
[in a new window]
  		Figure 4 Parameter-rating in Phase I for DFT testing (A) and first permanent programme (B). Parameter rating category vs. mean (±SD) of the relative frequency per patient and category (n=178 cases). For DFT testing (A), a total of 13 669 parameters (86±8.2 per case) were evaluated; for the first permanent programme, a total of 20 051 parameters (126±17.8 per case). Rating categories: Category A, identical to physician; Category B, acceptable; Category C, unacceptable; Category D, potentially harmful.

 
In addition, the physicians were asked to rate the ICD programmes for each patient as a whole. These results are displayed in Fig. 5. None of the ICD programs were considered harmful, neither for DFT testing nor for the first permanent programming suggestion.


Figure 0285
View larger version (28K):
[in this window]
[in a new window]
  		Figure 5 Programme-rating in Phase I for DFT testing (A) and first permanent programme (B). Programme rating category vs. rating frequency per category (n=178).

 
The ratings for the handling of the implant-assistant are shown in Fig. 6. The majority of physicians considered the handling to be easy, as 72% of all ratings were within the ‘easy’ range 1–3 and only less than 1% were in the ‘difficult’ range 8–10. The mean handling rating for the 19 first cases of all 19 physicians was 4.1 and fell to 2.6 for their respective 19 last cases, suggesting a steep learning curve. The mean rating for the ease of understanding the terminology used in the implant-assistant was 2.5±1.5. Eighty-two per cent of the ratings were within the easy range (1–3) and only once was a rating in the difficult range (8–10) chosen.


Figure 0286
View larger version (19K):
[in this window]
[in a new window]
  		Figure 6 Rating for handling the implant-assistant. Rating of handling for each case (n=178) vs. absolute frequency of each rating. Possible ratings range from 1 (very easy) to 10 (very difficult).

 
Adaptation of implant-assistant
On the basis of the comments of the physicians on the implant-assistant's suggestions, the following changes were implemented:
  1. Additional preferences for ventricular ATP schemes and for enforcing maximum energy shocks were included.
  2. The rules for determining the ventricular shock energy, upper tracking rate, and VT-1 zone limit were adapted.
  3. The rules handling the long QT syndrome and the atrial therapy schemes were expanded with more specific suggestions.

Phase II
In Phase II, only 159 of the 178 patient files could be re-evaluated, as two complete centres were lost to follow-up owing to technical reasons. The Phase I results for the remaining 159 patients did not significantly differ from the complete set of 178 patients (Table 1).


View this table:
[in this window]
[in a new window]
  		Table 1 Comparison of parameter and programme ratings of Phase I and Phase II

 
When comparing the Phase I results of the 159 patients with the corresponding Phase II results, a significant improvement of the implant-assistant performance after the adaptation was observed. In none of the 159 cases was the rating ‘potentially harmful’ used by the evaluating physicians, neither for the parameter rating nor for the ICD programme rating (Table 1): this was valid for both DFT and first permanent programming.

Only one parameter in all 159 ventricular DFT test programme suggestions was rated ‘unacceptable’. The physician requested to change the VF zone limit to shorter cycle lengths (310–280 ms) to include fast but haemodynamically stable VT into the activated VT zone with ATP.

In the corresponding final programmes, 38 parameters were rated unacceptable: the most frequently criticized parameter was the shock energy of the second shock in the VT-1 zone (n=7), followed by VT-1 zone limit (n=5), VT ATP time out (n=5), bradycardia pacing rate (n=4), and the VT-2 zone limit (n=3). The other 14 criticisms were distributed among 11 parameters. However, in contrast to Phase I, none of the unacceptable ratings demanded a significant adaptation of the implant-assistant, as the majority of the experts considered the rules chosen by the implant-assistant in each case as safe.

The overall acceptance improved significantly for the DFT test programme from 93.7 to 100%. The improvement in the first permanent programme (from 85.5 to 90.6%) did not reach significance (P=0.072). Of the 15, 11 (73%) ratings of unacceptable came from only 2 of the 19 centres. In addition, it was observed that the criticized settings, on which the ratings were based, had been accepted in Phase I.


   Discussion
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
Limitations
 Conclusions
 Acknowledgements
 References
 
Medical knowledge-based systems have a long-standing history.3Go,4Go Though the field of pacemakers and ICDs is very technical in its nature, only few attempts have been made to establish such systems for pacemakers and none to our knowledge for ICDs. These developments focused on supporting diagnosis, trouble-shooting, or offered only limited support with respect to patient-adapted programming suggestions.5Go–9Go None of these systems were capable of providing a complete clinical interface for programming an electrotherapy device. The implant-assistant was specially developed to fulfil this goal. The idea of a clinical interface was well accepted by the experts in our study. This was reflected by the ratings of handling and ease of understanding the terminology used by the implant-assistant. In less than 1% of cases, the expert had difficulties in using or understanding the interface.

Any medical supporting tool must prove that its suggestions are safe, reproducible, and of good quality. For this purpose, the opinion of selected experts was used as ‘gold standard’ in the present study. The suggestions of the implant-assistant were subjectively rated by the experts, assuming that the experts represented a very high level of knowledge for optimal ICD programming.

To balance the differences between different expert opinions two different strategies were used. The first strategy influenced the design of the implant-assistant by introducing the concept of physician preferences. These preference choices were created whenever competitive opinions with similar significance were identified in the ICD community (e.g. short vs. long ATP schemes). The second strategy influenced the planning of the study. By recruiting many centres from different countries the gold standard was expanded beyond regional standards. Data on different regional standards are scarce, but one characteristic example for parameters with high regional dispersion is the use of the rate adaptation, whereas in Denmark about 95% of first implants of pacemakers are programmed in a rate adaptive mode, and in Germany the rate adaptive mode is only used in about 50%, which cannot be explained by the variations in indications.10Go,11Go This preference-oriented design resulted in an overall high acceptance rate, which is independent of the expert's educational and professional provenance and suggests the possibility of enhancing quality by standardizing ICD programming.

The implant-assistant has displayed its potential to simplify the programming of complex electrotherapy devices, while providing an overall standard, which is in great part independent of the expertise of a single ICD programming physician. It is a first major step in development towards physician supporting devices. This becomes more and more important, as the number of ICD indications is increasing. The major centres will most likely not be capable of handling the rapidly increasing number of ICD patients implying that other facilities will be obliged to treat a significant number of patients. The assistant concept would permit provision of a high standard to the patient across centres.

The daily routine can be further simplified by connecting the implant-assistant to the electronic patient file of the hospital information system. This will provide the patient data in an electronic form directly to the implant-assistant and allows easy documentation of ICD programming, measurements, and the underlying rationale in the hospital information system.


   Limitations
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Limitations
Conclusions
 Acknowledgements
 References
 
One limitation of this study is that the performance of the suggested programme was not tested by actual programming into patients' ICDs with an adequate period of follow-up. This would certainly be a more objective way of judging the quality of the suggestion. However, the effort would be tremendous, as not only the episodes with their detection details and therapy would have to be evaluated but also the patient's physical capacity, the quality-of-life, hospitalizations, atrial fibrillation occurrence, etc.

It is also noteworthy, that the results of Phase II are derived from physicians re-evaluating their own cases from Phase I. However, this design, which is usually associated with a bias towards improvement, was necessary, as the study should provide a verification in Phase II, that the changes in the knowledge base demanded by Phase I reflect the physicians' opinions. An independent validation of the implant-assistant would anyhow be outside the possibilities of the present study, because, as a minimum, a prospective, clinical study would be mandatory for this task.

Another limitation is the unblinded evaluation of the suggestions by the clinical experts, which introduces both sympathy and antipathy towards the idea of knowledge-based systems into the rating process. In other settings, e.g. the MYCIN expert system, this problem was overcome by separating the creation of alternative suggestions and the rating process, with the experts divided into two groups.3Go In this study, each member of the first group was given only the case data without the expert system's suggestions and was asked to create their own suggestions for each case. All suggestions were blinded for their origin and then rated by the members of the second group. In this setting, the expert systems scored as well as or even better than the experts included in the first group. Owing to logistic and resource limitations, this study design could not be followed for the first evaluation of the implant-assistant.

A knowledge-based system like the proposed one must not only be evaluated for security and quality of its suggestions but also for ergonomics in the daily routine. This is planned for a future prospective study after implementation of the implant-assistant in the standard ICD programming device.


   Conclusions
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Limitations
 Conclusions
Acknowledgements
 References
 
The implant-assistant, a knowledge-based supporting tool for programming ICDs, provides a promising support for the difficult task of ICD programming. It will develop its full potential when extended from implant-only support to a broad support including follow-up. It might replace the commonly used technical interface because it adapts ICD programming to a more patient-oriented clinical interface, which is based on the language of its user, the physician.


   Acknowledgements
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Limitations
 Conclusions
 Acknowledgements
References
 
The development of the implant-assistant and the evaluation study were supported by Biotronik GmbH, Berlin, Germany.


   References
Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Limitations
 Conclusions
 Acknowledgements
 References
 
[1] Schaldach M. Expert pacing system. Prog Biomed Res 2000; 5: 336–48.

[2] Revishvili AS. Dual-chamber automatic implantable cardioverter-defibrillator. Prog Biomed Res 1998; 3: 9–13.

[3] Jackson P. Evaluation of MYCIN. In Jackson P (Ed.). Introduction to Expert Systems 1998; 3rd ed. Reading, MA Addison-Wesley pp. p53–55 Chapter 3.4.1.

[4] Buchanan BG and Shortliffe EH. Rule-based Expert Systems: The MYCIN Experiments of the Stanford Heuristic Programming Project 1984; Reading, MA Addison-Wesley.

[5] Lucas P, Tholen A, van Oort G. An intelligent system for pacemaker reprogramming. Artif Intell Med 1999; 17: 249–69.[Medline]

[6] Lascault G, Nitzsche R, Ritter P, et al. Assistant programming software: a new tool for an improved programming of pacemakers. Pacing Clin Electrophysiol 1990; 13: 1732–6.[Medline]

[7] Sack S, Mouton E, Defaye P, et al. Improved detection and analysis of sensed and paced events in dual chamber pacemakers with extended memory function—a prospective multicenter trial in 626 patients. Herz 2001; 26: 30–9.[Medline]

[8] Funck RC, Adamec R, Lurje L, et al. Atrial overdriving is beneficial in patients with atrial arrhythmias: first results of the PROVE study. Pacing Clin Electrophysiol 2001; 23: 1891–3.

[9] Defaye P, Dornaux F, Mouton E. Prevalence of supraventricular arrhythmias from the automated analysis of data stored in the DDD pacemakers of 617 patients: the AIDA study. The AIDA multicenter study group. Automatic interpretation for diagnosis assistance. Pacing Clin Electrophysiol 1998; 21: 250–5.[CrossRef][Medline]

[10] Irnich W.Jahresbericht. 1999; des Deutschen Zentralregisters für Herzschrittmacher http://www.uniklinikum-giessen.de/zrhsm/index.html.

[11] .Møller M and Arnsbo P. Danish pacemaker register. 2001; Denmark Department of Cardiology, Odense University Hospital http://www.pacemaker.dk/stat2000.pdf.


Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


This article has been cited by other articles:


Home page
 Eur Heart J SupplHome page
C. Stellbrink and H.-J. Trappe
The follow-up of cardiac devices: what to expect for the future?
Eur. Heart J. Suppl., December 1, 2007; 9(suppl_I): I113 - I115.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
8/3/161    most recent
euj028v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (1)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Diem, B. H.
Right arrow Articles by Hanrath, P.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Diem, B. H.
Right arrow Articles by Hanrath, P.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?