32
edits
Accuracy of Computer Interpretation: Difference between revisions
Accuracy of Computer Interpretation (view source)
Revision as of 06:23, 1 August 2008
, 1 August 2008no edit summary
mNo edit summary |
No edit summary |
||
| Line 5: | Line 5: | ||
}} | }} | ||
Modern ECG equipment often includes some form of | Modern ECG equipment often includes some form of algorithm that performs a computer interpretation of the electrocardiogram. It is useful to have some insight into the reliability and pitfalls of computer interpretation as to be able to assess the value of these interpretations. | ||
First of all it is important to realize that the algoryhtm used often interpret the raw ECG-signal i.e. the signal before it is filtered. This means that the ECG that is printed by the machine is not the ECG that is | First of all it is important to realize that the algoryhtm used often interpret the raw ECG-signal i.e. the signal before it is filtered. This means that the ECG that is printed by the machine is not the ECG that is interpreted by the machine. Therefore one has to be very careful in making sure that the signal is as 'clean' as possible by: | ||
* cleaning the skin with e.g. alcohol before placing the electrodes | * cleaning the skin with e.g. alcohol before placing the electrodes | ||
* shave skin when necessary | * shave skin when necessary | ||
| Line 13: | Line 13: | ||
* carefully checking the electrodes position | * carefully checking the electrodes position | ||
Different manufacturers of ECG equipment have produced different | Different manufacturers of ECG equipment have produced different algorithms for ECG interpretation. Not surprisingly the accuracy of algorithms therefore differs. One study comparing nine computer algorithms with eight cardiologists found that overall the percent of correct classifications of computer programs was 69.7% as compared to 76.3% for the cardiologists.<cite>Willems</cite> However this study was performed in 1991 and algorithms have developed since. Especially some 'self learning' algorithms have been developed that improve themselves by learning from adjustments that cardiologists make to their diagnosis. | ||
Much attention has been aimed at the diagnosis of acute cardiac ischemia, as this can be challenging if subtle ST segment shifts are present and ST segment shifts can also be present in healthy persons. A systematic review on the subject found a sensitivity of 76% (95% CI, 54% to 89%) and a | Much attention has been aimed at the diagnosis of acute cardiac ischemia, as this can be challenging if subtle ST segment shifts are present and ST segment shifts can also be present in healthy persons. A systematic review on the subject found a sensitivity of 76% (95% CI, 54% to 89%) and a specificity of 88% (95% CI, 67% to 96%) for the correct interpretation of acute cardiac ischemia by a computer algorithm. For acute myocardial infarction the sensitivity was 68% (95% CI, 59% to 76%) and the specificity 97% (95% CI, 89% to 92%).<cite>Ioannidis</cite> One study investigating the Marquette 12SL system in the diagnosis of patients eligible for thrombolytic therapy found a sensitivity of 61.5% and specificity of 100%. The authors conclude that reliance on computer interpretation would result in inappropriate underuse of thromobolytic therapy.<cite>Massel</cite> | ||
A recent scientific statement by a consortium of AHA / ACC and the HRS also advice to be prudent when using computer interpretations. The official recommendation reads: "Computer-based interpretation of the ECG is an adjunct to the electrocardiographer, and all computer-based reports require physician overreading. Accurate individual templates should be formed in each lead before final feature extraction and measurement used for diagnostic interpretation." <cite>Kligfield</cite> | A recent scientific statement by a consortium of AHA / ACC and the HRS also advice to be prudent when using computer interpretations. The official recommendation reads: "Computer-based interpretation of the ECG is an adjunct to the electrocardiographer, and all computer-based reports require physician overreading. Accurate individual templates should be formed in each lead before final feature extraction and measurement used for diagnostic interpretation." <cite>Kligfield</cite> | ||